Datasets:

haritzpuerto

/

the_pile_00_PubMed_Central

like 1

**PURPOSE:** The desire to avoid flexor tendon injury after volar plating of distal radius fractures has lead to a myriad of new plate designs. However, the rate of innovation has outpaced the data addressing whether these new models actually decrease the rate of tendon injury. The purpose of this study is to determine whether the designs of three representative distal radius plates impact the amount of force exerted on the FDP and FPL tendons when the plates are placed proximal and distal to the watershed line. **METHODS:** Three commercially available plates, including a 2.0 mm thick FPL sparing plate, a 2.0mm adaptive plate, and a 1.6mm babyfoot plate were applied to ten fresh, matched-pair upper extremity specimens fixed to a custom jig to mimic pinch grip. Plates were placed at various positions relative to the watershed line. External cyclical loading was applied to the FPL and FDP tendons, and the force generated between each tendon and plate was measured with a thin, flexible pressure sensor. Linear mixed effect models were used to evaluate differences in maximum and mean force by plate position and plate design. **RESULTS:** The maximum and mean force on the FPL and FDP tendons differed significantly between positions relative to the watershed line, as measured by Soong grade. For the FPL tendon, the maximum force experienced with a plate in Soong 2 was 4.50 \[95% CI: 2.8, 7.3\] times higher than when the plate was in a Soong 0 placement, and 4.63 \[95% CI: 2.82, 7.61) times higher for the FDP tendon. There was a consistent trend towards a decreased maximum and mean force with the babyfoot plate, likely due to the thinner plate profile, although this difference did not achieve statistical significance. **CONCLUSION:** The main determinant of plate prominence and therefore flexor tendon injury potential is placement in relation to the watershed line. While not statistically significant, decreased observed force values with a thinner plate indicate that plate thickness could also represent a clinically meaningful difference in plate prominence. Due to the variability in distal radii, use of the Y-shaped, FPL sparing plate does not lead to a consistent decrease in force on the FPL tendon in comparison to other plate designs. Our results do not clearly show superiority of one plate type, but suggest that decreasing the overall plate profile may have greater utility in preventing flexor tendon injury than anatomic plate design.

Background {#Sec1} ========== Although accreditation has been widely used as a means to improve quality in secondary and tertiary care, its presence in primary care is limited \[[@CR1]\]. This is particularly evident in the Eastern Mediterranean Region (EMR). Accreditation systems originally developed to set standards and enhance the quality of care in acute care settings. With the expansion of primary care and the heavy emphasis placed on this sector of healthcare, accreditation organizations started to put more focus on means to improve the quality of service in primary care organizations. In Canada for instance, primary care-specific accreditation standards were developed \[[@CR2]\]. Accreditation of primary care settings was shown to strengthen quality control and improvement \[[@CR3]\]. Results of a study on the effectiveness of quality-improvement in primary care showed management progress in the practices that applied organizational standards \[[@CR4]\]. A study done by Braun et al. \[[@CR5]\] revealed that accredited primary care centers were more committed to risk management, environmental safety and quality improvement \[[@CR5]\]. Similarly, literature review showed that accreditation of primary care organizations showed improvements in the areas of teamwork, access to care, patient safety, care processes and quality of care \[[@CR2]\]. EMR studies revealed similar findings, a study that was conducted on primary health care centers in Lebanon emphasized on the benefits of documentation, reinforcement of quality standards and improvements in staff, and patient satisfaction \[[@CR6]\]. Alternatively, findings of a study conducted in 2015 showed that there were gaps in the evidence on quality in primary care in the EMR region \[[@CR7]\]. Another primary care - focused research revealed no major difference in compliance with accreditation standards between health services that went for accreditation and their counterparts. Indicators showed no such difference in efficiency as well in the areas of immunization, maternal care services, and family planning \[[@CR8]\]. The practice of accreditation programs in primary care should be further established and assessed to be able to evaluate if participation is worthwhile or not \[[@CR9]\]. This study aimed at filling this gap in the literature by assessing the implications of adopting accreditation in primary health care organizations. The purpose of this research was to quantitatively examine the changes resulting from the implementation of accreditation at Primary Health Care Corporation (PHCC) in the state of Qatar by studying the impact of applying Accreditation Canada International's (ACI) standards on quality improvement and organizational learning in 21 primary health centers. Methods {#Sec2} ======= Setting {#Sec3} ------- At the time of the research, PHCC operated through 21 health centers distributed all over the country across three main geographical regions: Central, Western, and Northern. Twelve of the existing health centers were located in Doha city (the capital and main city of the country), while the rest of the centers were located in less populated areas in other parts of the country \[[@CR10], [@CR11]\]. Study design and population {#Sec4} --------------------------- This cross-sectional study was conducted in 2015 by utilizing a descriptive correlational methodology to gather data from PHCC employees who were present at the organization during the accreditation process. Data was collected from both front-line and management staff 12 months after the accreditation survey. All PHCC employees who declared to have good English language competency were the target population; that is 750 English competent employees out of close to 4000 employees. These employees worked at the 21 health centers as well as the headquarters as the organization implemented the accreditation program. Employees were managers, administrative staff, doctors, nurses, pharmacists, radiologists, technicians, support staff, and clerks. This diverse composition enabled the researcher to assess the accreditation impact from a variety of angles, from both frontline and management levels as well as from the perspectives of both healthcare providers and administrators. Data collection {#Sec5} --------------- Data was collected anonymously, without direct and indirect identifiers, using structured questionnaires that were previously used and tested for reliability. The questionnaires were adopted from Shortell's (1995) \[[@CR12]\] and Quinn's and Kimberly's (1984) \[[@CR13]\] and amended to include the *Accreditation* and *Information about Yourself* sections \[[@CR14]\]. The same tool was also used to measure the impact of accreditation in selected Canadian hospitals Weber (2005) \[[@CR15]\]. This study utilized the same approach and questionnaire with minor changes to reflect primary care setting. The survey questions were in the English language; thus, the English competent employees in PHCC were considered as eligible subjects. In order to get a representative sample, simple random sampling was used to give every subject an equal chance to be included and to reduce the risk of selection bias. The survey was sent out to the 500 randomly selected PHCC employees (out of the 750 English competent employees) and participants were given two weeks to respond. The target population staff and managers were informed through an email of the research study and were provided a link to complete the survey electronically using the *SurveyMonkey* online survey tool. The email message included all elements of an informed consent. The questionnaires were divided into two categories (a) *Management Questionnaire*, for assessing staff perception of quality improvement, and (b) *Culture Questionnaire*, for studying the organizational learning processes. The management questionnaire contained three main scales; *Quality of Care* (to gather information on employees' involvement in quality improvement across seven areas or subscales: leadership, information and analysis, strategic quality planning, human resources utilization, quality management, quality results, and customer satisfaction), *Accreditation Impact* (to examine the impact of the accreditation process on dynamics of change) and *Demographics* (gender, age, occupational category, and years of experience in the corporation). The culture questionnaire gave insight on the organizational culture in four dimensions; the organization's character (employees perception of the organization such as being dynamic, entrepreneurial, productive, etc....), the organization's managers (the way managers treated employees), the organization's cohesion (the loyalty and commitment of the organization), and the organization's emphasis (areas related to human resources, performance, and achievements). The dependent variables were quality improvement and organizational learning. The independent variable was accreditation process. Statistical analysis {#Sec6} -------------------- A correlation design using Spearmen's correlation guided the analysis of this study. A *P*-value \< 0.05 was considered significant. Data was entered and analyzed using the statistical software IBM-SPSS (Statistical Package for Social Sciences), version 21.0 (Chicago, IL, USA). Results {#Sec7} ======= A total of 285 questionnaires were submitted electronically out of 500, for a response rate of 57%, which is relatively adequate for this type of survey \[[@CR16]\]. A total of 253 questionnaires were included in the analysis (50.6%), and 38.8% of the submitted questionnaires (that is, 194 questionnaires) were complete. The description of the sample is summarized in Table [1](#Tab1){ref-type="table"}. The response rate from the health centers (48%) and the headquarters (50%) was very close indicating a roughly equal representation from both frontline and management perspectives. More females than males responded to the questionnaire (57%) which is close to the proportion of female employees in the organization, being 58% in 2015 at the time of the survey as reported by human resources department at PHCC. The majority (79%) of the respondents was under the age of 45 years, and close to two thirds of the respondents had more than ten years' experience with the organization (63%). Out of the respondents, 15% identified themselves as occupying a managerial position. This is close to the proportion of head quarter's employees in the organization, being 17% in 2015 at the time of the survey as reported by the human resources department at PHCC.Table 1Demographic DetailsN (253)PercentageGender Male10943 Female14457Age \< =4519979 \> 455421Years in the Organization \< =1018071 \> 107329Managerial Position Yes3715 No21685Clinical Team Yes13252 No12148Member of QMD Yes4116 No21284Involved in last Accreditation Yes18674 No6726Work Location Headquarters12248 Health Centers13152QMD*:* Quality Management Directorate The main clinical occupation category representing respondents was nursing (23.4%). Physicians' rate was 8.6%. Radiology (6.6%) and dental (7%) were almost the same. For the non-clinical, the majority were administrative (21.5%), followed by coordinator level employees (12.1%). Managers were 6.6% and project managers and heads were both 8% (data not shown). Quality of care and accreditation impact {#Sec8} ---------------------------------------- Data analysis of the quality of care part generated the results presented in Table [2](#Tab2){ref-type="table"}. Interpretation of the total mean scores suggested that the areas of strengths in the quality of care variable were quality results (4.03) and leadership (4.01). The areas of weakness relative to other scales were customer satisfaction (3.79) and human resource utilization (3.67). Interpretation of findings under the seven components; leadership, information and analysis, strategic quality planning, human resources utilization, quality management, quality results, and customer satisfaction showed that employees provided considerably high ratings since all scores had a high value ranging between 3.67 and 4.03 (Table [2](#Tab2){ref-type="table"}).Table 2Employees Perception of Quality Improvement and AccreditationMeanStandard DeviationRangeQuality Scales Leadership4.010.693.64 Information and Analysis3.940.663.00 Strategic Quality Planning3.830.723.00 Human Resources Utilization3.670.813.75 Quality Management3.930.613.00 Quality Results4.030.633.25 Customer Satisfaction3.790.744.00Accreditation Scales Overall Impact4.170.573.00 Preparations4.210.693.50 Recommendations4.110.653.00 Internal Changes4.220.663.00 Externally Oriented Changes4.090.704.00 Valuable Tool4.320.593.00 For the accreditation impact section, as presented in Table [2](#Tab2){ref-type="table"}, results showed that, overall employees agreed on the positive impact of accreditation on the organization. Interpretations of the findings showed that: (a) employees were aware of and involved in the changes that were happening in preparation for accreditation (a score of 4.21). (b) They were aware of the recommendations (a score of 4.11). (c) they saw the benefit of accreditation in improving the quality of care, in the values shared in the organization, as well as in the use of internal resources (a score of 4.22). (d) They were confident in accreditation's positive impact on addressing issues brought in by external factors like population needs and working with external stakeholders (a score of 4.09). Moreover, (e) they believed that the organization was more responsive to change due to accreditation (a score of 4.32). Correlational analysis between accreditation and quality of care {#Sec9} ---------------------------------------------------------------- A correlation analysis was carried out to assess the relationship between accreditation and quality of care, as perceived by PHCC employees. As shown in Table [3](#Tab3){ref-type="table"}, the results showed a significantly positive correlation between staff perception of accreditation and the perception of quality of care for all seven scales.Table 3Correlation between Accreditation and Quality of Care scalesLeadershipInformation and AnalysisStrategic Quality PlanningHuman Resources UtilizationQuality ManagementQuality ResultsCustomer SatisfactionAccreditation ImpactCorrelation Coefficient^a^0.5720.5670.5280.5090.5870.6200.537*P*-value\< 0.001\< 0.001\< 0.001\< 0.001\< 0.001\< 0.001\< 0.001Overall Accreditation ImpactPreparationsRecommendationsInternal ChangesExternally Oriented ChangesValuable ToolQuality of CareCorrelation Coefficient^a^0.6150.3510.4820.5550.6210.509*P*-value\< 0.001\< 0.001\< 0.001\< 0.001\< 0.001\< 0.001^a^: Spearmen's correlation Coefficient The correlation analysis also showed that employees who had positive perception about accreditation for all accreditation sub- scales (preparations, recommendations, internal changes, externally-oriented changes and valuable tool) were also positive about the quality of care (p- value \< 0.001 for all sections as presented in Table [3](#Tab3){ref-type="table"}). Culture {#Sec10} ------- Results of the culture questionnaire revealed the two dominant cultures at PHCC to be group, with a score of 28.61, and hierarchical, with a score of 26.59 as depicted in Table [4](#Tab4){ref-type="table"}. Employees perceived PHCC to have affiliations, team work, and participation (group culture) but also had certain embedded norms and values that were associated with bureaucracy (hierarchical culture). The other two culture types, developmental and rational were 21.82 and 22.98, respectively.Table 4Employees Perception of CultureCulture TypeMeanRangeStandard DeviationGroup28.6110014.01Developmental21.82508.46Hierarchical26.5978.7511.93Rational22.987510.06 Correlational analysis between accreditation and culture {#Sec11} -------------------------------------------------------- A correlation analysis was carried out to assess the relationship between accreditation and culture (see Table [5](#Tab5){ref-type="table"}). The results showed a positive correlation between staff perception of accreditation and their perception of culture type whenever the culture was identified as group (*r* = 0.182, *p*-value = 0.011). For the hierarchical culture, there was non-significant negative correlation between the perception of accreditation and the perception of culture type (*r* = − 0.132, *p*-value = 0.067).Table 5Correlation between Accreditation Impact and CultureGroupDevelopmentalHierarchicalRationalAccreditation ImpactCorrelation Coefficient^a^0.1820.093−0.132−0.070*P*-value0.010.200.070.33^a^ Spearmen's correlation Coefficient Discussion {#Sec12} ========== Results of this study indicated that accreditation triggered major changes in the organization at both quality improvement and organizational learning levels, thus emphasizing the importance of accreditation as a main endeavor towards improving quality in primary care. Findings also added to the body of literature that the positive impact of accreditation in relation to quality improvement and organizational learning was definite in a primary care setting, since most of the evidence found in current literature related to acute care settings. Quality results indicated that the organization recently achieved significant improvements in quality and performance as well as in administrative areas like finance and human resources, as reported by the employees. Leadership high scores showed that the leaders of the organization had strong focus and emphasis on quality values and that quality values were integrated in the management system of the organization. As observed through data interpretation, most of the scales under quality of care had high scores, which meant that employees perceived the organization with significant progress in the areas of quality improvement and performance. Employee involvement data indicated that whenever employees were involved in accreditation work, they had a better perception of quality in areas relating to leadership, finance, continuous quality improvement efforts, and collection of data and measurements and they were more confident about the positive changes accreditation brought during preparation, implementation, and recommendation phases. Moreover, the correlation analysis between accreditation and quality of care sections was very strong confirming that employees perceived accreditation to be a valuable tool that triggered recent quality improvements at the organization. These results led to the conclusion that accreditation did influence the development of quality improvement practices and thus had a positive impact on quality. Results of this study complemented what was stated in the literature about the positive impact of accreditation on quality improvement. \[[@CR14], [@CR17]--[@CR21]\] A systematic review on healthcare accreditation reported a positive correlation between accreditation and quality programs especially during the 3 years preceding accreditation \[[@CR22]\]. Improved compliance of healthcare organizations with the requirements of accreditation was shown to be as a tangible indication of the organizations' effectiveness \[[@CR21]\]. It was also reported that hospitals who were working on accreditation showed a higher compliance rate with quality standards in comparison to hospitals that were not \[[@CR20]\]. On the other side, some discrepancies were highlighted in few studies, for example a study that was conducted in Egypt indicated that accreditation had no effect on quality, thus contradicting the findings of this research \[[@CR8]\]. Findings of this study parallel those reported on the effect of accreditation on causing changes that influence relational and strategic changes in organizations \[[@CR19]\] as well as on accreditation being a motivation tool that supports the quality of health services \[[@CR17]\]. Similar to this study, research showed that accreditation was effective whenever there was strong involvement and commitment from staff \[[@CR23]\]. In addition, healthcare professionals who were supporters of accreditation considered the process as an effective quality improvement tool that reinforced transparency and team work \[[@CR24]\]. El Jardali et al. (2014) stated that accreditation did show improvements in the quality of health services in a study conducted in Lebanon in the primary care health centers in the country. It was found out that accreditation did have a positive impact on quality as well as on customer satisfaction. These results were in support of this research inferences on quality; however, the results contradicted with the customer satisfaction's results since it was found in this study that accreditation did not significantly affect customer satisfaction \[[@CR6]\]. Moreover, it was argued that improvement initiatives were only observed when organizations were preparing for the survey. The initiatives did not have a long lasting effect over time, which contradicted what is generated in this study especially that this research was conducted after one year of attainment of accreditation \[[@CR25]\]. In this study, the predominant group culture revealed the organization as a personal place where employees had high commitment and loyalty, and managers were very caring and focusing on employees' growth and development. The hierarchical culture was the second predominant choice and this set the organization as a very formalized and structured place governed by bureaucratic procedures and rigid policies, and characterized by permanence and efficient operational procedures. This showed that employees who were positive about accreditation perceived the culture to be of a group type, that is, they were part of a team, and they had the potential to affect quality, patient care, policy and management. They felt that they belonged to the organization. Analysis of accreditation and culture correlation, showed a positive association between staff perception of accreditation and their perception of culture type whenever the culture was identified as group. For the hierarchical culture, there was negative correlation between the perception of accreditation and the perception of culture types. This showed that employees who were positive about accreditation perceived the culture to be of a group type. Further analysis on the scores showed that employees who were involved in accreditation had a higher score for group culture in comparison to those employees who were not involved in accreditation (data not shown). These findings lead the researchers to conclude that accreditation had a positive impact on culture and, thus, on organizational learning. Then again, correlation between accreditation and the seven scales under quality of care was positive indicating a positive impact on areas relating to leadership, information and analysis, strategic quality planning, human resources utilization, quality management, quality results and customer satisfaction. This discussion lead to the conclusion that accreditation did influence organizational learning in those areas as well. This positive impact of accreditation on organizational learning indicated that the organization had enhanced its capabilities to produce certain desirable outcomes, and that it had the potential to institutionalize those capabilities through its employees. As a result learning and becoming accustomed to changing conditions had become monotonous, and as a result the organization developed a culture of learning \[[@CR26]\]. Findings on organizational learning complemented what is stated in the literature about the ability of healthcare organizations that embrace certain characteristics like teamwork, communication, group affiliation, to demonstrate a broader adoption of quality improvement strategies \[[@CR27]\]. Accreditation was also shown to create an organizational culture committed to continuous improvement of skills, teamwork, processes, product and service quality, which complemented what this research generated in relation to accreditation's positive impact on organizational learning \[[@CR28]\]. The study findings also lent evidence to what was stated in the literature about accreditation being a tool that aimed at both the acquisition of knowledge and the enhancement of the quality of services \[[@CR18], [@CR29], [@CR30]\]. Similar to this study, it was reported that accreditation was a tool of change that fits into organizational change and the factors affecting change \[[@CR31]\]. In the same context, accreditation was found to be a management tool that provoked change in the same sense that a quality program or a new strategic plan would \[[@CR32]\]. The study conflicted with the findings of Sack et al. (2011) who argued that implementation of accreditation standards did not provide evidence of improvement in quality, which likewise was an absolute opposition of the findings of this study \[[@CR33]\]. This research was distinctive because it addressed an under-researched area of accreditation in the EMR with a growing interest in studying accreditation in primary care settings. What made this study even more unique is that it addressed the primary care sector in the State of Qatar and there was no evidence on any kind of research that was conducted for the same purpose in the country. Some limitations were identified in this study. Quality of care was assessed based on employees' perceptions and not on measures that related to management processes or clinical outcomes. What could add more strength to the research is supplementing this set of data with measures of performance and outcomes of quality projects. There was a risk of selection bias as the population was limited to English competent employees. Another limitation was the possibility that misunderstandings might arise from the context, culture, and different interpretations of words and sentences especially that the study was conducted in English and for many of the respondents, English was a second language. Misinterpretations could be due to different understandings based on employees' language competency and/or on previous experiences in other settings and cultures since most of employees were expatriates coming to Qatar from different countries, including Canada, the UK, South Africa, Lebanon, India, Philippines, Jordan, Syria, Egypt and other counties. An additional limitation is social desirability bias, respondents may have provided answers they considered desirable to the researchers. However, it can be safely assumed that the results should not be overstated since staff were briefed about the confidentiality and anonymity of the study. Generalizability of the study could also be considered as a limitation since the research was limited to 21 health centers in one country. Results of the study may not be applicable to other types of healthcare organizations. A cross-sectional design was adopted in this research, in which data was collected at a single point of time. It is recommended that the research is replicated using a longitudinal study design, so that changes over time are observed. In addition, it would be very valuable to run the study in an organization prior and post to attainment of accreditation. It is also recommended that the study be conducted in both accredited and non-accredited primary care organizations and comparisons are generated to assess the differences in staff perceptions towards accreditation. Conclusions {#Sec13} =========== High quality health care systems lead to better health outcomes and improved patient safety. Accreditation has offered a potential solution to improve the quality and learning of healthcare systems in countries all over the world including the EMR. This study emphasizes on the positive impact of accreditation on the quality of care, as reported by PHCC employees, including areas relating to leadership, information and analysis, strategic quality planning, human resources utilization, quality management, quality results. In addition, the results supported the notion that accreditation is a drive for organizational learning. In particular, this research gives an indication on the importance of employee involvement in accreditation and in promoting a culture that supports quality improvement and that allows employees to feel that they have a sense of belonging to the organization. Primary care in the EMR in general and in Qatar in particular has a potential to expand services and improve outcomes. To achieve these objectives, there should be a stronger leadership focus on quality and accreditation and primary care organizations should continue to receive the necessary support and encouragement from healthcare leaders in the country. There is also a need to build the capacity of primary care organizations to allow them to embed a culture of quality improvement and thus have a continuous state of accreditation readiness for accreditation. ACI : Accreditation Canada International EMR : Eastern Mediterranean Region PHCC : Primary Health Care Corporation SPSS : Statistical Package for Social Sciences The authors thank all Primary Health care corporation employees participated in the current study. The abstract section of this manuscript is based on the doctoral dissertation that can be accessed through this <https://search.proquest.com/docview/1759630525> Availability of data and materials {#FPar1} ================================== The datasets used and analyzed during the current study are available from the corresponding author on reasonable request. AG, HS, and MZ were involved in designing the study and interpreting results. AG and MZ have performed the statistical analysis. AG has drafted the manuscript, and HS and MZ have commented and revised the final manuscript. All authors read and approved the final version of the manuscript. Ethics approval and consent to participate {#FPar2} ========================================== Ethical approval for the study was granted by the Walden University Institutional Review Board (IRB) and by the Research Committee at Primary Health Care Corporation. Data was collected anonymously. The survey was sent out to randomly selected employees through an email that included necessary elements of consent. The completion of the survey was considered implied consent to participate. Consent for publication {#FPar3} ======================= Not applicable. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar5} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction {#s1} ============ In a eukaryotic cell, the presence of cis-regulatory elements ensures expression of genes at an appropriate level and in appropriate cells. Cis-regulatory elements not only include transcription factor binding motifs within the promoters but also DNA sequences located several kilobases away from the promoter that can positively or negatively influence the transcription rate of a gene [@pone.0093561-Wittkopp1]. Except for certain genes that are involved in housekeeping functions most genes are expressed in a tissue-specific manner. This tissue-specific regulation of genes is in turn achieved by interplay of the various cis-regulatory elements and their associated trans-acting factors. The importance of these regions in gene function can also be gauged by the fact that many of the disease causing mutations have been mapped to these cis-regulatory elements [@pone.0093561-Kleinjan1]. A well-studied example of cis-regulatory elements is the Polycomb/Trithorax Response Element (PRE). It was first identified in Drosophila but is present in most eukaryotic organisms and controls gene expression by recruiting Polycomb and Trithorax groups of regulatory proteins [@pone.0093561-Francis1]--[@pone.0093561-Woo1]. *DNMT3L* is a member of the Dnmt3 family of de novo DNA methyltransferases that includes *DNMT3A* and *DNMT3B*. DNMT3L lacks the catalytic domain and cannot methylate DNA on its own [@pone.0093561-Chedin1]. But it can influence DNA methylation by a non-specific mechanism through its interaction with DNMT3A and DNMT3B [@pone.0093561-Gowher1]. It also interacts with histone H3 at lysine 4. This interaction was found to be specific to the unmethylated form of lysine 4, indicating that DNMT3L could be the epigenetic effector that can read the histone code and postulated as a link between DNA methylation and histone modifications [@pone.0093561-Ooi1], [@pone.0093561-Hashimoto1]. Functionally, it has been shown to be involved specifically in setting up of DNA methylation during gametogenesis [@pone.0093561-Bourchis1], [@pone.0093561-Arima1]. Coincident with its function, *Dnmt3l* is expressed in mice during early embryogenesis and in the germ cells. It is also expressed at a very high level in ES cells [@pone.0093561-Hata1]--[@pone.0093561-Aapola1]. In all other tissues, it is kept transcriptionally silent and this inactivity has been attributed to the epigenetic status of DNA sequences within and around the *DNMT3L* promoter [@pone.0093561-Aapola1], [@pone.0093561-Gokul1]. We had previously shown loss of DNA methylation for a CpG island spanning the human *DNMT3L* promoter/exon1 region (promoter for *DNMT3L* variant 2 and first exon in case of *DNMT3L* variant 1) for cervical and ocular cancer samples [@pone.0093561-Gokul2], [@pone.0093561-Manderwad1]. Since there is a genome-wide nuclear reprogramming associated with carcinogenesis, the loss of DNA methylation observed in the CpG island around the *DNMT3L* promoter could either be coincidental to the process of carcinogenesis or has a role to play during carcinogenesis. In addition, the loss of DNA methylation at this CpG island could be indicative of a role for this region in the regulation of the *DNMT3L* expression. To examine this, we sought to analyse the functional role of this region in regulating transcription. In the present study, we have shown, by performing reporter gene assays in mammalian cell lines and Drosophila, that the DNA sequence present within the *DNMT3L* promoter/Exon1 acts to repress transcription. This region acts as a Polycomb/Trithorax Response element (PRE) and mediates repression by adopting inactive-chromatin-specific histone modifications through its interaction with Polycomb group of proteins. Materials and Methods {#s2} ===================== Transient transfection assay in mammalian cell lines {#s2a} ---------------------------------------------------- The CMV promoter fragment was PCR amplified from pEGFPC3 vector (Clontech) and cloned between the *Eco*RI and *Bam*H I sites of the promoter-less pAcGFP1-1 vector (Clontech) to derive the pCMV-AcGFP1-1 plasmid. 3L-S and 3L-L regions from the *DNMT3L* locus ([figure 1](#pone-0093561-g001){ref-type="fig"}) were PCR amplified from the HeLa genomic DNA using the primers 3LF (5′-CCTGAGGGCCCCATCCTCTG-3′) and 3LSR (5′-AAGGATCCAGGCCCACCTGGGAC-3′) for 3L-S and 3LF and 3LLR (5′-CAGGGACCCCTGGGGATGGTCTTGGCC-3′) for 3L-L. The fragments were cloned in both orientations upstream of the CMV promoter in pCMV-AcGFP1-1 vector between *Xho*I and *Eco*RI. As a negative control we cloned a 1 kb region from human chromosome 1, which was previously shown to have no transcriptional potential upstream of the CMV promoter [@pone.0093561-rom1]. A 1.5 kb region from the H19 ICR, which has been shown to be a transcriptional repressor [@pone.0093561-Ginjala1] was also cloned upstream of the CMV promoter ([Figure 1](#pone-0093561-g001){ref-type="fig"}). As a control for transfection efficiency we used the pG5luc vector (Promega) containing the luciferase reporter gene. Transfections were done in HEK293 cells using Lipofectamine 2000 (Invitrogen) in duplicates following the manufacturer\'s protocol. {#pone-0093561-g001} Expression analysis of the Reporter gene {#s2b} ---------------------------------------- 48 hrs post transfection, cells were harvested and analysed for AcGFP and Kan/Neo^R^ transcript levels by Real-time PCR using the Mesa Green qPCR Mastermix plus (Eurogentec cat no 05-SY2X-03+WOU) in the ABI Prism SDS 7500 system with the following primer sets: For *AcGFP* AcGFPF 5′TCCTGGGCAATAAGATGGAG 3′ AcGFPR 5′ TGGGGGTATTCTGCTGGTAG 3′ For *Neo^R^* Neo^R^F 5′ GGAGAGGGCTATTCGGCTATG 3′ Neo^R^R 5′ GACAGGTCGGTCTTGATAAA 3′ For *GAPDH* GAPDHF 5′ TGCTGGCGCTGAGTACGTCG 3′ GAPDHR 5′ GGGTGGCAGTGATGGCATGG 3′ *GAPDH* was used as an internal control in Real-Time PCR. To examine the levels of AcGFP protein, the respective constructs were co-transfected and Western blotting was performed as per established protocol [@pone.0093561-Hames1]. The transfection efficiency was controlled by co-transfecting the Luciferase gene containing pG5luc vector. β-TUBULIN was used as loading control. GFP antibody was from Sigma and the β-TUBULIN antibody was from Abcam. The band intensities were determined by performing a densitometric scan using the Alpha Ease FC software (Alpha Innotech Corporation). Luciferase assay was done using the Luciferase Assay System Kit (Promega) following the manufacturers instruction and the light intensity was measured in the Varioskan Flash Multimode Reader (Thermo Scientific). The GFP protein level was calculated by normalizing the GFP/β-TUBULIN ratio for each construct with their respective Luciferase activity (Relative Light Unit at 0.1 sec per reading integration time). The final value was calculated as a percentage of GFP expression for each construct relative to the expression of GFP in the control CMV only construct (which was taken as 100%). Each experiment was performed thrice and was done in duplicates. Reporter gene assay in Drosophila {#s2c} --------------------------------- Both 3L-S and 3L-L, cloned between the two *LoxP* sites of the pLML vector were subcloned in the Drosophila vector, pCaSPeR, upstream of the reporter *miniwhite* gene that was under the control of *hsp70* promoter. The sequence as well as the orientation was confirmed by sequencing. The constructs for microinjection in Drosophila were purified by midi-prep kit (Qiagen). Microinjection was done in the Drosophila white eyed mutants *w^1118^* as per established protocols [@pone.0093561-Voie1]. All the lines were balanced using the double balancer flies to establish independent balanced stocks. The 3L-L fragment was excised from various transgenic lines by crossing transgenic lines to the *Cre* recombinase-expressing Drosophila flies as described previously [@pone.0093561-Siegal1]. The transgenic lines as well as the respective flipped out lines were checked by PCR to ensure that 3L-L region had been successfully flipped out. The primers used for checking out the flipped out lines were 3L-LF-5′GCAGTATGCCGTTTACTGTGTG3′ and 3L-LR-5′CCCCGATCCCCCTAGAATCCCAAA 3′ To examine the localisation of the construct in the genome, inverse PCR was performed using the following primers: Plac1- 5′ CACCCAAGGCTCTGCTCCCACAAT 3′ Pwht1- 5′ GTAACGCTAATCACTCCGAACAGGTCACA 3′ Drosophila Eye pigmentation assay {#s2d} --------------------------------- 20 heads from flies of similar age and sex were homogenised in 1∶1 mixture of Chloroform and 0.1% Ammonium hydroxide mix. The homogenate was centrifuged at low speed to pellet down the debris and absorbance of the supernatant was taken at 485 nm using Chloroform/Ammonium Hydroxide mixture as the blank [@pone.0093561-Sultana1]. The experiment was repeated thrice. Crosses with Drosophila mutants {#s2e} ------------------------------- To determine the epigenetic effectors that interact with the 3L-L region in Drosophila, we set up a series of crosses with the various Polycomb and Trithorax mutant flies. The transgenic lines were crossed to various mutant stocks that were in white eye background. The Trithorax mutants used in these crosses were: *Ash2^1^*, *Mor^1^*, *Brm^2^*, *Trl^R85^* and *Trx^1^* while *Pc^1^*, *esc^2^*, *Asx^XF53^*, *Psc^1^*, *Pcl^T1^*, *Scm^R5-13B^*, *Phd* and *Pho* were the Polycomb mutants used. Mutants for the epigenetic modulators that enhance or suppress position effect variegation including *Su(var)2-5^01^* and *Su(var)3-9^06^* were also used [@pone.0093561-Schotta1], [@pone.0093561-Reuter1]. Chromatin immunoprecipitation {#s2f} ----------------------------- Isolation of chromatin from Drosophila larvae was done following established protocol (see [Supplementary Information S1](#pone.0093561.s006){ref-type="supplementary-material"}). To isolate chromatin for ChIP analysis from transfected cells, 1×10^6^ cells transfected cells were incubated with 1% formaldehyde at 37°C for 10 minutes. The reaction was quenched by incubation with 125 mM Glycine. The medium was removed and the cells were washed twice with PBS containing protease inhibitors. Cells were collected centrifuged at 300 g and resuspended in 400 µl of lysis buffer (1% SDS, 10 mM EDTA and 50 mM Tris pH 8). The lysate was sonicated at maximum power for 12.5 minutes with 30 sec off/on cycle to obtain DNA fragments between 200--1000 bp in a bioruptor™ (Diagenode). For Chromatin Immunoprecipitation with antibodies against various histone modifications and other epigenetic effectors we performed the protocol described by antibody manufacturer (Millipore). Antibodies to H3K4me3, H3K9ac, H3K9me3 and H3K27me3 (Millipore) were used in our ChIP analyses. Rabbit IgG (Sigma) was used as a control. 1% of chromatin was taken as input and processed separately. The enrichment of the various proteins or Histone modification at the region of interest was evaluated by performing Real time PCR using Mesa Green qPCR Mastermix plus (Eurogentec) in the ABI Prism SDS 7500 system. To evaluate the efficiency of our ChIP assays, we performed Chromatin Immuno-precipitation for known regions in the human genome with the antibodies that we used in our assay ([Figure S1](#pone.0093561.s001){ref-type="supplementary-material"}, based on literature provided by Abcam, the manufacturer of the antibodies used). The primers used in our assay were *In Drosophila* For 3L-L region ChIP3L-LF- TGGATCCTTCCAGCTCATTC and ChIP3L-LR-GCATGGTCAGGGAGTAGGAA For *hsp70* promoter Hsp70F 5′ TGCAGTTGATTTACTTGGTTGC 3′ Hsp70R 5′ ATAGAGCGCTTCGTCTACGC 3′ *In mammalian transfection study* For 3L-L region pAcGFP3LLF 5′ TAGCGCTACCGGACTCAGAT 3′ pAcGFP3LLR 5′AAAGAATCTGGGCGTGTCC 3′ For CMV promoter CMVF 5′ TCCGCGTTACATAACTTACGG 3′ CMVR 5′ GTCAATAGGGGGCGTTCTTG 3′ For GAPDH control oligos GAPDH F 5′ TACTAGCGGTTTTACGGGCG 3′ GAPDH R 5′ TCGAACAGGAGGAGCAGAGAGCGA 3′ For SAT2 control oligo SAT2 F 5′ CTGCAATCATCCAATGGTCG 3′ SAT2 R 5′ GATTCCATTCGGGTCCATTC 3′ For MyoD control oligo MyoD F 5′ CCTCTTTCGGTCCCTCTTTC 3′ MyoD R 5′ TTCCAAACCTCTCCAACACC 3′ Each ChIP followed by Real-time PCR was performed thrice and done in duplicates. DNA Methylation analysis by bisulfite sequencing {#s2g} ------------------------------------------------ Sodium bisulphite modification of DNA was performed using BisulFlash DNA modification kit (Epigentek) according to the manufacturer\'s protocol using 1 µg of genomic DNA. Modified DNA was collected in 20 µl of Elution Buffer. Methylation-specific PCR was set up using the following primers: CMV promoter BisCMVF 5′ ATTTGGTAGTATATTAAGTGTATTATATGT 3′ BisCMVFR 5′ AACTCTACTTATATAAACCTCCC 3′ 3L-L region Bis3LLF 5′ TTGAGTTATTTTTTGATTTTGTGGATAA 3′ Bis3LLR 3′ CAAAAATCCAAACCCACCTAAA 3′ Endogenous 3L-L region Bis 3LL F 5′ TGTTTTTATAGTTTGGTGATTTTTGG 3′ Bis 3LL R 5′ CCCCTAAAAATAATCTTAACCAACC 3′ The PCR products obtained were cloned using pCR2.1-TOPO cloning kit (Invitrogen). For each region a minimum of 15 clones were sequenced to confirm the methylation pattern. Gene knock-down by siRNA {#s2h} ------------------------ The siRNA knock-down of the polycomb proteins *EZH2* and *EED* and suppressor of varigation protein, *SUV39H1* was performed in HEK293 cells using the siRNA Smart Pools for these proteins (Thermo Scientific Dharmacon) according to the manufacturer\'s protocol. Expression of the endogenous *EZH2*, *EED* and *SUV39H1* was assayed by qRT-PCR. The downregulation of the individual genes in specific-siRNA transfected cells was quantitated in comparison to that observed for cells transfected with scrambled siRNA (from Thermo Scientific Dharmacon). The endogenous expression level of *DNMT3L* was measured in the knock-down and the untreated cells by qRT PCR thrice and in duplicates. RT-PCR primers for the respective genes are as follows: For Suv39H1 Suv39H1F -- 5′GCGTATCCTCAAGCAGTTCC3′ Suv39H1R -- 5′TCCAGGTCCACCTCATTCTC3′ For EZH2 EZH2F- 5′ATGCGACTGAGACAGCTCAA3′ EZH2R- 5′TGGGATGACTTGTGTTGGAA3′ For EED EED F- 5′GAGAGGGAAGTGTCGACTGC3′ EED R- 5′CATTTTCCCTTTCCCCAACT3′ For CYCLOPHILIN B CYCF- 5′CTTCCCCGATGAGAACTTCA3′ CYCR- 5′TTATCCCGGCTGTCTGTCTT3′ For endogenous *DNMT3L* DNMT3L F- 5′CTCTCAAGCTCCGTTTCACC3′ DNMT3L R- 5′GTACAGGAAGAGGGCATCCA3′ Results {#s3} ======= Functional analysis of the *DNMT3L* promoter/Exon 1 region {#s3a} ---------------------------------------------------------- In our previous study, we had identified a CpG island spanning the promoter (DNMT3L-002, Ensembl Transcript ID ENST00000431166) and Exon 1 (DNMT3L-001, Ensembl Transcript ID ENST00000270172) of the human *DNMT3L* gene that showed loss of DNA methylation in different cancers [@pone.0093561-Gokul2], [@pone.0093561-Manderwad1]. The loss of DNA methylation within this region was also shown to correlate with increased *DNMT3L* expression [@pone.0093561-Gokul1]. As this region showed differential DNA methylation profiles in normal and cancer samples we have referred to it as DNMT3L DMC (Diffferentially Methylated in Cancer) in this manuscript. To examine whether this CpG island within the *DNMT3L* promoter/Exon1 exerts a cis-regulatory role, we performed reporter gene assays in mammalian cell lines and Drosophila. Transient transfection assay in cell lines {#s3b} ------------------------------------------ For functional analysis of the DNMT3L DMC, we cloned two overlapping fragments from this region into the pAcGFP-CMV vector ([Figure 1](#pone-0093561-g001){ref-type="fig"}, see [Materials and Methods](#s2){ref-type="sec"}). The smaller region 3L-S contained the 11CpGs that were examined in our previous study [@pone.0093561-Gokul2]. The larger fragment 3L-L contained 19 CpG (including all the 11 present in 3L-S, [Figure 1](#pone-0093561-g001){ref-type="fig"}). Both the fragments were cloned upstream of the CMV promoter in both orientations. A 1 kb region from human chromosome 1, which was previously shown to have no effect on transcriptional potential of a promoter and a 1.5 kb region from the H19 ICR, which has been shown to be a transcriptional repressor were also examined in our assay (see [Materials and Methods](#s2){ref-type="sec"} and [Figure 2A](#pone-0093561-g002){ref-type="fig"}, [@pone.0093561-rom1], [@pone.0093561-Ginjala1]). The constructs were transfected into HEK293 cells and 48 hours after transfection, the GFP expression for the various constructs was examined either by Real-time RT-PCR or Western. *GAPDH* was used as control in each Real-Time RT-PCR experiment. To control for the transfection efficiency, ratio of *GFP* transcription and *Kan^R^/Neo^R^* expression (part of the same pAcGFP-CMV vector) was calculated. As can be seen in [figure 2B](#pone-0093561-g002){ref-type="fig"}, the presence of 3L-L and 3L-S in both orientations significantly decreased the expression level of GFP. The approximately 70% decrease in the *GFP* expression was comparable to that observed for the H19 ICR, a known transcriptional repressive element [@pone.0093561-Ginjala1]. For Western analysis of GFP expression, transfection efficiency was controlled by co-transfecting pG5luc vector containing the Luciferase reporter gene. A representative blot is shown in [figure 2C](#pone-0093561-g002){ref-type="fig"}. Ratio of band intensity for GFP and β-TUBULIN proteins was calculated and the values were normalised for the transfection efficiency with the Luciferase activity. As was observed for real-time RT-PCR analysis, the presence of 3L-L and 3L-S in both orientations significantly decreased the expression level of GFP by more than 70% ([figure 2D](#pone-0093561-g002){ref-type="fig"}). {#pone-0093561-g002} Identification of the minimal region within the DNMT3L DMC that can repress transcription {#s3c} ----------------------------------------------------------------------------------------- To identify the minimum region within the DNMT3L DMC that can repress transcription we cloned three overlapping sub-fragments of 3L-S (termed as s1, s2 and s3) in to the pAcGFP-CMV vector ([figure 3A](#pone-0093561-g003){ref-type="fig"}). The GFP expression for the various constructs was examined 48 hours after transfection into HEK293 cells by Real-time RT-PCR. The s2 and s3 constructs showed approximately 70% decrease but s1 construct showed only 20% decrease in GFP expression ([figure 3B](#pone-0093561-g003){ref-type="fig"}). Since s2 and s3 were overlapping fragments, it was decided to examine whether the overlap region (designated as s4) was enough to cause transcriptional repression. As can be seen in [figure 3C](#pone-0093561-g003){ref-type="fig"}, s4 construct also showed approximately 70% decrease in GFP expression. {ref-type="fig"} for the details of the vector). Base-pair position of each fragment with respect to the 3L-S fragment is given within brackets. Vertical lines denote the minimal region that can repress reporter gene transcription. Subfragments of 3L-S have been named s1, s2, s3 and s4. B & C- Quantitative RT-PCR was performed to determine relative transcriptional level of the reporter AcGFP gene in the various constructs. The relative transcriptional levels were measured with respect to the AcGFP expression in the CMV only construct. See [figure 2A](#pone-0093561-g002){ref-type="fig"} for information on the Control and H19 constructs. Error bars represent Standard Deviation (S.D.). Asterisks indicate significant difference (Student\'s t test, \* - p\<0.05, \*\*\* - p\<0.001, \*\*\* - p\<0.005).](pone.0093561.g003){#pone-0093561-g003} Transgene reporter assay in Drosophila {#s3d} -------------------------------------- To examine the role of DNMT3L DMC in a *mini-white* reporter gene assay in Drosophila the 3L-L fragments flanked by the *lox*P site was cloned upstream of the *hsp*70 promoter in the Drosophila cloning vector pCaSPeR and injected into *w^1118^* Drosophila embryos ([Figure 4B](#pone-0093561-g004){ref-type="fig"} and [Materials and Methods](#s2){ref-type="sec"}). It was decided to continue further with only the 3L-L Drosophila transgenic line as we had observed similar levels of transcriptional repression for both 3L-L and 3L-S in transient transfection assay in mammalian cells. Even after several rounds of injections, wherein we injected the 3L-L construct into around 1500 embryos, we obtained only one transgenic line for 3L-L. To take in consideration the effect for integration site on the expression of a transgene in Drosophila, we mobilised the 3L-L fragment from the only transgenic line which was obtained for 3L-L by crossing it to the flies carrying Δ2-3 transposase gene and expressing transposase endogenously. Upon, mobilisation, we obtained 12 independent heterozygous 3L-L transgenic lines ([Table 1](#pone-0093561-t001){ref-type="table"}). The eye colour varied from very light yellow to deep red due to position effect. To check the effect of the 3L-L on the *mini-white* gene expression, we flipped-out the 3L-L region in these lines by crossing them to the *cre*-transgenic Drosophila line [@pone.0093561-Siegal1]. Eye colour comparison was done both visually under the microscope and by the eye pigmentation assay between the 12 transgenic lines and their respective flipped-out lines ([figure 4A](#pone-0093561-g004){ref-type="fig"}). For 8 out of the 12 3L-L transgenic lines, a statistical significant increase in eye colour pigmentation was observed in the flipped-out counterparts ([figure 4C](#pone-0093561-g004){ref-type="fig"}, [Table 1](#pone-0093561-t001){ref-type="table"}) suggesting that the 3L-L region was inhibiting transcription of the *mini-white* reporter gene. We confirmed the genomic localisation of the 3L-L construct in these 8 transgenic lines by inverse PCR. Except for 25.2.12 and 25.2.30, all the transgenic lines had 3L-L construct at different loci in the genome ([Table S1](#pone.0093561.s005){ref-type="supplementary-material"}). {#pone-0093561-g004} 10.1371/journal.pone.0093561.t001 ###### Eye color phenotype of the various DNMT3L DMC transgenic lines and their flipped-out counterparts. {#pone-0093561-t001-1} LINE Eye color (3LL) Eye color (Δ3LL) Relative Pigmentation (%\[Δ3LL/3LL\]) ------------- ----------------- ------------------ --------------------------------------- **25.2.7** Light Yellow Light Yellow 96.05 **25.2.8** Yellow Yellow 102.65 **25.2.11** Light Yellow Orange 318.57 **25.2.12** Light Yellow Orange 333.33 **25.2.13** Light Yellow Light Orange 258.73 **25.2.16** Light Yellow Light Orange 208.43 **25.2.24** Yellow Orange 203 **25.2.25** Red Red 101.76 **25.2.29** Yellow Light Orange 175.45 **25.2.30** Yellow Light Orange 171.42 **25.2.31** Yellow Light Orange 188.49 **25.2.35** Red Red 105.07 The approximate eye color for each line is mentioned. Relative eye pigmentation was calculated as percentage change between each transgenic line and its flipped-out counterpart. Characterisation of interaction between DNMT3L DMC with Polycomb and Trithorax proteins {#s3e} --------------------------------------------------------------------------------------- Polycomb and Trithorax group of proteins are known regulators of gene expression. The functions performed by the two groups are antagonistic. Polycomb group of proteins are involved in repressing gene expression and creating a highly condensed chromatin conformation whereas Trithorax group of proteins help in enhancing gene expression and keeping a gene in active state of chromatin organisation [@pone.0093561-Ringrose1], [@pone.0093561-Schuettengruber1]. Polycomb group (PcG) of proteins are known to bind to the Polycomb Response Elements (PRE) thereby preventing the binding of transcription factors to the DNA [@pone.0093561-Mller1]. It is also believed that the PcG proteins bring about the deacetylation of the histones [@pone.0093561-Vlag1]. To test whether Polycomb or Trithorax group are involved in the observed transcriptional repression by the DNMT3L DMC, two of the eight 3L-L transgenic Drosophila lines (25.2.12 and 25.2.29) were crossed with the following mutant Drosophila lines: *Ash2^1^*, *Mor^1^*, *Brm^2^*, *Trl^R85^* and *Trx^1^* (Trithorax group); *Pc^1^*, *esc^2^*, *Asx^XF53^*, *Psc^1^*, *Pcl^T1^*, *Scm^R5-13B^*, *Phd and Pho* (Polycomb group). We also set up crosses with the histone methyltransferases, *Su(var)2-5^01^* and *Su(var)3-9^06^*, as they are known to suppress position effect variegation and hence can modulate epigenetic circuitry [@pone.0093561-Schotta1]. Comparison of the eye colour of the progeny from these crosses with the original 3L-L transgenic lines was done visually under the microscope as well as by eye pigmentation assay ([figures 5](#pone-0093561-g005){ref-type="fig"}). The results are also tabulated in [Table 2](#pone-0093561-t002){ref-type="table"}.We observed an increase in the eye colour of progeny for both the 3L-L transgenic Drosophila lines when they were crossed to one or more of the Polycomb mutants. Both, 25.2.12 and25.2.29, transgenic lines showed eye colour increase in crosses with *Pc^1^*, *Asx^XF53^* and *Pho*. In addition, 25.2.29 also showed eye color increase when crossed with *Phd*, *esc^2^* and *Su(z)2*. Progeny of both the 3L-L transgenic Drosophila lines also showed increase in eye colour upon crossing with *Su(var)2-5^01^*. On the other hand, none of the Trithorax mutants showed any change in the eye colour when crossed to the 3L-L transgenic Drosophila lines ([figure 5A](#pone-0093561-g005){ref-type="fig"} and [Table 2](#pone-0093561-t002){ref-type="table"}). {#pone-0093561-g005} 10.1371/journal.pone.0093561.t002 ###### Interaction of DNMT3L DMC with Polycomb and Trithorax proteins. {#pone-0093561-t002-2} LINE 25.2.12 25.2.29 ---------------------- --------------------------- ------------------ --------- --------- ***Pc^1^*** Polycomb PRC1 ✓ ✓ ***Psc^1^*** Polycomb PRC1 × × ***Scm^R5-13B^*** Polycomb PRC1 × × ***Phd*** Polycomb PRC1 × ✓ ***su(z)2^1.a1^*** Polycomb PRC1 × ✓ ***esc^2^*** Polycomb PRC2 ✓ ✓ ***Pcl^T1^*** Polycomb PRC2 × × ***Asx^XF53^*** Polycomb Polycomb related ✓ ✓ ***Pho*** Polycomb PhoRC ✓ ✓ ***Su(var)3-9^06^*** Suppressor of variegation × × ***Su(var)2-5^01^*** Suppressor of variegation ✓ ✓ ***Ash2^1^*** Trithorax MLL complex × × ***Mor^1^*** Trithorax Swi/snf complex × × ***Brm^2^*** Trithorax Swi/snf complex × × ***Trl^R85^*** Trithorax Trx recruiter × × ***Trx^1^*** Trithorax TAC1 complex × × Comparison of eye color of DNMT3L DMC transgenic lines with progeny of their crosses with Polycomb, Trithorax and Suvar mutants was performed. Progeny from the crosses showing eye color change are denoted by ✓. × represents no change. To confirm that the effect on *mini-white* gene expression was due to the interaction of Polycomb proteins with DNMT3L DMC and did not reflect the chromatin organisation of the genomic loci where the transgene had got integrated, crosses of Polycomb mutants were set up with the 3L-L flipped out lines. As shown in [Figure S2](#pone.0093561.s002){ref-type="supplementary-material"}, no significant difference in eye colour was noted in the progeny of the crosses between Polycomb mutants and 3L-L flipped out lines indicating that the repression of mini-white gene was being mediated by the interaction of polycomb group of proteins with the 3L-L region. To further confirm the binding of Polycomb protein to the 3L-L region in the transgenic lines, we performed ChIP with Pc′ antibody and analysed for Pc′ binding with 3L-L region in the two independent 3L-L transgenic lines (25.2.12 and 25.2.29). As can be seen in [Figure S3](#pone.0093561.s003){ref-type="supplementary-material"}, Polycomb protein indeed binds to the 3L-L region. Histone modification profile of the Hsp70 promoter in the mini-white reporter construct in the presence of DNMT3L DMC {#s3f} --------------------------------------------------------------------------------------------------------------------- Several of the Polycomb group of proteins and Suvar proteins are involved in modulating chromatin conformation either by modifying or binding to specific modified residues on histone proteins [@pone.0093561-Ringrose1], [@pone.0093561-Imhof1], [@pone.0093561-Peterson1]. The modifications of histone that are associated with a locus reflect its chromatin conformation [@pone.0093561-Peterson1]. To examine whether the interaction of Polycomb group of proteins with the DNMT3L DMC translated in to change in chromatin conformation at the *hsp70* promoter and 3L-L region of the reporter gene construct, we probed the association of some histone modifications to these regions. Using antibodies to H3K4me~3~ and H3K9ac (specific to active chromatin) and H3K9me~3~ and H3K27me~3~ (usually associated with inactive chromatin) we performed Chromatin Immunoprecipitation analyses on the 25.2.12 DNMT3L DMC transgenic line [@pone.0093561-Schuettengruber2], [@pone.0093561-Mahmoudi1]. As can be seen in [figure 6A](#pone-0093561-g006){ref-type="fig"}, the 3L-L region was preferentially associated with H3K9me~3~ and H3K27me~3~ as compared to the H3K4me~3~ and H3K9ac modifications (p\<0.05). Comparison of these modifications between the *hsp70* promoter in the 3L-L transgenic lines and their flipped out counterpart showed that H3K9me~3~ and H3K27me~3~ association with the *hsp70* promoter was significantly more in the 3L-L lines than their flipped-out counterpart (p\<0.005 for H3K9me~3~ and p\<0.01 for H3K27me~3~, [figure 6B](#pone-0093561-g006){ref-type="fig"}). H3K4me~3~, an active chromatin specific histone modifications was significantly less associated with the *hsp70* promoter in the 3L-L line as compared to its flipped-out counterpart (p\<0.05, [figure 6B](#pone-0093561-g006){ref-type="fig"}). No change was observed for the association of H3K9ac with *hsp70* promoter in the two lines. {#pone-0093561-g006} Effect of DNMT3L DMC on the interaction between the CMV promoter and Polycomb/Trithorax proteins in mammalian cells {#s3g} ------------------------------------------------------------------------------------------------------------------- Based on our observation that the 3L-L region interacted with the Polycomb proteins in the Drosophila reporter gene assay, we sought to examine the interaction of various Polycomb and Trithorax proteins with the CMV promoter in the mammalian transient transfection reporter assay by ChIP analysis using antibodies to these proteins. Polycomb proteins EZH2, EED, ASXL1, CBX2 and PHF1; Trithorax proteins MLL and WDR5; Histone methyltrasferase SUV39H1 and the heterochromatin proteins HP1α (CBX5, product of *Su(var)2-5* in Drosophila) and HP1β (CBX1) were analysed for their association with the CMV promoter [@pone.0093561-Schotta1]. Similar to our observations in the Drosophila experiments, the association of Polycomb proteins EZH2, EED, ASXL1, CBX2 and PHF1 was significantly more with the CMV promoter in presence of the 3L-L region ([figure 7 A & B](#pone-0093561-g007){ref-type="fig"}). The association of both the Trithorax proteins WDR5 and MLL were significantly less with the CMV promoter in the presence of 3L-L region in the construct ([figure 7A](#pone-0093561-g007){ref-type="fig"}). SUV39H1 and HP1α also associated significantly more with the CMV promoter in the presence of the 3L-L region ([Figure 7C](#pone-0093561-g007){ref-type="fig"}). However, HP1β showed decreased association with CMV promoter in presence of the 3L-L region in the construct ([Figure 7C](#pone-0093561-g007){ref-type="fig"}). {ref-type="fig"} for graphical representation of the constructs). Quantitation was done by Real-time PCR. Enrichment of the CMV region in the bound fraction is represented as percentage of Input. The names of the chromatin proteins examined are mentioned below the X-axis. D--G. Histone modifications and DNA methylation profile of the 3L-L region in the *AcGFP* reporter gene construct after transient transfection in HEK293 cells. C- ChIP analysis for the 3L-L region with the indicated histone H3 modifications. D- Comparison of histone modifications associated with the CMV promoter in CMV only (vector) and 3L-L (3LL) constructs after transfection into HEK293 cells. Enrichment in the bound fraction is represented as percentage of Input. IgG - control ChIP with rabbit IgG. Error bars represent Standard Deviation (S.D.). Asterisks indicate significant difference (Student\'s t test, \* - p\<0.05, \*\* - p\<0.01, \*\*\* - p\<0.005). F, G- Methylation profile of the 3L-L region (F) and CMV promoter (G) in the construct after bisulfite sequencing. Compare the methylation profile of the CMV promoter in CMV only (vector) and the 3L-L (3LL) constructs. Each horizontal line indicates a single clone from the 3L-L PCR products after bisulfite treatment. Circles denote CpG dinucleotides present within the sequence. The positions are not drawn to scale. Open circles indicate no methylation. Filled circles represent methylated cytosine.](pone.0093561.g007){#pone-0093561-g007} Histone modification and DNA methylation profile of the CMV and 3L-L region in the transient transfection assay {#s3h} --------------------------------------------------------------------------------------------------------------- To examine if the epigenetic status of the 3L-L region and the CMV promoter is also influenced by the presence of the 3L-L region, we performed Chromatin immunoprecipitation (ChIP) analysis on transiently transfected HEK293 cells using antibodies against various Histone modifications. As shown in [figure 7D](#pone-0093561-g007){ref-type="fig"}, the 3L-L region in the 3L-L+ construct showed a preferential association with H3K9me~3~ and H3K27me~3~ modifications as compared to H3K9ac. Comparison of ChIP analysis for the CMV promoter in the 3L-L+ and the CMV only constructs showed a gain of H3K9me~3~ and H3k27me~3~ modifications at the CMV promoter in the 3L-L+ constructs ([figure 7E](#pone-0093561-g007){ref-type="fig"}). No difference was observed between the 3L-L+ and control constructs for the H3K4me~3~ and H3K9ac modifications ([figure 7E](#pone-0093561-g007){ref-type="fig"}). Unlike Drosophila, DNA methylation is an important part of the epigenetic circuitry in mammalian cells [@pone.0093561-Zachary1]. To assess if the presence of 3L-L in the construct has any effect on the DNA methylation status of the CMV promoter we performed DNA methylation analysis by Bisulfite sequencing. No significant difference was observed in the DNA methylation profile of the CMV promoter between the cells transfected with the control CMV and the 3L-L+ construct ([figure 7G](#pone-0093561-g007){ref-type="fig"}). The 3L-L region ([figure 7F](#pone-0093561-g007){ref-type="fig"}) in the 3L-L+ construct also showed no appreciable gain of DNA methylation in the transient transfection assay. Epigenetic profile of the DNMT3L DMC at the endogenous locus in mammalian cells {#s3i} ------------------------------------------------------------------------------- Next it was sought to assess whether the epigenetic status of the DNMT3L DMC region in its endogenous locus matched with what was observed in the reporter gene assays. As was observed for the 3L-L region in the transfection construct, endogenous 3L-L region in the HEK293 cell line (used for the transient transfection assays) also preferentially associated with H3K9me~3~ and H3K27me~3~ as compared to H3K9ac ([figure 8A](#pone-0093561-g008){ref-type="fig"}). In addition, DNA methylation status of the DNMT3L DMC region at the endogenous locus was found to be unmethylated in HeK cells, similar to what was observed for the 3L-L region in the construct ([figure 8B](#pone-0093561-g008){ref-type="fig"}). {#pone-0093561-g008} To test whether the histone modifications observed at the endogenous 3L-L region correlated with its interaction to relevant Polycomb and Suvar group proteins, we performed ChIP with antibodies against a few Polycomb, Suvar and Trithorax group of proteins for the DNMT3L DMC region at the endogenous locus in HeK cells. As can be seen from [figure 8C](#pone-0093561-g008){ref-type="fig"} for HeK cells, DNMT3L DMC preferentially associated with EZH2, EED, CBX2 (Polycomb) and SUV39H1 (Suvar) proteins. Both the Trithorax proteins MLL and WDR5 did not show any association with DNMT3L DMC. This interaction profile of DNMT3L DMC with Polycomb and Suvar proteins was not dependent of its DNA methylation status as we observed the same profile for HeLa cells that showed higher level of DNA methylation ([Figure S4A](#pone.0093561.s004){ref-type="supplementary-material"}, B). However, as reported previously [@pone.0093561-Gokul1], *DNMT3L* expression seems to be partially dependent on DNA methylation as its expression was found to be more in HeK293 cells as compared to HeLa cells ([Figure S4C](#pone.0093561.s004){ref-type="supplementary-material"}). Effect of Polycomb proteins on the transcription of the *DNMT3L* gene from the endogenous locus {#s3j} ----------------------------------------------------------------------------------------------- To examine whether Polycomb proteins have an effect on the transcription of the *DNMT3L* gene from its endogenous locus in HEK293 cells, siRNA based knock-down of *SUV39H1*, *EZH2* and *EED* was performed. Approximately 70% decrease in expression of *SUV39H1*, *EZH2* and control *CYCLOPHILIN B* gene and approximately 35% decrease in *EED* expression was observed ([figure 9A](#pone-0093561-g009){ref-type="fig"}). Comparison of *DNMT3L* expression between untransfected HEK293 cells and HEK293 cells transfected with the various siRNAs was performed by Real-time RT-PCR. As can be seen in [figure 9B](#pone-0093561-g009){ref-type="fig"}, statistically significant increase in *DNMT3L* expression was observed in HEK293 cells transfected with siRNA against *SUV39H1*, *EED* (p\<0.01) and *EZH2* (p\<0.05) as compared to untransfected cells. *CYCLOPHILIN* B siRNA had no effect on *DNMT3L* expression. {#pone-0093561-g009} Discussion {#s4} ========== The role of *DNMT3L* in modulating the DNA methylation at several imprinted loci and its interactions with various epigenetic modifiers like the de novo DNA methyltransferases DNMT3A & DNMT3B and histone H3 at Lysine 4, confers it an important role in regulation of mammalian development [@pone.0093561-Ooi1], [@pone.0093561-Hata1]. Previous results from our laboratory indicated that overexpression of the human *DNMT3L* gene was correlated with carcinogenesis [@pone.0093561-Gokul1], [@pone.0093561-Gokul2]. This would indicate that *DNMT3L* transcription needs to very tightly regulated so that it is kept silent in most somatic cell types and expressed at appropriate levels only in germ cells and during early embryogenesis [@pone.0093561-Hata1], [@pone.0093561-Aapola1]. In our present study we show a Polycomb/Trithorax Response Element (PRE), present within the CpG island that encompasses the *DNMT3L* promoter and first exon region (DNMT3L DMC), can modulate its expression. DNMT3L DMC is a repressor of transcription {#s4a} ------------------------------------------ DNMT3L DMC is the same region which was shown to be hypomethylated in cervical and ocular cancer samples [@pone.0093561-Gokul2], [@pone.0093561-Manderwad1]. The loss of DNA methylation observed for this region was found to be correlated with increased expression of *DNMT3L*, which in turn was found to be a cause of nuclear reprogramming, a salient characteristic of carcinogenesis [@pone.0093561-Gokul1]. There is a possibility that the loss of methylation at DNMT3L DMC observed in cancer samples was merely a reflection of nuclear reprogramming observed in cancer [@pone.0093561-Jones1]. But it is also possible that the changed DNA methylation profile of the *DNMT3L* locus indicated its role in carcinogenesis. If latter was the case then the reason for the loss of DNA methylation only within the promoter/first exon of *DNMT3L* could indicate an important role for this region in the regulation of *DNMT3L*. We show through both Drosophila and mammalian reporter gene assays, that the presence of DNMT3L DMC in cis acts to repress transcription of the reporter gene. The extent of the repression of GFP expression due to the presence of DNMT3L DMC in the transient transfection of mammalian cell was similar to that observed for H19 ICR, a known transcriptional repressor [@pone.0093561-Ginjala1]. The ability of DNMT3L DMC to repress transcription and the fact that *DNMT3L* is kept transcriptionally silent in most somatic cell types would suggest a role for DNMT3L DMC in the tight regulation of *DNMT3L* transcription [@pone.0093561-Hata1], [@pone.0093561-Aapola1]. Dissection of the DNMT3L DMC into smaller regions showed that a 80 bp region was sufficient to repress reporter gene expression ([figure 3](#pone-0093561-g003){ref-type="fig"}). We could identify one binding site for the polycomb protein Pho and 5 binding sites for YY1, a mammalian homolog of Pho within the DNMT3L DMC region but none in the minimal 80 bp region [@pone.0093561-Wang1]. Interestingly, all the overlapping region of 3L-S that were used in our reporter gene assay showed varying levels of repression. Transcriptional repression by s1+ was not to the same extent as 3L-S but it still was able to weakly repress the reporter gene transcription (approximately 20% repression). This would indicate presence of multiple cis-elements or motifs within DNMT3L DMC that are capable of transcriptional repression. DNMT3L DMC is a Polycomb Response Element (PRE) {#s4b} ----------------------------------------------- In Drosophila, the role of Polycomb and Trithorax group of proteins has been well characterised in the developmental control of Hox gene expression [@pone.0093561-Mahmoudi1]. Polycomb proteins are involved in inhibiting Hox gene expression while Trithorax proteins are involved in maintaining Hox genes in an active state [@pone.0093561-Schuettengruber1]. Both the transgenic DNMT3L DMC Drosophila lines, the 3L-L region in the reporter gene construct and the endogenous 3L-L showed interaction with several members of Polycomb proteins that we tested. This suggested that the inhibition of the *mini-white* reporter gene in the DNMT3L DMC transgenic flies was being mediated through Polycomb proteins. The role of Polycomb proteins in repression of *DNMT3L* gene expression was further confirmed by knock-down of Polycomb proteins in mammalian cells which resulted in increased expression of endogenous *DNMT3L*. Polycomb proteins can be sub-classified into at least three groups, PRC1, PRC2 and PhoRC, based on which repressive complex they are part of. While *Pho* is the DNA binding component of the Polycomb complexes, PRC1 and PRC2 complexes achieve chromatin compaction and repression by different mechanisms [@pone.0093561-Schwartz1],[@pone.0093561-Surface1]. DNMT3L DMC in Drosophila transgenic lines and transient transfection assay was able to interact with members of all the three Polycomb subgroups ([Table 2](#pone-0093561-t002){ref-type="table"}, [Figure 7](#pone-0093561-g007){ref-type="fig"}). Apart from Polycomb and Trithorax, proteins like HP1α (su(var)2-5) and SUV39H1 are also involved in chromatin organisation. SUV39H1 is a H3K9 methytransferase and Heterochromatin associated HP1α is generally associated with gene repression and bind to H3 at lysine K9 in the trimethylated form [@pone.0093561-Schotta1], [@pone.0093561-Nestorov1], [@pone.0093561-Grewal1]. HP1α also form a complex with SUV39H1 [@pone.0093561-Krouwels1]. Our results show that HP1α and SUV39H1 also interact with DNMT3L DMC leading to inhibition of the reporter gene expression. It has been suggested that regions that show interaction with both PRC1 and PRC2 complexes have epigenetic regulatory properties and are part of large CpG islands [@pone.0093561-Ku1]--[@pone.0093561-Lynch1]. DNMT3L DMC, which is part of a CpG island, is methylated in most somatic tissues [@pone.0093561-Gokul1]--[@pone.0093561-Manderwad1] and we show here it interacts with PRC1, PRC2 and PhoRC polycomb repressive complexes. This would indicate that DNMT3L DMC possess essential attributes of a regulatory element. That this CpG island within the *DNMT3L* gene possesses multiple cis-elements/motifs, is normally methylated and interacts with multiple repressive elements also underlines its importance in keeping a tight control on the transcription level of *DNMT3L* in various cell types. It was interesting to note decreased association of HP1β (CBX1) with the CMV promoter in presence of DNMT3L DMC. While both the HP1 (HP1α and HP1β) homologs have been correlated with repression of gene expression especially in heterochromatin context, there have been a few reports that have indicated a role of HP1β (or Drosophila HP1b) in transcriptional activation [@pone.0093561-Hwang1]--[@pone.0093561-Shiota1]. Presence of DNMT3L DMC causes the adjacent promoter to adopt an inactive chromatin conformation {#s4c} ----------------------------------------------------------------------------------------------- Regulation of gene activity is achieved by the interplay of DNA with histones and non-histone proteins within the chromatin context [@pone.0093561-Peterson1]. Polycomb group of proteins are either enzymes that can modify specific residues of the histone tails or are proteins that bind to these modified residues [@pone.0093561-Ringrose1]. Concordant with the observation of its interaction with Polycomb proteins, our results also showed that DNMT3L DMC adopts an inactive chromatin conformation in both the Drosophila transgene reporter gene assay and the mammalian transient transfection assay. In both assays, DNMT3L DMC region associated preferentially with inactive chromatin associated histone modifications H3K9(me)~3~ and H3K27(me)~3~ as compared to H3K9ac, a modification linked to active chromatin [@pone.0093561-Schuettengruber2]. Importantly, the presence of the DNMT3L DMC in the construct influenced the chromatin organisation of the reporter gene promoter. In both Drosophila and mammalian cells, the promoter for the reporter gene showed a gain of inactive chromatin-specific histone marks. This indicated that the inactive chromatin conformation adopted by DNMT3L DMC had either spread to the promoter or had influenced the promoter into adopting an inactive chromatin conformation. The inactive chromatin profile at the DNMT3L DMC region within the reporter gene constructs was also mirrored at the endogenous locus in HEK293 cells. Finally, the importance of identifying a cis-regulatory element (a PRE) with in the DNMT3L gene should be viewed in light of the loss of DNA methylation at the DNMT3L DMC observed in certain cancers [@pone.0093561-Gokul2], [@pone.0093561-Manderwad1]. *DNMT3L* is an epigenetic effector that provides specificity to the DNA methylation activity of de novo methyltransferases, plays an important part in setting up DNA methylation imprints in the germ cells and is associated with nuclear reprogramming during carcinogenesis [@pone.0093561-Bourchis1], [@pone.0093561-Gokul1]. Because of its important role in regulating the epigenetic circuitry, our findings that multiple layers of epigenetic modifications at DNMT3L DMC are being utlilised to repress *DNMT3L* expression support our hypothesis that tight regulation of *DNMT3L* expression is required to avoid nuclear reprogramming and initiation of carcinogenesis. Therefore, epigenetic changes at the *DNMT3L* locus leading to its overexpression could be an important event in carcinogenesis. However, further work would be required to identify the regulatory framework for *DNMT3L* in cancer patients and establish the role of this regulatory element during carcinogenesis. Supporting Information {#s5} ====================== ###### **ChIP analysis for known region in the human genome with the antibodies against histone modifications that we used in our study.** The efficacy of the antibodies in our Chromatin immuneprecipitation assay was checked by performing ChIP for genomic loci that are known to be associated with the tested histone modifications. Enrichment in the bound fraction is represented as percentage of Input. The loci tested and the histone modifications examined are mentioned below the X-axis. IgG - control ChIP with rabbit IgG. Error bars represent Standard Deviation (S.D). (TIF) ###### Click here for additional data file. ###### **DNMT3L DMC interacts with Polycomb group of proteins.** Comparison of eye color pigmentation between Δ3L-L transgenic lines (25.2.12 and 25.2.29) and their counterpart lines after crossing with the respective Polyomb (*Pc^1^*, *esc^2^*, *Asx^XF53^*, *Psc′*, *Pcl^T1^*, *Scm^R5-13B^*, *Phd*, *Pho*), Trithorax (*Ash2^1^*, *Mor^1^*, *Bbrm^2^*, *Trl^R85^*, *Trx^1^*) and Supressor of Variegation (*Su(var)2-5^01^*, *Su(var)3-9^06^*) mutants. Δ P/+, heterozygous Δ 3L-L transgenic lines; Δ 3LL/−, their respective counterparts after crosses with the respective mutant (− is the name of the Polycomb, Trithorax or Suvar mutant). Each bar represents eye color pigmentation for progeny from crosses of individual transgenic lines with a particular mutant, the details of which are provided below the X-axis. As the assays were done in batches, the eye pigmentation for the control 3L-L transgenic line was done for each batch and is shown as white bars (ΔP/+ male with *W^1118^*) in the graphs. Error bars represent Standard Deviation (S.D.). Asterisks indicate significant difference (Student\'s t test, \* - p\<0.05, \*\* - p\<0.01). (TIF) ###### Click here for additional data file. ###### **ChIP analysis for the 3L-L region in the reporter gene construct from transgenic Drosophila.** ChIP analysis was done using Polycomb protein antibodies for the 3L-L region in the reporter gene construct. Quantitation was done by Real-time PCR. Enrichment of the 3L-L region in the bound fraction is represented as percentage of Input. IgG - control ChIP with rabbit IgG. Error bars represent Standard Deviation (S.D.). (TIF) ###### Click here for additional data file. ###### **Epigenetic profile of endogenous DNMT3L DMC in HeLa cells.** A- ChIP analysis for the 3L-L region at the endogenous *DNMT3L* locus using antibodies to the various Polycomb, Trithorax and Suvar proteins mentioned below the X-Axis. Enrichment in the bound fraction is represented as percentage of Input. IgG - control ChIP with rabbit IgG. B. DNA methylation profile of the 3L-L region at the endogenous *DNMT3L* locus in HeLa cells. C. *DNMT3L* gene expression from the endogenous locus in HeLa and HEK293 cells was quantitated by Real-time RT-PCR. Error bars represent Standard Deviation (S.D.). (TIF) ###### Click here for additional data file. ###### **Genomic location of the 3L-L construct in transgenic Drosophila.** Inverse PCR followed by Sequencing of the PCR product was done to determine the Drosophila genomic loci harbouring the 3L-L construct. The coordinates mentioned are with reference to the Drosophila genomic sequence available in NCBI (provided in column 5). (XLSX) ###### Click here for additional data file. ###### **Isolation of chromatin from Drosophila larvae.** Section 1 describes protocol to isolate Drosophila Larvae chromatin. Section 2 describes protocol to perform Chromatin Immunoprecipitation on Drosophila Larvae. (DOC) ###### Click here for additional data file. We thank Narendra, Shreekant, Bharathi for their help in fly work. A.B. is the recipient of Junior and Senior Research Fellowships of the Council of Scientific and Industrial Research towards the pursuit of a PhD degree of the Manipal University. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RKM SK. Performed the experiments: AB VD. Analyzed the data: AB VD RKM SK. Contributed reagents/materials/analysis tools: AB RKM SK. Wrote the paper: AB SK.

Introduction {#s1} ============ The lower urinary tract (LUT, bladder and urethra) is responsible for the storage and timely elimination of urine and is controlled by autonomic, sensory, and central motor innervation. The autonomic components are comprised of the pelvic ganglia (PG), which harbor both sympathetic and parasympathetic neurons (Wanigasekara et al., [@B40]). The coordination of these neural components allows the bladder smooth muscle to relax as it fills with urine while the internal urethral sphincter contracts to prevent urine leakage, and vice versa when the micturition reflex is initiated (de Groat et al., [@B5]). When any of these neural components do not function properly, urinary incontinence or retention can occur and severely diminish quality of life. The nerves supplying bladder innervation are prone to damage in instances of spinal cord injury and during surgical procedures in the lower abdomen (Zullo et al., [@B48]; Wyndaele, [@B44]). Additionally, LUT dysfunction can arise during the course of natural aging (Wagg et al., [@B39]). Despite the widespread prevalence of these conditions, treatment options remain limited. Regenerative strategies aimed at restoring LUT innervation are promising (Kim et al., [@B17]), and such efforts require a deeper understanding of the molecular processes governing the development of LUT innervation (Keast et al., [@B15]). In an effort to characterize patterns of gene expression in the development of LUT sensory innervation, we previously examined expression of the serotonin receptor 5-HT3A (encoded by the *Htr3a* gene) in developing dorsal root ganglia (Ritter and Southard-Smith, [@B28]). In these analyses we also observed pronounced expression of *Htr3a* in fetal mouse pelvic ganglia that supply autonomic innervation to the bladder. This finding is of particular interest in the context of a previous study reporting severe voiding dysfunction in *Htr3a*^*vs*/*vs*^ mice that express a constitutively active form of 5-HT3A (Bhattacharya et al., [@B1]). However, the role of 5-HT3A signaling in the development of LUT autonomic innervation has not been previously examined in detail. Given the evidence of 5-HT3A receptor activity in adult LUT function and expression of this receptor early in fetal LUT development, we tested the hypothesis that 5-HT3A is required for the development and function of bladder innervation. Using the *Htr3a*^−/−^ knockout line, we demonstrate here that 5-HT3A is required for normal micturition in male mice. Conducting the spontaneous void spot assay in pre-pubescent *Htr3a* mutant mice revealed aberrant urinary voiding in male, but not female, knockout animals. Anesthetized bladder cystometry revealed that in males, the loss of 5-HT3A receptor activity diminishes bladder voiding efficiency. At the cellular level we found that 5-HT3A is required for the maturation of autonomic and sensory neuronal fibers innervating bladder smooth muscle. Collectively our results demonstrate a novel role for 5-HT3A in peripheral nervous system development and further highlight this receptor as a critical signaling component in the regulation of micturition. Materials and methods {#s2} ===================== Animals ------- All experimental protocols were approved by the Institutional Animal Care and Use Committee (IACUC) at Vanderbilt University and University of Wisconsin-Madison. Tg(*Htr3a*-EGFP)DH30Gsat/Mmnc ([RRID: MMRRC_000273-UNC](https://scicrunch.org/resolver/RRID:MMRRC_000273-UNC)) transgenic mice were obtained from the Mutant Mouse Resource & Research Center and maintained as heterozygotes on an outbred Swiss Webster background. Tg(*ChAT*-EGFP)GH293Gsat/Mmucd (hereafter *ChAT*-EGFP) transgenic reporter mice were obtained from the Mutant Mouse Resource and Research Center and maintained as homozygotes on a C57BL6/J background. B6.129X1-*Htr3a*^*tm*1*Jul*^/J ([RRID: IMSR_JAX:005251](https://scicrunch.org/resolver/RRID:IMSR_JAX:005251); hereafter referred to as *Htr3a*^−/−^) mice were acquired from The Jackson Laboratory and maintained on a C57BL6/J background. To generate *ChAT-*EGFP; *Htr3a-*KO animals for pelvic ganglia gene expression studies, *ChAT*-EGFP/EGFP homozygotes were bred to *Htr3a*^−/−^ mice to generate *ChAT-*EGFP/+; *Htr3a*^+/−^ animals. These progeny were then mated with *Htr3a*^+/−^ mice to produce *ChAT*-EGFP/+; *Htr3a*^+/+^, *Htr3a*^+/−^, and *Htr3a*^−/−^ animals. This breeding strategy was adopted to avoid comparing expression patterns of heterozygous vs. homozygous expression of the *ChAT-*EGFP transgene. All mice were housed in a modified barrier facility at approximately 25°C and 25--40% humidity. Mice were maintained on a 14-h on, 10-h off light cycle and housed in rectangular polypropylene micro-isolator cages (Lab Products Inc., \#10025, 29.2 × 19.0 × 12.7 cm) and provided standard diet (Purina PicoLab Laboratory Rodent Diet \#5L0D) and water *ad libitum*. Males and females were paired for timed matings to obtain fetal tissues at specific developmental stages; the morning of observation of a seminal plug was designated as 0.5 days post coitus (dpc). Tissue collection ----------------- Male and female mouse fetuses were dissected in ice-cold 1X Phosphate Buffered Saline (PBS) and fixed in 10% Neutral Buffered Formalin (NBF, Sigma Aldrich HT501128) overnight at 4°C. Following fixation, tissues were washed three times in 1XPBS for 15 min each and then finally washed for 1 h. Tissues were then infiltrated with 30% sucrose in 1XPBS and stored in the same solution at 4°C until the day of embedding and cryo-sectioning. To obtain pelvic ganglia and bladders at adult stages, mice were anesthetized by isoflurane inhalation and euthanized by cervical dislocation. A lower abdominal incision was made to reveal the LUT and the pelvic ganglia were carefully teased away from the neck of the bladder and pinned out flat on petri dishes lined with Sylgard (Ellsworth Adhesives, \#4019862). The pelvic ganglia were fixed in 10% NBF overnight at room temperature while gently shaking. Bladders were removed by cutting the proximal urethra and the ureters. To remove urine and allow bladder relaxation so that smooth muscle was not stretched thin, bladders were flushed with 1XPBS and subsequently emptied using a needle and syringe before fixation in 10% NBF overnight at 4°C. Bladders and pelvic ganglia were washed three times in 1XPBS for 15 min each, followed by 1 h of washing in 1XPBS and infiltrated with 30% sucrose at 4°C. Immunohistochemistry -------------------- ### Cryo-sections Fixed tissues for cryo-sectioning were embedded in Tissue Freezing Medium (TFM, General Data, \#TFM) and immediately sectioned with a Leica Cryostat (CM1900-UV). Sagittal sections of 20 μm thickness were cut and mounted onto 3-APES treated slides (Sigma Aldrich, A3648) and dried for 30 min at 37°C on a slide warmer. Slides were then submerged in 1XPBS-0.3% Triton X-100 to dissolve residual TFM for 5 min. Blocking solution (1XPBS-0.3% Triton X-100, 10% Bovine Serum Albumin (Sigma Aldrich, A2153), 5% Normal Donkey Serum (Jackson ImmunoResearch, 017-000-121), 0.45 μm sterile filtered) was applied to sections for at least 30 min at room temperature. Sections were then incubated in primary antibodies diluted in blocking solution overnight at 4°C. Primary antibody information can be found in Table [1](#T1){ref-type="table"}. On the following day, sections were rinsed with sterile 1XPBS and incubated in secondary antibodies diluted in blocking solution for 1 h at room temperature. Secondary antibody information can be found in Table [2](#T2){ref-type="table"}. After rinsing with sterile 1XPBS, sections were incubated in 0.5 mM cupric sulfate in 50 mM ammonium acetate buffer, pH 5.0, for 10 min at room temperature to quench autofluorescence (Potter et al., [@B24]). Sterile water was used to quench the cupric sulfate reaction. Slides were finally mounted with AquaPolyMount (PolySciences, Inc., 18606) and coverslipped. ###### Primary antibodies used in immunohistochemistry experiments. **Antibody** **Host** **Vendor** **RRID** **Dilution** --------------------------------------------- ---------- ----------------------------------- ------------------------------------------------------------ -------------- Tyrosine Hydroxylase (TH) Sheep Millipore, \#AB1542 [AB_90755](https://scicrunch.org/resolver/RRID:AB_90755) 1:1,000 Vesicular Acetylcholine Transporter (vAChT) Rabbit Synaptic Systems, \#139-103 [AB_887864](https://scicrunch.org/resolver/RRID:AB_887864) 1:1,000 Calcitonin Gene Related Peptide (CGRP) Rabbit Sigma Aldrich, \#C8198 [AB_259091](https://scicrunch.org/resolver/RRID:AB_259091) 1:1,000 Neurofilament 200 (NF200) Rabbit Sigma Aldrich, \#N4142 [AB_477272](https://scicrunch.org/resolver/RRID:AB_477272) 1:500 β-III Tubulin (Tuj1) Rabbit Covance, \#PRB-435P [AB_291637](https://scicrunch.org/resolver/RRID:AB_291637) 1:10,000 Hu C/D Human Gift of Vanda Lennon, Mayo Clinic N/A 1:10,000 ###### Secondary antibodies used in immunohistochemistry experiments. **Antibody** **Vendor** **RRID** **Dilution** ----------------------------------- --------------------------------------- -------------------------------------------------------------- -------------- Donkey anti-Rabbit Cy3 Jackson ImmunoResearch, \#711-165-152 [AB_2307443](https://scicrunch.org/resolver/RRID:AB_2307443) 1:1,000 Donkey anti-Rabbit AlexaFluor 647 Jackson ImmunoResearch, \#711-605-152 [AB_2492288](https://scicrunch.org/resolver/RRID:AB_2492288) 1:250 Donkey anti-Rabbit AlexaFluor 488 Jackson ImmunoResearch, \#711-545-152 [AB_2313584](https://scicrunch.org/resolver/RRID:AB_2313584) 1:400 Donkey anti-Sheep Cy3 Jackson ImmunoResearch, \#713-165-147 [AB_2340728](https://scicrunch.org/resolver/RRID:AB_2340728) 1:1,000 Donkey anti-Human AlexaFluor 647 Jackson ImmunoResearch, \#709-605-149 [AB_2340578](https://scicrunch.org/resolver/RRID:AB_2340578) 1:200 ### Whole-mount pelvic ganglia Following fixation and rinsing, pelvic ganglia were blocked at least 24 h at 4°C in the same blocking solution used for cryo-sections. Tissues were then incubated in primary antibody dilutions for 48 h at 4°C in a 24-well tissue culture plate and rinsed three times, 20 min each, in sterile 1XPBS. Pelvic ganglia were then incubated in secondary antibody dilutions for 1 h at room temperature while rocking. Tissues were rinsed three times, 20 min each, in sterile 1XPBS and then incubated in 0.5 mM cupric sulfate in 50 mM ammonium acetate buffer, pH 5.0, for 10 min at room temperature to quench autofluorescence. The cupric sulfate quenching reaction was quenched by transferring the tissues to sterile water. Tissues were mounted onto slides with AquaPolyMount and coverslipped and allowed to dry before imaging. Confocal microscopy of immunohistochemical experiments was conducted using a Zeiss LSM 510-Meta confocal microscope, equipped with a 20x/0.75 plan-apochromat objective. Spontaneous void spot assay --------------------------- Spontaneous void spots assays (VSA) were conducted as published (Keil et al., [@B16]), but with modifications as described here. Mice of the same sex were housed in groups of five animals or fewer per cage. Whatman filter paper (3 mm, chromatography grade, GE Healthcare \#3030-347) was cut into sheets to precisely fit the bottom of a standard mouse cage (see section Animals in Materials and Methods) and autoclaved in foil-wrapped packets. The assay was conducted at the same time each day (12:00--1:00 p.m.) to avoid circadian variability in micturition behavior. To conduct the assay, each mouse was placed in its own void assay cage (a cage identical to the animals\' home cage, with no food, water, or bedding) with a piece of autoclaved Whatman filter paper on the floor of the cage. The void assay cages were then returned to the cage rack and left undisturbed for the duration of the assay (1 h). Voiding was assessed for 1 h to minimize stress and anxiety from prolonged food and water deprivation. At the end of the hour, the mice were carefully lifted out of the trial cages and returned to their home cages. Prior to the VSA trial days, 5 days of assay acclimation that entailed placing the animals in individual cages with Whatman paper were conducted to minimize the effects of novel environment (i.e., the assay cage and Whatman paper) on micturition behavior. Following acclimation and 2 days of rest, three consecutive trials were carried out. For each acclimation and trial, the same void assay cage was used for each animal. Acclimation and testing took place while the animals were 4--5 weeks old to minimize the effects of sex hormones and estrous cycling on micturition behavior. The soiled sheets of Whatman paper were stained with Ninhydrin (2% w/v dilution in 190-proof ethanol, Sigma Aldrich, 151173) and allowed to air-dry overnight. Dried Ninhydrin-stained sheets were scanned to a computer as TIF files. Urine void spots were identified by their purple coloration due to Ninhydrin staining and round shape; non-urine marks were identified as either fecal material (brown spots/smudges) or paw prints (purple paw-print shapes). These non-urine artifacts were erased in Adobe Photoshop prior to analysis. The particle analysis tool in ImageJ (<http://imagej.nih.gov/ij/>) was used to measure the total number of void spots \>0.02 cm^2^ and the surface area of each individual void spot. Staining with Ninhydrin allows clear visualization of the borders of overlapping voids (Supplementary Figure [1](#SM1){ref-type="supplementary-material"}). Overlapping void spots were manually counted by drawing a line on the border between overlapping spots and using the freehand selection tool to count them separately. These measurements were then incorporated into the analysis. The total number of individual voids and the combined area of total voids were averaged over the 3 days of trial. Anesthetized bladder cystometry ------------------------------- Cystometric recordings were made in mice under urethane anesthesia using established methods (Uvin et al., [@B37]; Boudes et al., [@B3]; Bjorling et al., [@B2]; Roman et al., [@B29]). Briefly, mice were weighed and anesthetized with urethane (1.43 g/kg, subcutaneous injection). Thirty minutes later, the bladder was carefully exposed through a lower midline abdominal incision. A PE 50 cannula with the end flared was inserted into the dome of the bladder and secured with a 6.0 silk purse string suture. Muscle and skin layers were then closed separately with a 5.0 silk suture. One hour after cannula implantation, the distal part of the cannula was connected to a physiological pressure transducer (Memscap AS, Norway) and an infusion pump (Harvard Apparatus, Holliston, MA) via a 3-way stopcock. Room temperature saline was infused at a rate of 0.8 mL/h for 60--90 min to stabilize voiding patterns and elicit repetitive micturition cycles. Intravesical pressure was recorded continuously using a PowerLab data collection system (ADInstruments, Colorado Springs, CO) connected to a PC computer. The voided volume of saline was quantified by collecting voided saline on small filter papers weighed before and after voiding. Additionally, void volume was simultaneously measured using a force transducer (Model FT03, Grass Instruments) placed underneath the distal tip of the urethra to collect voided infusate. The force transducer was calibrated using known volumes of infusate to generate a pressure-volume curve. At least six reproducible micturition cycles were recorded and used to quantify multiple parameters. Specifically, intercontractile interval (ICI, time between micturition events), maximal voiding pressure (VP), bladder capacity (BC, defined as the saline infusion rate mu ltiplied by the intercontractile interval), void volume (VV), and non-voiding bladder contraction (NVC, intravesical pressure increases \>5 mmHg during the filling phase, without the release of fluid from the urethra) and voiding efficiency (VE, voided volume/bladder capacity × 100) were determined (Vera and Nadelhaft, [@B38]; Füllhase et al., [@B10]; Lai et al., [@B19]; Wu et al., [@B43]). Estrous cycle phase in female mice was not recorded prior to conducting cystometry. Analysis of bladder innervation density --------------------------------------- Sagittal cryo-sections were obtained of *Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^ 18.5 dpc and adult male bladders were stained for markers of various neuronal subtypes: Tuj1 (all neurons), vesicular acetylcholine transporter (vAChT, cholinergic), CGRP (peptidergic nociceptive sensory), and NF200 (myelinated Aδ sensory). Three to five cryo-sections per animal, for each genotype class, were imaged. For each cryo-section, five images (665 × 508 μm) were captured: superior and inferior bladder neck (referred to as "Bladder Neck"), and superior, inferior, and apex bladder dome (referred to as "Bladder Dome"). All five images collectively are referred to as "Total Bladder." Imaging was conducted on a Leica DMI6000B fluorescent microscope equipped with a 20x objective and software zoom set to 1.0x. Using ImageJ software (<http://imagej.nih.gov/ij/>) the grayscale images were processed using the threshold function to set the minimum gray value of the image that would be counted as true fluorescence and not background or auto-fluorescence. The same minimum gray value threshold was used for all images captured within one replicate of immunohistochemistry. To quantify innervation density in bladder smooth muscle, three non-contiguous 75 μm-diameter regions equivalent to 150 pixels were selected in each image and the ImageJ measure function was used to calculate the percentage of pixels in the selected area that were greater than or equal to the minimum gray value threshold. The percentages of pixels above the minimum gray value obtained from each randomly selected region were averaged for each image. Then, these values were averaged for each designated bladder domain (bladder dome, bladder neck, or total bladder). Statistical analysis -------------------- Average voiding frequency assessed by spontaneous void spot assay was averaged across 3 days of experiments and across animals segregated by sex and *Htr3a* genotype. The means were then statistically compared by one-way ANOVA followed by Tukey\'s Honest Significant Difference (HSD) *post-hoc* test to correct for multiple comparisons of pairs of means. A *p* \< 0.05 was considered statistically significant. Anesthetized bladder cystometry data was averaged across animals segregated by sex and *Htr3a* genotype and then statistically compared by one-way ANOVA followed by Tukey\'s HSD *post-hoc* test. A *p* \< 0.05 was considered statistically significant. Bladder smooth muscle innervation density was measured by calculating the percentage of pixels within a selected area that correspond to neuronal fiber staining. These measurements were averaged across male animals of each genotype (*Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^), and means were statistically compared by one-way ANOVA followed by Tukey\'s HSD *post-hoc* test. A *p* \< 0.05 was considered statistically significant. All numerical data are presented as the mean ± standard error of the mean (SEM). All study authors had access to study data, and all authors reviewed and approved the manuscript. Results {#s3} ======= Serotonin receptor 5-HT3A is expressed in mouse fetal and adult major pelvic ganglia ------------------------------------------------------------------------------------ In our previous study examining expression of *Htr3a* in developing dorsal root ganglia (Ritter and Southard-Smith, [@B28]), we also observed transgene fluorescence in the autonomic pelvic ganglia during fetal stages. To further characterize the expression of this serotonin receptor in developing autonomic LUT innervation, we evaluated the distribution and cell type specificity of *Htr3a* in *Htr3a-*EGFP transgenic tissues (Tg(*Htr3a-*EGFP)DH30Gsat/Mmnc, hereafter referred to as *Htr3a-*EGFP) by immunohistochemistry and confocal microscopy. We detected widespread *Htr3a-*EGFP co-expression with pan-neuronal marker Hu C/D in fetal pelvic ganglia at 14.5dpc (Figures [1A--A"](#F1){ref-type="fig"}). Expression of *Htr3a* is maintained through later fetal development in 18.5dpc pelvic ganglia neurons (Figures [1B--B"](#F1){ref-type="fig"}) and into adulthood (Figures [1C--C"](#F1){ref-type="fig"}). From these experiments we conclude that serotonin receptor 5-HT3A is expressed early in developing autonomic LUT innervation, and its expression is maintained in many pelvic ganglia neurons through sexual maturity in adult mice. {#F1} *Htr3a^−/−^* males, but not females, exhibit increased urinary voiding frequency -------------------------------------------------------------------------------- Given previous reports indicating a role for 5-HT3A receptor signaling in micturition (Bhattacharya et al., [@B1]; Schwen et al., [@B30]) and the widespread expression of *Htr3a* in pelvic ganglia neurons, we sought to determine if 5-HT3A is required for normal urinary voiding. To do this we employed the B6.129X1-*Htr3a*^*tm*1*Jul*^/J (hereafter *Htr3a*^−/−^) mouse line, which is a global knockout of *Htr3a* gene expression that has been bred to congenicity on the C57BL/6J inbred strain (Zeitz et al., [@B47]). Bladder function was initially assessed in *Htr3a* mutants and wild type littermates by spontaneous void spot assays (VSA) that have been shown in prior studies to be a reliable assessment of micturition in mice (Yu et al., [@B46]; Bjorling et al., [@B2]; Keil et al., [@B16]). In 1 h assessments conducted across 3 consecutive days, we observed male wild type and heterozygous animals had very similar voiding frequencies (9.63 ± 0.98 voids and 10.69 ± 1.64 voids, respectively, *p* = 0.8894, one-way ANOVA with Tukey\'s HSD *post-hoc* test), while homozygous mutant males exhibited an increased number of voids that averaged 18.21 ± 2.19 (Figure [2](#F2){ref-type="fig"}). The difference in *Htr3a*^−/−^ male voiding frequencies was significantly increased compared to both wild type and *Htr3a*^+/−^ littermates (*p* = 0.0046 and *p* = 0.0074, respectively, one-way ANOVA with Tukey\'s HSD *post-hoc* test, *n* = 15--25 animals per genotype). However, wild type, heterozygous, and *Htr3a* female homozygous mutants did not differ significantly in their average voiding frequencies despite a trending increase in void frequency in female *Htr3a*^−/−^ mice that trends in the same direction as that seen for *Htr3a*^−/−^ males (6.97 ± 1.29, 6.42 ± 0.99, and 9.98 ± 1.27 voids, respectively, *p* = 0.09108, one-way ANOVA, *n* = 13--19 animals per genotype). Total void surface area averaged over three trial days was not significantly different among *Htr3a* genotypes in males or females (data not shown). From these experiments we conclude that male *Htr3a*^−/−^ mutant mice have more frequent urinary voids. {#F2} Male *Htr3a^−/−^* urinary voiding efficiency is impaired -------------------------------------------------------- The 5-HT3A receptor is expressed not only in the autonomic pelvic ganglia and sensory dorsal root ganglia supplying bladder innervation, but also in other aspects of the micturition circuitry. Recent work has shown that 5-HT3A is expressed in Barrington\'s nucleus, the locus coeruleus, periaqueductal gray, and entorhinal cortex (Koyama et al., [@B18]), all of which have been previously implicated in relaying autonomic output and cortical regulation of voiding behavior (Fowler et al., [@B9]). Because the *Htr3a*^−/−^ mutant line used in our studies is a global knockout, micturition centers in the brain expressing 5-HT3A may also be affected. To determine if the effects of 5-HT3A loss observed in the spontaneous void spot assay are due to autonomic dysfunction or cognitive control of voiding behavior, we conducted anesthetized bladder cystometry *in vivo* in female and male *Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^ adult mice. Anesthesia of the animals during cystometry allows isolation of involuntary urinary voiding and eliminates the behavioral component of micturition. Parameters measured during anesthetized cystometry included bladder volume-to-body weight ratio, intercontractile interval, maximal voiding pressure, void volume, bladder capacity, and non-voiding bladder contractions. There were no significant differences between any of the *Htr3a* genotype classes for these measures among either male or female mice (Table [3](#T3){ref-type="table"}). Representative cystometric traces that illustrate comparable values are shown in Figures [3A,B](#F3){ref-type="fig"}. In addition, we measured voiding efficiency, which is defined as the percentage of bladder capacity expelled in one voiding event (Figure [3C](#F3){ref-type="fig"}). Normally, an efficient bladder will expel its entire bladder capacity in one voiding event, which would be reflected in a voiding efficiency value of nearly 100%. Wild type male mice exhibited an average voiding efficiency of 97.24 ± 3.01% while heterozygous mice had a voiding efficiency of 94.39 ± 1.98% and homozygous mutant voiding efficiency was 87.30 ± 3.12% (*n* = 7--14 animals per genotype). This level of voiding efficiency was significantly different for male *Htr3a*^−/−^ mice compared to wild type littermates (*p* = 0.0357, one-way ANOVA with Tukey\'s HSD *post-hoc* test). Despite the significant difference in voiding efficiency for homozygous mice, voiding efficiency of heterozygous mice was not significantly different from wild type or homozygous littermates (*p* = 0.7068 and *p* = 0.0826 respectively, one-way ANOVA with Tukey\'s HSD *post-hoc* test). Average voiding efficiencies of *Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^ females were similar, with values of 90.19 ± 6.60, 90.35 ± 3.01, and 86.43 ± 4.49%, respectively (*p* = 0.7701, one-way ANOVA, *n* = 6--10 animals per genotype). Given the detrimental effect of *Htr3a* gene loss on the ability of homozygous males to efficiently empty bladder content, we conclude that 5-HT3A is required for maintaining normal LUT function in male mice. ###### Summary of anesthetized bladder cystometry measurements. **Cystometric parameter** **Female *Htr3a^+/+^*** **Female *Htr3a^+/−^*** **Female *Htr3a*^−/−^** **One-way ANOVA p-value** **Male *Htr3a^+/+^*** **Male *Htr3a^+/−^*** **Male *Htr3a*^−/−^** **One-way ANOVA *p*-value** -------------------------------------------- ------------------------- ------------------------- ------------------------- --------------------------- ----------------------- ----------------------- ----------------------- ----------------------------- Voiding Efficiency 90.19%±6.60% 90.35 ± 3.01% 86.43 ± 4.49% 0.7701 97.24%±3.01% 94.39 ± 1.98% 87.30%±3.12% **0.0287^\*^** Bladder Volume/Body Weight Ratio (mm^3^/g) 33.96 ± 6.92 32.25 ± 5.71 34.88 ± 7.99 0.9618 78.48 ± 22.23 64.51 ± 10.89 38.97 ± 7.90 0.2000 Bladder Capacity (μL) 54.43 ± 8.01 63.69 ± 6.55 80.97 ± 17.33 0.3427 77.55 ± 11.55 69.81 ± 11.18 75.51 ± 16.66 0.9153 Intercontractile Interval (sec) 244.92 ± 36.06 286.59 ± 29.49 364.44 ± 77.99 0.3423 348.41 ± 51.98 314.17 ± 50.31 339.80 ± 74.96 0.9154 Void Volume (μL) 50.38 ± 8.99 58.32 ± 6.75 67.17 ± 12.94 0.5490 77.84 ± 11.61 66.58 ± 10.21 64.60 ± 13.61 0.7609 Maximum Void Pressure (mmHg) 26.62 ± 3.75 28.56 ± 1.09 32.09 ± 1.26 0.1556 20.53 ± 1.38 21.40 ± 1.36 23.38 ± 2.20 0.5388 Non-Voiding Contractions 0.083 ± 0.083 0.280 ± 0.207 0.478 ± 0.329 0.5979 1.39 ± 0.41 1.40 ± 0.38 1.01 ± 0.33 0.7178 *Anesthetized bladder cystometry was conducted to determine the effects of 5-HT3A receptor loss on unconscious voiding activity. Mean values ± SEM are given for female and male wild type, heterozygous, and homozygous Htr3a knockout adult mice. One-way ANOVA was conducted to determine if differences in means were significant among female mice and male mice. The statistically significant p-value (\<0.05) is presented in bold with an asterisk. Voiding efficiency is defined as the percent bladder capacity expelled in a voiding event. Bladder volume-to-body weight ratio was calculated by dividing the bladder volume (mm^3^) by the animal\'s total body weight. Bladder capacity is defined as the intervoid interval multiplied by the infusion rate of saline during the assay. Intercontractile interval is defined as the time (in seconds) between voiding events. Void volume is defined as the volume of infusate voided during a micturition event. Maximum void pressure is defined as the peak intravesical bladder pressure immediately prior to micturition. Non-voiding contractions are defined as rhythmic increases in intravesical pressure that occur between micturition events*. {#F3} Loss of 5-HT3A disrupts autonomic neuronal marker expression patterns in fetal pelvic ganglia --------------------------------------------------------------------------------------------- Given the consequences of *Htr3a* gene loss on adult bladder function and our observation of *Htr3a* expression in fetal mouse pelvic ganglia, we examined the effects of *Htr3a* loss on the development of pelvic ganglia neurons. To assess the composition of pelvic ganglia during development, we monitored the appearance of cholinergic neurons that comprise the majority of pelvic ganglia neuronal subtypes (Wanigasekara et al., [@B40]) as well as noradrenergic neurons labeled by tyrosine hydroxylase (TH) expression. *Htr3a* mutants were crossed with the Tg(*ChAT-*EGFP)GH293Gsat/Mmucd (hereafter *ChAT*-EGFP) transgenic reporter mouse line (Gong et al., [@B11]) to facilitate imaging of cholinergic neurons, as direct visualization by GFP labeled discrete cholinergic neurons more readily than several commercially available ChAT antisera (data not shown). Progeny from intercrosses of *ChAT-*EGFP; *Htr3a*^+/−^ mice were sectioned sagittally and stained for TH and the pan-neuronal marker Hu C/D. At 14.5 dpc, we observed a visible reduction in *ChAT*-EGFP expression in *Htr3a*^+/−^ and *Htr3a*^−/−^ pelvic ganglia compared to *Htr3a*^+/+^ littermates, along with a concurrent increase in TH expression (*n* = 4 animals per genotype, Figure [4](#F4){ref-type="fig"}). However, when we examined expression of these markers in adult (6--8 weeks old) male and female pelvic ganglia, the differences between *Htr3a* genotypes seen at 14.5 dpc were no longer apparent among multiple replicates examined (*n* = 6 animals per genotype, Figure [5](#F5){ref-type="fig"}). Expression of TH in adult pelvic ganglia across all three genotype classes was comparable to prior reports of noradrenergic neurons in adult mice (Yan and Keast, [@B45]). From these experiments we conclude that loss of 5-HT3A receptor produces a transient disruption of expression of autonomic neuronal subtype markers in fetal pelvic ganglia. {#F4} {#F5} *Htr3a* mutant bladder detrusor is more densely innervated compared to wild type -------------------------------------------------------------------------------- The 5-HT3A receptor has been previously implicated in mediating neuronal morphological complexity in the developing mouse brain (Chameau et al., [@B4]; Oostland et al., [@B22]). Specifically, *Htr3a*^−/−^ mice exhibit hyper-elaboration of pyramidal neuronal dendritic fibers in the developing cortex; however, its role in peripheral neuronal development *in vivo* has not been reported. To determine if loss of 5-HT3A alters aspects of peripheral neuronal development or neuronal morphology, we conducted immunohistochemistry and measured innervation density of bladder smooth muscle on sagittal sections from 18.5dpc *Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^ mice (*n* = 5 animals per genotype). After staining for a variety of neuronal markers, we measured the percentage of pixels of a minimum intensity level within a fixed area of bladder smooth muscle for three domains (total bladder, bladder neck, and bladder dome) as a proxy for average area innervated by labeled neuronal fibers. These measurements, along with statistical *p*-values, are listed in Table [4](#T4){ref-type="table"}. ###### Measurements of 18.5dpc bladder smooth muscle innervation density. **Neuronal Marker** **Percentage Innervation Density in 18.5 dpc Fetal Mice (average** ± **SEM)** **One-way ANOVA** **One-way ANOVA with Tukey\'s** ***post-hoc*** **HSD** ***p*****-values** --------------------- ------- ------------------------------------------------------------------------------- ------------------- --------------------------------------------------------------------------- ------------ ------------ -------- ------------ Tuj1 Total 9.87 ± 1.15 16.22 ± 1.57 17.82 ± 1.58 **0.0052** **0.0232** 0.7193 **0.0057** Neck 10.67 ± 1.76 18.27 ± 1.52 18.99 ± 2.17 **0.0135** **0.0315** 0.9595 0.1928 Dome 9.34 ± 0.82 14.76 ± 1.80 17.04 ± 1.38 **0.0061** **0.0432** 0.4997 **0.0055** vAChT Total 7.23 ± 1.71 15.06 ± 4.19 18.47 ± 2.33 **0.0316** 0.1601 0.6706 **0.0283** Neck 7.80 ± 1.94 13.95 ± 3.07 19.28 ± 3.14 **0.0335** 0.3081 0.4029 **0.0269** Dome 6.85 ± 1.63 14.67 ± 4.10 17.92 ± 1.99 **0.0248** 0.1327 0.6637 **0.0223** CGRP Total 1.29 ± 0.24 1.43 ± 0.34 2.38 ± 0.78 0.2986 N/A N/A N/A Neck 1.62 ± 0.35 1.94 ± 0.46 3.18 ± 1.05 0.2777 N/A N/A N/A Dome 1.08 ± 0.18 1.08 ± 0.28 1.84 ± 0.63 0.3612 N/A N/A N/A NF200 Total 8.05 ± 1.69 10.07 ± 1.92 11.42 ± 2.32 0.5118 N/A N/A N/A Neck 9.13 ± 1.85 10.29 ± 1.77 12.92 ± 2.35 0.4284 N/A N/A N/A Dome 7.33 ± 1.78 9.92 ± 2.08 10.42 ± 2.31 0.5488 N/A N/A N/A *Innervation density was measured by calculating the percentage of pixels in a selected region of 18.5 dpc bladder smooth muscle that exceeded the threshold intensity level for neuronal fiber staining. Average percentages were calculated from values obtained from images of superior and inferior bladder neck ("Bladder Neck"), and superior, inferior, and apex of bladder dome ("Bladder Dome"). "Total Bladder" innervated density was calculated by averaging the percentage values from both bladder neck and bladder dome. Overall differences in means were statistically compared with one-way ANOVA; Tukey\'s Honest Significant Difference (HSD) post-hoc test was conducted to compare pairs of means. Statistically significant p-values (p \< 0.05) are presented in bold*. We first asked if overall innervation patterns differed between wild type, heterozygous, and *Htr3a*^−/−^ fetal tissues by staining for the pan-neuronal marker Tuj1 (β-III tubulin) that strongly labels neuronal fibers. Interestingly, Tuj1+ staining was significantly increased in all areas of bladder smooth muscle examined in *Htr3a*^+/−^ and *Htr3a*^−/−^ samples compared to wild type littermates (*p* = 0.0232 and *p* = 0.0057 respectively, one-way ANOVA with Tukey\'s HSD *post-hoc* test). The average innervation density in wild type bladders was 9.87 ± 1.15%, while *Htr3a*^+/−^ and *Htr3a*^−/−^ Tuj1+ innervation densities were nearly doubled (16.22 ± 1.57 and 17.82 ± 1.58%, respectively) (Figures [6A--A"](#F6){ref-type="fig"}). The difference between *Htr3a*^+/−^ and *Htr3a*^−/−^ Tuj1+ innervation density measurements was not statistically significant (*p* = 0.7193, one-way ANOVA with Tukey\'s HSD *post-hoc* test). {ref-type="table"}. *p* \< 0.05 was considered significant. *n* = 5 animals per genotype. Scale bar = 100 μm. Asterisks denote the level of statistical significance: ^\*^ \< 0.05; ^\*\*^ \< 0.01; ^\*\*\*^ \< 0.001.](fnins-11-00690-g0006){#F6} Given the large proportion of cholinergic neurons found in the autonomic pelvic ganglia and their critical role in mediating bladder contractility (de Groat and Yoshimura, [@B6]), we next asked if 18.5 dpc *Htr3a* mutants would exhibit perturbed cholinergic innervation patterns in the bladder smooth muscle. Upon staining with vesicular acetylcholine transporter (vAChT), we observed that 7.23 ± 1.71% of the wildtype bladder smooth muscle received vAChT+ innervation, while average innervation density in *Htr3a*^+/−^ and *Htr3a*^−/−^ animals was 15.06 ± 4.19 and 18.47 ± 2.33% respectively. The increase in *Htr3a*^−/−^ vAChT+ bladder innervation density was significant compared to wild type bladder (*p* = 0.0283, one-way ANOVA with Tukey\'s HSD *post-hoc* test), but *Htr3a*^+/−^ innervation density was not significantly different compared to either wild type or homozygous bladder (*p* = 0.1601 and *p* = 0.6706 respectively, one-way ANOVA with Tukey\'s HSD *post-hoc* test) (Figures [6B--B"](#F6){ref-type="fig"}). The sensation of bladder filling and muscle stretch is primarily detected by sensory afferent neurons in the dorsal root ganglia (de Groat and Yoshimura, [@B6]). During our prior analysis of *Htr3a* expression in the developing dorsal root ganglia, we documented that many bladder-innervating sensory neurons expressing *Htr3a* also express neuropeptide Calcitonin Gene Related Peptide (CGRP) and mechanosensory marker Neurofilament 200 (NF200) (Ritter and Southard-Smith, [@B28]). We were interested to see if these populations would be affected by loss of *Htr3a*. Upon staining for CGRP, we did not observe any significant differences in CGRP+ innervation density of bladder at 18.5 dpc (*p* = 0.2986, one-way ANOVA) (Figures [6C--C"](#F6){ref-type="fig"}). CGRP+ innervation density in *Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^ averaged 1.29 ± 0.24, 1.43 ± 0.34, and 2.38 ± 0.78%, respectively. Similarly, fetal bladders across these three genotypes did not differ significantly in NF200+ innervation density (Figures [6D--D"](#F6){ref-type="fig"}). Innervation density of NF200+ fibers were 8.05 ± 1.69, 10.01 ± 1.92, and 11.42 ± 2.32% for *Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^, respectively (*p* = 0.5118, one-way ANOVA). From these experiments we conclude that in late fetal development, loss of *Htr3a* gene expression results in an overall increase in the presence of neuronal fibers in fetal bladder smooth muscle, with cholinergic fibers being particularly enriched while sensory fibers are not at this stage of development. Due to the refining of neural architecture that may occur over the postnatal developmental period (Purves and Lichtman, [@B25]; Tapia et al., [@B33]), we sought to determine if differences in innervation patterns would also be present in adult *Htr3a* mutants. We conducted the same staining and analysis procedures on adult male *Htr3a*^+/+^*, Htr3a*^+/−^, and *Htr3a*^−/−^ bladder sagittal sections (*n* = 5 animals per genotype, Table [5](#T5){ref-type="table"}). Quantification of Tuj1+ staining in adult bladder tissues similarly revealed a significant increase in innervation density in *Htr3a*^−/−^ mutant bladder compared to wild type (*p* = 0.0033, one-way ANOVA with Tukey\'s HSD *post-hoc* test) (Figures [7A--A"](#F7){ref-type="fig"}). The average adult wild type Tuj1+ innervation density was 6.32 ± 0.72%, while *Htr3a*^+/−^ bladders were 9.01 ± 0.81% innervated and *Htr3a*^−/−^ bladders exhibited a two-fold increase in innervation density, with average percent innervated values of 12.54 ± 1.49%. The difference between *Htr3a*^+/−^ and wild type Tuj1 innervation density was not significant (*p* = 0.2082, one-way ANOVA with Tukey\'s HSD *post-hoc* test), nor was the difference between *Htr3a*^+/−^ and *Htr3a*^−/−^ Tuj1 innervation density (*p* = 0.0828, one-way ANOVA with Tukey\'s HSD *post-hoc* test). ###### Measurements of adult bladder smooth muscle innervation density. **Neuronal Marker** **Percentage Innervation Density in Adult Mice (average** ± **SEM)** **One-way ANOVA** **One-way ANOVA with Tukey\'s** ***post-hoc*** **HSD** ***p*****-values** --------------------- ------- ---------------------------------------------------------------------- ------------------- --------------------------------------------------------------------------- ------------ ------------ ------------ ------------ Tuj1 Total 6.32 ± 0.72 9.01 ± 0.81 12.54 ± 1.49 **0.0044** 0.2082 0.0828 **0.0033** Neck 6.89 ± 0.57 9.89 ± 1.09 12.38 ± 0.94 **0.0034** 0.0832 0.1631 **0.0026** Dome 5.94 ± 0.87 8.51 ± 1.01 12.64 ± 2.42 **0.0351** 0.5106 0.2016 **0.0294** vAChT Total 3.74 ± 0.54 6.94 ± 1.36 11.98 ± 1.57 **0.0018** 0.2041 **0.0344** **0.0014** Neck 4.19 ± 0.95 7.56 ± 1.84 12.04 ± 1.74 **0.0131** 0.3130 0.1477 **0.0102** Dome 3.43 ± 0.39 6.59 ± 1.15 11.95 ± 1.62 **0.0008** 0.1789 **0.0183** **0.0006** CGRP Total 2.32 ± 0.43 4.49 ± 0.89 7.43 ± 1.17 **0.0052** 0.2323 0.0867 **0.0039** Neck 2.53 ± 0.43 5.07 ± 0.85 8.71 ± 1.97 **0.0153** 0.3605 0.1459 **0.0122** Dome 2.18 ± 0.49 4.10 ± 1.03 6.59 ± 0.77 **0.0069** 0.2426 0.1082 **0.0053** NF200 Total 3.01 ± 0.29 6.06 ± 0.83 7.92 ± 0.79 **0.0009** **0.0219** 0.1759 **0.0007** Neck 4.26 ± 0.68 5.45 ± 0.80 9.59 ± 1.33 **0.0057** 0.6754 **0.0279** **0.0061** Dome 2.18 ± 0.12 6.42 ± 1.08 6.81 ± 0.79 **0.0019** **0.0059** 0.9316 **0.0031** *Innervation density was measured by calculating the percentage of pixels in a selected region of adult bladder smooth muscle that exceeded the threshold intensity level for neuronal fiber staining. Average percentages were calculated from values obtained from images of superior and inferior bladder neck ("Bladder Neck"), and superior, inferior, and apex of bladder dome ("Bladder Dome"). "Total Bladder" innervated density was calculated by averaging the percentage values from both bladder neck and bladder dome. Overall differences in means were statistically compared with one-way ANOVA; Tukey\'s Honest Significant Difference (HSD) post-hoc test was conducted to compare pairs of means. Statistically significant p-values (p \< 0.05) are presented in bold*. {ref-type="table"}. *p* \< 0.05 was considered significant. *n* = 5 animals per genotype. Scale bar = 100 μm. Asterisks denote the level of statistical significance: ^\*^ \< 0.05; ^\*\*^ \< 0.01; ^\*\*\*^ \< 0.001.](fnins-11-00690-g0007){#F7} When we stained for vAChT to visualize cholinergic fibers in adult bladder tissues we noted that *Htr3a* mutants maintained higher than normal levels of cholinergic fiber density into adulthood, and the difference among genotypes was more pronounced than in fetal samples. Adult *Htr3a*^+/+^ bladder smooth muscle tissues had an average vAChT+ innervation density of 3.74 ± 0.54% (Figures [7B--B"](#F7){ref-type="fig"}). In contrast, *Htr3a*^+/−^ and *Htr3a*^−/−^ bladder tissues had an average innervation density of 6.94 ± 1.36 and 11.98 ± 1.57%, respectively. *Htr3a*^−/−^ vAChT+ innervation density was significantly increased compared to both wild type and heterozygous littermates (*p* = 0.0014 and *p* = 0.0344 respectively, one-way ANOVA with Tukey\'s HSD *post-hoc* test), but *Htr3a*^+/−^ vAChT+ innervation was not significantly different from wild type (*p* = 0.2041, one-way ANOVA with Tukey\'s HSD *post-hoc* test). Quantification of sensory innervation density by immunohistochemical localization of CGRP+ fibers revealed that, in contrast to density of sensory innervation in fetal tissues, bladder smooth muscle of adult *Htr3a*^−/−^ mutants is more densely innervated by sensory fibers than that of wild type (*p* = 0.0039, one-way ANOVA with Tukey\'s HSD *post-hoc* test) (Figures [7C--C"](#F7){ref-type="fig"}). Wild type bladder had an average CGRP+ innervation density of 2.32 ± 0.43%, while *Htr3a*^+/−^ innervation density levels were 4.49 ± 0.89% and *Htr3a*^−/−^ CGRP+ innervation was increased approximately three-fold, values of 7.43 ± 1.17%. *Htr3a*^+/−^ CGRP+ innervation density was not significantly different from either wild type or *Htr3a*^−/−^ tissues (*p* = 0.2323 and *p* = 0.0867 respectively, one-way ANOVA with Tukey\'s HSD *post-hoc* test). Staining adult bladder tissues with NF200 also revealed a highly significant increase in mechanosensory neuronal innervation of both adult *Htr3a*^+/−^ and *Htr3a*^−/−^ mutant bladder smooth muscle compared to wild type (*p* = 0.0219 and *p* = 0.0007 respectively, one-way ANOVA with Tukey\'s HSD *post-hoc* test) (Figures [7D--D"](#F7){ref-type="fig"}). *Htr3a*^+/+^ average NF200+ innervation density was 3.01 ± 0.29%, while *Htr3a*^+/−^ and *Htr3a*^−/−^ innervation densities were 6.06 ± 0.83 and 7.92 ± 0.79% respectively. The difference between *Htr3a*^+/−^ and *Htr3a*^−/−^ NF200+ innervation density measurements was not statistically significant (*p* = 0.1759, one-way ANOVA with Tukey\'s HSD *post-hoc* test). From our experiments on adult *Htr3a* mutant tissues, we conclude that loss of *Htr3a* gene expression has sustained effects on innervation densities beyond fetal and postnatal development, in which mutant bladder smooth muscle has an overabundance of neuronal fibers, including cholinergic, neuropeptidergic, and mechanosensory. Discussion {#s4} ========== While the effects of serotonin signaling in neural development have been studied for decades, the 5-HT3A serotonin receptor has only recently been recognized as a key player in developmental mechanisms (Engel et al., [@B8]). These recent studies of 5-HT3A signaling in neural development have been focused on the CNS. The goal of this study was to define the expression and role of 5-HT3A in the development of urinary bladder innervation and determine whether this serotonin receptor is required for normal voiding in adult mice. Using a knockout mouse model of 5-HT3A deficiency, we demonstrate a novel role for this serotonin receptor in the development of the peripheral nervous system architecture and show that perturbing 5-HT3A signaling has lasting effects on visceral organ function in adult animals. The analysis presented has important ramifications for the potential impact of pharmacological agents taken during pregnancy that may have long-lasting effects on bladder innervation and function in later life. Expression of 5-HT3A was first demonstrated in 14.5 dpc autonomic pelvic ganglia via *in situ* hybridization over two decades ago (Tecott et al., [@B35], [@B36]); surprisingly, these findings have not been further investigated. Our expression analysis using the *Htr3a*-EGFP transgenic reporter allele confirms expression of 5-HT3A in fetal pelvic ganglia at stages when neurogenesis is actively occurring (Wiese et al., [@B42]). Moreover, our results show for the first time that the 5-HT3A receptor is maintained in adult pelvic ganglia and is expressed in the majority of adult pelvic ganglia neurons (Figure [1](#F1){ref-type="fig"}). VSAs have been established as a means to quantify urinary voiding in an environment as similar to the animals\' home cage environment as possible and thus eliminate many confounding factors inherent to other methods of urinary voiding assessment (Yu et al., [@B46]; Keil et al., [@B16]). Our VSA study of *Htr3a* mutant mice revealed an increase in voiding frequency in male *Htr3a* mutants compared to wild type littermates (Figure [2](#F2){ref-type="fig"}). This is the first report of any urinary tract dysfunction in *Htr3a* homozygous loss of function mutants. Because detection of any urinary dysfunction requires unique experimental assays and careful analysis of large animal cohorts, this aspect of the *Htr3a* knockout phenotype was most likely overlooked by prior studies of these mutants that observed the animals in their home cage environment. Differences in voiding frequency measured in a 4-h VSA have not been previously reported for wild type post-pubescent male and female mice on a C57BL6/J background (Bjorling et al., [@B2]). We had anticipated that there would be no sex specific effects in our analysis of *Htr3a* mutants. Remarkably, we found that loss of the 5-HT3A receptor in *Htr3a* knockout mice differentially affects voiding patterns of males and females by VSA before the onset of reproductive maturity. The increased voiding frequency we observed in male *Htr3a* mutants suggests that 5-HT3A receptor activity normally functions in male mice to inhibit urinary voiding. It should be noted that female *Htr3a*^−/−^ animals in our VSA study exhibited a slight trending increase in urinary voiding frequency compared to wild type animals, but the increase in frequency was not statistically significant. Our observation suggests that the 5-HT3A receptor may be involved in inhibiting micturition in both males and females, but does so to a greater extent in male mice. In contrast, a pharmacological study conducted in cats demonstrated that 5-HT3A receptor signaling inhibits the micturition drive in both male and female cats (Schwen et al., [@B30]). A separate study examining the effects of 17alpha-estradiol and 17beta-estradiol *in vitro* found that these hormones can act as allosteric antagonists of 5-HT3A in an exogenous expression system (Wetzel et al., [@B41]; Oz et al., [@B23]), but the mechanisms by which sex hormones may alter 5-HT3A activity in regulating urinary voiding *in vivo* are unknown. Interestingly, estrogen receptors are expressed in parasympathetic pelvic ganglia neurons (Purves-Tyson et al., [@B26]) and sex hormones have been shown to influence pelvic ganglia neuron function (Keast, [@B14]), providing an avenue by which 5-HT3A may mediate urinary tract function in a sex-specific manner. Further studies focused on determining if sex hormones regulate 5-HT3A signaling *in vivo* and how this signaling contributes to sex differences in micturition circuitry would be valuable. Because *Htr3a* is expressed in many brain regions involved in cognitive control of micturition and our study relied upon a global *Htr3a* knockout mouse line, there is the possibility that the increase in urinary frequency observed in male *Htr3a* mutants could be due to abnormal cognitive control of when and where to void. However, the anesthetized bladder cystometry assay we applied addresses this issue by specifically measuring involuntary, autonomic urinary bladder voiding (Figure [3](#F3){ref-type="fig"}). The finding that male *Htr3a* mutants exhibit decreased voiding efficiency is consistent with the results obtained in our VSA testing. The results of both of these experiments together suggest that 5-HT3A receptor signaling is required in males for maintaining normal urinary tract function. There are multiple aspects of LUT function that could be affected by the loss of 5-HT3A activity, including bladder detrusor contractility and its coordination with urethral sphincter relaxation (detrusor-sphincter dyssynergia). Future studies combining the use of cystometry with electromyography recordings of the external urethral sphincter would be valuable for elucidating these mechanisms in *Htr3a* mutants. A prior study utilizing the *Htr3a*^*vs*/*vs*^ mouse model, which expresses a constitutively active hypersensitive form of the 5-HT3A receptor, also reported poor urinary bladder function in adult males (Bhattacharya et al., [@B1]) that were accompanied by bladder wall hyperplasia. However, in contrast to the deficits reported in the *Htr3a*^*vs*^ hypersensitive mutants, we did not find any evidence of bladder outlet obstruction or other overt LUT malformations in *Htr3a*KO mice. Our findings in concert with those of Bhattacharya et al. indicate that the control of micturition in mice is acutely sensitive to levels of 5-HT3A receptor activity and highlight the importance of interrogating gene function from multiple angles. Both the *Htr3a*^−/−^ and the *Htr3a*^*vs*/*vs*^ mutants are constitutive alleles, with the effect of the receptor variants exerted through the entire window of endogenous gene expression. Given that *Htr3a* is expressed during LUT development and into adulthood, our studies raise the intriguing question of whether 5-HT3A receptor activity is required at a specific time during the development of bladder innervation. Use of an inducible *Htr3a* mutant allele in conjunction with assays of urinary tract function will be needed to determine the developmental time point when 5-HT3A receptor activity is most critical for establishing normal voiding patterns in adults. The pelvic ganglia in mice are unique among autonomic structures in that they are comprised of both sympathetic and parasympathetic neurons (Keast, [@B13]; Wanigasekara et al., [@B40]). Given the widespread and sustained expression of 5-HT3A in these structures, we sought to determine if *Htr3a* knockout would affect expression of autonomic neuronal subtype markers. Interestingly, we observed expression of TH is visibly up-regulated in *Htr3a* homozygous mutant fetal pelvic ganglia compared to wild type and heterozygous littermates, with a concurrent down-regulation of cholinergic marker ChAT. However, by adulthood the discernable differences in TH+ and ChAT+ cells seen at 14.5dpc were no longer apparent (Figures [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"}). Prior work has established that TH and ChAT levels in pelvic ganglia neurons can be profoundly affected by circulating sex hormones (Melvin and Hamill, [@B21]; Keast, [@B14]); thus, it is possible that 5-HT3A receptor signaling is not required for post-pubescent TH and ChAT gene expression. Stage-specific effects of 5-HT3A knockout to discern the effects of this receptor on neurochemical profiles in the pelvic ganglia could be explored in the future with an inducible *Htr3a* mutant allele. The 5-HT3A receptor has been previously shown to be critical for the morphological maturation of neurons in the CNS (Chameau et al., [@B4]; Smit-Rigter et al., [@B32]; Oostland et al., [@B22]), but whether *Htr3a* expression is required for normal peripheral neuronal morphology has not been previously investigated. In the present study, we found that loss of *Htr3a* gene expression leads to an excess of neuronal fibers innervating the smooth muscle of the bladder, indicating the requirement of this receptor for the normal maturation of neural architecture in peripheral tissue. Regardless of *Htr3a* genotype, bladder smooth muscle intensity was reduced in adult tissues compared to 18.5dpc (with exception being CGRP+ fibers, Figures [6](#F6){ref-type="fig"}, [7](#F7){ref-type="fig"}). This is likely due to the pruning of neuronal branches that normally occurs in postnatal development (Purves and Lichtman, [@B25]; Riccomagno and Kolodkin, [@B27]). The exception of CGRP+ fibers suggests that there are neuronal subtype-specific differences in the time course of neuronal process outgrowth and maturation. It is also possible that expression of CGRP in neuronal fibers within the bladder changes over postnatal development. Our prior study examining gene expression profiles in developing dorsal root ganglia indicates that CGRP expression is somewhat dynamic through development (Ritter and Southard-Smith, [@B28]), but additional work is needed to explore these possibilities further. The observation that loss of one or both copies of the *Htr3a* gene leads to an increase in bladder smooth muscle innervation density of all four neuronal markers examined suggests that 5-HT3A may act to either inhibit neuronal fiber outgrowth, or may possibly be involved in the process of fiber pruning in postnatal development. Other research groups have reported similar effects of 5-HT3A loss; the apical dendrites of layer II/III cortical pyramidal neurons in the brain exhibit significantly more complex branching morphology in *Htr3a*^−/−^ mice compared to wild type littermates (Chameau et al., [@B4]; Smit-Rigter et al., [@B32]). However, other groups have reported contradictory results from *in vitro* studies of this receptor---in PC12 cells, 5-HT3A receptor activation promotes neurite outgrowth via increased calcium flux (Homma et al., [@B12]). Interestingly, the study by Bhattacharya et al. using the hyperactive *Htr3a*^*vs*^ variant described reduced nerve fiber density in bladder smooth muscle whereas our analysis looking at loss of *Htr3a* function identified increased innervation. Together, these findings suggest that the density of bladder innervation is dependent on the levels of 5-HT3A receptor activity. In light of these results, future studies are needed to identify the molecular factors downstream of 5-HT3A activation that govern neuronal architecture in peripheral neurons as a pathway to defining mechanisms that control bladder innervation. Our results collectively have important implications for human neural development. The changes in adult male voiding function that derive from loss of 5-HT3A in mice may occur as a result of deficient 5-HT signaling in the fetus, the adult, or both. While mouse studies cannot be directly translated to humans, the results of altered 5-HT3A receptor signaling that we describe suggest more thorough investigation focused on the outcomes of pharmacological intervention at this receptor during gestation is warranted. Ondansetron, a potent 5-HT3A receptor antagonist, is commonly prescribed to pregnant women to treat severe nausea (Mcparlin et al., [@B20]; Taylor et al., [@B34]). Additionally, the selective serotonin reuptake inhibitor (SSRI) fluoxetine, which is also prescribed frequently to pregnant women for the treatment of depression, has been shown by other groups to interact with 5-HT3A (Eisensamer et al., [@B7]; Smit-Rigter et al., [@B31]). Overall, the current study highlights the importance of 5-HT3A receptor signaling in adult urinary bladder function and demonstrates a novel role for this receptor in autonomic and sensory nervous system development. Author contributions {#s5} ==================== KER: designed research, performed research, analyzed data, and wrote the paper; ZW: performed research and analyzed data; CMV: designed research and analyzed data; DEB: designed research and analyzed data. EMS-S: designed research, analyzed data, and wrote the paper. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. This work was supported by grants (NIH U01 DK101038 and Pilot Award Support from the Vanderbilt Conte Center for Neuroscience Research P50 MH096972) to EMS-S, grant NIH U54 DK103410 to DEB, ZW, and CMV, and NIH F31-DK097938 to KER. Imaging performed in Vanderbilt University\'s Cell Imaging Shared Resource (CISR) is supported by NIH grants CA68485, DK20593, P50-DK58404, HD15052, DK59637, and EY08126. Two of the mouse strains used in this work, Tg(*Htr3a-*EGFP)DHGsat/Mmnc (stock number 000273-UNC) and Tg(*ChAT-*EGFP)CH293Gsat/Mmucd were acquired from the Mutant Mouse Research Resource Center (MMRRC), a NCRR-NIH funded mouse line repository, and were donated to the MMRRC by the NINDS-funded GENSAT BAC transgenic project. We are grateful for the Hu C/D antibody generously donated by Dr. Vanda Lennon at the Mayo Clinic. Supplementary material {#s6} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fnins.2017.00690/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Joel C. Bornstein, University of Melbourne, Australia [^2]: Reviewed by: Russ Chess-Williams, Bond University, Australia; Margaret A. Vizzard, University of Vermont, United States [^3]: This article was submitted to Autonomic Neuroscience, a section of the journal Frontiers in Neuroscience

Background {#Sec1} ========== Malocclusion is a common worldwide dental problem, often causing aesthetic and functional concerns, which can lead to a negative impact on the quality of life \[[@CR1], [@CR2]\]. Malocclusion is normally corrected using orthodontic treatment which includes fixed and/or removable braces. There are a number of fixed appliance systems used in contemporary orthodontics which include multiple pre-adjusted edgewise fixed appliance systems. Edward H Angle is regarded as the father of modern orthodontics and introduced the 'edgewise' bracket into orthodontic treatment in 1925 as a development of the 'ribbon arch' appliance. The interaction between the edgewise bracket slot and the orthodontic archwire can determine the nature of the orthodontic forces. In order that the gold archwires were sufficiently rigid, Angle proposed the bracket slot dimensions be 0.022 by 0.028 inches. The advent of stainless steel alloys facilitated the use of smaller dimension wires with the same rigidity as that of the larger gold alloy archwires. This allowed the bracket slot size to be reduced, and as a result the 0.018-inch bracket slot was introduced into orthodontics. However, the introduction of the 0.018-inch bracket slot did not eliminate fixed appliance systems using 0.022-inch bracket slots from clinical practice \[[@CR3]\]. In the last 20 years, a number of other alloy archwires have been developed. These include standard nickel titanium, thermally activated nickel titanium, superelastic nickel titanium and titanium-molybdenum archwires. Despite these advances in metallurgy, no high quality studies have sought to investigate the effectiveness of 0.018-inch compared with 0.022-inch bracket slots in orthodontic treatment. Only two retrospective studies have compared the duration of treatment for patients treated with 0.018-inch slot and 0.022-inch slot fixed appliance systems \[[@CR4], [@CR5]\]. It is interesting to note that these two studies reported a statistically significant reduction in the mean orthodontic treatment duration for the 0.018-inch slot system; however, this difference was not considered by the authors to be clinically significant. A number of biomechanical advantages and disadvantages have been suggested for both 0.018-inch and 0.022-inch bracket slots. It has been postulated that overbite reduction (a key stage in fixed appliance orthodontic treatment) and closure of any residual extraction space may be more efficient with 0.022-inch bracket slots as the space between the working archwire (0.019 by 0.025-inch stainless steel) and the 0.022-inch bracket slot (known as slop) allows exaggerated bite-opening bends to be placed whilst still being able to be fitted with relative ease \[[@CR6]\]. Conversely, the working archwire (0.016 by 0.022-inch stainless steel) for 0.018-inch bracket slots is presumed to deliver third order movement (known as torque) more effectively and earlier in treatment without additional wire bending \[[@CR6]\]. All fixed appliance orthodontic treatment results in adverse effects on the roots of the teeth, the most common being orthodontically induced inflammatory root resorption (OIIRR). Excessive forces during treatment have been associated with iatrogenic damage including OIIRR, periodontal destruction, loss of crestal bone height and loss of pulp vitality. It is not known if these are actually significantly reduced when 0.018-inch bracket slots are used, although different appliances are associated with differing levels of biological trauma during treatment \[[@CR7]\]. A retrospective study by Sameshima and Sinclair \[[@CR8]\] which investigated multiple treatment factors for the prediction and prevention of OIIRR reported that difference in bracket slot size was not significantly associated with OIIRR. The preference for each slot size varies throughout the world. In the United Kingdom, the majority of orthodontists continue to use 0.022-inch bracket slots. However, in the rest of mainland Europe, 0.018-inch bracket slots are more commonly used, whilst in the United States, the majority of orthodontists work with 0.022-inch bracket slots. It has been estimated that approximately 10% of Americans relocate to another state each year and this is likely to also be true for Europe and the United Kingdom \[[@CR9]\]. When patients undergoing orthodontic treatment transfer to a different orthodontist, the dichotomy in orthodontic bracket sizes is not only inconvenient for the new orthodontist if appropriate inventory is not kept in stock, but unnecessary potential side-effects can occur if active treatment is prolonged. These biological and clinical management problems could all be avoided or minimized if one slot size was shown to be superior and subsequently adopted by the profession as the international standard \[[@CR3], [@CR9]\]. The aim of this trial is to provide reliable evidence for clinicians as to whether the use of 0.018-inch or 0.022-inch bracket slot systems has any impact on the effectiveness of orthodontic treatment. We will do this by determining if the use of 0.018-inch and 0.022-inch bracket slot systems has any effect on the duration or quality of result of orthodontic treatment, patient's perception of treatment or the biological side effects associated with orthodontic treatment. We hypothesize that there is no difference in the outcomes of orthodontic treatment (as listed above) using either a 0.018-inch or 0.022-inch slot fixed appliance system. Methods/Design {#Sec2} ============== Trial design {#Sec3} ------------ The current trial is a multicenter, two-arm, parallel group, randomized controlled trial (Figure [1](#Fig1){ref-type="fig"}) designed following the CONSORT guidelines \[[@CR10]\]. Patients who give informed consent to participate in this study will be entered into the study database, which will be held on one of the computers within the Orthodontic Department at Dundee Dental Hospital and School. Each participating patient will be allocated the next available study number at the start of active treatment using a series of opaque envelopes containing the treatment group.Figure 1**Trial CONSORT flow chart with projected numbers of participants throughout the trial.** Intervention 1: 0.018-inch bracket slot system and intervention 2: 0.022-inch bracket slot system. This trial will recruit orthodontic patients from three sites in secondary care setting in NHS Tayside, Scotland: Dundee Dental Hospital and School, Perth Royal Infirmary and Springfield Medical Centre (Arbroath). Prior to study recruitment at any individual site, there will be a site initiation visit and site staff will receive training on the study design, methodology, clinical intervention and maintaining trial documentation, including the case report forms questionnaires and investigator site file. Participant inclusion criteria {#Sec4} ------------------------------ Patients scheduled to undergo dual arch fixed appliance orthodontic treatment in any of the three trial centers will be invited to participate in this trial by the operator planning to conduct their orthodontic treatment. Patients who are eligible to receive treatment in the NHS Scotland should have a significant need of treatment which requires an Index of Orthodontic Treatment Need (IOTN) score of 4 or 5 for the dental health component, or a score of 3 combined with a high aesthetic component score (\>5) \[[@CR11]\]. All types of malocclusions (Class I, II and III malocclusions) will be recruited to the study including cases which require extraction or non-extraction as part of the treatment plan. Participant exclusion criteria {#Sec5} ------------------------------ Participants will be excluded if they fulfil any of the following criteria: patients who have undergone previous orthodontic treatment including fixed, removable and functional appliances; patients less than 12-years-old at the start of orthodontic treatment; patients with orofacial clefting, severe hypodontia and special needs patients; and patients undergoing orthognathic (jaw) surgery as part of their orthodontic treatment plan. Intervention protocol {#Sec6} --------------------- All teeth will undergo pumice and water prophylaxis immediately before bonding and banding. The teeth will then be prepared using self-etching primer (Transbond™ Plus Self Etching Primer, 3 M Unitek, Monrovia, United States). Depending on the card inside the participant's envelope, either 0.018-inch or 0.022-inch slot Victory Series™ McLaughlin, Bennett, Terevisi (MBT) prescription adhesive pre-coated (APC) brackets and buccal tubes (or bands, where appropriate) (APC™ II Victory Series Twin MBT™, 3 M Unitek, Monrovia, United States\] will be placed. A standardized wire sequence for each bracket slot system will be followed whenever clinically appropriate. The 0.018-inch bracket slot system archwire sequence will be: 0.016-inch nickel titanium alloy archwire, followed by 0.016 × 0.022-inch nickel titanium alloy archwire followed by 0.016 × 0.022-inch stainless steel archwire. The 0.022-inch bracket slot system sequence will be: 0.016-inch nickel titanium alloy archwire, followed by 0.019 × 0.025-nickel titanium alloy archwire followed by 0.019 × 0.025-inch stainless steel archwire. A flow chart for the study (Figure [2](#Fig2){ref-type="fig"}) showing the key steps will be attached to the participant's clinical case notes to help remind clinicians to adhere to study protocol. A standardized clinical treatment regime will be used throughout the trial, accepting that clinical circumstances may necessitate deviations from the standard protocol.Figure 2**Trial key steps diagram.** Showing the pre-treatment, mid-treatment and end of treatment records collected. Smile better questionnaire is completed by the study participants with regards to their experience wearing the fixed appliances. Index of Orthodontic Treatment Need Aesthetic Component (IOTN AC), Orthopantomograph (OPT). Outcome measures {#Sec7} ---------------- At the end of the trial, the standardized records will be anonymized before any scoring will be undertaken. Thus the investigator will be blind to the allocation before any observations are recorded in order to eliminate investigator bias. ### Primary outcome {#Sec8} The primary outcome will be the duration of orthodontic treatment (measured in months) required to finish orthodontic treatment in both the maxillary and mandibular arches, including the number of visits required to complete treatment. ### Secondary outcomes {#Sec9} The first secondary outcome will be the treatment outcome. This will be measured by evaluating the quality of treatment, the effectiveness of the appliance in delivering torque and the improvement in dental attractiveness. The quality of the treatment result will principally be measured by assessing the occlusal outcome using the Peer Assessment Rating Index (PAR Index) from study models recorded at the start and end of treatment. The torque of the maxillary incisors will be used to evaluate the effectiveness of appliance in delivering effective torque by measuring the angle between the long axis of the central incisors and the maxillary base from lateral cephalograms towards the end of treatment. The improvement in the dental attractiveness of the patient will be calculated using the aesthetic component of the IOTN. This will be evaluated by comparing the patient rating to the attractiveness of his or her teeth before and after treatment. The second secondary outcome will be the biological side effects of the treatment, as measured by the amount of OIIRR. OIIRR will be established from periapical radiographs recorded at nine months into active treatment. Patients who show evidence of marked OIIRR will be treated in accordance with current recommendations. The third secondary outcome will be the patient's perception of wearing the fixed orthodontic appliance and of treatment outcome. This will be recorded using a questionnaire (Smiles Better) at six months after the start of treatment. Patient perception of the of fixed appliance treatment outcome will be measured by comparing patients' pre-treatment and after-treatment questionnaires. Target sample size {#Sec10} ------------------ The sample size calculation is based on the primary outcome of duration of orthodontic treatment. Using nQuery Advisor 7.0 sample size was calculated to detect a difference in mean treatment duration of three months (clinically important difference). Standard deviation was estimated using the studies of Amditis and Smith \[[@CR5]\] and Eberting et al. \[[@CR12]\]. A sample size of 92 in each group will have 80% power to detect a difference in means of three months assuming that the common standard deviation is 7.2 using a two group Student's *t*-test with a 0.05 two-sided significance level. Assuming a 15 to 20% dropout rate, we will recruit a total of 216 patients. Participant recruitment {#Sec11} ----------------------- Patients attending the orthodontic department in any of the trial centers for orthodontic treatment who meet the inclusion criteria will be assessed by the clinician planning the orthodontic treatment. The patients will be invited to participate in the study and will be given a participant's and/or parents' information sheet to take home with them. Patients will be asked to decide about joining the trial on the next appointment (after at least two weeks). Sequence generation and randomization {#Sec12} ------------------------------------- Simple randomization will be accomplished with no stratification using a restricted (10 number block) random number using <http://www.graphpad.com/quickcalcs/randomn2.cfm> to ensure equivalence of numbers in each group. The odd numbers will be allocated to group 1 and the even numbers to group 2. In every 10 number block from the random table, the sequence will be checked to ensure the even numbers are equal to the odd numbers. Each number in the random table will be given a study number and assigned into one of the study groups. A table for the allocation of the participants in the study will be composed and kept in a sealed envelope. All the documents used for the randomization and allocation sequence generation will be kept in a box in a locked office away from the clinical environments. Participant allocation {#Sec13} ---------------------- Numbered, identical, opaque sealed envelopes will be used for patient allocation in the current trial. The allocation envelope contains the treatment allocation card showing either group 1 or group 2. The allocation envelopes will be kept in a labelled box in an agreed location in the clinical environments. After the clinician obtains informed consent from the patient, a dental nurse will be asked to identify the next allocation envelope in sequence. The allocation will only be revealed at the time of appliance placement. Then the allocation envelope will be opened in front of the participant. Both the participant and the clinician will be informed about the group allocation for the participant. Blinding {#Sec14} -------- Due to the nature of the trial it is not possible to blind the clinician to the group once it is allocated. Once the opaque numbered envelope used for random allocation is opened and the clinician and participant know which appliance type (either 0.018-inch slot or 0.022-inch slot) will be used, the appliance will be clearly specified on the pro forma kept within the patient's case notes. This will allow the clinician to adhere to the recommended standard archwire sequence for each type of appliance. Data and records to be collected {#Sec15} -------------------------------- The following standardized data will be recorded for the trial participants at three stages of orthodontic treatment. Before start of orthodontic treatment: full orthodontic diagnostic clinical assessment, orthodontically trimmed study models at the start of orthodontic treatment, IOTN (aesthetic component score), long-cone paralleling technique periapical radiographs of maxillary central incisors, lateral cephalometric radiographs recorded before the start of treatment and patient questionnaire (pre-treatment). During orthodontic treatment: treatment duration, number of appointments, dates of key stages in fixed appliance orthodontic treatment, number of fixed appliance components replaced during active treatment, patient questionnaires (Smiles Better) regarding their perception of the impact of treatment (completed by the patient at six months from the start of orthodontic treatment), long-cone paralleling technique periapical radiographs of maxillary incisors recorded at nine months after the start of orthodontic treatment using digital radiography and lateral cephalometric radiograph recorded before the completion of active treatment. End of orthodontic treatment: orthodontically trimmed study models at the end of treatment, patient questionnaire (post-treatment) and IOTN (aesthetic component score). Data management {#Sec16} --------------- A data collection sheet designed specifically for the current trial will be used to collect data from all participants' records including study models, patient questionnaires, case notes and radiographs. The data collection sheet will be coded to allow for the blinding of the investigator to the type of appliance used during data analysis. The data will then be uploaded into a validated excel sheet specially designed for the current trial complying with Good Clinical Practice regulations. At the end of the trial the data will be exported into SPSS (Statistical Package for Social Sciences, Inc., Chicago, Illinois, United States) v.18 for statistical analysis. Post-trial care {#Sec17} --------------- All study documentation will be retained for at least five years post final data lock. The end of study is defined as the last participant's last visit. The end of the study will be reported to the Research Ethics Committee (REC) within 90 days, or 15 days if the study is terminated prematurely. A summary report of the study will be provided to the REC within one year of the end of the study. Statistical analysis plan (SAP) {#Sec18} ------------------------------- ### Descriptive statistics {#Sec19} Descriptive statistics for demographic data (such as age and gender) and baseline measurements will be tabulated by treatment group. Data will be checked for normality of distribution and any evidence of skewness. ### Statistical analysis {#Sec20} Table [1](#Tab1){ref-type="table"} shows the planned statistical tests to be used for each variable investigated. The significance level will be set to α = 0.05 (two-tailed).Table 1**Statistical tests**Outcome measureType of variableStatistical test ***P***\<0.05**Primary outcome**Duration of alignment stage for the upper and lower archesContinuous, assuming normal distributionOne-way ANOVA test**Secondary outcomes**Number of visits requiredContinuous, assuming normal distributionOne-way ANOVA testOcclusal outcome PARContinuous, assuming normal distributionOne-way ANOVA testPatient perceptionCategorical dataChi-square testSeverity of OIIRRCategorical dataChi-square test or Friedman testA list of statistical tests that will be used for data analysis depending on the type of data in the current trial. Peer Assessment Rating index (PAR), orthodontically induced inflammatory root resorption (OIIRR), analysis of variance (ANOVA). Intention-to-treat and imputation {#Sec21} --------------------------------- Participants will be considered part of the intention-to-treat population if they were randomized at the start of the trial. If a participant has missing data for the outcome, then the median value for the treatment group will be imputed. To provide evidence of the robustness of the study results obtained from the above analyses, the same analyses will be repeated on two other data sets; one where imputation is done using the 75th percentile instead of the median, and another using the 25th percentile. These three sets of results will then be compared. Stopping rule {#Sec22} ------------- During the nine-month periapical radiographic assessment, if severe OIIRR (more than one third of the root Malmgren et al. \[[@CR13]\]) is found in the majority of patients in one group, whilst the other group shows minor changes, then a trial monitoring committee will be convened to consider whether the study will be terminated. Operator standardization {#Sec23} ------------------------ All patients will be treated in the Dundee Dental Hospital, Perth Royal Infirmary or Springfield Medical Center (Arbroath) by specialist orthodontic staff. Patients will be treated according to the study protocol. No additional appointments will be required due to participation in the study. Trial monitoring {#Sec24} ---------------- The Tayside Medical Science Centre standard operating procedure on adverse event recording will be adhered to for reporting of harms in the current study. Any participant reporting an adverse event would discuss this with the clinic staff or may contact the trial team directly. All adverse events will be recorded from the time participants have the orthodontic appliance bonded until the date of debonding (end of treatment with appliance removal). The trial is considered to be low risk and no adverse events are expected from the fixed orthodontic appliances that are used commonly and safely in most orthodontic units around the world. Trial Monitoring Committee {#Sec25} -------------------------- A Trial Monitoring Committee will be established, composed of three researchers (DB, GM and AE). This committee will meet every three months to monitor and discuss data for this trial. If it is found that one intervention particularly is causing significant harm to the dentition then the committee will have to decide to terminate the trial and shift all the study participants to the other intervention. Investigator responsibilities {#Sec26} ----------------------------- The principal investigators are responsible for the overall conduct of the study at their site and compliance with the protocol and any protocol amendments. Responsibilities may be delegated to an appropriate member of study site staff, which in this case is a senior clinical dental staff member who has been nominated as a co-investigator. Delegated tasks must be documented on a delegation log and signed by all those named on the list. Emergency code-breaking procedure {#Sec27} --------------------------------- Emergency code-breaking in this randomized control trial is not applicable because the bracket slot size system used for each participant will be clearly detailed within the patient's clinical notes. The proposed intervention is unlikely to result in significant adverse events. However, should they occur, they will be recorded in keeping with routine clinical practice. Ethics and dissemination {#Sec28} ------------------------ ### Ethics approval and consent {#Sec29} The study will be conducted in accordance with the principles of Good Clinical Practice. Approval was obtained from the East of Scotland Ethics Committee (approval number: 09/S1401/56) and National Health Service Tayside Research and Development (approval number: 20092E07). Any changes in research activity, except those necessary to remove an apparent, immediate hazard to the participant, will be reviewed and approved by one of the co-chief investigators. Amendments to the protocol will be submitted in writing to the East of Scotland Ethics Committee and Tayside National Health Service Research and Development. Informed consent will be taken by the treating clinician (member of the research team) after clarifying any queries asked by the patient and/or the parents. In all trial sites, one of the trial team who is suitably trained will obtain consent from the participant and/or parent(s) or legal guardian(s). All research team clinicians will have had previous Good Clinical Practice training including obtaining informed consent. An independent clinician who is not part of the trial research group agreed to be an independent reference for participants and parents in case there are any further queries regarding the study. His contact information is included in the participant information sheet. The participant and/or parent(s)/legal guardian(s) in the study will provide written informed consent before any study procedures are carried out and a participant information sheet will be provided to facilitate this process. Those with English as an additional language will be invited to bring an interpreter with them to the subsequent treatment appointment, or to request an NHS interpreter, where this service is available. As part of the consent process participants and/or parent(s)/legal guardian(s) must agree to researchers and regulatory representatives having access to their medical records for monitoring and audit purposes. Participants and/or parent(s)/legal guardian(s) may withdraw their consent to participate at any time during the study. The investigator or delegated member of the trial team and the child should sign and date the assent form whereas the participant (and/or parent(s)/legal guardian(s)) and investigator should sign the informed consent form(s) to confirm that consent has been obtained. The participant should then receive a copy of this document and a copy should be filed in the investigator site file. Patients who refuse to enter the trial will also be recorded. Their demographic and clinical details will be compared at the end of the study to determine that volunteer bias was not present. ### Withdrawal procedures {#Sec30} Parent(s)/legal guardian(s) will be informed that they have the right to withdraw from the study at any time. The right to refuse to participate without reasons will be respected. After the participant has entered the study the clinician remains free to provide alternative treatment to that specified in the protocol at any stage if he/she feels that it is in the participant's best interest, but the reasons for doing so will be recorded. In these cases the participants remain within the study for the purposes of follow-up and data analysis. All participants will be free to withdraw at any time from the study without giving reasons and without prejudicing further treatment. ### Confidentiality {#Sec31} All participants' records will be identified in a manner designed to maintain participant confidentiality. All records will be kept in a secure storage area with limited access. Clinical information will not be released without the written permission of the participant, except as necessary for monitoring and auditing by the sponsor, its designee, regulatory authorities or the REC. All investigators and study site staff involved with this study will comply with the requirements of the Data Protection Act (UK) 1998 with regard to the collection, storage, processing and disclosure of personal information and will uphold the Act's core principles. Computers used to gather the data will have limited access via user names and passwords. Published results will not contain any personal data that could allow the identification of participants. ### Dissemination of results and publication policy {#Sec32} The ownership of the data from the current trial resides with the trial team. On competition of the trial the data will be analyzed and a clinical study report will be prepared. In addition, the results of the trial will be presented at orthodontic national and international conferences and meetings. All patients recruited in the trial will be given a summary of the findings after the final report is finalized. Discussion {#Sec33} ========== Orthodontic treatment with fixed appliances usually takes around two years to complete. Therefore this randomized controlled trial presents potential difficulties with participant dropout due to patients moving to other treatment centers and clinicians failing to obtain treatment records at key stages during treatment. Although it is not possible to prevent the former, the latter source of bias will be minimized by regular trial updates and reminders throughout the duration of the trial. The results from the current study will serve as evidence to guide clinicians in deciding whether the difference in bracket slot size has a significant impact on the effectiveness of orthodontic treatment. Trial status {#Sec34} ============ The current trial is open for recruitment since January 2010 and is expected to reach the target enrolment of 216 participants in September 2014. **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** DB and GM conceived of the study and AE contributed to its design. AE, DB and GM were involved in writing the study protocol. All authors read and approved the final manuscript. We acknowledge Dr. Tatiana Macfarlane (Senior Research Fellow at University of Aberdeen) for her statistical advice. This trial is sponsored by the University of Dundee, Nethergate, Dundee, DD1 4HN, Scotland, UK.

The International Society for Biocuration (ISB) was created in 2009 specifically to promote biocuration, the product of multidisciplinary teams of database curators, software developers and bioinformaticians. Biocurators, whose work facilitates research and education across the life sciences, create and maintain a wide variety of online tools and databases, covering topics as diverse as biochemical structures, chromosomal features and phenotypes of mutant genes. These resources are now so essential to the biological community that databases have become integral to the daily work of most researchers. But biological databases are far from static: in addition to the need to constantly capture new knowledge (from the literature, from other databases, from analysis tools, etc.), data representation must also keep pace with current research---new data types must be modeled---and improved methods of data storage, representation and analysis must be continually developed. Nevertheless, such important efforts have not always been recognized, and have often not been published in full, owing to the lack of a suitable journal. DATABASE, *The Journal of Biological Databases and Curation*, was launched in 2010 to support the growing need of the research community to discuss a range of issues related to the creation, development and maintenance of biological databases, and to strengthen communication between database developers, curators and users. As this resonates strongly with the mission of the ISB, we are delighted to announce that DATABASE has now become the Society's official journal. The scope of DATABASE includes many areas relevant to the endeavors of the biocuration community. Moreover, DATABASE is an open-access journal, which is critical for biocuration worldwide and one of the core values that the ISB promotes. Since its creation, DATABASE has published more than 250 articles, 50 of which have appeared in the *Biocuration Virtual Issue*, a special collection of articles describing work presented at the annual International Biocuration Conference. Scholarly exchanges among scientists are invaluable for helping a discipline to realize its full potential. To this end, DATABASE and the ISB are excited to be able to work together more closely, a collaboration that we expect will enhance the visibility and impact of biocurators' work, and hence to increase the value of the Journal to members both of the ISB and of the wider scientific community. The Executive Committee of the International Society for Biocuration ==================================================================== Pascale Gaudet (Chair), Monica Munoz-Torres (Secretary), Marc Robinson-Rechavi (Treasurer), Teresa Attwood, Alex J. Michael Cherry, Renate Kania, Claire O\'Donovan and Chisato Yamasaki.

Introduction {#s1} ============ The endoplasmic reticulum (ER) mediates three-dimensional folding of newly synthesized proteins that are destined for membrane insertion or extracellular secretion ([@bib36]; [@bib15]; [@bib37]). Excess demand for protein assembly in the ER causes accumulation of unfolded proteins -- a condition known as ER stress. The unfolded protein response (UPR) is an intracellular sensing-signaling network that detects ER stress and orchestrates ER adaptation to reestablish cellular homeostasis. The UPR drives physical and biochemical expansion of the ER, while temporarily abating protein-translational load, and promoting disposal of unfolded proteins through ER-associated degradation (ERAD). In metazoan cells the UPR consists of three ER-transmembrane proteins: inositol-requiring enzyme 1 (IRE1), protein kinase R-like kinase (PERK), and activating transcription factor 6 (ATF6) ([@bib36]; [@bib15]; [@bib37]). These proteins directly or indirectly detect ER stress via their ER-lumenal and transmembrane domains; in response, they transduce signals to the cytosol and nucleus to promote cellular adaptation. The ER chaperone BiP/GRP78 plays an important role in keeping UPR activation in check ([@bib2]; [@bib18]). IRE1 possesses tandem cytoplasmic serine/threonine kinase and endoribonuclease (RNase) enzymatic modules, tethered to the ER membrane through an 80-amino-acid linker region ([Figure 1A](#fig1){ref-type="fig"}). Upon sensing ER stress, IRE1 forms homodimers, which perform trans-autophosphorylation of the kinase moiety, leading to RNase engagement. The RNase activates the transcription factor spliced X-box protein 1 (XBP1s) through non-conventional mRNA editing. In turn, XBP1s activates numerous genes that promote ER adaptation, ERAD, and cytoprotection ([@bib1]). The RNase module also helps reduce ER-associated translational load through regulated IRE1-dependent mRNA decay (RIDD) ([@bib16]), which also can contribute to cytoprotection by regulating specific, functionally relevant genes ([@bib24]; [@bib5]). IRE1α (sometimes called ERN1) is the most evolutionarily conserved of the UPR sensors, displaying structural and functional homology in yeast, flies, worms, fish and primates. PERK harbors a similar lumenal domain and a cytoplasmic serine/threonine kinase module, but lacks an RNase moiety. In response to ER stress, PERK phosphorylates eukaryotic translation-initiation factor α (eIF2α) ([@bib36]). This suppresses general translation, yet induces expression of activating transcription factor 4 (ATF4). In turn, ATF4 stimulates a number of genes that support ER adaptation, including one encoding the transcription factor C/EBP homologous protein (CHOP). The main function of the eukaryotic UPR is to adapt the ER to dynamic changes in demand for protein folding. However, the metazoan UPR performs an important additional function: it triggers apoptotic cell elimination in the event that ER-stress mitigation fails ([@bib32]). Teleologically, this mechanism probably evolved to limit the threat to the whole organism posed by potentially rogue cells with a severely damaged ER. {#fig1} Apoptosis is a built-in cell-suicide program that is shared by most metazoans ([@bib19]; [@bib9]; [@bib13]). The core intracellular apoptotic machinery entails a family of cysteine-dependent, aspartate-specific proteases, called caspases ([@bib35]; [@bib29]). Two interconnected signaling cascades control caspase activation in response to severe cellular stress or damage: the intrinsic (or mitochondrial) pathway and the extrinsic (or death-receptor) pathway ([@bib25]; [@bib3]; [@bib7]). These signals often converge on the proapoptotic BCL2 family protein BAX (and/or its relative BAK), which mediates mitochondrial outer membrane permeabilization (MOMP) through oligomerization and pore formation ([@bib38]). This releases cytochrome C from the mitochondrial intermembrane space into the cytosol. In the cytosol, cytochrome C helps nucleate the apoptosome complex, which sets off an enzymatic cascade involving the initiator protease caspase-9 and the executioner proteases caspase-3 and caspase-7 ([@bib31]). It is believed that these caspases bring about the cell's apoptotic demise through 'death by a thousand cuts" ([@bib25]; [@bib34]). The PERK branch of the UPR plays a key role in driving apoptosis in the context of irresolvable ER stress, through mechanisms that involve ATF4 and CHOP and converge on BAX ([@bib24]; [@bib32]; [@bib28]; [@bib11]). An additional proapoptotic signal involves the ER-resident BCL2 family protein BOK, which is kept at low levels by ERAD but accumulates when ERAD is diverted toward unfolded proteins ([@bib23]). In contrast to PERK, physiological levels of IRE1 suppress apoptosis activation, by mediating degradation of DR5 mRNA via RIDD during the early phase of the UPR ([@bib24]). If ER stress persists, PERK-dependent attenuation of IRE1 via the phosphatase RPAP2 inhibits RIDD, allowing DR5 levels to rise and thereby driving apoptosis ([@bib6]). While the UPR exerts tight regulation over apoptotic cell commitment in the face of ER stress, it remains poorly studied whether the caspase machinery exerts reverse controls over the UPR. In mouse liver cells, BAX and BAK support activation of IRE1 by direct interaction with its cytoplasmic region ([@bib14]). Conversely, the ER-resident antiapoptotic protein BAX inhibitor-1 inhibits IRE1 activation through cytoplasmic association ([@bib21]). In addition, there is preliminary evidence that caspases may cleave PERK ([@bib30]) and IRE1 ([@bib33]); however, this has not been further investigated at the molecular and cellular levels. In the present study, we uncover a novel, unexpected mechanism of cross-regulation between apoptotic caspases and the UPR, which regulates cell survival during ER stress. Results {#s2} ======= ER stress promotes caspase-mediated cleavage of IRE1 {#s2-1} ---------------------------------------------------- Human IRE1α (herein IRE1) is a 977 amino-acid protein ([Figure 1A](#fig1){ref-type="fig"}), composed of an N-terminal ER-lumenal domain (LD), a single-pass transmembrane (TM) domain, and a C-terminal cytoplasmic domain (CD). The CD segment consists of three parts: a membrane-proximal 'linker' region, followed by a serine/threonine kinase domain, and an endoribonuclease (RNase) moiety. In exploring the biochemical fate of the IRE1 protein during ER stress, we obtained western blot (WB) evidence suggesting that full-length IRE1 undergoes proteolytic processing in cell lines derived from cancer patients with a B cell malignancy called multiple myeloma (MM) (see below). To follow up on these initial findings, we immunized mice with a purified recombinant human IRE1 LD protein and isolated a mouse IgG2a monoclonal antibody that selectively recognizes a specific epitope within the LD (anti-IRE1α LD). We then used this antibody, alongside a commercially available rabbit polyclonal antibody that specifically detects IRE1's RNase domain (anti-IRE1α CD), to further characterize the apparent processing of IRE1. We first interrogated two MM cell lines, KMS11 and OPM2, in which ER stress can be induced with classical pharmacological agents, i.e., thapsigargin (Tg), tunicamycin (Tm), brefeldin A (BfeA), and dithiothreitol (DTT), or the pathophysiological bacterial toxin subtilase AB5 (SubAB) ([@bib27]). WB analysis with the anti-IRE1 LD antibody revealed not only the presence of the full-length \~105 kDa IRE1 protein, but also the emergence under ER stress of an additional \~55 kDa band ([Figure 1B and C](#fig1){ref-type="fig"}), indicating the formation of an N-terminal, LD-containing IRE1 fragment. Parallel analysis with the anti-IRE1α CD antibody also revealed the full-length IRE1 protein, as well as the formation in response to ER stress of one or more additional bands of \~50 kDa ([Figure 1B and C](#fig1){ref-type="fig"}), indicating the production of one or more C-terminal, CD-containing fragments. Some variation in gel mobility of the IRE1 species occurred; although further investigation is needed, this size variation may be due to differences in N-linked glycosylation ([@bib22]) or other post-translational modifications of IRE1 under different ER stress conditions. Unmitigated ER perturbation can lead to caspase-dependent apoptosis ([@bib36]; [@bib32]). We therefore reasoned that caspase activation might underlie the apparent proteolytic cleavage of IRE1 during ER stress. Treatment of KMS11 or OPM2 cells with the pan-caspase inhibitor zVAD blocked generation of both the LD- and CD-containing IRE1 fragments upon ER stress ([Figure 1D and E](#fig1){ref-type="fig"}), indicating that this cleavage of IRE1 requires caspase activity. We first attempted to identify the processing site(s) by N-terminal sequencing through mass spectrometry, using either purified recombinant IRE1 proteins of various lengths or immunoprecipitated endogenous IRE1 polypeptide from ER-stressed cells; however, this approach proved unsuccessful (data not shown). Instead, we turned to the CaspDB database (<http://caspdb.sanfordburnham.org/>) to examine the IRE1 polypeptide sequence for potential caspase recognition sites. Although a number of sites were predicted, in light of the estimated molecular mass of the observed fragments and the sequence conservation of IRE1 between diverse species, we postulated that a major cleavage site(s) resides within the linker region, after aspartic acid 507 (0.879 probability), or aspartic acid 512 (0.921 probability), or both ([Figure 1A](#fig1){ref-type="fig"}). To test this prediction, we replaced these two aspartic acids---either individually or simultaneously---with alanine residues by site-directed mutagenesis. We then transfected each mutant or a wild type (WT) IRE1 control construct into KMS11 cells harboring CRISPR/Cas9-based knockout of endogenous IRE1 and tested for caspase-mediated cleavage. Although alanine substitution of either aspartic acid 507 or 512 alone did not prevent IRE1 processing in response to ER stress ([Figure 1---figure supplement 1A](#fig1s1){ref-type="fig"}), replacement of both residues blocked most of the cleavage ([Figure 1F](#fig1){ref-type="fig"} and [Figure 1---figure supplement 1B](#fig1s1){ref-type="fig"}). Some residual bands were detectable by the anti-IRE1α LD antibody even with the double mutation, suggesting that cleavage can shift to alternative, less efficient sites. A band of \~85 kDa was further detected by the anti-IRE1α CD but not the anti-IRE1α LD antibody, independent of mutation ([Figure 1F](#fig1){ref-type="fig"}), suggesting the existence of another cleavage site that disrupts the LD epitope. However, given its location in the ER lumen, this site is unlikely to be targeted by caspases, which reside mainly in the cytoplasm ([@bib29]). An additional band that ran beyond the 49 kDa marker could be detected by the anti-IRE1α CD antibody ([Figure 1B-E](#fig1){ref-type="fig"}). However, this band was absent in cells expressing the double mutant ([Figure 1F](#fig1){ref-type="fig"}), suggesting that it may be a secondary product of cleavage at 507 and 512. Regardless, although different caspase-susceptible sites within IRE1 may be cleaved under diverse stress conditions, our data maps two prominent, adjacent cleavage sites to aspartic acids 507 and 512, within the cytoplasmic linker region of IRE1. We therefore focused our investigation on further elucidating this event and its cellular consequences. Executioner caspase activation in response to ER stress often involves the proapoptotic protein BAX ([@bib24]; [@bib28]). To verify whether the processing of IRE1 under ER stress requires BAX, we disrupted the BAX gene in KMS11 cells via CRISPR/Cas9 technology. We obtained two independent BAX knockout clones, which produced no BAX protein as expected; in contrast to the WT cells, these clones showed substantially less generation of cleaved (activated) caspase-3 and lacked any detectable IRE1 fragments in response to several ER stressors ([Figure 1G](#fig1){ref-type="fig"}). Thus, caspases operating downstream to BAX in the apoptotic signaling cascade induced by ER stress in KMS11 cells mediate the proteolytic processing of IRE1. Consistent with this conclusion, purified caspase-3 and caspase-7 performed concentration-dependent cleavage of a purified recombinant protein comprising the linker, kinase and RNase domains of IRE1 (IRE1 LKR), with slightly more efficient processing by caspase-3 ([Figure 1---figure supplement 1C](#fig1s1){ref-type="fig"}). To verify that the cleavage of IRE1 was not a peculiarity of KMS11 and OPM2 cells, we expanded our analysis to other cell types. We observed a similar pattern of caspase-mediated IRE1 processing in additional MM cell lines, as well as in several different types of lymphoma cells ([Figure 1---figure supplement 1D--F](#fig1s1){ref-type="fig"}). In contrast, a number of non-hematopoietic cancer cell lines examined, including NCI-H441 (non-small cell lung carcinoma), JHH-1 (hepatocellular carcinoma), and A2058 (melanoma), did not display appreciable IRE1 cleavage upon ER stress (data not shown). Aspartic acids 507 and 512 are conserved between human, mouse, rat and fruit-fly IRE1, while position 512 is occupied by glutamic acid in zebrafish IRE1 ([Figure 1A](#fig1){ref-type="fig"}). Consistent with the homology between the human and mouse orthologs, ER stress induced caspase-mediated cleavage of mouse IRE1 in two types of murine lymphoma cells ([Figure 1---figure supplement 1D and F](#fig1s1){ref-type="fig"}). Similar processing occurred also in mouse primary bone marrow-derived dendritic cells (BMDC), which already exhibited some IRE1 cleavage at baseline ([Figure 1H](#fig1){ref-type="fig"}). Thus, the results obtained so far suggest that ER-stress-induced caspase-mediated IRE1 processing occurs more readily in hematopoietic cells, including both malignant and non-malignant types. The major fragments of IRE1 differ in their cellular disposition {#s2-2} ---------------------------------------------------------------- The presence of prevalent caspase cleavage sites in the cytoplasmic linker region of IRE1 suggested that their hydrolysis splits the protein into an N-terminal fragment that contains both the LD and TM segments (LDTM) and a C-terminal cytoplasmic domain fragment(s) containing the kinase and RNase moieties (CD). Subcellular fractionation of OPM2 cells confirmed that the LDTM polypeptide was associated primarily with the membrane compartment, whereas the CD products were found mostly in the cytosolic fraction ([Figure 2A](#fig2){ref-type="fig"}). To investigate the cellular persistence of these products, we tracked their levels in cells undergoing Tg-induced ER stress over an 8 hr period. We performed these studies in the presence of cycloheximide (CHX) to inhibit further biosynthesis of IRE1 precursor. In both OPM2 and KMS11 cells, the LDTM fragment increased in abundance over 8 hr after Tg addition, while the CD fragment accumulated over the first 4 hr and then declined ([Figure 2B and C](#fig2){ref-type="fig"}, [Figure 2---figure supplement 1A and B](#fig2s1){ref-type="fig"}). Remarkably, the amount of full-length IRE1 protein substantially decreased over this period of ER stress, while caspase inhibition by zVAD blocked the drop, preserving most of the initial full-length IRE1 protein ([Figure 2B and C](#fig2){ref-type="fig"}). Thus, in the absence of de novo protein synthesis, caspases can cleave much of the available cellular pool of IRE1 in response to ER stress. {#fig2} Upon sensing of unfolded proteins by IRE1, the LD directs activation of the cytoplasmic kinase and RNase modules ([@bib20]). Cleavage within the linker region separates the lumenal and cytoplasmic moieties and therefore should disrupt IRE1 function. To test this prediction, we used zVAD, aiming to block IRE1 processing. OPM2 cells showed detectable XBP1s protein at baseline ([Figure 2---figure supplement 1C](#fig2s1){ref-type="fig"}), perhaps due to immunoglobulin overproduction by this malignant plasma cell line. Treatment with zVAD before Tg addition indeed stabilized the full-length IRE1 protein and augmented IRE1 phosphorylation. Although XBP1s protein levels declined over time, likely due to the addition of CHX, zVAD treatment enhanced XBP1s production ([Figure 2---figure supplement 1C](#fig2s1){ref-type="fig"}), consistent with the stabilization of IRE1. BMDC showed no detectable XBP1s at baseline; nonetheless, their treatment with zVAD before Tg addition also stabilized the full-length IRE1 protein and augmented XBP1s production under ER stress ([Figure 2---figure supplement 1D](#fig2s1){ref-type="fig"}). Thus, caspase-mediated IRE1 cleavage dampens IRE1 activation by ER stress, suggesting that the apoptotic cascade feeds back onto IRE1 to reduce its cytoprotective activity. LDTM attenuates apoptotic caspase activation {#s2-3} -------------------------------------------- The LDTM fragment of IRE1 persisted longer than the CD fragment. Therefore, we turned to investigate whether LDTM performs some specific cellular function. To examine this, we introduced (via stable transfection) a cDNA plasmid construct encoding a Flag-epitope-tagged LDTM protein, driven by a cytomegalovirus promoter, into the KMS11 cell line. (Because this experiment was performed before we identified the precise cleavage sites, we first used a construct encoding amino acids 1--470, which ends within the linker region upstream to the 507 and 512 positions.) Whereas the parental cells displayed bands that matched the endogenous full-length IRE1 and its Tg-induced LDTM product, the transfected cells showed an additional band corresponding to the (shorter) ectopic LDTM protein, independent of ER stress ([Figure 3A](#fig3){ref-type="fig"}). Surprisingly, ectopic LDTM expression improved viability and diminished caspase-3/7 activation in KMS11 cells as compared to untransfected counterparts under Tg-induced ER stress ([Figure 3A and B](#fig3){ref-type="fig"}), indicating apoptosis attenuation. Removal of the C-terminal Flag tag from the cDNA construct did not alter LDTM's ability to attenuate caspase-3/7 activation in response to Tg ([Figure 3---figure supplement 1A](#fig3s1){ref-type="fig"}). Moreover, similar to the pooled transfected KMS11 cells, single-cell-derived clones overexpressing LDTM also displayed diminished caspase-3/7 activation as compared to parental cells upon ER stress induction by Tg or SubAB ([Figure 3C](#fig3){ref-type="fig"} and [Figure 3---figure supplement 1B](#fig3s1){ref-type="fig"}). To test an LDTM construct that more faithfully reflects the precise caspase cleavage site, we generated a cDNA plasmid encoding amino acids 1--507. Reassuringly, expression of this LDTM version also reduced caspase-3/7 activation in response to diverse ER stressors, including SubAB, Tg, Tm, and BfeA, in KMS11 or JJN3 cells ([Figure 3---figure supplement 1C, D, F and G](#fig3s1){ref-type="fig"}). These results demonstrate that LDTM inhibits apoptotic caspase activation upon ER stress. By extension, caspase-mediated cleavage of IRE1 during ER stress produces a membrane-tethered, lumenal product of IRE1 that in turn feeds back onto the apoptotic signaling cascade to inhibit further caspase activation. Consistent with this localization, immunofluorescence analysis showed co-staining of the transfected LDTM fragment with the ER membrane marker Calnexin in MDA-MB-231 cells ([Figure 3---figure supplement 1E](#fig3s1){ref-type="fig"}). {#fig3} LDTM operates independently of the UPR to attenuate apoptosis {#s2-4} ------------------------------------------------------------- To investigate mechanistically how LDTM inhibits caspase activation, we first asked whether its antiapoptotic activity requires the presence of full-length IRE1. To test this, we engineered KMS11 cells expressing an inducible shRNA construct that targets the 3' UTR of the IRE1 mRNA. We subsequently transfected these cells with a plasmid encoding Flag-tagged LDTM cDNA designed to resist the IRE1 shRNA. As expected, the parental cells (which harbor the shRNA but not the LDTM construct) expressed autologous full-length IRE1 and generated a corresponding LDTM product upon treatment with Tg ([Figure 3D](#fig3){ref-type="fig"}). Moreover, these parental cells showed complete depletion of the full-length IRE1 protein upon treatment with Doxycycline (DOX) to induce the shRNA, which blocked Tg-induced production of the endogenous LDTM fragment. As further expected, LDTM-transfected cells expressed the ectopic LDTM protein across all conditions. These cells similarly displayed complete depletion of the endogenous full-length IRE1 and its LDTM product upon DOX treatment. Importantly, cells expressing ectopic LDTM showed improved viability and diminished caspase-3/7 activation under Tg- or Tm-induced ER stress, regardless of DOX treatment ([Figure 3D and E](#fig3){ref-type="fig"}, [Figure 3---figure supplement 1F and G](#fig3s1){ref-type="fig"}). This data demonstrates that apoptosis attenuation by LDTM is equally effective in the presence or absence of endogenous IRE1. Thus, LDTM suppresses apoptotic caspase activation independently of full-length IRE1. The UPR governs not only cellular adaptation but also apoptosis activation in response to protein misfolding. Accordingly, a conceivable mechanism that might mediate LDTM's antiapoptotic activity is that it binds to unfolded proteins in the ER lumen, in a manner similar to full-length IRE1 ([@bib10]), and sequesters them away from activating the ER-resident UPR sensors, thereby abating apoptotic signaling. However, consistent with its IRE1-independent function, ectopic LDTM expression did not alter IRE1 activity in response to ER stress, as measured by the levels of XBP1s mRNA ([Figure 4---figure supplement 1A](#fig4s1){ref-type="fig"}) or protein ([Figure 4---figure supplement 1B](#fig4s1){ref-type="fig"}), and by mRNA levels of the known RIDD target, DGAT2 ([Figure 4---figure supplement 1A](#fig4s1){ref-type="fig"}). In contrast to LDTM, ectopic expression of the CD fragment of IRE1 (LKR) did not affect caspase-3/7 activation by Tg or SubAB, nor did it impact LDTM's attenuation of caspase activation by these ER stressors ([Figure 4---figure supplement 1C and D](#fig4s1){ref-type="fig"}). Of note, LKR expression did not alter XBP1s levels ([Figure 4---figure supplement 1D](#fig4s1){ref-type="fig"}). Regardless, this data indicates that the LDTM product acts independently of the CD fragment to curb apoptosis signaling. Furthermore, LDTM expression did not inhibit ATF6 activation by ER stress, as gauged by levels of the key UPR chaperone BiP---a transcriptional target of ATF6 ([@bib36]) ([Figure 4---figure supplement 1B](#fig4s1){ref-type="fig"}); nor did it decrease PERK activation, as judged by the levels of the PERK-regulated transcription factors ATF4 and CHOP, both of which are implicated in driving apoptosis downstream to PERK ([@bib32]; [@bib11]). LDTM also provided cell protection against SubAB---which induces ER stress by proteolytically cleaving BiP ([@bib24]; [@bib27]) ([Figure 3C](#fig3){ref-type="fig"}, [Figure 3---figure supplement 1C and D](#fig3s1){ref-type="fig"}). Thus, LDTM functions independently of the key UPR sensors, IRE1, ATF6 and PERK, as well as the major ER chaperone, BiP, to attenuate apoptotic caspase activation during ER stress. LDTM inhibits key mitochondrial apoptotic events {#s2-5} ------------------------------------------------ To further define how LDTM may attenuate caspase activation, we asked whether it acts upstream or downstream to specific mitochondrial events known to be crucial for an irreversible apoptotic cell commitment ([@bib25]; [@bib7]). LDTM-transfected KMS11 clones displayed approximately 1.5-fold higher baseline levels of BAX as compared to parental controls ([Figure 4A](#fig4){ref-type="fig"}), perhaps reflecting a compensatory upregulation of BAX in response to the cytoprotection provided by LDTM. Nevertheless, ectopic LDTM expression attenuated the relative increase in mitochondrial BAX and the coordinated decrease in cytosolic BAX upon ER stress ([Figure 4A](#fig4){ref-type="fig"}). Consistent with its ability to decrease mitochondrial recruitment of BAX, LDTM expression also inhibited mitochondrial depolarization in cells undergoing ER stress in response to SubAB or Tg ([Figure 4B](#fig4){ref-type="fig"}). Furthermore, LDTM attenuated the uptake of calcium by mitochondria -- a characteristic apoptotic event which was similarly curtailed by BAX KO ([Figure 4C](#fig4){ref-type="fig"}). LDTM also attenuated the Tg-induced drop in mitochondrial cytochrome C levels and the corresponding gain in cytosolic cytochrome C ([Figure 4D](#fig4){ref-type="fig"}). Moreover, LDTM attenuated Tg-induced activation of caspase-9, known to be triggered in the cytosol through cytochrome C-induced assembly of the apoptosome, as well as the activation of caspase-3/7 downstream ([Figure 4E](#fig4){ref-type="fig"}). Taken together, these results indicate that LDTM attenuates apoptotic caspase activation at the level of, or upstream to, BAX recruitment to mitochondria, thereby blocking the consequent steps of the apoptotic cascade; i.e., MOMP, mitochondrial calcium uptake, release of cytochrome C into the cytosol, and activation of caspase-9 and caspase-3/7. Further supporting the conclusion that LDTM regulates apoptosis via BAX, the BCL2 small molecule inhibitor ABT-199, which prevents BAX blockade by BCL2 ([@bib4]), reversed LDTM's attenuation of mitochondrial calcium uptake and of caspase-3/7 activation in cells undergoing Tg-induced ER stress ([Figure 4---figure supplement 1E and F](#fig4s1){ref-type="fig"}). {#fig4} Ectopic LDTM expression enhances MM tumor progression {#s2-6} ----------------------------------------------------- To establish whether the antiapoptotic effect of LDTM leads to a biologically significant consequence, we examined LDTM's impact on the growth of MM cells in vitro and in vivo. Ectopic LDTM expression augmented proliferation of KMS11 cells restricted on matrigel, as measured by confluence analysis over several days using an Incucyte S3 instrument ([Figure 5A](#fig5){ref-type="fig"}). To test whether this growth augmentation could still occur in the context of an in vivo microenvironment, which is expected to be more stringent, we subcutaneously xenografted the cells into SCID mice. Importantly, ectopic LDTM expression significantly enhanced the growth of KMS11 tumor xenografts in vivo ([Figure 5B](#fig5){ref-type="fig"}). Furthermore, while IRE1 depletion by shRNA inhibited KMS11 cell growth both in vitro and in vivo, in keeping with the cytoprotective role of this UPR sensor ([@bib12]), expression of LDTM nevertheless accelerated growth even in the context of IRE1 knockdown ([Figure 5A and B](#fig5){ref-type="fig"}), consistent with LDTM's ability to improve cell viability independent of full-length IRE1. A single-cell-derived KMS11 clone expressing the same LDTM construct exhibited a similar enhancement in tumor growth ([Figure 5---figure supplement 1A](#fig5s1){ref-type="fig"}), as did a clone expressing the more precise 1--507 cleavage product ([Figure 5---figure supplement 1B](#fig5s1){ref-type="fig"}). Of note, KMS11 tumors constitutively produced an endogenous LDTM fragment, suggesting that cell stress in the tumor microenvironment drives caspase-mediated cleavage of IRE1. Taken together, these results suggest that the N-terminal product of caspase-mediated IRE1 cleavage -- modeled here by ectopic LDTM -- exerts sufficient biological impact on KMS11 cells to enhance their fitness to proliferate under growth-limiting conditions in vitro and in vivo. {#fig5} Discussion {#s3} ========== The primary function of the eukaryotic UPR is to help cells adapt to dynamic changes in demand for ER-mediated protein folding. The metazoan UPR has acquired an additional role, which may be equally as important: to eliminate cells that have sustained irreparable ER stress and as such pose a threat to the whole organism. The UPR performs this latter function by engaging the cell's apoptotic caspase machinery. However, this requires tight control, so that cells do not commit prematurely to an irreversible apoptotic fate. We know from earlier work that PERK acts through ATF4 and CHOP to drive apoptosis in response to excessive ER stress, that is, through Bim ([@bib28]), DR5, and/or DR4 ([@bib17]; [@bib8]), triggering apoptotic signals that converge on BAX. BOK further drives apoptosis when ERAD is diverted toward other ER client proteins ([@bib23]). In contrast to PERK, IRE1 opposes apoptotic signaling during the initial phase of the UPR by suppressing DR5 mRNA levels through RIDD ([@bib24]). If stress persists, PERK activity attenuates IRE1 by driving its dephosphorylation through the RPAP2 phosphatase, thereby releasing the brake on DR5 to promote apoptosis during the terminal UPR ([@bib6]). These mechanisms exemplify stringent UPR control over apoptosis activation. However, whether the cell's apoptotic caspase machinery feeds back onto the UPR to affect ultimate cell fate has been unclear. Our present studies uncover a novel mechanism of reciprocal cross-regulation between the UPR and the apoptotic caspase cascade during ER stress ([Figure 5D](#fig5){ref-type="fig"}). We show that caspase activation in response to diverse ER-stress stimuli leads to proteolytic processing of IRE1. Caspases operating downstream to BAX cleave IRE1 at two adjacent sites within the cytoplasmic linker region, thus dividing the protein into two major products. Given that activated apoptotic caspases often overlap in selectivity toward intracellular substrates, we did not attempt in this study to identify which specific BAX-driven caspases are involved in the cellular cleavage of IRE1. Nevertheless, both caspase-3 and caspase-7 were capable of cleaving a purified recombinant IRE1 LKR protein. In cells, the N-terminal product of IRE1, consisting of the ER-lumenal domain, transmembrane segment, and some residual linker sequence (LDTM), remains membrane-anchored and persists for at least 8 hr after ER stress exertion. In contrast, the C-terminal product(s), which contains most of the cytoplasmic region, displays a more transient nature. By uncoupling the sensing and signaling domains of IRE1, caspase-mediated processing within the linker dampens activity of this key UPR sensor in response to ER stress. Because the N-terminal fragment appeared more stable, we chose to focus on exploring its potential function. We found unexpectedly that upon overexpression, this product of IRE1 mediates significant negative feedback onto the apoptotic signaling cascade. In contrast, the more labile C-terminal fragment did not affect key apoptotic indicators, nor did it functionally influence the N-terminal fragment. LDTM consistently restricted caspase activation in the face of ER stress. Furthermore, it acted without requiring full-length IRE1 or altering the engagement of PERK and ATF6, and it abated apoptosis even in response to SubAB, which causes ER stress by cleaving the central ER chaperone BiP. Together, these results indicate that LDTM operates independently of the UPR to inhibit further caspase activation. Additional mechanistic studies demonstrated that LDTM inhibits the recruitment of BAX from cytosol to mitochondria, thereby attenuating a number of critical mitochondrial and cytosolic events required for cell commitment to apoptosis. Precisely how LDTM suppresses BAX recruitment at the molecular level remains an open question. We have begun to explore this through proteomics approaches, seeking to identify specific LDTM interaction partners that might mediate its regulatory effect on the apoptotic cascade. The analysis so far has excluded the possibility that LDTM directly interacts with BAX (data not shown), in contrast to the earlier reported interaction of full-length IRE1 (via its cytoplasmic region) with BAX ([@bib14]). Of note, the BCL2 inhibitor ABT-199 overcame the cytoprotective effect of LDTM, thus reinforcing the importance of BAX regulation by caspase-dependent IRE1 cleavage. The processing of IRE1 by caspases during ER stress seems paradoxically to have two opposite outcomes ([Figure 5D](#fig5){ref-type="fig"}): on the one hand it enhances apoptotic signaling by disrupting IRE1's activity, which is cytoprotective, while on the other hand it attenuates further caspase activation by restricting BAX translocation to mitochondria. A simple and plausible explanation for this apparent paradox is that the generation of LDTM through IRE1 cleavage provides a fail-safe checkpoint to ensure that cells with still remediable ER stress do not commit unnecessarily to apoptotic suicide. Only if ER stress exceeds a cell's capacity for mitigation does sufficient caspase activity develop to override this checkpoint and fully execute apoptotic elimination. The cytoplasmic location of the prominent cleavage sites within IRE1 is consistent with the fact that caspase activity resides primarily in the cytosol. Furthermore, the linker region of IRE1 may be more accessible to caspases than other, perhaps more buried, cytoplasmic domains. This could explain why caspase recognition sites within the linker are favored over four additional theoretical consensus sites found within the kinase and RNase domains of human IRE1. Of interest, earlier work characterizing ectopically expressed IRE1 in COS7 cells showed that presenilin-1 cleaves IRE1 within the linker ([@bib26]), which suggests that this region of IRE1 can be targeted also by other types of proteases. IRE1 processing by apoptotic caspases occurred primarily in hematopoietic cell types, of both malignant and non-malignant origin. Because hematopoietic cells tend to express relatively high levels of IRE1 mRNA, it is possible that caspase-mediated processing requires abundant IRE1 protein: however, this warrants further investigation. Remarkably, in the absence of de novo IRE1 biosynthesis, most of the available cellular pool of IRE1 protein was subject to caspase-dependent cleavage in response to ER stress. This suggests that substantial amounts of LDTM can be generated within the cell, with significant consequences. Indeed, our biological studies showed that LDTM overexpression enhances restricted MM cell growth in vitro, as well as tumor progression in vivo. Thus, the production of LDTM upon caspase-mediated cleavage of IRE1 can significantly improve the fitness of cancer cells to survive and grow in stressful microenvironments. This conclusion has important potential implications for the role of IRE1 in hematopoietic malignancies such as MM, because it suggests that IRE1 promotes tumor growth not only in its general role as a cytoprotective UPR mediator ([@bib12]) but also in particular through its caspase-driven antiapoptotic LDTM product. Our studies open the door to further investigating caspase-mediated processing of IRE1 and other UPR sensors in response to proteotoxic as well as other types of cell stress in settings of health and disease. Materials and methods {#s4} ===================== ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Reagent type\ Designation Source or\ Identifiers Additional\ (species) or\ reference information resource ----------------------------------- ---------------------------------------------------------- ------------------------------------- ------------------------------------------------------------ ---------------------------------------------------------------------- Strain, strain background (mouse) CB-17 SCID Charles River Laboratories Strain, strain background (mouse) C57BL/6 Charles River Laboratories Cell line (human) KMS11 Genentech RRID:[CVCL_2989](https://scicrunch.org/resolver/CVCL_2989) Multiple myeloma Cell line (human) OPM2 Genentech RRID:[CVCL_1625](https://scicrunch.org/resolver/CVCL_1625) Multiple myeloma Cell line (human) JJN3 Genentech RRID:[CVCL_2078](https://scicrunch.org/resolver/CVCL_2078) Multiple myeloma Cell line (human) RPMI-8226 Genentech RRID:[CVCL_0014](https://scicrunch.org/resolver/CVCL_0014) Multiple myeloma Cell line (human) LP-1 Genentech RRID:[CVCL_0012](https://scicrunch.org/resolver/CVCL_0012) Multiple myeloma Cell line (human) U266B1 Genentech RRID:[CVCL_0566](https://scicrunch.org/resolver/CVCL_0566) Multiple myeloma Cell line (human) Ramos Genentech RRID:[CVCL_0597](https://scicrunch.org/resolver/CVCL_0597) Burkitt's lymphoma Cell line (human) Raji Genentech RRID:[CVCL_0511](https://scicrunch.org/resolver/CVCL_0511) Burkitt's lymphoma Cell line (human) BJAB Genentech RRID:[CVCL_5711](https://scicrunch.org/resolver/CVCL_5711) Burkitt's lymphoma Cell line (human) Daudi Genentech RRID:[CVCL_0008](https://scicrunch.org/resolver/CVCL_0008) Burkitt's lymphoma Cell line (human) Maver-1 Genentech RRID:[CVCL_1831](https://scicrunch.org/resolver/CVCL_1831) Mantle Cell Lymphoma Cell line (human) Jurkat Genentech RRID:[CVCL_0367](https://scicrunch.org/resolver/CVCL_0367) Acute T-Cell Leukemia Cell line (human) MDA-MB-231 Genentech RRID:[CVCL_0062](https://scicrunch.org/resolver/CVCL_0062) Triple negative breast cancer Cell line (mouse) BW5147.3 Genentech RRID:[CVCL_4135](https://scicrunch.org/resolver/CVCL_4135) Thymic Lymphoma Cell line (mouse) ABE8.1/2 Genentech RRID:[CVCL_3487](https://scicrunch.org/resolver/CVCL_3487) Pre-B Cell Lymphoma Cell line (human) BAX knockout Clone one and Clone 2 This paper Isolated clones of CRISPR/Cas9 deletion of BAX gene in KMS11 Cell line (human) IRE1 knockout KMS11 Genentech [@bib12] Biological sample (mouse) BMDC This paper Isolated from the tibia and femur bones of C57BL/6 mice Transfected construct (human) LDTM (1-470) This paper IRE1 aa1-470 with or without C-terminal Flag tag under CMV promoter Transfected construct (human) 1-507/1-507F This paper IRE1 aa1-507 with or without C-terminal Flag tag under CMV promoter Transfected construct (human) IRE1 shRNA Genentech [@bib12] Transfected construct (human) IRE1 wt This paper Full-length wild-type IRE1 expressed from CMV promoter Transfected construct (human) IRE1 D507A This paper Asp at 507 was mutated to Ala Transfected construct (human) IRE1 D512A This paper Asp at 512 was mutated to Ala Transfected construct (human) IRE1 D507A, D512A This paper Both Asp at 507 and 512 were mutated to Ala Transfected construct (human) IRE1 LKR This paper IRE1 aa468-977 with N-terminal His tag expressed from CMV promoter Transfected construct (human) BAX gRNA_1 This paper GCGGTGATGGACGGGTCCG Transfected construct (human) BAX gRNA_2 This paper TTCATGATCTGCTCAGAGC Antibody GAPDH-HRP Cell Signaling Technology Cat. \#: 2118 (1:5000) Antibody XBP1s (rabbit monoclonal) Genentech\ (1:1000) ([@bib6]) Antibody IRE1α LD (mouse monoclonal, IgG2a) This study (1:1000) Antibody IRE1α CD (rabbit monoclonal) Cell Signaling Technology Cat. \#: 3294 (1:1000) Antibody BAX (rabbit polyclonal) Cell Signaling Technology Cat. \#: 2772 (1:1000) Antibody Cytochrome-C (rabbit monoclonal) Cell Signaling Technology Cat. \#: 11940 (1:1000) Antibody COXIV (rabbit monoclonal) Cell Signaling Technology Cat. \#: 4850 (1:1000) Antibody ATF4 (rabbit monoclonal) Cell Signaling Technology Cat. \#: 11815 (1:1000) Antibody BiP (rabbit monoclonal) Cell Signaling Technology Cat. \#: 3177 (1:1000) Antibody Cleaved caspase-3 (rabbit monoclonal) Cell Signaling Technology Cat. \#: 9664 (1:1000) Antibody CHOP (mouse monoclonal IgG2a) Cell Signaling Technology Cat. \#: 2895 (1:1000) Antibody Flag (mouse monoclonal IgG1) Sigma-Aldrich Cat. \#: F1804 (1:1000) Antibody Calnexin (rabbit polyclonal) Abcam Cat. \#: ab22595 (1 μg/ml) Antibody Anti-rabbit IgG HRP Jackson ImmunoResearch Laboratories Cat. \#: 711-035-152 (1:10,000) Antibody Anti-mouse IgG2a HRP SouthernBiotech Cat. \#:1080--05 (1:10,000) Antibody pIRE1α (rabbit monoclonal) Genentech ([@bib6]) (1:500) Recombinant DNA reagent pRK.TK.Neo-IRE1 WT Genentech IRE1 wild-type cDNA in pRK.TK.Neo backbone Recombinant DNA reagent pRK.TK.Neo-IRE1 D507A Genentech IRE1 cDNA with Asp 507 mutated to Ala in pRK.TK.Neo backbone Recombinant DNA reagent pRK.TK.Neo-IRE1 D512A Genentech IRE1 cDNA with Asp 512 mutated to Ala in pRK.TK.Neo backbone Recombinant DNA reagent pRK.TK.Neo-IRE1 D507A, D512A Genentech IRE1 cDNA with Asp 507 and 512 mutated to Ala in pRK.TK.Neo backbone Recombinant DNA reagent pRK.TK.Neo-IRE1 1--470-Flag Genentech IRE1 cDNA aa1-470 with C-terminal Flag in pRK.TK.Neo backbone Recombinant DNA reagent pRK.TK.Neo-IRE1 1--470 Genentech IRE1 cDNA aa1-470 in pRK.TK.Neo backbone Recombinant DNA reagent pRK.TK.Neo-IRE1 1--507-Flag Genentech IRE1 cDNA aa1-507 with C-terminal Flag in pRK.TK.Neo backbone Recombinant DNA reagent pRK.TK.Neo-IRE1 1--507 Genentech IRE1 cDNA aa1-507 in pRK.TK.Neo backbone Recombinant DNA reagent pcDNA3.1.Zeo-IRE1 6xHis 468--977 Genentech IRE1 cDNA aa468-977 with N-terminal His tag in pcDNA3.1.Zeo backbone Commercial assay or kit Caspase-Glo 3/7 Assay Promega G8090 Commercial assay or kit Caspase-Glo 9 Assay Promega G8210 Commercial assay or kit CellTiter-Glo Luminescent Cell Viability Assay Promega G7570 Commercial assay or kit MitoProbe JC-1 Assay Kit for Flow Cytometry ThermoFisher Scientific M34152 Commercial assay or kit Subcellular Protein Fractionation Kit for Cultured Cells ThermoFisher Scientific 78840 Commercial assay or kit Mitochondria Isolation Kit for Cultured Cells ThermoFisher Scientific 89874 Recombinant protein LKR This paper Purified N-terminally His-tagged IRE1 aa468-977 Recombinant protein Caspase 3 Enzo Life Sciences ALX-201--059 Recombinant protein Caspase 7 BioVision 1087 Chemical compound, drug Rhod-2, AM, cell permeant Invitrogen R1244 Chemical compound, drug Thapsigargin, Tg Tocris 1138 Chemical compound, drug Tunicamycin, Tm Tocris 3516 Chemical compound, drug Brefeldin A, BfeA Tocris 1231 Chemical compound, drug DTT ThermoFisher Scientific R0861 Chemical compound, drug Cycloheximide, CHX Sigma-Aldrich C4859 Chemical compound, drug Z-VAD-FMK, zVAD R and D Systems FMK001 Chemical compound, drug Doxycycline, DOX Clontech NC0424034 Chemical compound, drug ABT-199 Genentech G00376771 Chemical compound, drug Subtilase toxin AB5, SubAB [@bib27] Software, algorithm Prism 7 GraphPad Software, algorithm ImageJ NIH Software, algorithm FlowJo 10.4 FlowJo, LLC ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Cell culture and experimental reagents {#s4-1} -------------------------------------- All cell lines were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum, 2 mM glutamine and penicillin plus streptomycin. Tm, Tg, DTT and BfeA were purchased from Sigma and Tocris. zVAD-FMK was from R and D Systems. Caspase-Glo 3/7, Caspase-Glo 9, and CellTiter-Glo assay kits were from Promega. Antibodies to the RNase domain of IRE1α, (anti-IRE1α CD), ATF4, GAPDH, BAX, cleaved Caspase 3, BiP, and CHOP were from Cell Signaling Technology. Antibodies to phospho-IRE1α, XBP1s and IRE1α LD were generated at Genentech. Doxycycline was from Clontech. Geneticin selective antibiotic was from GIBCO. Cell lines {#s4-2} ---------- A full list of cell lines used in this study are shown in the Key Resources Table. All cell lines were obtained or generated from an internal repository maintained at Genentech. ### Cell line authentication/quality control {#s4-2-1} Short Tandem Repeat (STR) Profiling: STR profiles are determined for each line using the Promega PowerPlex 16 System. This is performed once and compared to external STR profiles of cell lines (when available) to determine cell line ancestry. Loci analyzed: Detection of sixteen loci (fifteen STR loci and Amelogenin for gender identification), including D3S1358, TH01, D21S11, D18S51, Penta E, D5S818, D13S317, D7S820, D16S539, CSF1PO, Penta D, AMEL, vWA, D8S1179 and TPOX. ### SNP fingerprinting {#s4-2-2} SNP profiles are performed each time new stocks are expanded for cryopreservation. Cell line identity is verified by high-throughput SNP profiling using Fluidigm multiplexed assays. SNPs were selected based on minor allele frequency and presence on commercial genotyping platforms. SNP profiles are compared to SNP calls from available internal and external data (when available) to determine or confirm ancestry. In cases where data is unavailable or cell line ancestry is questionable, DNA or cell lines are re-purchased to perform profiling to confirm cell line ancestry. SNPs analyzed: rs11746396, rs16928965, rs2172614, rs10050093, rs10828176, rs16888998,rs16999576, rs1912640, rs2355988, rs3125842, rs10018359, rs10410468, rs10834627, rs11083145, rs11100847, rs11638893, rs12537, rs1956898, rs2069492, rs10740186, rs12486048, rs13032222, rs1635191, rs17174920, rs2590442, rs2714679, rs2928432, rs2999156, rs10461909, rs11180435, rs1784232, rs3783412, rs10885378, rs1726254, rs2391691, rs3739422, rs10108245, rs1425916, rs1325922, rs1709795, rs1934395, rs2280916, rs2563263, rs10755578, rs1529192, rs2927899, rs2848745, rs10977980. ### Mycoplasma testing {#s4-2-3} All stocks are tested for mycoplasma prior to and after cells are cryopreserved. Two methods are used to avoid false positive/negative results: Lonza Mycoalert and Stratagene Mycosensor. Cell growth rates and morphology are also monitored for any batch-to-batch changes. Transfection with cDNA {#s4-3} ---------------------- Cell lines were transfected with cDNA using Lipofectamine 2000 (ThermoFisher Scientific) according to manufacturer's protocol. Cells were treated and analyzed by WB after 48 hr or stably selected with either G418 or Zeocin for 2 weeks. CRISPR/Cas9 knockout of BAX {#s4-4} --------------------------- The gRNAs were cloned into pUC57_AIO_CMV_Cas9_T2_GFP, enabling co-expression of each sgRNA, Cas9, and an GFP-based selection marker following transient transfection into target cells. Transfection was with Lipofectamine 3000 (Invitrogen) according to manufacturer's protocol. At 24 hr after transfection, cells were washed once in PBS and resuspended in PBS media containing 3% BSA Fraction V. The cell suspension was then filtered through a 35 mm membrane followed by immediate FACS sorting using the GFP+ selection marker. Single cell clones (n = 96) were plated and grown. Clones producing colonies were tested for proper BAX disruption by immunoblot. Monoclonal antibody generation {#s4-5} ------------------------------ A recombinant protein encoding the lumenal domain (LD) of human IRE1α (amino acids 1--443) was generated via a baculovirus expression system in SF9 cells and purified to homogeneity using a TEV-protease cleavable His6 tag. Mice were immunized using standard protocols and monoclonal antibodies were screened by western blot against recombinant purified IRE1α lumenal and cytoplasmic domain proteins or lysates from MDA-MB-231 cells expressing wild type IRE1 or harboring CRISPR/CAS9 knockout of IRE1. A mouse IgG2a monoclonal antibody that specifically and selectively detected the human IRE1α LD (Lum017) was thus isolated and cloned. Western blot analysis {#s4-6} --------------------- Cells were lysed in a buffer containing 1% Triton X-100, 150 mM NaCl, 50 mM Tris-HCl supplemented with protease and phosphatase inhibitors (ThermoFisher Scientific). Samples were cleared, analyzed by SDS-PAGE, electro-transferred, and membrane was blocked with 5% dried nonfat milk powder in PBST, washed and then reacted with antibody and analyzed using ECL reagent (GE Healthcare or Invitrogen). Typically, equal protein amounts, as measured by a BCA assay (ThermoFisher Scientific), were loaded onto each lane of the SDS-PAGE gel. Caspase cleavage assay {#s4-7} ---------------------- Recombinant IRE1 LKR (aa 468--977) was expressed using Baculovirus in insect cells and was purified with nickel resin using the His-tag at the N-terminus of the protein. 75 ng of purified LKR was incubated with either 0.05 or 0.5 units of active recombinant caspase 3 (Enzo Life Sciences) or caspase 7 (BioVision) for 2 hr at 37°C. Reactions were TCA precipitated, resuspended in SDS loading buffer and then analyzed by western blot. Caspase activity assays {#s4-8} ----------------------- Cells were lysed as described for western blot analysis. Equal protein amounts, as measured by a BCA assay (ThermoFisher Scientific), were diluted in PBS to a total volume 100 μl. An equal volume of Caspase-Glo buffer was added and luminescence was measured after 1 hr. Isolation and differentiation of bone marrow-derived dendritic cells (BMDC) {#s4-9} --------------------------------------------------------------------------- The tibia and femur bones of C57BL/6 mice were thoroughly flushed internally with PBS to extract bone marrow cells. Cells were cultured in RPMI-1640 medium, containing 10% fetal bovine serum, 50 U/ml streptomycin/penicillin, 50 mg/ml L-glutamine, and 50 mM β2-mercaptoethanol (Sigma-Aldrich). The medium was supplemented with 20 ng/ml granulocyte--macrophage colony-stimulating factor (Biolegend, San Diego, CA) and 10 ng/ml interleukin-4 (Biolegend) for 9 days, with growth media being replenished on days 3 and 6 of culture, as described previously (Fernandez et al., Nature Medicine, 1999). BMDCs were verified to be ≥90% CD11c^+^ MHC class II^high^ by flow cytometric analysis. Immunofluorescence {#s4-10} ------------------ For immunostaining, cells cultured on Lab-TekII Chamber slides were washed three times in PBS, fixed for 20 min in 4% paraformaldehyde (EMS) at room temperature, washed, and permeabilized with 0.2% Triton X-100 for 10 min at room temperature. The slides were then blocked with 5% goat serum (Jackson ImmunoResearch) in 3% BSA/PBS for 30 min at room temperature. Flag (Sigma) or calnexin (Abcam) antibodies were diluted in 3% BSA/PBS and incubated with cells at 4°C overnight. After three washes with PBS, cells were incubated with secondary antibodies conjugated to Alexa Fluor 488 or Alexa Fluor 647 (Jackson ImmunoResearch) for 1 hr at room temperature. Slides were mounted with ProLong Gold Antifade Mountant with DAPI (Invitrogen) and viewed with Leica SP5 inverted confocal microscope using a 100X objective. Mitochondrial calcium assay {#s4-11} --------------------------- Rhod-2, am cell permeant dye (ThermoFisher Scientific) was prepared according to manufacturer's protocol. Cells were incubated with 10 μM dye for 45 min. After trypsinization, cells were washed once with PBS and analyzed by FACS in the FL-2 channel using FACSCalibur (BD Biosciences) flow cytometer. Mitochondrial outer membrane depolarization (MOMP) assay {#s4-12} -------------------------------------------------------- MOMP was analyzed by FACS using the MitoProbe JC-1 assay kit from Molecular Probes according to the manufacturer's protocol. Mitochondrial isolation {#s4-13} ----------------------- Mitochondrial isolation kit (ThermoFisher Scientific) was used to isolate mitochondrial fractions following the manufacturer's protocol. Subcutaneous xenograft tumor growth studies {#s4-14} ------------------------------------------- All procedures were approved by and conformed to the guidelines and principles set by the Institutional Animal Care and Use Committee (IACUC) of Genentech and were carried out in an Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC)-accredited facility. For tumor growth studies, 10 × 10^6^ KMS11 IRE1 sh8-9, LDTM pool, LDTM C13, 1--507 C5, BAX Knockout Clone 1 or Clone 2, respectively, were suspended in HBSS, admixed with 50% Matrigel (Corning) to a final volume of 100 μl, and injected subcutaneously in the right flank of 6 to 7 week old female CB-17 SCID mice. Tumor size and body weight were measured twice per week. Subcutaneous tumor volumes were measured in two dimensions (length and width) using Ultra Cal-IV calipers (model 54--10 − 111; Fred V. Fowler Co.). The tumor volume was calculated with the following formula: tumor size (mm^3^) = (longer measurement × shorter measurement^2^) × 0.5. For IRE1 knockdown studies, tumors were monitored until they reached a mean tumor volume of approximately 150 mm^3^, and then to induce knockdown of IRE1α randomized into the following treatment groups: (i) 5% sucrose water (provided in drinking water, changed weekly), or (ii) doxycycline (0.5 mg/ml, dissolved in 5% sucrose water, changed 3x/week). When mice reached endpoint criteria (see below) or after a 21 day treatment cycle, mice were euthanized by cervical dislocation and subcutaneous xenografts harvested for immunoblot analysis. Animals in all studies were humanely euthanized according to the following criteria: clinical signs of persistent distress or pain, significant body-weight loss (\>20%), tumor size exceeding 2500 mm^3^, or when tumors ulcerated. Maximum tumor size permitted by the IACUC is 3000 mm^3^ and in none of the experiments was this limit exceeded. Statistical analysis {#s4-15} -------------------- Graphs depict the mean ± SEM or SD of triplicates. Statistical analysis was performed by Student's *t* test in Prism. A single asterisk indicates p\<0.05 in comparison to the relevant paired value (e.g., in [Figure 2A](#fig2){ref-type="fig"}, p\<0.05 for the comparison of Tg-treated LDTM-transfected cells versus Tg-treated parental cells). Similarly, two asterisks indicate p\<0.01 and three asterisks indicate p\<0.001. We thank Ariel Chen for generating cell lines, members of the Antibody engineering department for mAb generation, Peter Liu and Wendy Sandoval for mass spectrometry assistance, and the Ashkenazi lab for discussions. Additional information {#s5} ====================== Anna Shemorry is affiliated with Genentech Inc. The author has no other competing interests to declare. Jonathan M Harnoss is affiliated with Genentech Inc. The author has no other competing interests to declare. Ofer Guttman is affiliated with Genentech Inc. The author has no other competing interests to declare. Scot A Marsters is affiliated with Genentech Inc. The author has no other competing interests to declare. László G Kőműves is affiliated with Genentech Inc. The author has no other competing interests to declare. David A Lawrence is affiliated with Genentech Inc. The author has no other competing interests to declare. Avi Ashkenazi is affiliated with Genentech Inc. The author has no other competing interests to declare. Conceptualization, Data curation, Formal analysis, Methodology, Writing---original draft, Writing---review and editing. Conceptualization, Data curation, Formal analysis, Methodology. Data curation, Methodology. Conceptualization, Data curation, Methodology. Data curation, Formal analysis, Methodology. Conceptualization. Conceptualization, Supervision, Methodology, Writing---original draft, Writing---review and editing. Animal experimentation: All procedures were approved by and conformed to the guidelines and principles set by the Institutional Animal Care and Use Committee (IACUC) of Genentech (protocol \#16-3257) and were carried out in an Association for the Assessment and Accreditation of Laboratory Animal Care (AAALAC)-accredited facility. Additional files {#s6} ================ 10.7554/eLife.47084.012 Data availability {#s7} ================= All data generated or analysed during this study are included in the manuscript and supporting files. 10.7554/eLife.47084.014 Decision letter Mukherjee Shaeri Reviewing Editor University of California, San Francisco United States In the interests of transparency, eLife includes the editorial decision letter and accompanying author responses. A lightly edited version of the letter sent to the authors after peer review is shown, indicating the most substantive concerns; minor comments are not usually included. Thank you for submitting your article \"Caspase-mediated cleavage of IRE1 controls apoptotic cell commitment during endoplasmic reticulum stress\" for consideration by *eLife*. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Vivek Malhotra as the Senior Editor. The reviewers have opted to remain anonymous. The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission. Summary: The reviewers find that the paper describes a very interesting study of the mechanism of how IRE1 negatively regulates apoptosis during UPR stress. The authors follow up the recent discoveries on the mechanisms by which cells die in response to persistent ER stress. The work presents a novel and exciting finding by which caspase-driven processing of the ER stress sensor IRE1 leads to production of two proteolytic fragments, one of which is shorter-lived and pro-apoptotic and the other of which is longer-lived and -- surprisingly -- anti-apoptotic. Overexpression of the pro-survival fragment protects stressed cells from death, both normal and tumor cells. The effect seems to be limited to cells of hematopoietic origin for reasons that are unclear. Most of the data are of excellent quality and the findings greatly increase our knowledge of the mechanisms for the decision of living versus dying in cells encountering ER stress. This manuscript is appropriate for publication in *eLife* provided the issues are addressed. Essential revisions: 1\) The actual caspase required for IRE1 cleavage has not been identified. Even though the authors used z-VAD as a broad spectrum caspase inhibitor, it may have significant off-target effects. Thus, it is essential that the authors provide direct evidence that IRE1 is cleaved by caspases. The authors should either knockdown or knockout Caspase 3/7 in their hematopoietic cell lines and show that it abolishes the generation of LTDM. It is also possible to complement these experiments in vitro with incubation of purified Caspase 3 or 7 with purified IRE1 and IRE1 mutants lacking the cleavage sites to show the generation of the LTDM. 2\) Provide mechanism on how LTDM may inhibit BAX from localizing to mitochondria. Sub-cellular localization analyses (not just cytosol vs. mitochondria) addressing how BAX pool is redistributed would be useful. 3\) Both fragments of IRE1 should be expressed in cells to see if the LTDM exerts its effect even in the presence of the pro-apoptotic fragment. 4\) Compare the effects of BAX-inactivation on KMS11 tumor growth with that of LTDM. In addition, please tone down on claims of LTDM as most of it is from over expression studies and the physiological relevance is yet to be established. 10.7554/eLife.47084.015 Author response > Essential revisions: *1) The actual caspase required for IRE1 cleavage has not been identified. Even though the authors used z-VAD as a broad spectrum caspase inhibitor, it may have significant off-target effects. Thus, it is essential that the authors provide direct evidence that IRE1 is cleaved by caspases. The authors should either knockdown or knockout Caspase 3/7 in their hematopoietic cell lines and show that it abolishes the generation of LTDM. It is also possible to complement these experiments* in vitro *with incubation of purified Caspase 3 or 7 with purified IRE1 and IRE1 mutants lacking the cleavage sites to show the generation of the LTDM.* This is a valid comment. The involvement of caspases in IRE1 cleavage is also corroborated by the effect of BAX knockout, which prevented both caspase activation and IRE1 cleavage (Figure 1G). We attempted to disrupt caspase-3 and -7 by RNAi but were unsuccessful due to poor knockdown efficiency. Nevertheless, to examine more directly whether caspases can cleave IRE1, we generated a purified, recombinant IRE1 protein comprising the linker, kinase and RNase domains (LKR). Incubation of this protein with commercially available purified caspase-3 or -7 revealed its direct cleavage by these proteases (new Figure 1---figure supplement 1C). This demonstrates a direct capacity of executioner caspases to cleave IRE1. > 2\) Provide mechanism on how LTDM may inhibit BAX from localizing to mitochondria. Sub-cellular localization analyses (not just cytosol vs. mitochondria) addressing how BAX pool is redistributed would be useful. We attempted immunofluorescence studies with anti-BAX and the mitochondrial marker TOM20; however, this proved difficult due to the small size and minimal ER-excluded cytoplasm in multiple myeloma cells. Instead, we now provide data showing that LDTM inhibits calcium uptake by mitochondria similar to BAX knockout (new Figure 4C). We further show that the BCL2 small molecule inhibitor ABT-199 overrides the effects of LDTM on mitochondrial calcium uptake and caspase-3/7 activation (new Figure 4---figure supplement 1E and 1F), confirming that the antiapoptotic function is exerted over BAX. > 3\) Both fragments of IRE1 should be expressed in cells to see if the LTDM exerts its effect even in the presence of the pro-apoptotic fragment. As noted above, it appears there is a misunderstanding that the cytoplasmic fragment is pro-apoptotic\--we did not propose this in the original manuscript. Regardless, we now performed the suggested experiment, which showed that ectopic expression of the cytoplasmic LKR fragment by itself did not alter caspase activation under ER stress; nor did this fragment affect the anti-apoptotic activity of the lumenal fragment upon co-transfection (new Figure 4---figure supplement 1C and 1D). > 4\) Compare the effects of BAX-inactivation on KMS11 tumor growth with that of LTDM. We have performed an additional in vivo study to examine the effect of BAX inactivation and observed that BAX knockout and ectopic LDTM expression in KMS11 cells similarly augment in vivo tumor growth (new Figure 5C). > In addition, please tone down on claims of LTDM as most of it is from over expression studies and the physiological relevance is yet to be established. Please note that the identification of LDTM as a stable product of caspase-mediated IRE1 cleavage is based primarily on the detection of [endogenous]{.ul} IRE1 and its cleavage products. Furthermore, the endogenous IRE1 fragment accumulates during ER stress while full-length IRE1 isstabilized by caspase inhibition. Nevertheless, to acknowledge the limitation that the reviewer alluded to we added the term "overexpression" to the Discussion text.

INTRODUCTION {#sec1-1} ============ The most important of the dental diseases, which requires prevention as first approach, is dental caries. Although the prevalence has markedly declined, it still poses a major challenge for children and adults. Caries is perceived to be a prolonged imbalance in the oral cavity with alternating episodes of demineralization and remineralization, occurring without any loss of tooth structure.\[[@ref1][@ref2]\] *Streptococcus mutans*, one of the most important organism in the plaque biofilm, possesses a unique array of putative cariogenic trait and is the most virulent of the caries producing organisms.\[[@ref3][@ref4]\] Longitudinal studies have shown a relative rise of *S. mutans* in plaque samples from tooth surfaces that become carious at a later stage.\[[@ref4][@ref5][@ref6]\] Microbial monitoring has been considered as an alternative method for evaluating current caries activity and future caries risk methods aiming at the estimation of the chief pathogen, i.e., *S. mutans*. The mainstay of caries prevention revolves around a number of vehicles which have proved themselves to be potent caries inhibiting factors, one of these being fluorides in the form of varnish, whose anticariogenecity has been consistently demonstrated.\[[@ref7]\] Fluoride works primarily via topical mechanisms by inhibition of demineralization, enhancement of remineralization at the crystal surfaces, and inhibition of bacterial enzymes.\[[@ref8]\] Although it is present in low concentrations, the fluoride accumulated in plaque decreases microbial acid production. With thorough knowledge on dynamics of caries, demineralization, and remineralization cycles, the quest of mankind for newer materials to overcome the carious challenge has led to the introduction of casein phosphopeptide--amorphous calcium phosphate (CPP--ACP).\[[@ref2]\] An amorphous form of calcium phosphate (ACP) stabilized by a phosphopeptide from the milk protein casein (CPP) may be used to localize ACP in dental plaque, maintaining a state of super-saturation with respect to tooth enamel, thus reducing demineralization.\[[@ref9]\] Incorporation of CPP into the salivary pellicle substantially reduces the adhesion of *S. mutans*, and this relative selective inhibition would eventually produce a non-cariogenic plaque.\[[@ref10]\] The present study was conducted to compare the efficacy of the fluoride varnish and CPP--ACP complex visa viz. *S. mutans* in plaque, and thereby the role these two agents could play in the prevention of dental caries. MATERIALS AND METHODS {#sec1-2} ===================== The present study was conducted in the Department of Pedodontics and Preventive Dentistry between July 2015 and September 2015 in association with the Department of Microbiology, Swargiya Dadasaheb Kalmegh Smruti Dental College and Hospital, Nagpur, with approval from the Ethical Committee, Swargiya Dadasaheb Kalmegh Smruti Dental College and Hospital, Nagpur, and consent from the parents of the children participating in the study. Sample selection {#sec2-1} ---------------- Of the 568 kindergarten children screened, 280 caries inactive children with complete set of primary dentition, aged 3--5 years, were randomly selected. The children\'s personal details and details of past medical and dental history including any recent antibiotic and fluoride exposure were obtained. The children were included in the present study based on the following criteria: Caries inactive primary dentitionNo history of antibiotics for the past 3--4 weeksNo physically and mentally challenged patients or children with any systemic disordersNo history of fluoride treatment for the past 2 weeksChildren with any intraoral appliance were not included. Among the 280 caries inactive children selected, 120 children, who after microbiological evaluation carried out by the first author (95% confidence limit, 7% confidence interval), satisfied the criteria of belonging to the moderate and high caries risk group proposed by Krassee in 1985 were chosen for the study,\[[@ref11]\] that is, Class 0: \<10000 CFU/ml (no risk) (CFU: colony forming unit)Class 1: \<100,000 CFU/ml (low risk)Class 2: 100,000--100,0000 CFU/ml (moderate risk)Class 3: \>100,0000 CFU/ml (high risk) The nature and objective of the clinical trial as well as the possible side effects were explained and informed consent was obtained from the participants. Study groups {#sec2-2} ------------ The children were randomly allocated to four groups, namely, fluoride varnish, CPP--ACP complex, and a mix of CPP--ACP complex and fluoride was painted on the primary teeth of the children belonging to Group I (*n* = 30), Group II (*n* = 30), and Group III (*n* = 30), respectively. A control group (*n* = 30) was considred to be group IV who were to follow their routine oral hygiene procedures. Materials {#sec2-3} --------- The materials used for this study were Fluor Protector Fluoride varnish (Ivoclar Vivadent, Germany), Tooth Mousse (CPP--ACP) (Global Care Asia Dental Pte Ltd), Tooth Mousse Plus (CPP--ACP + Fluoride) (Global Care Asia Dental Pte Ltd), and Mutans Sanguis Agar (Himedia). Plaque sample collection {#sec2-4} ------------------------ The sample was collected 1--2 hours after brushing to minimize any effect on the growth of the microorganism. The plaque samples were collected from the following four sites using separate toothpicks for the four sites for all the groups: Maxillary right molar -- buccal surfaceMaxillary incisor -- labial surfaceMandibular incisor -- labial surfaceMandibular left molar -- lingual surface. The four tooth picks were then placed in a single test tube with 1 ml of sterile saline to maintain the accuracy of the number of bacteria present from the time of collection until the processing procedure in the laboratory. Lottery method of sampling was implemented for allocating the participants into the control group and the study groups. Every participant had an equal chance of being in any of the groups. Groups I, II, III (test groups) and Group IV (control group) consisted of 30 participants each. The participants were blinded to the group to which they belonged. The plaque samples were obtained from the abovementioned four sites and stored in test tubes containing 1 ml saline. The test tubes were then labeled as per their lot number and placed in an upright position in a cool storage device using dry ice. Within 30 min of sample collection, the test tubes were transported to the laboratory. The samples were then agitated in a vortex mixture for 45 s in order to disperse bacterial aggregates and to facilitate extraction of bacteria from the tooth pick. The dispersed samples were spread over the culture media plates containing Mutans Sanguis Agar supplemented with Bacitracin (300 units) and 20% sucrose (mitis salivarius-bacitracin Agar), and the plates were incubated under aerobic conditions with a minimum incubation time of 24 h. On the same day, after obtaining the plaque samples, fluoride varnish, CPP--ACP complex, and a mix of CPP--ACP complex and fluoride was applied to the participants belonging to Groups I, II and III, respectively, by the examiner, and the control group was asked to follow the routine oral hygiene procedures. Application of fluoride varnish {#sec2-5} ------------------------------- Teeth were driedWith help of a small tufted brush, the varnish was first applied on the lower arch starting with proximal surfacesThe child was asked to keep the mouth open for 2--3 min before spitting to let the varnish set on the teethThe child was requested not to rinse or drink anything at all for 1 h and not to eat anything solid but take liquids and semisolids only till next the morning following application. Application of CPP--ACP complex Excess saliva was removedA smear of tooth mousse was applied on the tooth surface using a microbrush or interproximal brushIt was left undisturbed for 2 minThe child was then asked to use the tongue to spread the paste throughout the mouthThe child was encouraged to hold the paste in the mouth for 1--2 minutes, avoiding expectorationThe child was requested not to eat or drink for 30 min following application. Application of fluoride and CPP--ACP complex mix Excess saliva was removedA smear of tooth mousse plus was applied on the tooth surface using a microbrushIt was left undisturbed for 1--2 min, and the child was asked to distribute the paste throughly with the help of the tongueThe patient was instructed not to rinse or drink anything at all for 1 h and not to eat anything solid but take liquids and semisolids only till the next morning following application. The children belonging to group IV (control group) were asked to follow their routine oral hygiene procedures. Following a period of 24 h, the plaque samples were again obtained from the mentioned sites in the oral cavity of all the participants. These were incubated for the same time and the same methodology was used as before; the same examiner evaluated the results. After incubation for 24 h, the presence of *S. mutans* was confirmed by the typical colony morphology on the selective culture media and further by gram staining. The efficacy all the three study groups was assessed by recording *S. mutans* count prior and after the application, and comparing the results with the control group. The *S. mutans* count (CFU/ml) was recorded at 48 h in order to check the efficacy of the action of these materials over a short period of time. The results were evaluated according to the criteria given by Krassee in 1985.\[[@ref11]\] The results were interpreted by single examiner, who was blinded to the group division. The *S. mutans* growth was identified as raised colonies sideways against a light or with a magnifying glass. To differentiate *S. mutans* colonies and epithelial cells on the culture media, a gloved finger was passed along it to identify the smooth and rough colonies. Only the rough colonies were allotted as the growth of *S. mutans*. RESULTS {#sec1-3} ======= A considerable difference was documented in all the groups in the pre application and post-application scores of *S. mutans* (*P* \< 0.01) \[[Table 1](#T1){ref-type="table"}\] using Pearson\'s Chi-square test. The intergroup comparison using Wilcoxon-matched pairs signed ranks test proved CPP--ACP plus fluoride group to be the most proficient \[[Table 2](#T2){ref-type="table"}\]. ###### Cross tabulation of pre and post-application scores for each study group  ###### Comparison of mean scores between pre-application and post-application for each study group  DISCUSSION {#sec1-4} ========== The structure of enamel is unique, in that it has no residual cellular components that effect repair when enamel is damaged by a cariogenic episode.\[[@ref12]\] Demineralization and remineralization (repair or healing) of enamel are continual and constant processes occurring on the availability of cariogenic flora and refined carbohydrates.\[[@ref1][@ref12][@ref13]\] The plaque biofilm encloses numerous microenvironments that can be disrupted through chemomechanical systems\[[@ref13]\] such as applications of topical fluoride, CCP--ACP, and tooth brushing. The plaque biofilm facilitates the attachment and spread of *S. mutans*, the bacteria most implicated in caries. Plaque fluid, which forms the plaque system, can harbor more concentrations of fluoride, calcium, and phosphate complexes than any other fluid. Hence, in this study, the effect of fluoride varnish and CPP--ACP on *S. mutans* count in plaque of caries inactive individuals was analyzed. Vogel and Ekstrand\[[@ref14]\] found that there exist a marked variation between the plaque fluid fluoride concentrations at various sites of the oral cavity. Considering this fact, plaque samples were simultaneously obtained from different sites of oral cavity. In the present study, Krasse criteria was used as a tool for evaluating the pre and post-application microbial count because it is a simple, non-invasive and patient friendly technique.\[[@ref3][@ref4][@ref11]\] In the present study, brushing was advised with toothpaste containing fluoride prior to sample collection because it enhances the uptake of CPP--ACP.\[[@ref15]\] Topical fluorides significantly contributes in reducing the prevalence of caries, and it has been established that its anticaries efficacy against plaque micro-organisms and remineralization potential substantially increases when available in high concentrations in oral cavity locally.\[[@ref16][@ref17]\] A significant difference observed in the pre and post-application scores \[[Table 1](#T1){ref-type="table"}\] (*P* \< 0.01) of Group I, where the mean score before treatment being 2.4 ± 0.5 decreased to a mean score of 1.3 ± 1.0 after the application of fluoride varnish, emphasizing the efficacy of fluoride against Streptococcus. Hydroxyapatite crystals, when pre-treated with fluor protector, produced significantly reduced lactic acid formation by *S. mutans*.\[[@ref18]\] Hence, the reduction of bacterial count in this study could be due to high concentration of fluoride from fluor protector, which might have entered the bacterial cell and resulted in the inhibition of various cellular processes such as:\[[@ref19][@ref20]\] Inhibition of enolase that indirectly affects the formation of ATP, which is central to cell maintenance and growthInhibition of H+/ATPase in the bacteria which compromises the maintenance of cellular pH and makes the internal environment more acidic and unsuitable for other enzymes to actInhibition of exogenous glycerol uptake into lipoteichoic acid which is believed to play a crucial role in membrane stability as well as in colonization of *S. mutans* on hydroxyapatiteReduction of the peptidoglycan macromolecule in the cell membrane and thereby causing partial lysis of cell membrane. However consumption of fluoride above the therapeutic dose over an extended period of time particularly in younger age group may lead to dental fluorosis.\[[@ref17]\] To overcome the drawbacks of fluorides, one of the newer materials that have shown effective protection against caries by promoting lesion remineralization and inhibiting enamel demineralization is CPP--ACP. Caseinophosphopeptides inhibit dental caries lesions by influencing the demineralization/remineralization process of dental enamel. The active CPP is a naturally occurring molecule that can bind to calcium and phosphates and at the same time stabilize ACP. In the oral cavity, calcium and phosphate ions are released from CPP as the pH in the plaque decreases resulting in supersaturation which reduces demineralization and promotes remineralization.\[[@ref7][@ref9]\] A mean decrease of 1.6 ± 0.9 for group II which was statistically significant (*P* \< 0.0001) was observed. The mean change in Group II was significantly higher than the mean change in Group I (*P* \< 0.05) showing that CPP--ACP had a better anticariogenic potential compared to fluoride varnish. Caseinophosphopeptides stabilizes ACP, which in turn localizes ACP in dental plaque, acting as large calcium reservoir within plaque that slows the diffusion of free calcium ions providing a source of calcium for remineralization, and thus restricting mineral loss during a cariogenic episode.\[[@ref21]\] Therefore, the mechanism of anticariogenicity for CPP--ACP is that this bioactive peptide substantially increases the level of ACP in plaque, thus depressing enamel demineralization and enhancing remineralization. Because the concentration of CPP--ACP in contact with tooth enamel increases so does remineralization. Reynolds\[[@ref9]\] showed an increased concentration of free and bound calcium in 0.5 and 1.0% CPP--ACP preparations, which was most effective at remineralization of the dental enamel. There is reduced enamel subsurface demineralization when enamel plaque is exposed to solutions of tryptic peptides of casein. Thus, incorporating casein peptides into enamel plaque increases the plaque\'s content of calcium and phosphate.\[[@ref2]\] The tryptic peptides responsible for caseinate\'s anticariogenic activity are the calcium phosphate stabilizing CPPs. These CPPs contain a specific sequence, which markedly increases the apparent solubility of calcium phosphate by stabilizing ACP, forming solutions that are supersaturated with respect to calcium phosphates. In a caries study, using 0.5% weights per volume (w/v) solution of CPP--ACP nanocomplexes with a 500 ppm fluoride solution, the anticariogenecity of CPP--ACP and fluoride together was additive, when compared to either CPP--ACP or fluoride alone.\[[@ref22]\] The advantage of using CPP--ACP complexed with fluoride is the simultaneous availability of calcium, phosphate, and fluoride. The calcium phosphate in these complexes is biologically available for remineralization of subsurface lesions in tooth enamel. In addition, in an *in vitro* study, Reynolds *et al*. found that the enamel remineralization potential of mouthwash containing CPP--ACP and 220 ppm fluoride was superior than CPP--ACP and fluoride mouth rinse.\[[@ref23]\] The distribution of participants according to the *S. mutans* score pre-application as 12 in class 2 and 18 in class 3 in group III for application of CPP--ACP plus fluoride (*P* = 0.60) changed to 15, 12, 3, and 0 in class 0, 1, 2, and 3, respectively (*P* \< 0.0001). The mean score before treatment was 2.6 ± 0.5 which decreased to a mean score of 0.6 ± 0.7, drawing attention to the fact that CPP--ACP and fluoride together is more proficient than either of the materials used independently, which was also observed by Sudjalim *et al*.\[[@ref24]\] and Erdam *et al*.\[[@ref25]\] Further, in the present study, no significant differences were reported among experimental and control groups in relation to incidence of adverse reactions. This is in accordance with the work of Rao *et al*.\[[@ref26]\] and Sitthisettapong *et al*.\[[@ref27]\] Well-documented evidence of fluorides reducing caries by various mechanisms has been clearly established, however, its hazardous effects cannot be overlooked.\[[@ref16][@ref17]\] To overcome these limitations, researchers have developed various nonfluoride caries preventive and remineralizing agents including CPP--ACP and CPP--ACP plus fluorides. Literature regarding the caries preventive and remineralization action of CPP--ACP and CPP--ACP plus fluorides is scarce, and the current review papers and meta-analysis focus on the requirement of further well-designed clinical trials and also advocate that the use of these materials must be encouraged by practicing dentists in day-to-day oral hygiene practices and dental materials.\[[@ref2][@ref7][@ref8][@ref21][@ref28][@ref29][@ref30]\] The results of the present study indicate that the use of the materials fluoride varnish, CPP--ACP, and CPP--ACP plus fluoride prevents the loss of tooth structure, preserving the integrity of primary dentition with the most encouraging results with CPP--ACP plus fluoride. CONCLUSION {#sec1-5} ========== With the ever changing scenario of preventive dentistry, the efficacy of newly introduced materials such as CPP--ACP in preventing enamel demineralization against the established material fluoride warrants further clinical trials. Therefore, the present study using fluoride varnish, CPP--ACP, and CPP--ACP plus fluoride was conducted, and the results revealed that there was a statistically significant reduction in the counts of *S. mutans* in the plaque biofilm after a period of 24 hours. Although the clinical trials on CPP--ACP and CPP--ACP plus fluoride are still in the infancy stage, the results of the present study stress the continued need for the long-term efficacy of these materials in changing the dynamics of dental caries towards remineralization, consequent to significant reduction in the counts of cariogenic bacteria both in children and adults. Financial support and sponsorship {#sec2-6} --------------------------------- Nil. Conflicts of interest {#sec2-7} --------------------- There are no conflicts of interest.

Background {#Sec1} ========== The word cyanosis indicates the dark bluish discoloration of the skin, mucus membranes, and nail beds that becomes detectable when the levels of deoxygenated hemoglobin (Hb) in arterial blood exceed 3-5 g/dl \[[@CR1]\]. It is sometimes difficult to detect cyanosis in the newborn due to factors such as skin colour, exposure to light, or presence of jaundice \[[@CR2]\]. Since the relationship between cyanosis and pulse oximetry (SpO~2~) is directly related to the hemoglobin concentration, the percent desaturation required to produce the same grade of cyanosis is higher in anemic than in polycythemic infant \[[@CR2]\]. Cyanosis can appear early after birth or later as an acute episode, often during crying or feeding when tissue oxygenation can decrease further \[[@CR2]\]. The detection of cyanosis in infants is very important because it generally suggests inadequate peripheral tissue oxygenation and is one of the most accurate clinical signs of severe illness in infants during the first months of life \[[@CR3]\]. Although cyanosis, hypoxemia, and hypoxia are not synonymous, they are interrelated and in clinical practice cyanosis can disclose the occurrence of tissue hypoxia whose pathogenesis is frequently due to cardiovascular and respiratory diseases or to less frequent pathologies such as hemoglobinopathies \[[@CR4]\]. Therefore, the sudden appearance of an episode of cyanosis in a newborn in the hospital nursery who is expected to be healthy generally prompts his/her admission to neonatal care unit to monitor vital signs and carry out biochemical and instrumental diagnostic tests. Although cyanosis episodes are not rare in the neonatal period, there are few studies in the literature reporting on the epidemiology of this condition and on the diagnostic and therapeutic management of these patients \[[@CR5]\]. Thus, the purpose of this study was to collect data on all late preterm and term infants who were admitted to our neonatal care unit in a tertiary hospital for unexpected episodes of cyanosis that occurred during rooming-in with mothers to evaluate their frequency and the most frequent pathologies, and document the performed diagnostic clinical approach. Methods {#Sec2} ======= We carried out this retrospective study at the neonatal care unit of the Careggi University Hospital of Florence after the approval of the Tuscany pediatric ethics committee. Inborn infants with gestational age ≥35 weeks who were admitted from the nursery with the clinical diagnosis of cyanosis from January 2009 to December 2016 were included in the study. Infants with gestational age of 34 weeks were not studied because in our center they are always admitted to the neonatal care unit. Cyanosis was defined as the occurrence of dark blue discoloration of cutaneous and mucosal surfaces throughout the body (central cyanosis). Exclusion criteria were the occurrence of cyanosis limited to the extremities and lips (acrocyanosis) \[[@CR6]\], and the diagnosis of sudden unexpected postnatal collapse (SUPC) as defined by the British Association of Perinatal Medicine: SUPC is (1) a sudden and unexpected postnatal collapse occurring within the first postnatal week (2) in an infant born \>35 weeks gestational age, with a 5-min Apgar of \>7, appearing well at birth and considered healthy, (3) requiring cardiopulmonary resuscitation and intensive care (with mechanical ventilation), and resulting in death or encephalopathy \[[@CR7]\]. All infants were in complete rooming-in care where the mother and the baby stay together in the same room after birth for the duration of hospitalization \[[@CR8], [@CR9]\]. Thus, episodes of cyanosis might be detected by parents or caregivers. Infants' records were examined and the following data were recorded for each patient: gestational age, birth weight, gender, type of delivery, Apgar score at 1 and 5 min, patients' age at episode of cyanosis, mode of recovery (spontaneous or stimulation), recurrent episodes, association with other signs/symptoms (i.e.: hypothermia, cardiac murmur, arrhythmias, hypotony, seizures, regurgitation/vomitus), blood tests \[i.e.: blood gas analysis, cell counts, reactive C protein (RCP), C-troponin\], cultures (blood and liquor), instrumental tests \[chest X-rays, electrocardiogram (ECG), electroencephalogram (EEG), heart and/or cerebral ultrasound, cerebral nuclear magnetic resonance (NMR)\], diagnosis at discharge, treatments, and admission duration. Sepsis was diagnosed when patients developed clinical signs and symptoms associated with a positive blood and/or liquor culture. Suspected sepsis was diagnosed when patients developed clinical signs and symptoms with abnormal C-reactive protein without the confirmation of positive blood and/or liquor cultures. Statistical analysis {#Sec3} -------------------- Patients' clinical characteristics and recorded data were described as mean and standard deviation, median and range, or rate and percentage. Possible differences in the occurrence of cyanosis in subgroups of patients were evaluated with the *χ* ^2^ test for categorical variables. A *p* \<0.05 was considered statistically significant. Results {#Sec4} ======= During the study period 49 infants were eligible for the study and their characteristics are detailed in Table [1](#Tab1){ref-type="table"}. One patients was excluded for SUPC. The frequency of admission for cyanosis was 1.8/1000 live births. The mean gestational age was 38 ±2 weeks, ranging from 35 to 41 weeks with 43 patients (88%) being term infants and 6 (12%) preterm. During the study period 1428 preterm infants and 25294 term infants (*n* = 26722) were assisted in complete rooming-in with mothers. Thus, the occurrence of cyanosis episodes was higher in preterm than in term infants (4.2/1000 vs. 1.7/1000 live births) but the difference was not significant (*p* = 0.167). The mean birth weight was 3240 ± 470 g; 2 of our patients were low birth weight infants (\<2500 g) and 2 were macrosomic (\>4000 g). Only 17 infants were males (35%).Table 1Clinical characteristics of infants. Mean ± (SD), rate and (%), or median and (range)Clinical characteristics*n* = 49Gestational age (wks)38 ±2Birth weight (g)3240 ± 470Male gender17 (35)Cesarean section14 (29)Apgar score: -at 1 min9 (8-10) -at 5 min10 (8-10)Age at episode of cyanosis (h)24 (3-120)Mode of recovery: -spontaneous40 (82) -stimulation9 (18)Recurrent episodes7 (14)Associated signs/symptoms: -hypothermia2 (4) -cardiac murmur3 (6) -arrhythmias2 (4) -hypotony9 (18) -seizures3 (6) -regurgitation/vomitus14 (29)Hospital stay (d)24 (3-44)Maternal pregnancy disorders -gestational diabetes10 (20) -hypertensive disorders5 (10) Cyanosis occurred in infants at a median age of 24 (3-120) hours of life, with the majority of episodes occurring during the first 24 h of life (*n* = 28; 57%), and the remaining occurring at 25-48 h of life (n = 10; 20%), or after 48 h of life (*n* = 11; 23%). The lowest value of SpO~2~ in infants (*n* = 7; 14%) who had recurrent episodes of cyanosis was 83%. Twenty-nine infants (59%) exhibited signs/symptoms associated with cyanosis and 4 (8%) had more than one. There was a maternal pregnancy disorder in 15 cases (31%) (Table [1](#Tab1){ref-type="table"}). It was possible to diagnose 16 (33%) infants (Table [2](#Tab2){ref-type="table"}) and, interestingly, this occurred more frequently (71%) when they had recurrent episodes of cyanosis (14%). Appropriate therapies followed each diagnosis (unreported data).Table 2Diagnosis identified in our population. Rate and (%)Diagnosis*n* = 49Gastro-esophageal reflux5 (10)Suspected sepsis3 (6)Inter-atrial septal defect2 (4)Genetic syndrome2 (4)Intraventricular hemorrhage1 (2)Seizures1 (2)Monolateral choanal atresia1 (2)Transient central apnea1 (2) All patients were given diagnostic tests and between laboratory tests blood gas analysis (100%) and cell counts (92%) were most commonly performed, while between instrumental tests heart (80%) and cerebral (88%) ultrasounds were the most common (Table [3](#Tab3){ref-type="table"}).Table 3Diagnostic tests performed in our population. Rate and (%)Test*n* = 49Laboratory and microbiological testsGas analysis49 (100)Cell count40 (92)Reactive C Protein (RCP)37 (76)Procalcitonin (PCT)29 (59)Blood culture9 (18)Liquor culture3 (6)Instrumental testsChest X-rays27 (55)Electrocardiogram (ECG),28 (57)Electroencephalogram (EEG),15 (31)Heart ultrasound39 (80)Cerebral ultrasound43 (88)Cerebral nuclear magnetic resonance (NMR)4 (8) Discussion {#Sec5} ========== In this study we report the frequency of unexpected episodes of cyanosis that occurred in our hospital in late preterm and term infants who were in complete rooming-in with mothers and that caused their admission to the neonatal care unit. We found a frequency of 1.8/1000 live births admitted that is lower than that previously reported by Casanueva et al. who found an admission rate of 5.6/1000 live births \[[@CR5]\]. This difference might be because in this study infants who were born at 34 weeks of gestation were also included; moreover, the Casanueva's definition of cyanosis was not well detailed and might have been different from ours \[[@CR5]\]. It is interesting that the occurrence of cyanosis in our population was not significantly affected by the infants' prematurity, since those born at 35-36 weeks had a similar frequency as term infants. It can be questioned whether our population size might affect this finding, however sample size (*n* = 26722) is reassuring from this point of view. This is very important because it is well known that late preterm infants commonly have a higher rate of medical complications (i.e.: respiratory distress, hypoglycemia, hyperbilirubinemia, etc.), prolonged hospital stay and need more medical resources when compared with matched full-term infants \[[@CR10]\]. Nevertheless, the practice of assisting late preterm infants in complete rooming-in care is very diffused and our data seem to be reassuring regarding the safety of this procedure. Most of patients in our population were female (65%), but this finding is not confirmed by a previous paper in which 51% of patients were female \[[@CR5]\] and is probably due to the paucity of our sample. We found that episodes of cyanosis frequently (67%) developed during the first day of life and we can speculate that this happens because this is the most critical period for the cardiovascular, respiratory, and metabolic transition from fetal to neonatal life \[[@CR11]\]. The majority of infants exhibited signs/symptoms associated with cyanosis that sometimes were diagnosis-related (i.e.: cardiac murmur and inter-atrial defect, seizures and intraventricular hemorrhage), but sometimes were not (i.e.: regurgitation/vomitus). Therefore, unfortunately, it seems that the presence of these associations has limited usefulness in deciding or excluding the need for an infants' admission for further evaluations. It was possible to perform an exact diagnosis and provide proper treatment in a minority (33%) of infants in our population, in agreement with Casanueva' study in which only 21% of patients were found to be affected by a specific pathology \[[@CR5]\]. These results suggest that unexpected episodes of cyanosis of unknown etiology are often due to self-limiting transient functional events that do not require specific treatments. This speculation seems to be confirmed by the fact that when cyanosis recurs it is more probable (71%) to detect a pathology. The most frequent diagnosis in our study was gastro-esophageal reflux and was based on clinical criteria (an episode of regurgitation/vomitus caused the admission for cyanosis and was followed by further episodes, etc.) and on recovery or improvement with thickening of breast milk/formula feedings \[[@CR12]\] and/or H^2^-antagonists (i.e.: ranitidine) and/or alginate preparation treatment \[[@CR13]\]. This finding partially agrees with a previous study \[[@CR5]\] that found a gastro-esophageal reflux in a percentage of infants similar to ours (10 vs. 14%). On the contrary, we found that seizures were a rare cause of cyanosis (2%) which was the most frequent diagnosis (33.3%) in the Casanueva' study \[[@CR5]\]. Our patients were checked with biochemical, cultural, and instrumental tests to detect any possible infectious, metabolic, respiratory, cardiac, or neurological diseases as appropriate for their clinical condition, confirming that the development of an unexpected episode of cyanosis is resource-consuming \[[@CR5]\] although efforts to ascertain a diagnosis are unsuccessful in the majority of cases. We wonder if the crises of cyanosis of unknown origin that we observed in our patients might actually represent episodes of SUPC \[[@CR7], [@CR14]\] that were precociously and casually detected before the development of more severe symptoms requiring need of resuscitation, or if they represent a low risk condition that may be referred to as a brief resolved unexplained event (BRUE) requiring simple interventions such as positional changes, brief stimulation, or procedures to resolve airway obstruction \[[@CR15]\]. We are not able to answer this question but we can report that, to the best of our knowledge, none of our patients developed subsequent episodes of cyanosis or more severe life-threatening events after their discharge from hospital. However, the issue of neonate' surveillance during the first days of life is urgent and requires special and close attention to improve their safety while they are rooming in. Conclusions {#Sec6} =========== We found 1.8/1000 live births unexpected episodes of cyanosis in infants without differences between late preterm and full-term who were in complete rooming-in with mothers which prompted their admission to neonatal care unit. We found that the majority of episodes of cyanosis occurred during the first day of life and that the majority of these infants were discharged without a known diagnosis. Not applicable Funding {#FPar1} ======= We had no specific fund for this study. Availability of data and material {#FPar2} ================================= The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request. Authors' contributions {#FPar3} ====================== CD conceive and wrote the study. LV and GB collected data. CD, GB, CP, and SP analyzed and interpreted patients' data. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== The present study was carried out after the approval of the Tuscany pediatric ethics committee. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

INTRODUCTION {#SEC1} ============ The uncovering of genome-wide gene regulatory networks is a key tool in systems biology allowing researchers to understand the complex mechanisms of transcriptional gene regulation and to model the expressional behavior of genes in accordance to changing internal and environmental conditions. Collectively, these models form the basis of our current understanding of the complex systems biology of cells and multicellular organisms. The inference of regulatory interactions through computational methods has become a standard tool and with the rapidly growing body of available gene expression studies currently available, the reconstruction of transcriptional gene regulatory networks (GRNs) using gene expression data is becoming more and more promising. The problem of inferring a gene regulatory network from gene expression data has received significant attention. It was the focus of four separate DREAM challenges, with DREAM5 in 2010 being the most recent one. A wide array of gene expression-based network inference methods have been developed ([@B1],[@B2]). Despite these efforts, previous evaluations found that gene expression-based inference methods achieve very modest performance when applied to real data, despite performing well on *in silico* generated data ([@B2],[@B3]). Additional methods incorporate multiple types of data such as network topology, sequence information and gene set enrichment ([@B4]) to improve predictive performance over a purely gene expression-based approach. Gene regulatory networks have commonly been modeled using Boolean networks ([@B9],[@B10]), Bayesian networks ([@B11],[@B12]) and ordinary differential equations ([@B13]). However, given that currently available regulatory information is provided as binary interaction networks, it is important to assess the validity of such networks regardless of the modeling approach taken. The consistency between GRNs and gene expression studies has previously been studied in *E. coli* for a set of well-studied genes under four different conditions ([@B17]). The sign consistency model introduced by Siegel *et al.* ([@B18]) provided a mathematical framework for evaluating inconsistencies in signed interaction graphs. Several methods, often Cytoscape ([@B19]) plugins, have subsequently been developed for automating the detection of inconsistencies based on the sign consistency model. COMA ([@B20]) uses a Boolean network model to detect inconsistent interactions in the network with respect to a single expression study; BioQuali ([@B21]) detects inconsistent genes in the network and can suggest changes to the expression profile in order to better explain the experimental data; CytoASP ([@B22]) uses an answer set programming approach to identify inconsistent genes and suggest possible ways to repair them, either by changing expression profiles, changing the influence of an interaction, or by introducing new interactions. Finally, the SigNetTrainer tool ([@B23]) uses an integer linear programming approach to repair inconsistent signs in a single gene expression experiment to match a given interaction network, or identify insertions and deletions of interactions in the network to best match a given set of experiments. Informed by our current understanding of transcriptional regulation it is generally assumed that an up- or downregulation of the genes coding for a transcriptional factor will result in a corresponding change in the expression of the genes regulated by that transcription factor. To investigate this assumption, we analyzed the consistency between a state of the art experimentally validated gene regulatory network of *E. coli* and a large compendium of gene expression profiles. We aimed to quantify how well the current gene regulatory network of *E. coli* corresponds to measured gene expression levels, and how well the transcriptional influence of TFs is reflected in the expression of their target genes. Our analysis revealed that both activating and repressing interactions were associated with positive correlation, which directly contradicted our preunderstanding. Furthermore, we show that, when evaluated using a sign consistency model, the regulatory network is not more consistent with measured gene expression than random network models. This implies that one cannot expect to see a causal relationship between the expression of transcription factors and their targets when evaluating the cell at a system-wide level. MATERIALS AND METHODS {#SEC2} ===================== Data set preparation {#SEC2-1} -------------------- We obtained a compendium of *E. coli* gene expression data from the DREAM5 challenge data set ([@B3]). The compendium contained expression for 4297 genes from 805 samples, all obtained with the Affymetrix *E. coli* Antisense Genome Array. The compendium contains expression from a wide range of experients including wild type, drug perturbations, environmental perturbations, gene knockout/knockdown and time series. The expression data was normalized using Robust Multi-array Averaging (RMA) and quantile normalization then log-scaled. We obtained a set of regulatory interactions from the RegulonDB database ([@B24]), consisting of 4564 experimentally validated transcription factor (TF) to gene interactions and 2154 TF to transcription unit (TU) interactions. A transcriptional gene regulatory network was constructed as a directed bi-partite graph containing a vertex for each TF, and one for each gene and TU, respectively. An edge between pairs of vertices (TF to gene/TU) was added for each transcriptional interaction reported in RegulonDB. Each interaction in the regulatory network was labeled as either an activation (↑) or repression (↓). Interactions with dual or unknown regulatory effect were removed (0.4% and 2%, respectively). Annotation of 214 transcription factors and 1036 transcription units were also obtained from from RegulonDB. TFs and TUs where one or more of its constituent genes had no expression profile were removed from the network (12% and 24%, respectively). The resulting network contained 175 TFs, 1865 targets and 4503 interactions, where 54% of interactions were activating and the remaining 46% were repressing. Genes in a TU are regulated together and thus we expect them to have similar expression. Because this is not always the case, we defined the expression of a TU as the mean expression of its constituent genes for each sample. For transcription factor complexes (i.e. several genes coding for proteins acting as TF complex together) we defined the expression level as the minimum level among the genes for each sample. For comparison we also obtained an *in silico* generated data set from the DREAM5 challenge containing a simulated regulatory network and corresponding gene expression compendium. The *in silico* regulatory network was generated to have topology similar to known regulatory networks of model organisms using the method described in ([@B25]). The simulated gene expression data was generated based on the regulatory network using GeneNetWeaver ([@B26]). The generated data simulates wild types, as well as drug/environmental perturbations, knockout/knockdown experiments and time series. The *in silico* network contained 178 TFs, 1498 targets and 4012 interactions, where 56% of interactions were activating and the remaining 44% were repressing. Inconsistency detection {#SEC2-2} ----------------------- We evaluated the consistency between the regulatory networks and the gene expression compendia using a sign consistency model. We first computed contrasts from raw expression values in order to identify genes that were up- or downregulated in each experiment with respect to some reference. For each experiment in each compendium we identified an unperturbed sample to use as reference. Contrasts were then computed as the log-ratio between the reference and each other case in the experiment. In both the *E. coli* and *in silico* data set this resulted in 655 contrasts. To assess the consistency between the regulatory network and the gene expression studies we identified a set of cases that we define to be *inconsistent*. We used an inconsistency model similar to the models used in BioQuali and COMA, but extended to include rules for genes that are unchanged. Each vertex (TF, gene or TU) was labeled as either upregulated (+), downregulated (−) or unchanged (0) for each contrast in the data set. Vertices were labeled according to the log-ratio between case and control and some threshold *t*, where vertices were considered downregulated when the log-ratio was less than −*t*, upregulated when greater than *t* and unchanged otherwise. Intuitively, when a TF (or TF complex) is upregulated, we expect its target genes (or TUs) to be upregulated if it is an activator, and downregulated if it is a repressor. Similarly, if a TF is downregulated, we expect its target genes to not be upregulated if it is an activator, and not downregulated if it is a repressor. When the TF is unchanged the interaction is considered consistent regardless of the labeling of the target gene/TU and type of interaction. When the target gene/TU is unchanged, we consider an interaction inconsistent if the TF is upregulated and consistent otherwise. One could make this model more strict by defining all interactions where one but not both interactors are unchanged to be inconsistent, however, it is not well-defined which of these cases are actually inconsistent. For this reason we chose a more conservative definition in order to reduce the number of interactions falsely labeled as inconsistent. Table [1](#tbl1){ref-type="table"} gives a overview of which interactions were considered consistent and inconsistent. ###### Overview of regulatory interactions considered inconsistent according to the expression of the transcription factor and target genes. Regulation TF exp. Target exp. Consistent ------------ --------- ------------- ------------ ↑ \+ \+ yes ↑ \+ − no ↑ \+ 0 no ↑ − \+ no ↑ − − yes ↑ − 0 yes ↓ \+ \+ no ↓ \+ − yes ↓ \+ 0 no ↓ − \+ yes ↓ − − no ↓ − 0 yes ↑, ↓ 0 +, −, 0 yes The symbols in column 1 describe whether the interaction is an activation (↑) or repression (↓). The symbols in columns 2 and 3 signify whether the vertex is labeled upregulated (+), downregulated (−) or unchanged (0). Let the *inconsistency vector* of an edge *e* be a vector *I*(*e*) ∈ {0, 1}^*n*^, where *n* is the number of contrasts and *I*~*i*~(*e*) = 1 if *e* is inconsistent with respect to contrast *i* and *I*~*i*~(*e*) = 0 otherwise. Let the inconsistency vector of a gene or TU \$v\$ be a vector *J*(*v*) ∈ {0, 1}^*n*^, where $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{upgreek} \usepackage{mathrsfs} \setlength{\oddsidemargin}{-69pt} \begin{document} }{}\begin{equation*} J_i(v) = \prod _{e \in N^-(v)} I_i(e), \end{equation*}\end{document}$$and *N*^−^(*v*) is the set of incoming edges of *v*. When a target gene/TU is subject to multiple regulators, it is considered consistent with respect to a specific contrast if there exists at least one incoming regulation that is consistent in that contrast. If several TFs of different types (i.e. activators and repressors) regulate the same target gene/TU, it is generally unclear which regulation is the dominant one. In our case we are looking for a conservative lower bound (i.e. the minimum inconsistency load) and resolve such cases by marking a gene/TU as 'explained' if any regulator explains its over-/underexpression. We define the number of inconsistencies for an interaction *e* as \|*I*(*e*)\|~1~. Similarly, we define the number of inconsistencies for a vertex *v* as \|*J*(*v*)\|~1~ (Figure [1](#F1){ref-type="fig"}). Furthermore, we define the *global inconsistency load* for a network as the total number of inconsistencies among all genes and TUs in the network. {#F1} Network perturbation methods {#SEC2-3} ---------------------------- In order to assess the consistency of our regulatory network compared to random data (the null model), two methods of perturbation were implemented. The broad distribution of node degrees suggests that the degree is an important characteristic for nodes in biological networks ([@B27]). For this reason, both perturbation methods preserve the degree of all nodes in the network to rule out the impact of node degree distribution on consistency. The first method perturbs the expression data by uniformly redistributing the entire expression profiles of genes using a Fisher-Yates shuffle. This method produces a regulatory network with identical topology to that of the original network, but with randomly assigned expression profiles. Entire profiles are exchanged in order to keep each profile internally consistent. If individual expression values of different genes were to be exchanged instead, we would no longer quantify how consistent a random set of interactions is wrt. the gene expression compendium, but instead how consistent random gene expression is wrt. the regulatory network. The second method perturbs the topology of the network while preserving both the in-degree and out-degree all nodes. This is achieved using a simple numerical algorithm proposed in ([@B28]). The algorithm selects two existing edges (*p, q*) and (*r, s*), and rewires their endpoints such that they become two new edges, (*p, s*) and (*r, q*). If one or both of these new edges already exist, the procedure is aborted and another pair is selected instead. This procedure is repeated 10 · \|*E*\| times, where \|*E*\| is the number of edges in the network. This method produces a network with a randomly perturbed topology but identical node degree distribution to that of the original network, while preserving the expression profiles of individual genes. RESULTS {#SEC3} ======= Correlation of expression profiles {#SEC3-1} ---------------------------------- First we assessed the correlation between gene expression profiles of known TF and target gene (or TU) interactions in the regulatory network of *E. coli*. The correlation was determined using Pearson's correlation coefficient. We first examined the distribution of correlations for known TF-target pairs and observed a mean correlation of 0.12 (Figure [2A](#F2){ref-type="fig"}). When evaluating all possible TF-target pairs in the network we observed a mean correlation of 0.02. Based on the assumption that greater expression of a repressor will lead to a lower expression of its targets, one would expect the the repressors to be associated with anti-correlation. However, when separating the interactions by regulatory influence, we observe a modest positive correlation in both cases (0.10 and 0.14 for repression and activation, respectively) (Figure [2B](#F2){ref-type="fig"}). {#F2} When looking at the correlation of interactions in isolation, we do not account for more complex relationships such as multiple TFs regulating the same gene. To investigate whether this has an impact, we examined the distribution of correlations for only those interactions where the TF is the sole regulator of the target gene or TU (17% of interactions, 40% of targets). For this reduced set of interactions, we observe a similar distribution, but with a greater mean correlation for both types of interactions (0.19 and 0.20 for repressors and activators, respectively) (Figure [2C](#F2){ref-type="fig"}). Surprisingly, for single-regulator interactions the difference in mean correlation between activations and repressions was in fact smaller than for the complete regulatory network. We repeated this evaluation for the *in silico* data set and found highly dissimilar results. The mean correlation of known interactions was close to zero (0.02), but with high variance compared to the background distribution (Figure [2D](#F2){ref-type="fig"}). When separating interactions by type we observed a clear separation between the two distributions, with repressors overall associated with negative correlation and activators associated with positive correlation (mean --0.23 and 0.21, respectively) (Figure [2E](#F2){ref-type="fig"}). When considering only single-regulator interactions we observed an even stronger separation between the distributions with almost no overlap (Figure [2F](#F2){ref-type="fig"}). Inconsistency between regulatory network in expression data {#SEC3-2} ----------------------------------------------------------- We used an inconsistency model to evaluate how consistent the regulatory network was with the gene expression data. We use a threshold of ±0.043 for the *E. coli* data and ±0.26 for the *in silico* data. These threshold were chosen such that ≈50% of contrast values were considered either up- or downregulated in the model. We first compared the global inconsistency load (total number of inconsistent cases among all targets) of the *E. coli* regulatory network to two random network models. The first method perturbs the network by random edge rewiring, while the second method redistributes gene expression profiles among all genes (see Methods). For each perturbation method the experiment was repeated 200 times. The global inconsistency load for the unperturbed network was 148 959. When perturbing the edges of the network the median global inconsistency load increased to 151 293 (1.6% increase) and when perturbing the expression profiles instead, the median inconsistency load decreased to 148 584 (0.3% decrease) (Figure [3A](#F3){ref-type="fig"}). The high similarity between the consistency load of the regulatory network and the null models suggests that the regulatory network is not significantly more consistent with the expression data than a random network. In fact, and surprisingly, when redistributing the expression profiles a majority of cases were more consistent than the unperturbed network. When separating the interactions by type we observed that repressive interactions were on average more inconsistent (mean 178.7) than activating interactions (mean 145.9) (Figure [3B](#F3){ref-type="fig"}). For single-regulator targets the mean inconsistency was slightly higher for repression and slightly lower for activation (181.1 and 135.5, respectively) (Figure [3C](#F3){ref-type="fig"}). {#F3} In the *in silico* data, we observed that perturbing the network greatly increased the overall inconsistency. The global inconsistency load for the unperturbed network was 55 247, increasing to median 84041.5 (52% increase) when rewiring interactions and median 95 710 (73% increase) (Figure [3D](#F3){ref-type="fig"}). When separating interactions by type, the mean number of interactions is nearly identical (128.2 and 130 for repression and activation, respectively) (Figure [3E](#F3){ref-type="fig"}). This was also the case when considering only single-regulator targets (Figure [3F](#F3){ref-type="fig"}). Association between inconsistency and experimental evidence {#SEC3-3} ----------------------------------------------------------- We examined the inconsistency load for each of the 655 contrasts to determine if some experimental conditions were associated with higher inconsistency. The number of inconsistencies ranged from 2 to 625, with mean 227.4 (Figure [4A](#F4){ref-type="fig"}). We observed that contrasts where the case condition was subjected to a perturbation (e.g. drugs or environmental perturbations) were associated with a greater inconsistency load (mean 267.9 versus 191.2, Mann--Whitney *U* *p* = 5.9e−14) (Figure [4B](#F4){ref-type="fig"}). However, we further observed that the perturbed experiments were associated with a higher variance in fold change and consequently a greater number of genes marked up- or downregulated in the sign consistency model (Figure [4C](#F4){ref-type="fig"}) and that the number of up- or downregulated genes in a contrast was highly correlated with number of inconsistencies which may explain this discrepancy (Figure [4D](#F4){ref-type="fig"}). {#F4} We also examined whether some interactions reported in RegulonDB were more inconsistent with the expression data than others. All interactions in RegulonDB are classified based on the type and amount of experimental evidence reported in the database. We observed that the mean inconsistency for interactions with strong experimental evidence was marginally lower than interactions with only weak evidence (mean 159.0 versus 163.7, Mann--Whitney *U* *p* = 5.8e−5) (Figure [4E](#F4){ref-type="fig"}). We further evaluated the overall inconsistency based on the type of experimental evidence (excluding uncommon methods, occurring \<100 times). Out of the six evidence types considered, the two methods considered strong evidence by RegulonDB, namely *binding of purified proteins* and *site mutation*, ranked first and third in consistency. However, mean inconsistency was very similar across methods, ranging from 158.5 to 164.0 (Figure [4F](#F4){ref-type="fig"}). DISCUSSION {#SEC4} ========== We investigated, in *E. coli*, the commonly held assumption, that a change in the expression of a transcription factor should have a positive or negative causal effect on the expression of its targets, depending on the type of regulation. This behavior was, however, not observed. Instead we on average observed a positive correlation between transcription factors and their targets, regardless of their regulatory influence, with similar distributions for both interaction types. In many cases a gene or TU is regulated by several transcription factors. For such targets, it may be too simplistic to expect each interaction to be reflected in the mRNA levels. However, when looking at only regulations of genes/TUs with only one regulator, we observed a similar pattern, again with repressive interactions being positively correlated. The set of regulatory interactions reported in RegulonDB is most likely far from complete, and thus some of the targets considered as having a single regulator may in fact be regulated through other undiscovered interactions. However, this does not explain why the mean correlation coefficient for repression was both greater and more similar to activation for single-regulator targets. As expected, no correlation between random pairs of transcription factors and genes/TUs was observed. This suggests that the correlation we observe in the unperturbed data is a property of the network and not simply because the gene expression experiments are correlated (e.g. genes having a general, experiment-specific chance of being more upregulated in one experiment than another one). In this study we measured correlation using the Pearson correlation coefficient. It could be the case that there is a causal but non-linear relationship between the expression of TFs and their targets, which would not be captured with Pearson correlation. To confirm this was not the case, we repeated the correlation analysis using Spearman's rank correlation coefficient which produced analogous results ([Supplementary Figure S1](#sup1){ref-type="supplementary-material"}). When comparing the total inconsistency load of the GRN to random network models, we observed a remarkably similar degree of consistency. This suggests that the experimentally validated network does not explain the transcriptome data better than a random network, at least with respect to our sign consistency model. Repressive interactions were on average associated with a higher degree of inconsistency than activating interactions. This is not surprising, given that the underlying assumption that repressors and their targets should be anti-correlated is contrary to what was observed in real gene expression. The choice of fold change threshold for when genes are considered up- or downregulated may have a significant impact on the results. To investigate this we repeated the inconsistency analysis with a lower and higher threshold ([Supplementary Figure S2--S5](#sup1){ref-type="supplementary-material"}). We observed that a lower threshold resulted in a higher consistency load, but the overall patterns were the same: the global inconsistency load in *E. coli* was mostly unchanged under random perturbation, but increased significantly in the *in silico* data, and repressive interactions were on average more inconsistent in *E. coli* but almost identical to activating interactions in *in silico*. In our signed model TFs, genes and TUs were classified as either upregulated, downregulated or unchanged. Previous methods have generally used a binary classification where genes were simply labeled 'up' and 'down' (or 'present' and 'not present'), but we included the unchanged state to avoid small changes in expression being considered a change in regulation. We implemented a binary consistency model as well and repeated the inconsistency evaluation for different thresholds ([Supplementary Text S1.1](#sup1){ref-type="supplementary-material"}). We observed that perturbing the network has little effect on the overall inconsistency in *E. coli* but greatly increases inconsistency in the *in silico* data ([Supplementary Figure S6--S8](#sup1){ref-type="supplementary-material"}). These results confirm that the poor consistency we observed in this study is not a result of our model being too conservative. We acknowledge that our model for evaluating the inconsistency is both simplistic and conservative. However, the stark contrast between the *E. coli* and *in silico* data and the fact that perturbing the *in silico* data greatly increased the inconsistency load demonstrates, that our model should be able to recognize it if the expression data more closely resembled the idealized model emulated in the *in silico* data. Several possible explanations exist for the observed lack of a causal relationship between TFs and targets. In previous studies on *E. coli* and *S. cerevisiae*, the regulatory response of target genes has, in some cases, been observed to be time delayed ([@B29]). When this is the case, a regulatory response cannot be detected by correlating expression of genes at the same time points. To determine such relationships, a time series analysis is necessary. The lack of correlation could also suggest that many TFs are not significantly transcriptionally regulated ([@B30]). If a TF is primarily regulated by post-translational modifications such as phosphorylation by two-component signaling ([@B31],[@B32]), its activity not will be reflected significantly in the expression level of its coding gene. Another reason for the poor correlation of regulator/target could be a generally modest correlation of transcriptome and proteome data ([@B33],[@B34]). In addition, even if a TF is transcribed and translated, it may be in an inactive conformation. The lack of correlation may thus also be due to TFs not bound by their activator metabolites, or bound by inactivators, both of which cannot be observed from transcriptome data. While these results may be surprising, they are well in line previous performance evaluations. In the DREAM5 challenge, even the best performing methods did not achieve high precision beyond very low recall values despite performing well on the *in silico* data ([@B3]). A similar result was reported by Madhamshettiwar *et al.* ([@B2]) who compared several methods on *in silico* data and microarray gene expression from ovarian cancer patients. We argue that our results bring into question the applicability of previously published methods based on the detection of inconsistencies between known regulations and gene expression, and possibly even methods based on correlation or mutual information as well. This does not imply that inconsistency detection cannot be utilized to produce meaningful results, but it does suggest that more sophisticated models are needed, for instance, involving mult-omics, more diverse network types, or temporality. Finally, the clear difference between the real and synthetic expression data brings into question the validity of using simulated data for evaluating methods. If simulated data is to be used for evaluation going forward, one must ensure that the assumptions underlying the simulated model accurately reflect gene expression patterns *in vivo*. CONCLUSION {#SEC5} ========== We evaluated the overall consistency between the experimentally validated regulatory network of *E. coli* found in RegulonDB and a large compendium of microarray gene expression data. We observed that both activating and repressing interactions were associated with a positive correlation between the expression of transcription factors and their targets, and that the distributions of correlation for activating and repressing interactions were remarkably similar. When evaluated using a sign consistency model the regulatory network was not significantly more consistent with measured gene expression than two random network models. Our results suggest that one cannot expect a causal relationship between the expression of transcription factors and their targets and, as such, currently available static gene regulatory networks do not adequately explain transcriptional gene expression -- at least not on a systems-wide level. As a way forward, we urge to researchers to reconsider this flawed view on the relationship between transcriptional regulation and gene expression, and we suggest to base future methods on more complex models, multi-omics data, using several network types and/or including temporality. Supplementary Material ====================== ###### Click here for additional data file. SUPPLEMENTARY DATA {#SEC6} ================== [Supplementary Data](https://academic.oup.com/nar/article-lookup/doi/10.1093/nar/gky1176#supplementary-data) are available at NAR Online. FUNDING {#SEC7} ======= Villum Foundation VILLUM Young Investigator Programme \[13154 to J.B. and S.J.\]; European Research Council Advanced Grant \[294683 to H.S.\]; European Cooperation in Science & Technology COST Action \[CA15120 to H.S.\]; European Union Horizon 2020 Research and Innovation Programme \[777111 to J.B. and H.S.\]. Funding for open access charge: VILLUM Young Investigator Programme. *Conflict of interest statement*. None declared.

Introduction ============ Polyarteritis Nodosa (PAN) is a systemic, necrotizing vasculitis that affects medium-sized and small muscular arteries resulting in microaneurysm formation. In the presteroid era, mortality was high and the diagnosis was exclusively made postmortem. Objectives ========== Our goal is to present very challenging, biopsy proven, case od PAN which resulted in massive necrotic skin lesions and deep venous thrombosis. Methods ======= We report the case of 8 year old boy who was referred to our hospital with 11 days history of high fever (39.5°C), fatigue, anorexia and weight loss. Two days prior to admission he began to complain of pain in the calves muscles, developed petechial skin rash and hematomas, gross haematuria and gastrointestinal (GI) bleeding. At admission was febrile, in very poor condition, hyperventilating, with calf muscle tenderness (mostly gastrocnemius and soleus muscles), scrotal pain, unable to walk. Skin findings: livedo reticularis, multiple hematomas, petechie, subcutaneous nodules. On the second day he developed cutaneous necrosis, five days after admission bilateral legs edema was observed. Doppler ultrasonography showed massive deep end superficial venous thrombosis. He was tested negative for hepatitis C antibodies, perinuclear antineutrophilic cytoplasmic antibody (pANCA) and cytoplasmic antineutrophilic cytoplasmic antibody (cANCA). A deep skin biopsy was performed and showed fibrinoid necrosis of small muscular arteries with leukocytic infiltrate; finding compatible with polyarteritis nodosa. Kidney biopsy was contraindicated because the boy had agenesis of the right kidney. Results ======= Patient responded well to the aggressive immunosuppressive therapy in terms of halting progression of skin necrosis and gradually resolving haematuria, GI bleeding, muscle and scrotal pain**.** Thrombosis were treated with intravenous and subcutaneous heparin with good respons. Cutaneous necrosis required skin transplantation. At present time, 18 months after disease onset, PAN is in remission and there are total recanalisation of deep and superficial lower extremities veins. Conclusion ========== PAN is life-threatening disease that often presents with multiorgan involment, but is rarely associated with deep venous thrombosis. Pain and swelling of the lower extremities is not always a symptom of the disease itself, but may indicate another very serious condition such as deep venous thrombosis. Disease requires promt diagnosis and aggressive immunosuppressive treatment to reduce morbidity and mortality. Disclosure of interest ====================== None declared.

Introduction {#S1} ============ Prescription opioid dependence has reached epidemic proportions in the U.S. and internationally \[[@R1]\]. Overdose deaths involving prescription opioids have increased significantly since 1999; indeed from 1999 to 2015, more than 183,000 people have died in the U.S. from overdoses related to prescription opioids \[[@R2],[@R3]\]. It is well known that drug addiction is a disease of the brain \[[@R4]\] and thus, there is a critical need to get a fundamental understanding of the effects of chronic prescription opioid use on the human brain specifically on brain function and structure. Research has identified brain regions specifically associated with psychological processes underlying craving and substance use. A few studies have investigated cue reactivity \[[@R5]--[@R7]\], and attentional bias \[[@R8],[@R9]\] to prescription opioid related picture cues in chronic users. Cue reactivity is defined as an observable, classically conditioned response to alcohol and drugs \[[@R10]\] that correlate modestly with self-reported craving \[[@R11]\], and attention bias refers to giving drug related cues an increased priority in cognitive processing \[[@R12]\]. Bunce et al. compared recently withdrawn Prescription Opioid Dependent (POD) patients with the patients in supervised residential care for 2--3 months using a prescription drug cue reactivity task. They monitored the prefrontal cortex with functional near-infrared spectroscopy during the cue reactivity task. The recently withdrawn patients showed increased activation to pill stimuli in the right dorsolateral prefrontal cortex relative to extended care patients. Also, POD chronic pain patients evidenced a significant attentional bias towards prescription opioid related cues whereas non-dependent users did not show any significant attention bias \[[@R8]\]. These studies suggest that prescription opioid cues in the environment acquire salience, initiate arousal, and bias attention, as do other types of drug related cues. Yet, critical to our understanding of the neurobiology of chronic prescription opioid use is defining brain connectivity, that is, how the brain regions interact as opposed to focusing on the functionality of individual regions of interest in isolation. There are only two brain connectivity studies that have examined functional connectivity in the brain of the POD patients \[[@R8],[@R13]\]. Resting State Functional Connectivity (RSFC), measured by the correlation of spontaneous fluctuations of Blood Oxygen Level-Dependent (BOLD) signals in different regions of the resting brain, is believed to provide a measure of the brain's functional organization \[[@R14],[@R15]\]. Individuals with more intense chronic pain showed decreased RSFC between the perigenual anterior cingulate and the default mode (cognitive control) network likely reflecting reduced ability to govern pain-related thought processes \[[@R8]\]. Bilateral structural volumetric loss in the amygdala, significantly decreased anisotropy in axonal pathways specific to the amygdala, and significantly decreased RSFC for seed regions that included the anterior insula, nucleus accumbens and amygdala subdivisions have been demonstrated in POD persons compared to healthy controls \[[@R13]\]. However, the effects of long term prescription opioid use on the large scale brain networks that have implications in drug addiction have not been examined. These large scale brain networks include default mode network \[[@R16]\]; salience network \[[@R17]\]; lateral visual network \[[@R17]\]; dorsal attention network \[[@R18]\]; and the Drug Cue Processing Network (DCPN) which is a part of the mesocorticolimbic and nigrostriatal system implicated in drug seeking behavior and continuation of drug use \[[@R19],[@R20]\]. The DCPN captures elements of the default, salience, and executive control networks. Drug addiction causes changes in specific regions in the mesocorticolimbic and nigrostriatal system and these changes are manifested through clinical features of compulsive drug seeking, relapses despite negative consequences of using a drug, inhibitory control deficit and reward disturbances \[[@R19]\]. One drug cue reactivity study \[[@R6]\] has examined isolated brain area activation in response to prescription drug cues, but no study has examined structural and functional connectivity among brain areas within the DCPN and other large scale brain networks in prescription opioid users. It is essential to conduct such a study as the results will be the key to our understanding about brain networks reorganization due to prescription opioid addiction. Directions for Future Research {#S2} ============================== The literature cited above suggests the following directions for future neurocognitive research on prescription opioid addiction. First, studies would do well to include a multimodal assessment of structural and functional brain changes in prescription opioid users' brain networks including DCPN. This would be achieved by using functional MRI and structural Diffusion Tensor Imaging (DTI) techniques and structural, functional and effective connectivity (which reveals a causal influence of one brain area on another) analysis to understand changes in brain networks due to long term prescription opioid use. It would also be important for future researchers to collect for the first time the structural, functional and effective connectivity data between the regions within different brain networks in PODs and matched controls. Data collection should be geared toward developing both pharmacologic and behavioral therapies in the treatment of relapse prevention in individuals with prescription opioid use disorder. Finally, translational constructs linking research to clinical applications should be explicated. Specifically, how therapies can be developed from brain connectivity research to change the neuroplasticity within the DCPN and other large brain networks. Restoring normal functioning of these networks holds promise in promoting abstinence from drug use in this high-risk population. Research supported by a National Institute on Drug Abuse grant (K01 DA029047).

The electronic version of this article is the complete one and can be found at: <http://F1000.com/Reports/Biology/content/1/92> Introduction and context {#s01} ======================== A major goal of genomic studies is to understand the regulatory architecture of the genome; how, for example, powerful regulatory elements, such as enhancers and silencers, which have the ability to act over large genomic distances, are constrained to act only on appropriate transcription units. There is considerable evidence for a domain model of genomic regulatory architecture where the actions of long-range regulatory elements are confined within chromosomal domains \[[@bib-001]-[@bib-003]\]. The promiscuity of enhancers is restricted to regulation of transcription units within the same domain. In this model of genome organisation, insulator elements are proposed to play a key role as domain gate-keepers, shielding genes within the domain from the influence of outside regulatory elements. How insulators function is still unclear but they have been proposed to associate together to form chromosomal loops providing a physical basis for chromosomal domains. Insulator elements have been shown to have appropriate activities for such a role; they are defined by their ability to block enhancer-promoter interaction and/or to provide boundaries between different chromatin states and have been demonstrated to be involved in chromosome looping. But does the segregation of the genome into domains through the formation of chromosomal loops really represent the endogenous function of insulator elements? Although this question has been asked in the past \[[@bib-004],[@bib-005]\], recent results in *Drosophila* emphasising links between insulators and the regulation of transcription have once again brought it into focus. Major recent advances {#s02} ===================== A number of proteins have been associated with insulator function in *Drosophila*, including the Suppressor of Hairy-wing \[Su(Hw)\] protein \[[@bib-006]\], CCCTC-binding factor (CTCF) \[[@bib-007]\], Centrosomal protein 190 (CP190) \[[@bib-008]\], Boundary element-associated factor (BEAF) \[[@bib-009]\], GAGA factor (GAF) \[[@bib-010]\] and Zeste-white-5 (Zw5) \[[@bib-011]\], and over the past few years, genomic binding sites for several insulator components have been mapped \[[@bib-012]-[@bib-018]\]. In support of the domain model, these studies reveal associations between particular insulator components and boundaries of chromatin state domains; for example, Bartkuhn *et al.* \[[@bib-014]\] document that CP190 and CTCF binding correlates strikingly with the boundaries of repressive chromatin domains marked by the H3K27me3 modification. This accords with genomic studies in vertebrates demonstrating that CTCF binding sites mark chromatin state boundaries \[[@bib-019],[@bib-020]\]. However, the *Drosophila* genomic mapping experiments also reveal a strong association between particular insulator components and transcription start sites. This is especially true for some components, CP190, GAF and BEAF, but not for Su(Hw), suggesting that there may be distinct types of insulator complexes \[[@bib-015]\]. BEAF 32A and B were initially identified as factors binding to an insulator element (scs\') associated with the 87A hsp70 heat shock locus \[[@bib-009]\]. Insulator activity is dependent on BEAF binding sites and BEAF is involved in the formation of a chromosomal loop through interaction with Zw5 bound to a second element (scs) \[[@bib-021]\]. However, despite its strong 'domain boundary' credentials, more than 85% of the 1820 BEAF binding sites in the genome are less than 300 bp from transcription start sites \[[@bib-016]\]. Many (50%) of these BEAF peaks are associated with head-to-head gene pairs, although these BEAF sites do not appear to function to insulate these adjacent promoter regions from each other. Rather, BEAF appears to be generally required to promote transcription. Sharing of components between insulators and promoters may fit with shared structural features of these elements (for example, both CTCF-binding sites and active transcription start sites are marked by the presence of specific histone variants and are associated with nucleosomes containing both the H3.3 and H2A.z variants \[[@bib-022]\]). While this may be interpreted as indicating a link between insulators and transcriptional regulation, it might also suggest that 'insulator' components may serve a variety of functions and not all their binding sites are necessarily associated with insulator function. However, other experimental approaches provide support for a fundamental role for insulator-promoter interaction. It has been known for some time that some promoters can behave as insulators in enhancer-blocking assays \[[@bib-010]\]. However, it was never clear whether the mechanism of enhancer-block was the same in both cases. Recently, Chopra *et al.* \[[@bib-023]\] have uncovered a connection between promoters with enhancer-blocking activity and the presence of stalled RNA polymerase II (Pol II). Testing the ability of several Hox gene promoters to exhibit enhancer-blocking, promoters associated with stalled Pol II showed blocking function whereas several unstalled promoters did not. The enhancer-blocking activity was dependent on the function of the elongation factors Negative elongation factor (NELF) and DRB sensitivity-inducing factor (DSIF), which are thought to stabilise Pol II at the pause site. However, the key observation that draws together insulators and promoters is that the enhancer-blocking activity of two bithorax-complex insulators, *Fab7* and *Fab8*, is also dependent on the elongation factors NELF and DSIF. As *Fab7* and *Fab8* do not seem to contain stalled polymerase, the implication is that the insulators interact with promoters to inhibit elongation. This fits with observations, using DamID (DNA adenine methyltransferase identification) or chromosomal conformation capture, that the *Fab7* and *Fab8* insulators can form loops that bring them into contact with promoter regions of the *Abd-B* Hox gene and that this configuration is associated with repression of *Abd-B* transcription \[[@bib-024],[@bib-025]\]. In support of the generality of interaction between insulators and the regulation of transcription elongation, Jiang *et al.* \[[@bib-016]\] found considerable overlap between BEAF and NELF binding sites. In addition, vertebrate studies using chromatin topology assays have identified insulator-promoter associations at several loci, including *Igf2-H19* \[[@bib-026],[@bib-027]\] and *CFTR* \[[@bib-028]\]. Future directions {#s03} ================= Overall, the above combination of genomic binding site location and functional analysis currently indicates a more intimate relationship between insulators and the regulation of gene transcription than is suggested by the domain boundary/chromosome loop model. Insulators may serve a diverse range of functions; as classical boundaries, as boundaries of regulatory regions \[[@bib-013],[@bib-029]\], and as regulators of transcription. Even the regulatory effects of insulators can apparently be diverse; BEAF appears to be predominantly required to facilitate transcription whereas the above interactions between the *Fab* insulators and *Abd-B* promoters occur in a repressive context. Is there a common thread in this diversity? It is feasible that insulator elements in general are involved in the specification of chromosomal topology; this could provide a common function but, depending on the context, could have diverse consequences for chromatin state and transcriptional regulation. RAHW acknowledges funding from the Biotechnology and Biological Sciences Research Council, the Newton Trust and the National Institutes of Health. Competing interests =================== The author declares that he has no competing interests. BEAF : Boundary element-associated factor CP190 : Centrosomal protein 190 CTCF : CCCTC-binding factor DamID : DNA adenine methyltransferase identification DRB : 5,6-dichloro-1-β-[d]{.smallcaps}-ribofuranosylbenzimidazole DSIF : DRB sensitivity-inducing factor GAF : GAGA factor NELF : Negative elongation factor Pol II : RNA polymerase II Su(Hw) : Suppressor of Hairy-wing Zw5 : Zeste-white-5

###### Strengths and limitations of this study - A key strength of this paper is the use of routinely collected data from hospital episode statistics linked to cancer registry data, providing a large data set to make accurate estimates of relative prostate cancer mortality. - Lack of prostate-specific antigen data and a relatively short follow-up period of 6 years are the key limitations of this study. - Given that this is an observational study, there is some uncertainty about the causes for the observed differences in prostate cancer mortality. Background {#s1} ========== Outcomes following a diagnosis of cancer vary markedly around the world. In the USA, cancer-related deaths have been demonstrated to be among the lowest. For example, US breast cancer mortality is 65% lower than the European average while death from colorectal cancer is 30% lower.[@R1] On the other hand, cancer mortality rates in England are among the highest in Europe.[@R2] The disparity in cancer outcomes appears greatest for prostate cancer (PCa) for which 5-year mortality has been reported to be six times higher in England compared with the USA.[@R1] A number of disease and treatment-related factors may account for the observed variation in PCa outcomes between the USA and England. These include variation in policy concerning PCa screening between the two countries together with variation in use of definitive PCa therapy. Other factors that may be at play include the methods by which data on cancer diagnoses and cancer-related deaths are both collected and processed. In the USA, the vast majority of men diagnosed with localised PCa have definitive therapy, either by radical radiation therapy or radical surgery. For example, three-quarters of men diagnosed with PCa between 1988 and 2006 were reported to have undergone definitive therapy for their disease.[@R3] This figure compares to only about one-third in England.[@R4] [@R5] We report differences in risk-adjusted prostate cancer mortality (PCM) between the USA and England. Furthermore, we investigate whether PCa outcomes are related to the use of definitive therapy between the two countries. This study is part of a programme of work assessing the value of procedure-specific and disease-specific metrics derived from English hospital admission records to assess the performance of English National Health Service (NHS) providers. Methods {#s2} ======= Study design {#s2a} ------------ We performed a population-based observational cohort study using patient-level cancer registry data from England and the USA. Data sources {#s2b} ------------ Data collected by the eight regional cancer registries[@R6] for all men diagnosed with PCa in England were linked to the Hospital Episodes Statistics (HES) database[@R7] and national mortality records provided by the Office for National Statistics. The Surveillance, Epidemiology and End Results (SEER) database was used to identify American patients with PCa from 18 regional cancer registries.[@R8] This database covers 28% of the US population and is linked to mortality data provided by the National Center for Health Statistics. Participants {#s2c} ------------ Men diagnosed with PCa between 2004 and 2008, and aged between 35 and 80 years at the time of diagnosis were identified from both countries. The years 2004--2008 were selected as comparable English, and American data were available for this period. Diagnosis of PCa was confirmed using the 'C61' International Classification of Diseases (ICD-10) diagnosis code in the HES and SEER databases. Follow-up data were available through to 16 April 2010 for the English cohort, and 31 December 2010 for the American cohort. Patients were included if PCa was histologically confirmed as their only primary malignancy. Patients with lymph node involvement or distant metastases were excluded, as they would not be candidates for primary definitive therapy. Where data on metastatic disease were missing, we considered the use of chemotherapy as a surrogate marker for metastases. Patients who underwent chemotherapy within 6 months of diagnosis were therefore also excluded. Twenty-one patients in the English data set were noted to have negative survival data (ie, date of diagnosis was chronologically after the date of death), and were therefore excluded. Those with missing data concerning pathological Gleason score (GS) or clinical tumour (cT) stage were excluded from the primary analysis, as they would not be amenable to risk stratification. Variable definition {#s2d} ------------------- English patients were considered to have undergone definitive therapy if their HES record contained the 'M61' Office of Population Censuses and Surveys Classification of Surgical Operations and Procedures (4th revision) code[@R9] indicating radical prostatectomy within 1 year of diagnosis, or alternatively if their cancer registry record indicated the use of radiotherapy. Patients from the SEER data set were considered to have undergone definitive therapy if they underwent radical prostatectomy or radiation therapy as part of their first course of therapy. American patients were considered to have undergone radical prostatectomy if they had undergone cancer-directed surgery, coded as any of the following: radical/total prostatectomy, or prostatectomy with resection in continuity with other organs/pelvic exenteration. All forms of radiotherapy were assumed to be definitive in nature, as treatment doses are not routinely recorded in the SEER or English cancer registries. Risk stratification {#s2e} ------------------- Patients were classified into risk groups using a modified version of the National Comprehensive Cancer Network (NCCN) PCa risk classification,[@R10] based on cT stage and GS. Risk groups were defined as follows: low risk (cT1 stage and GS 2--6), intermediate risk (cT2 stage or GS 7), and high risk (cT3-T4 stage or GS 8--10). Since prostate-specific antigen (PSA) levels are not recorded in the HES database or English cancer registries, this variable was not used for risk stratification in this study. Outcome measurement {#s2f} ------------------- The cause of death among English patients was extracted from national mortality records provided by the Office for National Statistics, which were linked to cancer registry and HES data. Similarly, cause of death is routinely recorded as part of the SEER data set for US patients. Where the cause of death was listed as the disease code for PCa, C61, it was classified as a PCa death. Statistical analysis {#s2g} -------------------- χ^2^ test was used to compare proportions between the two countries. A Cox regression model was used to calculate adjusted HRs for all-cause mortality (ACM), comparing mortality in England and the USA. Similarly, adjusted subhazard ratios (SHR) were calculated for PCM using a maximum likelihood competing risk regression model, according to the method of Fine and Gray.[@R11] Failure event for PCM was defined as death due to PCa, while death due to a cause other than PCa was defined as the competing event. All analyses were performed using STATA V.11 (StataCorp, College Station, Texas, USA). All regression models were adjusted for age group, year of diagnosis, ethnicity, cT stage and GS (model 1). Next, the impact of variation in use of definitive therapy was assessed by additionally including use of definitive therapy in a separate regression model (model 2). Separate regression models were built to test for differences between the two countries for each individual risk group. This resulted in 20 regression models in total: 5 patient groups (all eligible patients, all patients with complete data, low, intermediate and high risk)×2 adjustment models (model 1 and model 2)×2 outcomes (ACM and PCM). Sensitivity analysis {#s2h} -------------------- In order to investigate the influence of excluding patients for whom tumour stage and Gleason grade data were missing, we performed a sensitivity analysis where all eligible patients were included. Role of funding source {#s2i} ---------------------- The study benefited from a grant from the Academy of Medical Royal Colleges supporting a project assessing the value of procedure-specific and disease-specific metrics derived from routinely collected data to assess the performance of NHS providers. Sponsors were not involved in the study design; the collection, analysis or interpretation of data; in the writing of the report; or in the decision to submit the paper for publication. Results {#s3} ======= Participants {#s3a} ------------ Data were available on 328 182 men (111 917 from England and 216 265 from the USA) of which 301 989 (97 079 from England and 204 910 from the USA) met the selection criteria. Reasons for exclusion are described in [figure 1](#BMJOPEN2014006805F1){ref-type="fig"}. {#BMJOPEN2014006805F1} Complete data to enable risk stratification (ie, cT stage and GS) were available for 222 163 men (23 235 from England and 196 928 from the USA). These data were used to undertake the primary analysis. Men diagnosed with PCa in England tended to be older and less ethnically diverse, to present with higher cT stage, and to have higher pathological GSs ([table 1](#BMJOPEN2014006805TB1){ref-type="table"}, see online supplementary appendix 1), with each of these differences reaching statistical significance at p\<0.001. Among patients for whom complete data were available, men diagnosed with PCa in England were more likely to present with high-risk PCa according to our modified NCCN criteria (34.5% in England and 17.2% in USA, [table 1](#BMJOPEN2014006805TB1){ref-type="table"}). ###### Patient demographics by country (n=222 163) England USA p Value ----------------------------------- --------------- ---------------- --------- Year of diagnosis (%)  2004 5378 (21.3) 36 172 (18.4) \<0.001  2005 4959 (19.7) 34 403 (17.5)  2006 5172 (20.5) 40 531 (20.6)  2007 5009 (19.9) 43 800 (22.2)  2008 4717 (18.7) 42 022 (21.3) Age group (%)  35--59 3620 (14.4) 56 399 (28.6) \<0.001  60--64 4361 (17.3) 40 287 (20.5)  65--69 6104 (24.2) 42 439 (21.6)  70--74 6145 (24.4) 33 912 (17.2)  75--79 5005 (19.8) 23 891 (12.1) Ethnicity (%)  White 17 924 (94.8) 154 077 (80.4) \<0.001  African/Caribbean 571 (3.0) 28 361 (14.8)  Asian 318 (1.7) 8638 (4.5)  Other 105 (0.6) 626 (0.3)  Missing 6317 5226 cT stage (%)  cT1 9374 (37.2) 72 407 (36.8) \<0.001  cT2 9538 (37.8) 107 762 (54.7)  cT3 5577 (22.1) 15 482 (7.9)  cT4 746 (3.0) 1277 (0.7) Gleason score (%)  2--6 10 909 (43.2) 99 661 (50.6) \<0.001  7 9112 (36.1) 75 247 (38.2)  8--10 5214 (20.7) 22 020 (11.2) Modified NCCN risk (%)  Low risk 6151 (24.4) 45 045 (22.9) \<0.001  Intermediate risk 10 386 (41.2) 118 074 (60.0)  High risk 8698 (34.5) 33 809 (17.1) Treatment---all risk groups (%)  No definitive therapy 15 583 (61.8) 45 113 (22.9) \<0.001  Definitive therapy 9652 (38.2) 151 815 (77.1) Treatment---low risk (%)  No definitive therapy 3799 (61.8) 17 516 (38.9) \<0.001  Definitive therapy 2352 (38.2) 27 529 (61.1) Treatment---intermediate risk (%)  No definitive therapy 5696 (54.8) 21 999 (18.6) \<0.001  Definitive therapy 4690 (45.2) 96 075 (81.4) Treatment---high risk (%)  No definitive therapy 6088 (70.0) 5598 (16.6) \<0.001  Definitive therapy 2610 (30.0) 28 211 (83.4) cT, clinical tumour; NCCN, National Comprehensive Cancer Network. Men diagnosed with PCa in England were less likely to receive definitive therapy (38.2% in England and 77.1% in USA), and this difference was observed in all risk groups ([table 1](#BMJOPEN2014006805TB1){ref-type="table"}). Mortality {#s3b} --------- The median follow-up for the entire cohort was 43.3 months. Unadjusted 6-year ACM among English men was higher compared with American men (21.0% vs 9.6%). Similarly, unadjusted 6-year PCM among English men was also higher, as compared with American men (9.6% vs 2.6%). This trend was similar among patients with complete data, whose outcomes are described below ([table 2](#BMJOPEN2014006805TB2){ref-type="table"} and [figure 2](#BMJOPEN2014006805F2){ref-type="fig"}). ###### ACM and PCM according to country of treatment and modified NCCN risk (n=222 163) 6-year ACM Model 1 (age at diagnosis, year of diagnosis, ethnicity, clinical tumour stage and Gleason score) Model 2 (model 1 and definitive therapy) ------------------- ------------ --------------------------------------------------------------------------------------------------- ------------------------------------------ --------- --------------------- ------- n=196 928 n=25 235 All risk groups 9.3% 18.5% 1.60 (1.52 to 1.68) \<0.001 1.03 (0.97 to 1.08) 0.336 Low risk 8.7% 10.3% 1.30 (1.15 to 1.48) \<0.001 1.06 (0.93 to 1.21) 0.397 Intermediate risk 7.6% 12.5% 1.44 (1.32 to 1.58) \<0.001 0.98 (0.90 to 1.08) 0.740 High risk 16.3% 31.8% 1.92 (1.78 to 2.06) \<0.001 0.99 (0.92 to 1.08) 0.863 6-year PCM Model 1 (age at diagnosis, year of diagnosis, ethnicity, clinical tumour stage and Gleason score) Model 2 (model 1 and definitive therapy) ------------------- ------------ --------------------------------------------------------------------------------------------------- ------------------------------------------ --------- --------------------- ------- All risk groups 2.4% 7.6% 1.88 (1.72 to 2.05) \<0.001 0.97 (0.88 to 1.07) 0.568 Low risk 0.9% 0.9% 1.57 (1.08 to 2.30) 0.018 1.31 (0.89 to 1.93) 0.169 Intermediate risk 1.4% 2.8% 1.71 (1.40 to 2.09) \<0.001 1.00 (0.81 to 1.23) 0.994 High risk 8.1% 18.8% 2.06 (1.87 to 2.28) \<0.001 0.96 (0.86 to 1.08) 0.537 ACM, all-cause mortality; Adj HR, adjusted HR; Adj SHR, adjusted subhazard ratios; NCCN, National Comprehensive Cancer Network; PCM, prostate cancer mortality. {#BMJOPEN2014006805F2} ### Primary analysis {#s3b1} The primary analysis was conducted using data from the 222 163 patients for whom cT stage and GS were available, to allow risk stratification. Unadjusted 6-year ACM among patients who had definitive therapy was 7.3% in England and 4.9% in the USA. Corresponding ACM figures among those who did not have definitive treatment were 19.5% in England and 15.5% in the USA. The greatest difference was observed in patients at high PCa risk undergoing definitive treatment with a 6-year ACM of 15.1% in England and 8.1% in the USA, with the smallest difference observed in patients with low-risk PCa who did not undergo definitive therapy (9.5% in England and 9.9% in the USA). Unadjusted 6-year PCM among patients from all risk groups who underwent definitive therapy was 2.4% in England and 1.2% in the USA. This compared with 8.8% among patients who did not receive definitive therapy in England and 4.5% in the USA. Differences in unadjusted 6-year PCM were smallest among patients with low-risk disease undergoing definitive therapy (0.4% in England and 0.5% in the USA), and greatest among patients with high-risk disease undergoing definitive therapy (7.6% in England and 3.7% in the USA). When comparing all patients with complete data amenable for risk stratification, following adjustment for age group, ethnicity, year of diagnosis and tumour characteristics (model 1), significantly higher ACM (adjusted HR=1.60, 95% CI 1.52 to 1.68) and PCM (adjusted SHR=1.88, 95% CI 1.72 to 2.05) were found in England than in the USA ([table 2](#BMJOPEN2014006805TB2){ref-type="table"}). Within each of the three risk groups, with adjustment for patient and tumour characteristics (model 1), the greatest difference in ACM and PCM was noted among the intermediate-risk and high-risk patients ([table 2](#BMJOPEN2014006805TB2){ref-type="table"}). PCM was not significantly different at 0.9% in both countries at 6 years among men with low-risk disease. When treatment allocation was included in the multivariate model (model 2), no difference in ACM and PCM was noted between the USA and England for all men (ACM: adjusted HR=1.03, 95% CI 0.97 to 1.08; PCM: adjusted SHR=0.97, 95% CI 0.88 to 1.07) or within each of the individual risk groups ([table 2](#BMJOPEN2014006805TB2){ref-type="table"}). ### Sensitivity analysis {#s3b2} Multivariate analysis for the entire cohort of 301 989 patients, including patients for whom data regarding either cT stage or GS were missing, revealed a similar trend (see online supplementary appendix 2). Adjustment for age group, ethnicity and year of diagnosis revealed higher ACM (adjusted HR=2.19, 95% CI 2.13 to 2.26) and PCM (adjusted SHR=3.67, 95% CI 3.50 to 3.85) among English patients. Additional adjustment for the use of definitive therapy appeared, in part, to account for variation in ACM (adjusted HR=1.55, 95% CI 1.50 to 1.59) and PCM (adjusted HR=2.37, 95% CI 2.25 to 2.50). Discussion {#s4} ========== PCa death in intermediate to high-risk cases is higher in England than it is in the USA. When we adjusted for the different rates of definitive therapy in the two countries, the rates of PCa death were similar. This suggests that the differences in mortality may be explained by a lower use of definitive therapy in England. Methodological considerations {#s4a} ----------------------------- First, the English data set contained a high proportion of missing data for cT stage and GS. The high proportion of patients with missing data in the English data set may be due to poor data capture. Excluded English patients tended to be older, to have more advanced disease, and they less frequently received definitive therapy (see online supplementary appendix 3). This limitation is unlikely to have had a marked influence on our results, as inclusion of these patients would have increased the observed difference in PCM noted between the two countries. Thus, these data provide a conservative estimate of the spread of PCa risk among the general English population. Nevertheless, it is worthwhile to note that these are the only population-wide data currently available for comparing management of PCa in the two countries. Furthermore, a sensitivity analysis was performed to investigate the influence of excluding patients with missing cT stage or GS. This showed that PCM is significantly higher in England than the USA, though this difference is partly explained on additional adjustment for the variation in use of definitive treatment in the two countries. Owing to the higher proportion of men with low-risk or intermediate-risk disease in the USA, the variation in use of definitive treatment becomes more apparent on risk stratification in our primary analysis. Second, the SEER data set did not contain information concerning patient comorbidity. We feel our findings remain valid despite this potential limitation as PCM is less strongly influenced by comorbid conditions than ACM.[@R12] In addition, there were also differences between England and the USA in the PCM of young patients aged between 35 and 59 years who are least likely to have comorbid conditions at the time of diagnosis (adjusted SHR=2.66, 95% CI 1.99 to 3.56, p\<0.001). Third, 'lead time bias' could be an explanation for PCM being lower in the USA than in the UK given that the uptake of PSA testing is much higher in the USA, the effect of which is likely to be that men in the USA are diagnosed with less advanced PCa at an earlier age. In an attempt to minimise the effect of this limitation, we adjusted for clinical stage at diagnosis and patient age at diagnosis together with GS in our primary analysis. Lastly, PSA levels were not available for English patients, and therefore they could not be used to adjust the differences in PCM between England and the USA. To investigate this limitation further, we evaluated if the inclusion of PSA into our risk stratification model resulted in significant recategorisation of a patient\'s PCa risk for the US patients. We found little movement between risk groups with, for example, only 7.4% US patients being reclassified as intermediate-risk having initially been assigned a low-risk status. Furthermore, Elliott *et al*[@R13] have previously shown that while it is advantageous to have all three clinical variables (including PSA, cT stage and GS) available for risk stratification, patients with high-risk disease can still be correctly identified even if one of these variable (such as PSA) is missing. Despite the aforementioned limitations, routinely collected data provide a rich resource to explain performance of healthcare providers in different countries. However, differences in coding practices and differences in healthcare frameworks must be acknowledged. Comparison with other studies {#s4b} ----------------------------- ### Mortality {#s4b1} PCM was found to be significantly higher in England compared with the USA among men with intermediate-risk and high-risk PCa. In the current study, we used SEER data of men diagnosed between 2004 and 2008 and found that 6-year ACM was 9.3% and PCM 2.4%. A study using SEER data of men diagnosed between 1992 and 2005 found very similar figures (5-year ACM 14.3% and PCM 1.7%).[@R14] Improvements in management of PCa and other comorbidities may explain why our figures for ACM are slightly lower. In comparison, our analysis of the English HES database found that 6-year ACM was 18.5% and PCM 7.6%. A study reporting outcome of 50 066 men diagnosed with PCa in the London area between 1997 and 2006 with a median follow-up of 3.5 years reported a PCM for men who had undergone definitive treatment of about 2%, which corresponds closely to the figures we found in this study.[@R15] The only two relevant randomised controlled trials[@R16] [@R17] demonstrated benefit of definitive therapy in patients with high-risk disease, which is consistent with the results of our study. ### Differences between England and the USA {#s4b2} A study using the EUROCARE and SEER registries including men diagnosed between 1985 and 1989 reported a 2.8 times relative excess risk of death among European men with PCa compared with their American counterparts.[@R18] A recent study using SEER data between 1975 and 2004 together with UK cancer mortality statistics found that age-adjusted PCM rates in the USA were significantly lower than in England with the decline in PCM being 4.2% per year since the 1990s, a figure about four times higher than that reported for England.[@R19] The investigators of both these studies suggested that difference in PCM between England and the USA is the result of variation in disease burden brought about by the higher incidence of PCa screening in the USA. However, neither study adjusted for PCa risk. In this study, we have identified for the first time that irrespective of PCa stage and GS, PCa outcomes in terms of ACM and PCM are better in the USA than in England, which does not support the increased use of PCa screening in the USA as an explanation for the difference in PCM. Instead, our data suggest that the better PCa outcome seen in the USA may be due to the more frequent use of definitive treatment. ### Clinical implication {#s4b3} The decision to offer definitive PCa therapy is influenced by both disease characteristics and patient characteristics. As noted in our results, variations in healthcare systems have direct and indirect effects on both these factors. The expected survival benefit of definitive PCa therapy must therefore also be balanced against the associated probability of side effects, including urinary incontinence and erectile dysfunction. Our analysis suggests that PCM in England may be improved by an increase in the use of definitive treatment. However, due to the retrospective nature of this analysis, there could be other factors such as lead time bias which account for this difference. Only randomised trials can address these differences directly. AS has received funding from the Isaac Shapera Trust for Medical Research, and is currently funded by a National Institute of Health Research Academic Clinical Fellowship. ME receives research support from the UK National Institute of Health Research University College London/University College London Hospitals Biomedical Research Centre. PJC receives funding from The Orchid Charity and Barts & The London Charity. **Contributors:** AS, JHvdM and PJC were involved in the study design. AS prepared and analysed the data. JHvdM advised on statistical issues and data presentation. All authors were involved in data interpretation. AS and PJC drafted the report. All authors revised the report critically for intellectually important content, and approved the final version to be published. AS and PJC had full access to all of the data and are guarantors. **Funding:** Isaac Shapera Trust for Medical Research, The Orchid Charity, Academy of Medical Royal Colleges. **Competing interests:** ME has received funding from GSK, Sonacare, STEBA Biotech and Sanofi-Aventis, outside the submitted work. He acts as a consultant to these companies and has received honoraria for speaking and organising and participating in educational activities. JHvdM has received a 1-year unrestricted research grant from Sanofi-Aventis, outside the submitted work. **Provenance and peer review:** Not commissioned; externally peer reviewed. **Data sharing statement:** No additional data are available.

During the COVID-19 outbreak, healthcare professionals are exposed to a high-risk of infection and mental health problems [@bib0001], but also fear of contagion and spreading the virus to their families [@bib0002]. In fact, considering them as individuals implies looking beyond their function as frontline responders and taking into account their societal role as parents, spouses and offspring. While work-family balance is a common challenge for caregivers, this crisis comes with unprecedented threats. Spouses of healthcare professionals live in a state of fear as they\'re well aware of the actual shortage of human resources and protective equipment, in addition to dealing with an overload of responsibilities. In fact, parenting children alone in the absence of school or daycare is being discussed as one of the burdens of the COVID-19 outbreak with organizations such as the WHO providing instructions to handling "parenting stress" [@bib0003], yet recommendations on parenting for healthcare professionals and their spouses remain undiscussed. Furthermore, as up to 40 % of healthcare workers marry each other, the childcare crisis may cause greater impact. Even relatives\' support may not be available because of strict quarantine. Furthermore, in low-income countries such as Morocco where public social services are lacking, the disruption of traditional solidarity systems may worsen the situation. In this context, children of healthcare workers may undergo, besides the negative effects of confinement [@bib0004], additional emotional instability caused by the absence of both parents. Some Moroccan hospitals suggested in case of both spouses being health workers, to only deploy one of them for the COVID management in order to help alleviate strain on their households. Another burden families are facing is social stigma and dehumanization as they are seen as a source of contagion. Harassment has been widely reported [@bib0005], with repeated incidents of healthcare workers being evicted from their own apartments, deprived services or being excluded. Discrimination has even prompted local medical organizations to requisition hotels, lodges and guesthouses to provide shelter for the healthcare professionals facing these tribulations. Solidarity with healthcare professionals should not only be manifested through "clapping campaigns", but also mitigate the burden on their families, as they are subjected to stress, isolation and emotional distress of exceptional intensity. Several countries including Morocco recognize the contamination of healthcare workers as a professional disability. Further measures such as childcare, social assistance to the spouses and psychological support should also be implemented on a global level in order to protect these families and their future. Declaration of competing interests ================================== We declare no competing interests.

An effective simulation is one that is able to place a user in a situation which is close, if not completely identical to the scenario of which the system is attempting to simulate. The correct term for providing a user with a sense of presence, or "being there", is immersive, where in the world of virtual interaction, is defined as a complex technology that replaces real-world sensory information with synthetic stimuli such as 3D visual imagery, spatialised sound, and force or tactile feedback[@b1]. In manufacturing, the advent of computer numerical control (CNC) machining creates a form of ubiquitous computing, and provides an effective simulation as it becomes a necessity. CNC simulations have been developed in virtual environments for numerically controlled (NC) tool path verification and machining process optimisation[@b2]. However, limitations still exist even though virtual reality (VR)-based systems have already been applied broadly in the manufacturing industry. Firstly, the system is usually costly, requires powerful hardware, and separates the simulation aspect from the machining aspect, meaning the user has to adjust the experience gathered from the 3D graphic environment to the real machining environment[@b3]. Secondly, the system is so tightly integrated that it is difficult to support continuous improvement and lessens flexibility, considering that a fully autonomous CNC manufacturing environment from start to finish involves many steps, not just the machining aspect[@b4]. For example, how does one determine the placement of machines to accommodate the CNC machining aspect, and how does one place the stock material onto the worktable in an autonomous system? This brings forward the demand for a new technology, dubbed augmented reality (AR). AR is a rapidly growing field of research that aims to fully integrate virtual with real environment. AR has been developed since the early 90s, but has only recently been emerging as one of the forefront of technology, mainly due to the rise of popularity in smartphones and tablets[@b5]. Some of the systems that have been developed for production process are related to layout planning[@b6][@b7][@b8][@b9][@b10], product design[@b11][@b12][@b13][@b14][@b15], assembly[@b16][@b17][@b18][@b19][@b20], robot programming[@b21][@b22][@b23][@b24][@b25][@b26], and autonomous machining[@b3][@b27][@b28][@b29]. This proves that AR can be applied in many fields of research and even in consumer products due to the lower system requirement being one of the main contributing factors. By enhancing the users' understanding and interaction with the manufacturing environment, shorter lead time and lower manufacturing costs can be achieved[@b30]. In addition, providing simulation in a real environment partly removes the time-consuming geometric and kinematic modelling of the machine tools and accessories in the real environment, which helps to improve the simulation efficiency. Related Studies =============== The field of visualisation is stepping into a new era of development, with emerging hardware support by large corporations due to the realisation of a promising future associated with this technology. It has been applied in mobile robotics, entertainment industries, and even injury treatment[@b31][@b32], where each of them is derived from enhanced computer graphics that often uses multi-agent systems[@b33]. This is due to the degree of immersion it provides, which is a key factor in any form of simulation. A successful simulation is when the system is able to fully immerse the user in its surroundings for a better understanding. Immersion comes in many ways, and visualisation, as the name implies, improves that through sight. In the manufacturing industry, the use of robotic arms or computer-controlled machines always require a form of programming to teach them the actions that need to be taken depending on the situation. If this task is improved or enhanced visually[@b34][@b35], then it is safe to say that simulating the automated system will yield higher productivity due to higher immersion and intuitiveness. This section will discuss the past and present developments of the application of virtual or augmented reality in layout planning, robot programming, and machining, to determine the existing gaps. Layout Planning =============== The design of a facility's layout is associated with the allocation of machines, work cells, and departments which play a role in ensuring an efficient and effective operation[@b36]. Many researches were conducted to develop simulation models in the manufacturing system design, as engineers need to reduce any uncertainties present, such as assessment errors which are hard to determine in traditional facility layouts. A recent survey was conducted to establish the problems in today's layout planning process and it was found that graphical tools are able to create a more efficient and attractive environment that can replace existing planning processes[@b37]. Since flexible manufacturing system (FMS) is a system that integrates its elements tightly, the relation between them is often hard to compute[@b38]. Therefore, there is a demand for an analysis method that avoids any substantial loss in labour time, money, and resources. AR was used to aid the planning process of manufacturing systems with the key advantage of modelling 3D objects in the actual factory. By using AR as a form of user interface, any user will be able to freely manipulate the overall layout design on a table-top, which is extremely user-friendly[@b39]. A 2D view of a system is often not easy to understand and evaluate. Therefore, virtual systems offer a depth perspective that is not possible for a 2D view to provide, while at the same time it ensures the ability to re-layout existing factory layouts[@b6]. Moreover, it is worthwhile to mention that an AR system can achieve these benefits without any additional computing cost of a VR system. At this moment, AR technology has found a place in collaborative design work, maintenance, assembly, robot path planning, CNC simulation, and of course, plant layout planning[@b30]. Specifically, it was dubbed as AR-based factory layout planning (FLP) systems where it allows users to lay out virtual objects to integrate human intuitiveness with layout design process. Industrial Robots ================= Kinematic modelling of an industrial robot is a vital part of this study as the main aim of this module is to manipulate a virtual robotic arm effectively in an AR environment. The Denavit-Hartenberg-based (D-H) method is the most popular approach for kinematic analysis[@b40] and is used for the robotic arm manipulator in this study. The mathematics and notations of the robot's forward kinematics can be best determined through the D-H method regardless of its sequence or complexity, and the location of the joint is dependent on the previous joint's location which can be calculated using transformation matrices. Integration between layout planning and robot programming was achieved in the past with VR, where a modular system was developed that was able to pick and place objects that was planned earlier with the VR-layout system[@b9]. The system was similar in a sense that it was modular and encourages offline robot programming. However, a VR system is substantially different from AR in terms of execution, and although it has proven to be efficient, the system requires a higher computational capability and it does not cover the end product of what was being manipulated in the work cell. For example, the application and advantages of utilising AR in the field of industrial robot programming was discussed while presenting a novel approach towards planning a collision-free path in an unprepared environment[@b21]. The end-effector is always the main concern for robot programming; therefore, a Collision-Free Volume (CFV) was generated as a form of constraint to the position of the end-effector and to avoid collision with nearby objects. CNC Machining ============= Virtual CNC machining might be a new concept to some, but several studies have already analysed its possibility in application in industries, and some have already been applied. A study was made on the application of virtual machining in relation to its structural analysis as well as the challenges faced in the ongoing research in this field of technology[@b41]. The efficiency and capability of a machine tool highly depends on its structural dynamics, kinematics, CNC system, and the machining process itself. The technology we have today allows for the prediction of tool collision and path error checking by graphical means, but accuracy still remains a primary concern because a successful machining process is often the result of many more parameters that needs to be considered, such as chatter vibration. Existing virtual systems were reviewed in a recent article that covers from VR-based systems to mathematical modelling and NC-based simulation[@b42]. It was found that an unsolved issue until this very day is achieving flawless real-time simulation, even with the aid of web-based technologies. Much work is still required as human-computer interaction is a multi-disciplinary task with various approaches. NC-based simulations have found a place among the research community for some time[@b43], and image-space Boolean operations were even used to simulate cutting process in real-time since more than 20 years ago. A study then proved that shaded computer graphics were central in the design of solid models[@b44]. To achieve real-time shaded display, an extended Z buffer, or frame buffer structure was used. However, real-time changes in virtual models are still a challenge to overcome due to the low efficiency of Boolean operations in solid modelling. To further improve this, hardware inclusion like a 3D display is sorely needed, which also leads to the application of AR in CNC. The developed Augmented Reality Computer Numerical Control (ARCNC) system allows the operator to observe *in situ* simulation of the ball-end machining operation on a 3-axis vertical CNC machining centre, and the interaction between the real cutter and a virtual workpiece[@b3]. The system consists of the CNC machine, a camera, a display device, and a desktop PC for processing, where a physical simulation is performed with an enhanced dexel-based model. Integrating AR in the rendering of a cutting simulation on a 3-axis CNC milling machine was also discussed where the cutting force was estimated based on a volumetric model-based material removal rate (MRR) and a fully interactive panel[@b29]. This means that the system architecture is divided into the AR-assisted CNC simulation and a monitoring station. The simulation portion renders either online or offline cutting simulation and the monitoring module executes the real machining process and monitors the information. The versatility of AR is further explored when it is used for validation purposes[@b45]. Since ARToolKit can be programmed to recognize and detect problematic cases in machining, as well as overlay the process information in real-time to the operator, it can be used to validate the NC-path of complex 5-axis milling machines. CNC simulation is a form of automated machining, however, rarely does a system cover the steps and procedures taken before the machining is conducted. For example, the procedures involved include the methods of which the workpiece is placed on the worktable or the placement of machines in a work cell to facilitate a full production process. Therefore, the inclusion of a modular-based architecture greatly expands the scope of the system. For layout planning and industrial robot programming, both these systems can benefit greatly from AR. Planning a layout for machine placement can be difficult to visualise especially when factors such as the effects of the machine arrangements towards the production line, material travel time, and space required must be taken into account. With regards to robot programming, stopping a production line to programme a robot is extremely costly, and therefore a new form of offline programming can be more beneficial. The interaction between these systems creates a complete modular-based training system which are lacking for engineers currently. Therefore, the goal of this research is to develop a complete modular augmented reality based system that simulates the full planning, control and machining phase for training and education. Finally, the practicability of the developed system needs to be investigated by comparing it to commercially available software tools and through result validation. To summarise, [Table 1](#t1){ref-type="table"} compares the overall features and scope of the proposed work compared to previously developed related research. Methods ======= The research is divided into three separate programme modules with its own features which contribute to the overall system, as well as being able to stand alone as an independent system by itself. The planning module focuses on the initial planning stage, which is the application of AR technology in planning the layout of machine placement to optimize the material travel time and area required. The robotic arm module models an AR robot arm with inverse kinematics to pick a workpiece from a conveyor and place it into the 3-axis vertical CNC milling machine. The virtual CNC finally simulates the material removal with collision detection and generates a G-code programme for the actual machining operation. [Figure 1](#f1){ref-type="fig"} shows the full procedure for an engineer to follow should he or she apply this training system. ARToolKit is used to create a running program that can generate AR content through a marker-based tracking method[@b46][@b47]. It is essentially a software library for building an AR environment that is rendered with OpenGL. This is the basis of all AR-generated content in this study, where different markers perform different tasks based on the module's program. Microsoft Visual C++ 2008 express edition is used to compile and debug the C++ codes that will run synonymously with the aid of ARToolKit and OpenGL. On the hardware side, a personal computer is sufficient to run the program and AR environment. Complementing that is a head-mounted-display (HMD) that acts as the display system and a webcam for tracking and registration. The markers, which are symbols and patterns with specific functions which can only be recognised by the AR program, will be placed in its respective working environment. For the layout planning module, a tabletop system is sufficient to calculate the space and time required, but the robotic arm module markers are placed in the robotic lab with an up to scale virtual robotic arm superimposing a physical one for a proper distance estimation. The CNC simulation markers are placed on the CNC machine itself for an *in situ* simulation process. All of these markers must be placed in such a way that it is clearly visible to the camera under direct lighting, and the camera must be placed at a suitable height, as the images of the marker will be blurry if the camera is not adjusted to the correct height. The camera is mounted on an adjustable tripod facing the marker placements, and connected to the laptop. [Figure 2](#f2){ref-type="fig"} shows the physical setup for each of the module at its respective environment. The individual in this manuscript has given written informed consent (as outlined in PLOS consent form) to publish these case details. When the camera sees the marker in the real world and is captured in a real time video, the program calls out the specific function associated with the marker pattern. The virtual overlay of the model will appear as though on top of the marker, viewable on the HMD. This means that the virtual content has successfully merged with the real world imagery. AR Programming -------------- The fundamental aspect of AR programming is to generate 3D content, which all starts from a single point. This is especially useful in cases where a point such as the robot arm end-effector or the till of the milling bit needs to be visualised. The OpenGL function *glVertex3d* draws a vertex in space where the size and colour can be defined by the user. By creating at least 2 points, a line can then be visualised to connect them. This is used for cases where the path needs to be visualised, such as the material travel path between machines. *GL_LINE_STRIP* draws a line automatically by connecting two vertices together where the line width and colour again is fully adjustable. Creating full 3D models in the AR environment is achieved by importing a CAD file in stereolithography (STL) and read by the program[@b48]. STL files actually contain the coordinates of the numerous triangles used to build the model. Hence, the program needs to recognize and read these data. A *ReadSTL* and *DrawSTL* function is created which can recreate the CAD models as fully rendered AR content. Any form of data acquired from the simulation needs to be exported or generated in some way. Otherwise, it would be difficult to extract live data during the simulation, especially when it is being updated continuously. The program must be able to extract the data from the simulation through a user input and place them into a separate file that can be opened with a text editor. The user input is assigned to a mouse click, where the *GLUT_RIGHT_BUTTON* and *GLUT_DOWN* functions state the condition when the right mouse button is clicked once. This works in a decision making algorithm where it is paired with a *saveCoordinate* function that prints out the coordinate data to a separate output file. All three modules utilise this code function for the user to obtain and edit the parameters obtained in the simulation. The final set of codes that plays an important role in the entire system is the collision detection algorithm. This refers to the ability to detect objects which are within certain proximity and is achieved by first calculating the relative distance. A more advanced version of the code is applied in the CNC module and will be explained in later sections. Collision detection is extremely important in the manufacturing field because accidents that involve collision between human beings and machines can be fatal. In this system, it is applied to detect collision between machines during layout planning, to recognize the pick-and-place features for the robotic arm, and to simulate material removal. This is achieved with a series of decision making that sets a minimum acceptable distance value and checks if the distance between markers is equal or more than the said value. Otherwise, collision is said to have occurred and the models are rendered red in colour to depict that. Work Cell Layout Planning ------------------------- The first module, the work cell layout planning, incorporates AR to aid in the development of a flexible manufacturing cell (FMC) by superimposing 3D models of machines into the physical environment while taking into consideration spatial constraints and collision detection[@b10]. Each of the 3D models present in this module can in fact be scaled to any value. Therefore, the user is free to consider the actual dimension and area required for each machine. However, for ease of analysis and case study, the machines are scaled into a size that is appropriate for a commercial webcam to capture all the markers representing the machines, while still having the models present in the camera's field of view (FOV). A VR system may also freely scale objects, however, the high computational requirement and lack of spatial awareness of the real environment becomes the limiting factor. Four types of layout are analysed, namely the straight line, U-shaped, S-shaped (serpentine), and semi-circle-shaped environment. A data structure in extensible mark-up language (XML) is then developed to record the information regarding spatial relationship, material travel distance, area occupied, processing time, and sequence of operation. The first marker placed serves as the world coordinate, as well as the first machine in the production line. Every subsequent marker will act as the next machine, and the distance between that marker and the previous one will be the relative distance which exists between each virtual machine, as shown in [Fig. 3](#f3){ref-type="fig"}. Since the area required is also taken into consideration, the program must be able to identify which marker has the longest distance relative to the reference marker. By doing so, the system knows the largest possible area that the work cell will require; assuming the space given is rectangular or square in shape. The integrated collision detection code causes a change in colour of the virtual machines when collision is registered as shown in [Fig. 4](#f4){ref-type="fig"}, letting the user know that the machines are placed too closely together. The determination of the best layout will be conducted at the case study section, where the four different types of layout mentioned previously are tested to determine its effect on the cycle time in line balancing. Line balancing is the key method in designing the most efficient process that is in line with the expected volume or demand of a product, but it rarely takes into account the time required for materials to move between stages and how this affects the cycle time which focuses on processing time[@b49]. In fact, line balancing is computed by finding the required number of stages based on the cycle time, which by definition is the time taken for a product to emerge from a stage. Cycle time, *t*~1~ is computed based on the available time, *t*~*a*~ and the demand, *d*. The value of *t*~*a*~ and *d* is determined by the user based on their specific requirements. Next, a product requires several operations to manufacture or assemble, and each operation has its specific time, *t*~*operation*~ required to complete it depending on the operation's complexity and requirements. The total number of stages, *S* can be found by dividing the total work content, *T*~*n*~ with *t*~1~. Each machine in the AR environment is treated as a single stage, where the system then computes the material travel distance with the travel speed is decided by the user. Therefore, the material travel time, which is the time taken for a material to move from the first stage to the last stage, *t*~m~ can be found. *t*~m~ can then be added back into *T*~*n*~ to obtain the total operation time. The new cycle time, *t*~2~ which is computed by considering the material travel time represents a more accurate cycle time as it sums travel time with the actual work content. This allows engineers to carefully consider which layout arrangement is most suited for their required operations because the inclusion of standard time for material flow reduces the risk of late delivery of the final product[@b50]. The full formulation is shown below. If Then With the addition of *t*~m~, New cycle time, Robotic Arm Pick-and-Place Operation ------------------------------------ In a pick-and-place operation, the arm does not perform a specific operation like welding, soldering, and so on. Therefore, the focus on the kinematic study of the robot is important to obtain accurate modelling when creating a virtual robot arm[@b51]. This module emphasises the robot's kinematic study based on the KUKA KR 16 KS robot as well as the functions used to obtain a snapping visualisation to pick and place a virtual workpiece. Pro-Engineer is used to model the robot to scale and joint by joint, then assembled together in OpenGL to create a full virtual robot arm as shown in [Fig. 5](#f5){ref-type="fig"} where each joint can be manipulated at a variable angle. According to D-H kinematics, each compartment of the robot is first assigned a coordinate frame with the origin assigned to the top surface of the pedestal. The primary goal is to obtain the angle of each joint which results in the end effector position. These angles can then be used on the physical robot arm programming. A D-H coordinate frame consists of four parameters, *a*, *α*, *θ*, *d* which are the link length, link twist, joint angle, and link offset respectively. The linkage is illustrated in [Fig. 6](#f6){ref-type="fig"}, while [Table 2](#t2){ref-type="table"} demonstrates how the parameters are linked. The general equation for forward kinematics is the product of the matric transformations from frame 0 to frame 7. where each value of represents each joint. This gives us the formulation for the forward kinematics as well as the end effector position. *P*~*x*~, *P*~*y*~ and *P*~*z*~ represents the end effector coordinates. where *c*~*n*~ and *s*~*n*~ represents cosine and sine for the respective matrices. However, inverse kinematics is required to obtain the joint angle of the arms. Once the angle of each arm is determined, the robot can then use these values to obtain the desired end effector coordinate. A limitation is placed on *θ*~4~ and *θ*~6~ to reduce the probability for an error to take place, since these joints are twist joints which should not affect the consecutive joint's coordinate, and that the end effector will simply face downwards. [Figure 7](#f7){ref-type="fig"} shows the free body diagram of the other joints, where *θ*~1~ is shown in the X-Y plane rotating about the Z-axis, and *θ*~2~, *θ*~3~ and *θ*~5~ are shown in the X-Z plane rotation about the Y-axis. This method of computing the angles are detailed in a recent paper that explains the joint assumptions that were made[@b51]. Once the kinematic modelling of the robot is completed, the pick and place operation is initiated. The teach pendant must be able to manipulate the virtual stock in space, to show that the robot arm is picking and placing the stock around. This is called snapping, where an object immediately takes a position in space when an operation is performed. With a single mouse click, the virtual object takes the position of the tip of the manipulator with the condition that it is colliding. The user can also choose to drop the object anywhere in space simply by releasing the mouse button since the program is designed to continuously update the most recent position of the virtual object. (x, y, z) refers to the current position of the virtual object, while (X, Y, Z) refers to the position of the teach pendant tip at the moment the mouse button is clicked. The resulting effect is shown in [Fig. 8](#f8){ref-type="fig"}, whereas the function algorithm is shown in [Fig. 9](#f9){ref-type="fig"}. CNC Machining Simulation ------------------------ The main purpose of the final code module is to machine out the final product based on the design of the user. To achieve this, a modified collision detection system, machining parameters, heads-up-display (HUD) and G-code generation will be integrated together. The previously utilised collision detection algorithm simply calculated the distance between two points based on the formulation stated below. However, this formula implies that both of the points are considered as the centre point of a sphere-shaped object since the distance between them is constant. If a maximum allowable distance was set, such as a value of 100 cm, this will be equivalent to two spherical object of radius 50 cm touching each other at a single point. Therefore, this algorithm can only be applied to find collision between two points or spheres with no edges or corners present. In this study, it is assumed that the stock workpiece is a single block of material and the cutter is bounded by a rectangular box, which therefore requires a collision algorithm suitable for cuboid objects. Furthermore, the nature of vertical milling requires a variable depth from the top surface of the workpiece to visualize the depth of cutter engagement. The axis-aligned bounding box (AABB) algorithm is used to fulfil these requirements. As a bounding box or a typical 2D box is made of four sides, the routine requires four conditions which are the four corners. The intersection method is based on the simple logic in [Fig. 10](#f10){ref-type="fig"}. To apply this logic into the simulation, the boxes must first be transformed into 3D cubes. Both the stock and cutter will be treated as a bounding box. However, an inaccurate visualization will occur if the typical AABB method is used. If the entire cutter is placed into the workpiece, AABB collision will cause the visualization of only the intersection between the two boxes, which means a floating black box inside the workpiece. In an actual milling operation, this will result in a depth of the cut from the surface of the workpiece until the tip of the cutter, assuming that an operator actually cuts into a material until the depth of cutter engagement is higher than the actual length of the cutter. In other words, the depth of cut needs to be set as a variable, unlike the width and length. This is reflected in the code where the z-axis value only evaluates the lower surface and not the top of the cutter. Instead, the top is associated with the top surface of the workpiece instead. A variable *dcut* is defined as the depth of cut, or the difference in height between the top surface of the workpiece and the lower surface of the cutting tool. The uniform space decomposition (USD) method is also used to represent the stock workpiece so that it can be visualised as material being cut. In a USD-based method, the stock workpiece is represented as cubes, spheres, or any shape of the same size[@b52]. This means that the entire stock is made up of smaller cubes, where the size of each cube determines the resolution of the object. When the cutter or tool passes through the cubes, cubes which intersect with the tool during the process are rendered black, and eventually the volume of the black cubes represents the outcome of the machining process. [Figure 11](#f11){ref-type="fig"} shows the USD-based stock, together with the visualisation of depth of cut. The parameters involved in the simulation aids the user in understanding its effect through real-time visualisation depending on the current operation. These parameters are divided into the user input and the calculated output. Unlike the robot arm, kinematic modelling of the actual CNC machine is not included in this code module and thus, the axis movement and trajectory planning are borrowed from the physical machine by placing the markers directly on the machine itself. However, not all CNC machines calculate the machining parameters for the user, and thus are included in this program. N  =  RPM of Cutter, or Spindle Speed n  =  Number of Teeth on Cutter W = Width of cut (may be full cutter or partial cutter) T  = depth of cutter engagement V  = cutting speed (Handbook value) L  = Length of pass or cut f~m~  = Table (machine) Feed, or Feedrate f~t~  = feed per tooth of cutter, or Chip Load (Handbook value) D  = Cutter Diameter L~A~  = Approach Length L~O~  = Length of "OverTravel", where the turret moves beyond its boundaries Then, Spindle speed, Let Then, Cutting time, The *sscanf* function in the code allows the system to read values from an external file in the same directory in the program. For the user to enter the necessary parameters, a separate file called *machining_parameters.dta* is included where he or she simply needs to input the first initial six values which are the cutter diameter, workpiece thickness, width of cut, cutting speed, feed per tooth, and number of tooth, to compute all the necessary parameters. [Figure 12](#f12){ref-type="fig"} shows the file with a detailed explanation of each parameter and what they represent. Therefore, the user does not need to constantly input the values each time and just change them in the separate file should the need arise. The addition of a HUD is extremely useful when virtual content is involved in any context. It extends our knowledge of the current operation when it is performed, and continuously updates itself with the current situation as well. The information overlay covers the current tool state with related G-code, spindle rotation direction, coolant condition, and all of the aforementioned parameters. The live update feature ensures key information like current tool coordinate, MRR, and cutter depth engagement constantly changes to reflect the current machining conditions. To achieve this, the perspective of the model needs to negate the initial global origin which will cause the HUD to move around if the "Hiro" marker is moved. Additionally, a semi-transparent background is used to increase the visibility of the words without obscuring the operation much. The following functions were utilised: glMatrixMode (GL_PROJECTION); glMatrixMode (GL_MODELVIEW); glEnable (GL_BLEND); glBlendFunc (GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); Lastly, the *printw* function prints all the relevant information into the AR scene. The resulting effect is as shown in [Fig. 13](#f13){ref-type="fig"}. With the models accurately visualised, the system needs to generate G-codes which can then be used for the actual CNC machining. One of the key features of the simulation system is the ability to generate G-code blocks based on the virtual environment and the placement of the cutting tool relative to the workpiece coordinate system (WCS). Despite the system supporting 3-axis only, complex operations can still be carried out, evident by past applications even for non-uniform surfaces like sculpturing with NC machining[@b53]. Furthermore, extension to 4- and 5-axis CNC machines can be done once 3-axis machining is properly established[@b54]. The list of supported G-codes is shown in [Table 3](#t3){ref-type="table"}. These codes can all be seen on the HUD with visual cues, such as the workpiece becoming blue in colour when the coolant is switched on. The key values in a G-code programming, which are the X, Y, and Z values are tied to the *saveCoordinate* function which is specifically designed to operate with the mouse input to save the current coordinate when the mouse button is clicked. An example of the text file is shown in [Fig. 14](#f14){ref-type="fig"} which is generated based on a total of 12 mouse clicks. Case Study ---------- The case studies are designed as a form of validation process to observe how much the parameters deviate over conventional tools and to reflect the error present in the system[@b26][@b29][@b48][@b55][@b56][@b57]. This is to prove that the developed system has the potential to replace them, with the added benefit of it being more immersive, realistic, having a better sense of depth, with real-time information feedback, and a better simulation experience overall. Since the modularity of this system is emphasised, the case study is conducted in a fashion where each module is treated as a standalone system. Validation software tools like Mastercam and Kuka Sim Pro play a major role as they are primarily needed to validate the results generated by the AR environment. Layout Planning Case Study -------------------------- A case study is conducted based on the manufacture and assembly of a computer case. For ease of calculation, input values are kept at the lower range to reduce the computed value of the number of stages, since it can range from 0 to 100 stages in an actual production line. For this particular case study, *t*~*a*~ = 8 hours/day, *d* = 800 units/day and *n* = 6 operations. Since each operation has its own required time, *t*~*operation*1~ = 45 seconds, *t*~*operation*2~ = 18 seconds, *t*~*operation*3~ = 22 seconds, *t*~*operation*4~ = 32 seconds, *t*~*operation*5~ = 20 seconds and *t*~*operation*6~ = 43 seconds. The program uses these values to find *t*~1~ and *S*, which is equal to 36 seconds and 5 stages respectively. The goal of this case study is to acquire the best possible route with minimum material flow distance and the least space required. Several line shapes were evaluated, that includes the straight --line, S-shaped, U-shaped, and semi-circle-shaped as shown in [Fig. 15](#f15){ref-type="fig"}. Furthermore, two forms of orientation which represents an automated and a manual line, which are machine-centred and operator-centred respectively, are both analysed. Based on the result in [Table 4](#t4){ref-type="table"}, the lowest value of the total distance travelled is at 328.52 m, which is the operator-oriented method with a semi-circle-shaped arrangement. Additionally, this particular arrangement also scores the least distance travelled for a machine-centred oriented operation. Assuming the speed to be 0.33 m/second, this equals to a travel time of 108.4 seconds. We can then find *t*~*2*~, according to [equation (3)](#eq3){ref-type="disp-formula"}. Robotic Arm Case Study ---------------------- Validation of the developed module is conducted by utilising the virtual machines from the layout planning module to create a robotic work cell. The operation involves the picking and placing of a block of material around the various virtual machines. The user guides the end effector of the virtual robot arm by manually pointing at the locations for picking and placing using the teach pendant. The coordinate and angle of the arm at that point is then saved into a separate file. The saved angle values are input into Kuka Sim Pro, where each of the points can be compared and validated. [Figure 16](#f16){ref-type="fig"} shows multiple views of the work cell used for the case study. Kuka Sim Pro is an offline programming software that fully simulates the activities of the robot arm, and therefore is no different from the actual coordinate inputted into the teach pendant of the physical robot. The generated angles of each arm from the AR environment are input into Kuka Sim Pro to generate the corresponding coordinates of the end effector. The results are shown in [Table 5](#t5){ref-type="table"}, where the simulation is validated with a maximum error of 3.83 mm. CNC Machining Case Study ------------------------ The machining operation will be conducted in a fashion similar to Mastercam, as well as validated with Mastercam to determine the accuracy of the operation. A CAD model was created using Pro-Engineer with the overall dimensions being the same as the workpiece in the simulation. The imported 3D model carries over the dimensions of the CAD model with the same machined slots, as an overlay on top of the original virtual stock. This is essentially how Mastercam works, however Mastercam does not provide an *in situ* simulation system. The generated G-code is then compared with the Mastercam G-code and any inconsistencies were observed. [Figure 17](#f17){ref-type="fig"} shows how the simulation process was carried out on a physical milling machine to ensure an accurate axis movement of the cutter. The case studies will be separated to test the machining capabilities of each axis individually. The designed stock size is 200 mm × 200 mm × 100 mm, as well as the dimensions of the virtual stock in the AR system. The cutter marker can roughly be placed on the spindle area of the physical cutter as long as it remains sturdy. Similarly, the stock marker does not need to be placed on the actual vice with the mounting coordinates known. This is because its placement will not affect the generated G-code, as the values are calculated relative to the virtual workpiece and the virtual cutter. Conducting the case study on a physical milling machine is merely to provide an accurate axis movement for the cutter while at the same time, providing an actual machining or manufacturing environment to the operator. [Figure 18](#f18){ref-type="fig"} shows the machining simulation for cutting about the x and y-axis. From the G-codes generated through Mastercam, the main machining values are then extracted and compared side by side between both the simulation systems. The graph shown in [Fig. 19](#f19){ref-type="fig"} illustrates the error present in both the axes on its deviation from the Mastercam results, which is assumed to be ideal and error-free. The same procedure is used to find the error present when machining about the z-axis and finally complex machining is performed for all three axes, as shown in [Fig. 20](#f20){ref-type="fig"}. The z-axis machining case study drills four holes with different depths at an increment of 10 mm. The 3-axis machining case study is carried out because typically, a finished product requires machining of at least all three of the axes, therefore this validation is the most accurate representation simulated of a real product. The cutting is performed on the edges with a slope-like design for a variable depth. It is observed that the highest deviation from the Mastercam software for all the case studies is point 7 for the z-axis machining, with a value of 5.8 mm which is considerably higher when compared to the x-axis, y-axis, or 3-axis machining. Discussion ========== For the layout planning module, 57.68 seconds is a more accurate representation of cycle time using the semi-circle shaped arrangement. For the total area required, the straight line layout requires the least area, at 19833.14 m^2^. A smaller space would obviously be favourable; however, this depends on additional factors like the amount of allocated space in the first place, as well as the space required for other work cells. As for the robotic arm module and the CNC machining module, the highest deviation errors produced were 3.83 mm and 5.8 mm respectively. It needs to be understood that in both cases, the presence of these errors is unavoidable even in commercially available software programmed by first party developers. Simulation will always differ slightly with its physical counterpart; therefore, a certain error value needs to be decided upon as acceptable in this new form of simulation. Judging by the accuracy of ARToolKit itself and coupling it with the implemented algorithms, these errors were deemed acceptable at this point of time. The formulas for the transformation and rotation matrices are correct, because there would be a large deviation among the results if the supporting formulas are wrong. Additionally, with regards to the robotic arm module, hand jitter by the user also plays a role because the human hand is never 100% static[@b58]. Most of the illustrated figures are shown as being conducted with a simple computer monitor displaying the AR content. This is so that each generated model can be clearly illustrated on the monitor. However, using a HMD is highly recommended, as shown in [Figs 2](#f2){ref-type="fig"}(c) and [16](#f16){ref-type="fig"}. This provides the user with an immersive view of the AR content superimposing his or her FOV. The exact HMD model used is the Vuzix Wrap 920AR which is shown in [Fig. 21](#f21){ref-type="fig"} that provides a video see-through vision with built-in cameras and display. This essentially means that the HMD replaces both the camera and computer monitor for the user. While it is not a requirement, the usage of a HMD can greatly benefit the *in situ* simulation aspect of AR, where it is only natural and intuitive to look at the machine when performing the simulation, as opposed to referring to an external monitor. Tasks like pick-and-place and machining also requires viewing from different angles to assess the situation carefully, which is a benefit from utilising a HMD. Integration of all these modules as a complete system merely involves the steps illustrated in [Fig. 1](#f1){ref-type="fig"}. As of this moment, the case studies were meant to showcase the ability of each module to act as a standalone system. This freedom that is provided to the user is an important factor in this study, so that using them sequentially is always an option, as for using them as a standalone system or in any other sequences. Otherwise, inclusion of additional dependencies that ties these modules together in a program will become a limiting factor instead, and goes against its modular nature. A separate work is possibly required for this task. Conclusion and Future Works --------------------------- In this paper, a virtual modular-based automated factory operation system for teaching and learning has been developed. This is especially useful to bridge the gap between simulation and actual factory operations, as one of the key benefits of AR is superimposing the actual environment for a better sense of placement. The modular nature of the system allows the scope to be further expanded through additional modules where each plays a part for a complete training system that facilitates automatic machining. Modules like the layout planning and pick-and-place operation teaches the user about the effects of proper machine placement and robot programming respectively, and how it contributes to the production of an end product from the start to finish. Finally, with any simulation system, the developed AR system aims to eventually reduce the lead time and increase the productivity by being more intuitive and immersive. This study can be further improved by including a graphical user interface (GUI) which is an improvement over the HUD implementation. Further studies should also be conducted to identify and investigate the errors in the results[@b59] which can be due to several factors such as, lighting conditions, marker design and hand jitter. Currently, the errors present are still acceptable in the methodology used for validation or learning purposes, but they most definitely need to be addressed to increase user acceptance. Moreover, with regards to the CNC system, circular interpolation was not supported and the AABB algorithm has caused the visualisation of material removal to be only rectangular in shape, which can be both further improved. Finally, a tighter integration while maintaining the modular characteristics of the system would be a huge improvement, such as linking the layout planning module directly to the robotic arm and CNC machining simulation for a better material flow. With these improvements, previously developed methodologies for validation[@b60] can be used to determine the simulation accuracy. In terms of measuring immersion, a separate study is required to determine the level of which immersion can be quantified. It was stated that this depends on many factors like FOV, field of regard, display size and resolution, stereoscopy, head-based rendering, realism of lighting, frame rate, refresh rate, and so on[@b1]. It is best to acquire other forms of AR systems so a direct comparison can be made for the level of immersion each provides. Additional Information ====================== **How to cite this article**: Pai, Y. S. *et al.* Virtual Planning, Control, and Machining for a Modular-Based Automated Factory Operation in an Augmented Reality Environment. *Sci. Rep.* **6**, 27380; doi: 10.1038/srep27380 (2016). A million thanks to the Graduate School of Media Design, Keio University and the Department of Mechanical Engineering, Faculty of Engineering, University of Malaya, for providing the necessary facilities to support this study. This work was supported by the Ministry of Higher Education Malaysia with the grant number UM.C/HIR/MOHE/ENG/35(D000035-16001) and the Ministry of Science, Technology, and Innovation with the grant number (03-01-03-SF0980/SF007-2014). **Author Contributions** Y.S.P. conceived the idea and performed the main programming task, H.J.Y. and S.R. supervised the entire process and analysed the results, S.Z.M.D. provided the necessary facilities and tools, and S.Y.P. conducted the case studies and assisted with the programming. All authors contributed to the discussion, modifications, writing, and revision of this article. {#f1} {#f2} {#f3} {#f4} {#f5} {#f6} ![Free body diagram to compute *θ*~1~, *θ*~2~, *θ*~3~, and *θ*~5~ respectively[@b51].](srep27380-f7){#f7} {#f8} {#f9} {#f10} {#f11} {#f12} {#f13} {#f14} {#f15} {#f16} {#f17} {#f18} {#f19} {#f20} {#f21} ###### Comparison table between other related works.   VR simulation[@b2] ARCNC[@b3] Virtual factory layout[@b6] Mixed reality layout planning[@b7] VR-Rocell[@b9] AR Robot Programm-ing[@b21] The authors' research -------------------------------- -------------------------------------------- ---------------------------------- ----------------------------------------- ------------------------------------------- ----------------------- ----------------------------- ---------------------------------------------------------- Technology Used virtual reality augmented reality virtual reality virtual and augmented reality virtual reality augmented reality augmented reality Scope vertical milling only vertical milling only layout planning only layout planning and robot arm interaction robot arm interaction robot arm interaction layout planning, vertical milling, robot arm interaction G-Code -- acts as input -- -- -- -- Outputs G-code CAD Format VRML -- DXF Delmia software STL -- STL Axis/DOF 3-axis 3-axis -- 6-DOF robot arm 6-DOF robot arm n-DOF 3-axis machining, 6-DOF robot arm Material Removal Visualisation none due to high computational requirement supports real time visualisation -- -- -- -- supports real time visualization Layout Planning Method -- -- finds shortest material travel distance Delmia based layout-planning -- -- finds best formation and least area Robot operation -- -- -- Delmia-based simulation pick-and-place CFV for path planning pick-and-place ###### Denavit-Hartenberg parameters. Joint *i* Rotation *α*~(*i*−1)~ Link Length *a*~(*i*−1)~ Joint angle *θ*~*i*~ Link offset *d*~*i*~ ----------- ----------------------- -------------------------- ------------------------ ---------------------- 1 0 0 *θ*~1~ *d*~1~ = 235 *mm* 2 90° *a*~1~ = 450 *mm* *θ*~2~ = *θ*~2′~ + 90° 0 3 0 *a*~2~ = 680 *mm* *θ*~3~ = *θ*~3~ + 90° 0 4 90° *a*~3~ = 35 *mm* *θ*~4~ = 0 *d*~4~ = 670 *mm* 5 −90° 0 *θ*~5~ 0 6 90° 0 *θ*~6~ = 0 *d*~6~ = 158 *mm* ###### List of supported G-codes. G-Code Description ------------- --------------------------------- G00 Rapid Linear Interpolation G01 Linear Interpolation G21 Machine in mm G90 Absolute command M-Code Description M00 Program stop M03 Spindle On Clockwise M04 Spindle On Counter clockwise M05 Stop spindle from turning M08 Coolant On M09 Coolant Off Other Codes Description F Feed rate/Dwell time in seconds S Spindle speed X Code for the X-axis Y Code for the Y-axis Z Code for the Z-axis ###### Total travel distance and area for each layout. Layout Arrangement Total Travel Distance(m) Total Area (m^2^) -------------------- -------------------------- ------------------- ---------- Straight 662.61 671.74 19833.14 S-Shaped 696.79 494.58 43655.28 U-Shaped 553.71 432.40 34225.45 Semi-circle Shaped 476.26 328.52 31472.27 ###### Calculated root mean square error (RMSE) value.  [^1]: These authors contributed equally to this work.

{.220} {.221} {.222} {.223} {.224} {.225} {.226} {.227} {.228} {.229} {.230}

Introduction {#s2} ============ [](#mhcl-08-06-02-box01){ref-type="boxed-text"}With publication of the 5th edition of the *Diagnostic and Statistical Manual of Mental Disorders*, the terms *dementia*, *amnestic*, and *cognitive disorders* were included under the umbrella of neurocognitive disorders (NCDs). The term *NCD* covers a broad list of diagnoses that can affect both younger individuals and older adults. Neurocognitive disorders are labeled as either mild or major followed by their etiologic subtypes. Diagnoses considered to be NCDs include vascular NCD; frontotemporal NCD (Pick disease); and NCD due to Alzheimer disease, Lewy bodies, traumatic brain injury, human immunodeficiency virus infection, prion disease (eg, mad cow disease), Parkinson disease, or Huntington disease. The most prevalent types of NCDs are Alzheimer disease (70% of NCDs), vascular disease (10%-20% of NCDs), and Lewy bodies (3%-7% of NCDs).[@i2168-9709-8-6-284-b01] It is estimated that 50% of the NCD population has mixed dementia, which usually includes both an Alzheimer and a vascular presentation.[@i2168-9709-8-6-284-b02],[@i2168-9709-8-6-284-b03] Behavioral and psychological symptoms of dementia (BPSD) occur in approximately 80% of patients with major NCD.[@i2168-9709-8-6-284-b04],[@i2168-9709-8-6-284-b05] These symptoms lead to increased caregiver burden and early nursing home placement.[@i2168-9709-8-6-284-b06]-[@i2168-9709-8-6-284-b08] In addition, patients with BPSD have faster cognitive and functional decline and increased morbidity and mortality compared with dementia patients without these symptoms.[@i2168-9709-8-6-284-b06],[@i2168-9709-8-6-284-b09] *BPSD* is a broad term with a wide spectrum of possible symptoms, which are illustrated in [Table 1](#i2168-9709-8-6-284-t01){ref-type="table"}.[@i2168-9709-8-6-284-b10] These noncognitive symptoms are divided into 4 clusters: hyperactivity, psychosis, affective symptoms, and euphoria. Common hyperactivity symptoms are agitation, aggression, and irritability. Psychosis symptoms include hallucinations and delusions, and affective symptoms include depression, dysphoria, and anxiety. Symptoms that have been more difficult to categorize throughout the literature include aberrant motor behaviors, disinhibition, decreased appetite, sleep disturbances, and apathy.[@i2168-9709-8-6-284-b11] ###### Behavioral and psychological symptoms of dementia^10^ **Symptom** **Prevalence, %** ---------------------- ------------------- Disinhibition 9-35 Psychosis 10-35 Sleep disturbance 12-42 Anxiety 17-45 Irritability 20-55 Depression 20-57 Agitation/aggression 22-52 Apathy 29-76 Most patients with dementia will experience one or more BPSDs, which can cause distress for the patient and/or caregivers. Therefore, treatment of these symptoms is essential in community and institutional settings. Appropriate assessment of BPSD should involve noting the type, frequency, severity, pattern, and timing of the symptoms. Modifiable contributors, such as pain, constipation, infection, and medications, should be addressed. Nonpharmacologic methods should always be implemented first for BPSD, and these methods should be continued even if medication therapy is warranted.[@i2168-9709-8-6-284-b12],[@i2168-9709-8-6-284-b13] A few examples of nonpharmacologic interventions include gently redirecting the patient, providing structured activities during the day, creating a calm, quiet environment, and providing music therapy.[@i2168-9709-8-6-284-b14]-[@i2168-9709-8-6-284-b16] This article does not intend to dismiss the importance of nonpharmacologic treatments. Behavioral interventions have been shown to decrease medication use significantly among nursing home residents.[@i2168-9709-8-6-284-b17] However, the scope of this article is to review the medication options available when nonpharmacologic therapies are not effective. Patient Case 1 {#s3} ============== A 70-year-old with a diagnosis of major vascular NCD and history of significant alcohol use (currently in remission) is admitted to an inpatient psychiatric unit. The patient was transferred from the nursing home after experiencing increased aggressiveness toward various staff members and the spouse. The patient recently punched a nurse in the jaw and jumped up to strangle the spouse during a family visit. The nursing home refuses to readmit until the aggressive behaviors are under control. Upon admission, the patient is unresponsive to behavioral interventions and continues to be highly intrusive, irritable, and physically aggressive. The patient takes medications for hypertension and diabetes, but none of these medications appears to be a precipitant to the aggressive behavior. All laboratory exams, urinalysis, and vital signs are within normal limits, and the patient was noted to be otherwise healthy on physical exam. Take Home Points: {#sb1} ----------------- 1. Nonpharmacologic methods are the first-line recommendation when treating disruptive behaviors associated with dementia. However, psychotropic medications may be warranted if the behaviors are dangerous or cause distress to the patient. Despite their risks, antipsychotics are the most prescribed medications for treatment of dementia-related behaviors. When antipsychotic use is clinically warranted, dosing and subsequent titration should be approached conservatively. Once the disruptive behaviors are managed, an attempt to taper and discontinue the antipsychotic should be considered within 4 to 6 months. 2. The risks associated with antipsychotics are concerning; therefore, clinicians often look for other pharmacologic treatments for dementia-related behaviors when nonpharmacologic therapies fail. Other medications that have demonstrated small to moderate effects for reducing psychologic symptoms include cholinesterase inhibitors, memantine, and selective serotonin reuptake inhibitors. 3. The use of benzodiazepines and mood stabilizers cannot be recommended for treatment of behavioral and psychological symptoms of dementia because the published data demonstrate a high risk of adverse effects, with a low probability of being effective. Antipsychotics {#s4} ============== Antipsychotics are the most widely studied medications for the treatment of BPSD and appear to have modest efficacy.[@i2168-9709-8-6-284-b18]-[@i2168-9709-8-6-284-b20] The target symptoms for antipsychotic therapy are agitation, aggression, and psychosis, and the medications should only be initiated when these symptoms are assessed as dangerous or distressing for the patient.[@i2168-9709-8-6-284-b13] Second-generation antipsychotics (SGAs) with efficacy data include risperidone, olanzapine, quetiapine, and aripiprazole.[@i2168-9709-8-6-284-b18]-[@i2168-9709-8-6-284-b26] Based on meta-analyses, effect sizes for risperidone, olanzapine, and aripiprazole appeared to be small but statistically significant compared with placebo. Quetiapine was found to have a lower effect size.[@i2168-9709-8-6-284-b20] Overall, the SGAs have demonstrated a 35% improvement in Neuropsychiatric Inventory (NPI) scores, which is above the minimum number of 30% to be considered a clinical observable change. However, mean total improvement in NPI scores was 3.41 points, which is below the threshold of 4 points to be considered clinically significant.[@i2168-9709-8-6-284-b19] Haloperidol is the most studied first-generation antipsychotic (FGA).[@i2168-9709-8-6-284-b27]-[@i2168-9709-8-6-284-b29] However, recent treatment guidelines[@i2168-9709-8-6-284-b13] recommend against the use of haloperidol and other FGAs as first-line agents because of several studies showing a higher risk of mortality compared with SGA.[@i2168-9709-8-6-284-b30]-[@i2168-9709-8-6-284-b33] For patients who respond to antipsychotics, symptom improvement is usually seen within 1 to 4 weeks. Despite being effective, there are several concerns with antipsychotics in the dementia population. A US Food and Drug Administration (FDA) boxed warning exists for all antipsychotics regarding an increased risk of mortality in elderly patients with dementia-related psychosis. This warning was added after the FDA found SGA to be associated with an increased risk of death in a meta-analysis of 17 trials comparing risperidone, quetiapine, olanzapine, or aripiprazole to placebo for treatment of BPSD.[@i2168-9709-8-6-284-b34] In addition to the FDA review, a 2005 meta-analysis that included 15 SGA studies found incidence of death to be 3.5% in the SGA group compared with 2.3% in the placebo group (*P* = .02).[@i2168-9709-8-6-284-b35] Overall, these studies showed the risk of mortality was 1.5 to 1.7 times higher for dementia patients prescribed SGA compared with placebo, with the most common causes of death being stroke, myocardial infarction, and infection. First-generation antipsychotics were later studied and found to have equal or higher risk of mortality compared with SGAs.[@i2168-9709-8-6-284-b30]-[@i2168-9709-8-6-284-b33] Accelerated cognitive decline and increased thromboembolic risk have been associated with antipsychotic use in addition to the other common class effects, such as weight gain, hyperglycemia and dyslipidemia, extrapyramidal side effects, and orthostatic hypotension.[@i2168-9709-8-6-284-b36]-[@i2168-9709-8-6-284-b43] Additionally, several antipsychotics, such as olanzapine, clozapine, and chlorpromazine, antagonize muscarinic receptors, which can potentially worsen cognition and BPSD.[@i2168-9709-8-6-284-b44] One of the largest antipsychotic trials[@i2168-9709-8-6-284-b26] (n = 421), Clinical Antipsychotic Trials of Intervention Effectiveness--Alzheimer\'s Disease (CATIE-AD), found SGA modestly effective for BPSD (mean NPI improvement of 15 points), but 82% of the study participants stopped the antipsychotic before the 9-month study ended. Risperidone, olanzapine, and quetiapine had higher rates of discontinuation because of adverse effects compared with placebo (*P* = .009). Overall, time to discontinuation of treatment for any reason was no different for patients taking an antipsychotic versus those assigned to placebo. Rates of discontinuation were 77% for risperidone, 80% for olanzapine, 82% for quetiapine, and 85% for placebo.[@i2168-9709-8-6-284-b26] Because of the risks of adverse effects and mortality, antipsychotics should be used cautiously for the treatment of BPSD following failure of behavioral interventions.[@i2168-9709-8-6-284-b44] Interdisciplinary teams should consider alternative treatments, such as additional nonpharmacologic methods and safer medication regimens that target the specific behavioral symptoms. However, the benefits of antipsychotics outweigh the risks in cases where patients\' behaviors are dangerous or impede needed medical care, as noted in case 1. When choosing an antipsychotic, FGAs are generally avoided because of a higher risk of extrapyramidal side effects and mortality compared with SGA. Traditionally, clinicians decide between risperidone, olanzapine, and quetiapine when choosing an antipsychotic because these medications are the most studied for treatment of BPSD. The use of aripiprazole is limited in practice because of its longer half-life (75 hours) and titration schedule, but these factors should not prevent clinicians from using the drug completely. A patient\'s concurrent disease states and medications should be taken into consideration when choosing an initial agent. For example, dopamine antagonism can worsen neurologic symptoms for patients with NCD due to Lewy bodies or Parkinson disease; therefore, an antipsychotic such as quetiapine would be considered ahead of risperidone or olanzapine because of its lower affinity for dopamine receptors. Risperidone and aripiprazole would likely be considered before olanzapine and quetiapine for patients who are obese or have diabetes because these medications are associated with a lower incidence of hyperglycemia and weight gain. Antipsychotics should be dosed and titrated conservatively to the minimum effective dose ([Table 2](#i2168-9709-8-6-284-t02){ref-type="table"}). Signs of medication effectiveness include decreased aggression and agitation to where nursing staff can allow the patient to roam safely on the unit, the patient is not resisting medical care, or there is a decreased need for emergency medications. If benefit is not seen within 4 weeks, the medication should be tapered and discontinued. For antipsychotic responders, an attempt to taper and discontinue the medication should be considered within 4 to 6 months of starting therapy.[@i2168-9709-8-6-284-b13] Antipsychotic dose reduction attempts are mandated by the Centers for Medicare & Medicaid Services for patients residing in long-term care facilities. A minimum of 2 gradual dose reductions should be tried during the first year of therapy, with each attempt being 3 to 6 months apart.[@i2168-9709-8-6-284-b45] ###### Antipsychotics dosing recommendations for behavioral and psychological symptoms of dementia Antipsychotic Starting Dose, mg Titration Recommendations, mg --------------- --------------------- ---------------------------------------------------------------------------- Risperidone 0.25-0.5 at bedtime Increase by 0.25-0.5 every 5 d up to 2 per d Quetiapine 12.5-25 at bedtime Increase by 25 every 3 d up to 200 per d Olanzapine 2.5 at bedtime Increase by 2.5 every 5 d up to 10 per d Aripiprazole 2 daily Increase to 5 after 2 wk, then increase by 2.5-5 every 2 wk up to 15 per d Evidence suggests antipsychotic discontinuation can be done successfully. A medication withdrawal study led by Ruths and colleagues[@i2168-9709-8-6-284-b46] found no difference in neuropsychiatric scale scores 4 weeks after patients were tapered off their antipsychotic compared with those who continued antipsychotic therapy. In addition, 85% of patients tapered off their antipsychotic remained medication free 4 weeks later.[@i2168-9709-8-6-284-b46] In another study published the same year, Ballard et al[@i2168-9709-8-6-284-b47] observed no serious effects on behavior or cognition for most patients who tapered and discontinued their antipsychotic. However, a 2012 study by Devanand and colleagues[@i2168-9709-8-6-284-b48] found BPSD to worsen and relapse of symptoms to occur more frequently in patients tapered off risperidone (60%) compared with patients who continued taking the antipsychotic (33%). Patient case 1 is an example of a patient with dementia who is dangerously aggressive despite the implementation of nonpharmacologic treatments. In order to return to the nursing home, the aggression must be treated to where nursing staff and other patients will be safe. An antipsychotic is warranted in this situation. Based on treatment guidelines, SGAs are preferred over FGAs, such as haloperidol. Because of the patient\'s history of diabetes, risperidone and aripiprazole are safer options because they are less likely to worsen hyperglycemia compared with olanzapine and quetiapine. As noted earlier, aripiprazole\'s long half-life and titration schedule make it less practical in an acute clinical scenario; therefore, risperidone would be the antipsychotic of first choice for this patient case. Patient Case 2 {#s5} ============== A 73-year-old with a diagnosis of Alzheimer disease (moderate severity) was admitted to a nursing home 6 months ago. The geriatric psychiatry team has been asked to assess the patient because of concerns about increased depression, delusional thinking, and agitation. No physical or verbal aggression has been noted. The nurses report decreased appetite and refusal to participate in daily activities (which the patient usually enjoys) for the last 3 weeks. Recently, the patient has become more irritable and delusional. It appears the patient thinks the nursing home is their old home and the other residents are intruders. Vital signs and a full laboratory workup plus urinalysis were within normal limits, and the patient denies having any pain. A physical exam notes a thin-appearing person but is otherwise normal. Nonpharmacologic strategies were implemented to target behavioral symptoms as the initial intervention, but these have yielded minimal clinical improvement. The patient is currently prescribed donepezil 5 mg daily for cognitive decline. Sertraline 25 mg daily was added to the medication regimen within the last week to target depressive symptoms. Cognitive Enhancers {#s6} =================== Cholinesterase inhibitors and memantine are primarily used in dementias to help slow cognitive decline. Cholinesterase inhibitors are recommended for neurocognitive disorders due to Parkinson disease, Lewy body disease, and Alzheimer disease, whereas memantine is indicated only for moderate to severe Alzheimer disease.[@i2168-9709-8-6-284-b12] Cognitive enhancers may be useful in mixed dementia because many of these patients have an Alzheimer component to their disease.[@i2168-9709-8-6-284-b49] Most large, well-designed trials of cholinesterase inhibitors and memantine studied BPSDs as secondary outcomes. Statistical improvements in neuropsychiatric assessment scores have been observed (mean improvement of 2 points on the NPI), but the clinical significance is questionable.[@i2168-9709-8-6-284-b50]-[@i2168-9709-8-6-284-b52] Two additional studies[@i2168-9709-8-6-284-b53],[@i2168-9709-8-6-284-b54] where behavioral symptoms were the primary measurement found 5 to 6 months of donepezil use improved moderate to severe BPSD for patients with Alzheimer dementia, and the results also appeared to be clinically significant (mean NPI improvement of 12 points). Overall, studies[@i2168-9709-8-6-284-b50],[@i2168-9709-8-6-284-b53],[@i2168-9709-8-6-284-b54] suggest cholinesterase inhibitors are more effective for depression, dysphoria, apathy, and anxiety symptoms than for agitation or aggression. Memantine has been shown to improve agitation, aggression, and delusions.[@i2168-9709-8-6-284-b51],[@i2168-9709-8-6-284-b52] The benefit of cognitive enhancers may not be seen until 3 to 6 months after initiation; therefore, these medications will not have clinical utility in acute treatment of BPSD. However, clinicians can consider prescribing cholinesterase inhibitors and memantine to help slow cognitive decline, and eventually these medications may also assist with controlling disruptive behaviors. Memantine is usually well tolerated, with the most common adverse effects being headache, dizziness, and constipation.[@i2168-9709-8-6-284-b55] Cholinesterase inhibitors are associated with more concerning side effects, such as gastrointestinal upset, anorexia and weight loss, bradycardia, sleep disruption, and urinary incontinence. Recent studies[@i2168-9709-8-6-284-b55]-[@i2168-9709-8-6-284-b57] have concluded cholinesterase inhibitors may also increase risk of falls and hip fractures. In addition, withdrawal of cholinesterase inhibitors may lead to faster cognitive and functional decline.[@i2168-9709-8-6-284-b58],[@i2168-9709-8-6-284-b59] Selective Serotonin Reuptake Inhibitors {#s7} ======================================= Selective serotonin reuptake inhibitors (SSRIs) have also been studied for treatment of BPSD. Several trials that included sertraline (50-200 mg/d), citalopram (20-30 mg/d), or escitalopram (10 mg/d) have found these medications to be more effective than placebo or just as effective as antipsychotics for agitation and irritability.[@i2168-9709-8-6-284-b60]-[@i2168-9709-8-6-284-b65] Most studies were 6 to 12 weeks in duration. However, 1 small study (n = 85) comparing citalopram to perphenazine found BPSD to significantly improve by 17 days with both medications, and there was no difference in efficacy between the 2 drugs.[@i2168-9709-8-6-284-b63] The results from this study suggest SSRIs may be helpful for treating BPSD within a few weeks of initiation. Overall, authors conclude that antipsychotics have a quicker onset treating symptoms compared with SSRIs, but SSRIs are better tolerated. Even when antipsychotic therapy is warranted in cases of extreme aggression, the treatment team can consider starting an SSRI with the antipsychotic. The rationale for combination therapy is to increase the patient\'s chances of improvement and to allow eventual discontinuation of the antipsychotic. More literature is needed to determine the overall risk of mortality of SSRIs compared with antipsychotics. There are data to suggest SSRIs increase risk of mortality in older adults compared with nonusers.[@i2168-9709-8-6-284-b66] However, a large retrospective study comparing number needed to harm among patients with dementia prescribed antipsychotics, valproic acid, and antidepressants found mortality risk to be higher for antipsychotics and valproic acid compared with antidepressants.[@i2168-9709-8-6-284-b32] Risks associated with SSRIs include gastrointestinal upset, hyponatremia, tremor and akathisia, risk of falls, and bone density loss. Patients should also be monitored for signs and symptoms of bleeding, especially if nonsteroidal anti-inflammatory drugs, antiplatelet drugs, or anticoagulants are concurrently prescribed.[@i2168-9709-8-6-284-b66],[@i2168-9709-8-6-284-b67] In addition, citalopram was found to prolong QT interval and was associated with a small but statistically significant decline in cognition (−1.05 on the Mini Mental State Examination) in the Citalopram for Agitation in Alzheimer\'s Disease (CitAD) study.[@i2168-9709-8-6-284-b62] Based on the literature,[@i2168-9709-8-6-284-b68],[@i2168-9709-8-6-284-b69] QTc interval increases have ranged from 5 ms with low doses of citalopram up to 35 ms for maximum doses. The CitAD study[@i2168-9709-8-6-284-b62] observed a mean increase of 18 ms in QTc interval from baseline to week 3 with citalopram 30 mg daily, and more patients in the citalopram group had an increase of 30 ms or more in QTc interval compared with the placebo group. However, few patients actually met criteria for prolonged QTc interval (450 ms for males and 470 ms for females).[@i2168-9709-8-6-284-b70] Currently, citalopram labeling recommends patients 60 years and older not exceed 20 mg daily because of a risk of QTc prolongation.[@i2168-9709-8-6-284-b68] When choosing an SSRI, sertraline is often selected first because of a number of factors. Citalopram and escitalopram are associated with a dose-related risk of prolonged QTc,[@i2168-9709-8-6-284-b68]-[@i2168-9709-8-6-284-b72] and clinicians are able to titrate sertraline to higher doses in older adults if clinically warranted rather than being limited to low doses of citalopram and escitalopram. However, citalopram and escitalopram can remain a safe and efficacious choice when dosed conservatively in older patients without cardiac risk factors or ECG abnormalities. Due to lack of data, paroxetine and fluoxetine are not usually prescribed for BPSD. In addition, these 2 SSRIs are often not preferred for elderly patients because of their lower tolerability. Paroxetine has moderate antimuscarinic activity, and fluoxetine is associated with higher incidences of anxiety and agitation compared with the other SSRI.[@i2168-9709-8-6-284-b44],[@i2168-9709-8-6-284-b67] Both paroxetine and fluoxetine are significant inhibitors of P450 enzymes, which increases risk of drug-drug interactions in older adults.[@i2168-9709-8-6-284-b67] Based on studies, SSRIs should be initiated at low doses (eg, sertraline 25 mg daily) and titrated every 1 to 2 weeks. Symptom improvement will often be seen within 2 to 3 weeks of initiating therapy. There are currently no data to suggest how long to use an SSRI for BPSD after initiation. However, it is assumed these medications are not needed indefinitely for all patients. Gradual dose reductions of all psychotropic medications are required by the Centers for Medicare & Medicaid Services in the long-term care setting when being prescribed for BPSD. A minimum of 2 gradual dose reductions should be tried during the first year of therapy, with each attempt being 3 to 6 months apart. If a patient continues to require the SSRI after 1 year, then dose reductions should be attempted at least once yearly.[@i2168-9709-8-6-284-b45] These same recommendations can also be applied to community-dwelling patients. Regarding patient case 2, the delusions are distressing for the patient, despite the nursing home staff trying nonpharmacologic strategies. Therefore, medication intervention may be needed. An antipsychotic will likely treat the patient\'s delusional symptoms. However, the risks may outweigh the benefits in this scenario given the absence of dangerous behavioral symptoms, so alternative pharmacologic strategies should be attempted first. Clinicians may consider increasing the donepezil to 10 mg daily if tolerated and/or adding memantine to the regimen. Maximizing cognitive enhancers may provide more optimal long-term control of behavioral symptoms. However, the small effect observed in the current literature must be weighed against the risks of additional polypharmacy and adverse effects. Another option for this patient is to increase the sertraline to 50 mg daily to more optimally target depressive symptoms while potentially improving agitation and delusional thinking. For this particular patient, optimizing the sertraline dose is a good first choice because the patient is already taking a therapeutic dose of donepezil. The effects of sertraline are likely to be observed sooner than the effects of increasing the donepezil. To minimize the risk of adverse effects, one medication change should be made at a time. Therefore, the treatment team should avoid increasing the donepezil or adding memantine until the effects of the sertraline titration are assessed. Clinicians can consider increasing the donepezil and/or adding memantine if further symptom control is needed after optimizing the sertraline (6-12 weeks later). In addition, increasing the cholinesterase inhibitor or adding memantine will always be a future option as the patient\'s cognition continues to decline. For patients who cannot tolerate SSRIs, optimizing the donepezil or adding memantine would be the recommended treatment. Additional Agents With Limited Evidence {#s8} ======================================= Medications with limited evidence for the treatment of BPSD include trazodone, carbamazepine, and valproate (divalproex or valproic acid). One double-blind study showed trazodone 50 to 250 mg/d to be as effective as haloperidol 1 to 5 mg/d for agitation and verbal aggression.[@i2168-9709-8-6-284-b73] Trazodone 50 mg was also shown to increase sleep time for patients with Alzheimer disease and sleep disturbances.[@i2168-9709-8-6-284-b74] Carbamazepine was significantly better than placebo for improving agitation, aggression, and hostility in 2 randomized, controlled studies.[@i2168-9709-8-6-284-b75],[@i2168-9709-8-6-284-b76] Valproate was not found to be more efficacious than placebo for most BPSD symptoms in 5 randomized, controlled studies and a Cochrane review/meta-analysis.[@i2168-9709-8-6-284-b77] The authors of the meta-analysis[@i2168-9709-8-6-284-b77] concluded that valproate\'s risk of adverse effects outweighed any benefits of the medication. Observed adverse effects of valproate and their associated risk included sedation (odds ratio \[OR\] 2.48; confidence interval \[CI\] 1.37-4.47), gastrointestinal upset (OR 7.09; CI 1.73-29.02), urinary tract infections (OR 3.02; CI 1.04-8.08), and thrombocytopenia (OR 7.91; CI 1.92-32.57). A more recent study led by Tariot et al[@i2168-9709-8-6-284-b78] did not find valproate effective for treatment of BPSD compared with placebo, and patients taking valproate experienced more adverse effects, such as somnolence, gait disturbances, tremor, and gastrointestinal upset. Additionally, studies have found valproate to have a mortality risk similar to that of antipsychotics in the dementia population,[@i2168-9709-8-6-284-b79],[@i2168-9709-8-6-284-b80] and it has been associated with accelerated brain volume loss and further cognitive decline.[@i2168-9709-8-6-284-b78],[@i2168-9709-8-6-284-b81] Dextromethorphan/quinidine, which is FDA approved for treatment of pseudobulbar affect, was studied in 1 randomized, placebo-controlled trial of patients with Alzheimer disease and agitation. Neuropsychiatric Inventory agitation/aggression domain scores improved significantly in the treatment group (mean improvement of 2-3.3 points) compared with the placebo group (*P* \< .001). The most common adverse effects observed with dextromethorphan/quinidine compared with placebo included falls (8.6% vs 3.9%), diarrhea (5.9% vs 3.1%), and urinary tract infections (5.3% vs 3.9%).[@i2168-9709-8-6-284-b82] Data on the use of methylphenidate in dementia patients continue to emerge specifically for the treatment of apathy. Two studies,[@i2168-9709-8-6-284-b83],[@i2168-9709-8-6-284-b84] which each included 60 patients, compared methylphenidate to placebo for the treatment of apathy. Most patients had mild Alzheimer disease at baseline and randomized to methylphenidate or placebo for 6 to 12 weeks. The Apathy Evaluation Scale was used to evaluate the primary outcome in both studies. The 6-week study did not find methylphenidate 10 mg/d to statistically improve Apathy Evaluation Scale scores compared with placebo. However, there were statistically more methylphenidate patients assessed as having moderate or marked improvement based on the Clinical Global Impression of Change scale (21% vs 3%).[@i2168-9709-8-6-284-b83] Apathy Evaluation Scale scores did statistically improve compared with placebo in the 12-week trial, in which patients were titrated to methylphenidate 20 mg/d.[@i2168-9709-8-6-284-b84] Adverse events did not reach statistical significance in either study. A recent Cochrane review and meta-analysis concluded that methylphenidate may be useful for treatment of apathy, but there are limited data to predict how clinically meaningful its effect will be.[@i2168-9709-8-6-284-b85] Benzodiazepines are prescribed among the general population for anxiety, agitation, and sleep. However, these medications have not been shown to be consistently effective in treating agitation or sleep disturbances in dementia patients, and some studies have found benzodiazepines to be associated with cognitive decline.[@i2168-9709-8-6-284-b86] Additionally, benzodiazepines are known to be a high-risk drug class for older adults because of a risk of falls, orthostasis, oversedation, delirium, and respiratory depression.[@i2168-9709-8-6-284-b44] Conclusion {#s9} ========== When presented with a patient with disruptive behaviors due to NCD, clinicians should first search for any modifiable contributors, such as infection, pain, constipation, and medications. Nonpharmacologic methods should be implemented and continued even if medication therapy is needed. If a patient\'s behavioral symptoms, such as agitation and aggression, become dangerous, then SGA may be warranted. The goal should be to use the antipsychotic temporarily and to attempt a taper and discontinuation within 4 to 6 months of initiation. Maximizing the use of cholinesterase inhibitors and memantine may help reduce BPSD in the long term. Selective serotonin reuptake inhibitors can also be tried for agitation and irritability. Treatment onset of SSRIs appears to be slower than that of antipsychotics, but studies suggest SSRIs are better tolerated ([Table 3](#i2168-9709-8-6-284-t03){ref-type="table"}). Based on the current published data available, methylphenidate is recommended over cholinesterase inhibitors and antidepressants for treatment of apathy, but methylphenidate\'s clinical impact is still questionable. ###### Summary of benefits and risks of treatment options for behavioral and psychological symptoms of dementia (BPSD) **Medication** **Target Symptoms** **Benefits** **Risks** --------------------------------------------------------------------------------------------- --------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------ ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Antipsychotics  Risperidone Olanzapine Quetiapine Aripiprazole Haloperidol (not recommended as first-line) Severe agitation Aggression Psychosis Small to moderate efficacy Onset of efficacy is usually observed within the first 1-2 wk Increased risk of mortality Extrapyramidal side effects Metabolic adverse effects Risk of thromboembolism Possible cognitive decline Risk of falls Cholinesterase inhibitors  Donepezil Galantamine Rivastigmine Depression Dysphoria Anxiety Small improvements in BPSD Medications may already be appropriate to help slow cognitive decline Gastrointestinal upset Anorexia/weight loss Bradycardia Risk of falls Memantine Agitation Aggression Delusions Small improvements in BPSD Medication may already be appropriate to help slow cognitive decline Minimal risk of adverse effects • Headache • Dizziness • Constipation Selective serotonin reuptake inhibitors  Sertraline Citalopram Escitalopram Agitation Depression Irritability Some studies show selective serotonin reuptake inhibitors work as well as an antipsychotic Tolerability is often better compared with antipsychotics Gastrointestinal upset Hyponatremia Risk of bleeds Tremor/akathisia Risk of bone loss Risk of falls Possible cognitive decline (Citalopram for Agitation in Alzheimer\'s Disease study) Medications not recommended for the treatment of BPSD include benzodiazepines and anticonvulsants. Little evidence exists to support the use benzodiazepines for BPSD, but the risks of the medications in older adults have been well established. Valproate is associated with higher rates of adverse effects and does not appear more effective for BPSD compared with placebo. Because of limited data in the dementia population, carbamazepine should only be used as a last-line pharmacologic option. **Disclosures:** I have nothing personal to disclose. Psychopharmacology Pearls are review articles intended to highlight both the evidence base available and/or controversial areas of clinical care for psychiatric and neurologic conditions as well as strategies of clinical decision-making used by expert clinicians. As pearls, articles reflect the views and practice of each author as substantiated with evidence-based facts as well as opinion and experience. Articles are edited by members of the Psychopharmacology Pearls Editorial Board as well as peer reviewed by MHC reviewers. This article was developed as part of the 2018 Psychopharmacology Pearls product for BCPP recertification credit. The course information and testing center is at [cpnp.org/353267](http://cpnp.org/353267).

Introduction {#S1} ============ Sigma-1 receptor (S1R) is a ligand-operated protein that modulates activity of several client proteins from its position within the membrane of the endoplasmic reticulum (ER). It is widely expressed in multiple organs including the nervous system ([@B42]) and it has important roles in modulation of neuronal physiology ([@B100]) and synaptic plasticity ([@B177]; [@B184]; [@B82]). Autosomal recessive loss-of-function mutations in S1R are primarily associated with amyotrophic lateral sclerosis/frontotemporal dementia (ALS/FTD) ([@B92]; [@B4]; [@B77]; [@B89]; [@B188]; [@B38]; [@B62]; [@B193]), but polymorphisms in S1R also affect risk of developing Alzheimer's disease (AD) ([@B187]; [@B97]; [@B63]; [@B28]). Many S1R agonists are neuroprotective and loss of S1R accelerates neurodegenerative phenotypes ([@B102]; [@B131]; [@B167]; [@B109], [@B108]; [@B94]; [@B138]; [@B155], [@B156]; [@B103]). Neuroprotection from S1R activation is achieved by a diversity of signaling functions that promote cellular homeostasis and synaptic stability. In this review we summarize therapeutically relevant mechanisms by which the ligand-operated chaperone S1R modulates neurophysiology, counteracting its dysregulation from pathogenic stressors. Modulation of Neurophysiology by S1R {#S2} ==================================== Sigma-1 receptor is a 223 amino acid-long transmembrane protein residing in the ER membrane. S1R preferentially localizes to the specific microdomains of the ER called mitochondrial-associated membranes (MAM), where it can regulate InsP~3~R-dependent calcium flux from the ER to mitochondria ([@B56]), lipid dynamics ([@B55]), MAM stability ([@B193]), and the ER stress response ([@B124]). The MAM domain is also important for synthesis and transport of lipids and protein folding ([@B197]). Under resting conditions, S1R forms an inert complex with GRP78/BiP protein ([@B56]). When activated by agonists or ER calcium depletion, S1R dissociates from BiP and redistributes to the entire ER network ([@B56]), freeing it to interact with and modulate several client proteins including InsP~3~Rs inside and outside of the MAM domain as well as plasmalemmal ion channels, GPCRs, and kinases (summarized in [Table 1](#T1){ref-type="table"}). ###### S1R binding partners and biological outcomes mediated by these interactions. **S1R-interactng protein** **Method(s) revealing interaction** **Regulation by S1R expression level** **Regulation by S1R activation** **References** -------------------------------------------------------------- ------------------------------------------------ --------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------- ----------------------------------- **Plasmalemmal ion channels** Kv1.2 Co-IP Increased surface expression Agonists (cocaine) enhanced association between S1R and Kv1.2; increased surface expression [@B81]; [@B23] Kv1.3 Co-IP Co-expression increased Kv1.3 inactivation Inhibition by SKF-10047 [@B80] Kv1.4 Co-IP Overexpression of S1R dose-dependently increased Kv1.4 inactivation SKF-10047 reduced Kv1.4 outward currents [@B6] Kv1.5 Overexpression of S1R inhibited Kv1.5 currents SKF-10047 reduced Kv1.5 outward currents [@B100] Kv2.1 Imaging [@B107] L-type Ca^2+^ channels Co-IP Inhibition by S1R agonists (SKF-10047) [@B20]; [@B208]; [@B180]; [@B130] N-type Ca channels Co-IP Inhibition by S1R overexpression Inhibition by SKF-10047, PRE-084 [@B208]; [@B72]; [@B33]; [@B213] Nav1.5 Co-IP, pulldown, AFM Knockdown decreased Nav1.5 currents Agonists (PTZ) promoted dissociation of Nav1.5 [@B8] ASIC1a Co-IP, pulldown, AFM S1R activation decreased ASIC1a currents [@B58]; [@B17] hERG Co-IP, AFM Potentiation by S1R overexpression Ligands depressed hERG currents [@B21]; [@B7] STIM1/Orai1 Co-IP, imaging, AFM Overexpression inhibited SOC; S1R KD enhanced SOC Agonists inhibited SOC while antagonists enhanced SOC [@B170] **ER channels** InsP~3~R1 Calcium imaging Agonists suppressed ER calcium release mediated by InsP~3~R1 [@B155] InsP~3~R3 Co-IP Stabilization by S1R, overexpression increased IP3-induced Ca-release Activation of IP3-induced Ca-release by agonists [@B56]; [@B201]; [@B23] **ER resident proteins** BiP/GRP78 Co-IP, pull-down, NMR Stable complex formation with BiP Agonists dissociated S1R from BiP [@B56]; [@B140] Insig1; UDP-galactose:Ceramide Galactosyltransferase (CGalT) Co-IP Overexpression of S1R increased degradation of CGalT; S1R KO increased protein levels of CGalT Agonist (PTZ) increased association of S1R with Insig1 [@B50] IRE1a Co-IP, proximity biotinylation labeling S1R overexpression sustained IRE1 phosphorylation and signaling ([@B124]), S1R KO increased IRE1 activity ([@B150]) Fluvoxamine led to anti-inflammatory response [@B56]; [@B124]; [@B2]; [@B150] Ankyrin Co-IP S1R forms stable ternary complex with ankyrin and IP~3~R3s Agonists dissociated ankyrin from S1R and potentiated IP~3~-induced Ca-release [@B52] **PM receptors/proteins** CB1R Bimolecular fluorescence complementation assay S1R regulates formation of a CB1-HINT1-GluN1 complex S1R opposed CB1R-mediated suppression of NMDAR activity [@B161] D1R BRET Formation of D1R-S1R heteromers S1R agonists enhanced D1R signaling [@B133] D2R Co-IP, BRET Formation of D2R-S1R heteromers S1R agonist (cocaine) inhibited D2R signaling [@B134] MOR Co-IP, \[35S\]GTPγS binding Potentiation by S1R knockdown Potentiation by S1R antagonist [@B76] Integrin b1 Co-IP n/d Agonist (SKF-10047) reduced interaction and reduced cell adhesion [@B144] BDNF Knockdown suppressed secretion of mature BDNF Agonists promote secretion of mature BDNF [@B32] TrkB Co-IP Activation of S1R promoted TrkB signaling [@B79]; [@B73] PDGFR Pull-down, co-IP, FRET [@B205] Dopamine transporter (DAT) Co-IP, BRET, functional assays Agonists modulated stimulant binding to DAT and stimulant-evoked DA efflux via DAT and calcium signals [@B61]; [@B160] **Mitochondrial proteins** VDAC2; StAR Co-IP Reduction of cholesterol efflux under S1R KD conditions [@B95] **Proteins in the Cytosol** Rac1 Co-IP Interacts as part of multiprotein complex involving S1R, IP3R, Rac, BiP Agonist (PTZ) increased association [@B132] ELMOD 1-2 Co-IP Binding inhibited GAP activity [@B69] **Other** Emerin Co-IP, native gel electrophoresis Association with HDAC1/2, BAF, Emerin Increased association with HDAC1/2, BAF, emerin after cocaine treatment (*in vivo*) [@B182] Androgen receptor (AR) Co-IP Increased AR degradation under S1R KD conditions S1R inhibitors prevented nuclear transclocation and increased degradation of AR [@B181] Sigma-1 receptor agonists do not noticeably alter ER calcium signaling under resting conditions ([@B51]), but they can influence ER calcium release triggered by Gq-coupled receptors ([@B56]; [@B155]). S1R chaperones InsP~3~R3 to the MAM domain and prevents its degradation, enhancing Ca^2+^ flux into mitochondria ([@B56]). This augments ATP production ([@B39]), but, in excess, it could also trigger the mitochondrial permeability transition ([@B88]). By contrast, InsP~3~R1, which is the predominant InsP~3~R isoform in neurons and has important signaling functions outside of the MAM, is negatively regulated by agonist-stimulated S1R in certain cell types like striatal medium spiny neurons (MSNs) ([@B155]). Regulation ER calcium homeostasis and signaling by S1R has important implications for neurodegenerative diseases and this will be discussed later. Engagement of S1R with plasmalemmal channels and receptors is responsible for S1R-dependent fine-tuning of neuronal excitability ([@B82]). As many of S1R's interaction partners function in the plasma membrane, it was proposed that activated S1R translocates from the ER to the plasma membrane where it binds to client proteins and acts as a chaperone or an auxiliary subunit ([@B111]; [@B127]; [@B51]; [@B6]). However, this assumption is often based on experiments in which S1R and/or its interaction partner are overexpressed, raising several caveats that call this interpretation into question ([@B171]). Alternatively, S1R may interact with plasma membrane proteins from its position in the ER like STIM proteins ([@B107], [@B105]) and/or regulate the maturation and/or trafficking of certain proteins to the plasma membrane ([@B21]; [@B23]). Sigma-1 receptor activation alters neuronal excitability through its interactions with voltage-gated ion channels. Voltage-gated sodium (Nav) channels augment neuronal depolarization and mediate action potentials. S1R ligands dissociate S1R from Nav1.5, leading to suppressed Nav1.5 activity ([@B72]; [@B8]). This action can be evoked by the endogenous S1R agonist *N*,*N*-dimethyltryptamine (DMT) and is partially opposed by the endogenous S1R antagonist progesterone ([@B71]). S1R agonists also limit excitability by inhibiting other Nav channels including Nav1.2 and Nav1.4 ([@B72]; [@B33]). Voltage-gated potassium (Kv) channels respond to membrane depolarization during action potentials by releasing positively charged potassium ions from the cytosol to restore a hyperpolarized resting membrane potential and limit hyperexcitability. When S1R is activated by cocaine, S1R binds to the voltage-gated potassium channel Kv1.2 and enhances trafficking of Kv1.2 to the plasma membrane, decreasing excitability of dopamine D1 receptor (D1R)-expressing MSNs in the shell of the nucleus accumbens ([@B23]). S1R expression and activity also regulates the cardiac Kv channel hERG through control of maturation and trafficking ([@B21]) and this function appears to be dependent on cholesterol and not S1R ligands, possibly suggesting a role for lipid rafts in S1R client protein assembly and trafficking ([@B7]). S1R appears to basally regulate Kv1.3 and Kv1.4 independent of agonist-stimulation ([@B6]; [@B80]). By means of such interactions, S1R regulates neuronal excitability. Sigma-1 receptor also influences synaptic functions through modulation of the NMDA receptor (NMDAR) activity. Physiological NMDAR activation can induce hippocampal long-term potentiation (LTP) ([@B90]), spine maturation ([@B176]) and learning ([@B129]), but pathophysiological levels of NMDAR activity triggers excitotoxicity ([@B151]). S1R facilitates NMDA receptor signaling and neurotransmission in hippocampal neurons ([@B121], [@B120], [@B122]), possibly through altering responses to calcium signals (e.g., inhibiting Ca^2+^-activated SK channels) and promoting expression of NMDA receptor subunits and their trafficking to the plasma membrane ([@B96]; [@B141]). S1R can also obviate negative-regulation of NDMARs by cannabinoid 1 receptor (CB1R) ([@B161]). These interactions enhance neuronal firing and maturation of mushroom spines from NMDA receptor activation ([@B121]; [@B96]; [@B141]). Modulation of calcium signaling by S1R may regulate synaptic plasticity through stimulation of CaMKII, PKC, and ERK ([@B126]). Sigma-1 receptor agonists may promote synaptic plasticity and neuronal resilience in large part through their common ability to upregulate BDNF secretion and TrkB receptor signaling both *in vitro* and *in vivo* ([@B75]; [@B146]). For example, pridopidine, a potent S1R receptor agonist, promotes neurotrophic signaling via BDNF, ERK, and AKT pathways ([@B138]; [@B35]; [@B83]; [@B67]). S1R agonists appear to activate TrkB both through BDNF-dependent ([@B79]) and independent mechanisms ([@B73]). This may involve regulation of BDNF expression and processing as well as direct interactions of S1R with the TrkB receptor ([@B32]; [@B79]; [@B73]). S1R also stimulates signaling by other receptor tyrosine kinases including the epidermal growth factor receptor (EGFR) ([@B177]) and the platelet-derived growth factor receptor (PDGFR) ([@B205]). Stimulation of neurotrophic receptors confers neuroprotection through control of gene expression. Indirect regulation of transcriptional activity by S1R contributes to its neuroprotective properties. For example, S1R may prevent neuronal death by upregulating expression of the antiapoptotic mitochondrial protein Bcl-2 ([@B113]; [@B207]). S1R regulates transcription through interactions with inositol-requiring enzyme 1 (IRE1) and emerin. S1R facilitates dimerization of the ER stress sensor and endonuclease IRE1 at the MAM domain, leading to splicing-dependent activation of the transcription factor XBP1, which goes on to upregulate several ER chaperones ([@B124]). S1R also decreases IRE1-driven inflammation ([@B150]), which may be important for microglial reactivity and migration to and from injury sites ([@B128]). As the ER membrane is contiguous with the nuclear envelope, activated S1R can move to the nuclear envelope where it regulates transcription through its recruitment of emerin and then chromatin-remodeling factors ([@B182]). A microarray study involving knockdown of S1R in cultured hippocampal neurons revealed altered transcription in pathways controlling protein ubiquitination, sterol biosynthesis, oxidative stress, and actin dynamics ([@B185]). Knockdown of S1R reduces the size of dendritic spine size in hippocampal neurons, indicating that it actively supports stability of mature spines ([@B184]; [@B29]; [@B156]). This was initially proposed to involve its role in regulating oxidative stress and Rac-GTP signaling ([@B184]), but may also involve modulation of calcium homeostasis in conditions of disease ([@B156]). Knockout of S1R is associated with increased formation of reactive oxygen species (ROS) and decreased expression and activity of NRF2, which promotes expression and activation of antioxidant molecules under conditions of stress ([@B191]). This may explain how S1R suppresses generation of ROS ([@B113]). Interestingly, spine shrinkage from knocking down S1R was prevented by reducing oxidative stress ([@B184]). Finally, in addition to protein--protein modulation, S1R was shown to interact with lipids. *S*1R localizes to lipid rafts -- detergent-resistant microdomains of the ER -- where it binds cholesterol at sterol-binding sites and S1R agonists such as SKF-10047 displace S1R and its binding partners from lipid rafts possibly through out-competing cholesterol binding ([@B53]; [@B144]). S1R targets to galactosyl-rich microdomains of the ER and is potentially involved in regulation of the differentiation of oligodendocytes and myelination ([@B54]) as well as lipid transport to the myelin membrane ([@B196]). S1R also supports axonal growth through promoting myristoylation of p35, which increases its degradation and thereby decreases p25/CDK5-dependent hyperphosphorylation of Tau ([@B183]). This review only scratches the surface with regard to S1R's multiplicitous roles in neurophysiology/neuroprotection and provides a glimpse into the specificity of its actions in differing cell types. It is tempting to speculate that the nature of modulation by S1R depends on the levels of S1R and its interaction partners in a given cell type (e.g., preferential interactions of S1R with cell-type abundant effectors) as well as many circumstantial factors such as the current physiological/pathophysiological state of the cell and signaling events (e.g., presence/absence of S1R ligands). To better understand possible actions of S1R, we summarized S1R interaction partners and the effects of S1R ligands and knockdown/overexpression on these protein--protein interactions ([Figure 1](#F1){ref-type="fig"} and [Table 1](#T1){ref-type="table"}). ![Modulation of neurophysiology by S1R. Normally residing at the MAM, S1R is released from BiP upon activation from ER calcium depletion, ER stress or agonist stimulation, freeing it to interact with its client proteins. Within the MAM, S1R regulates lipid dynamics and chaperones InsP~3~R3 to the MAM, facilitating calcium flux from the ER to mitochondria. This enhances ATP production. S1R's actions on transcriptional pathways counteract oxidative stress through upregulation of antioxidants. Once activated, S1R redistributes to the entire ER network where it interacts with additional targets including InsP~3~R1, STIM1 and several plasma membrane ion channels and receptors. For example, S1R activation by pridopidine in striatal MSNs attenuates ER calcium release from InsP~3~R1 when it is hyperactive in HD from mutant Huntingtin protein, leading to suppression of synaptotoxic signals mediated by store-operated calcium entry channels (SOCCs). Conversely, nSOC pathway activity is important for mushroom spine stability in AD, but it is downregulated from reduced ER calcium leakage in AD models. In hippocampal neurons, S1R decreases ER calcium levels, possibly though positive regulation of presenilin leak channels (not shown). This restores nSOC pathway activity and promotes mushroom spine stability. S1R activation also limits excitotoxicity by decreasing activity of Nav and Cav channels, while promoting activity of some Kv channels. S1R enhances NMDAR activity, which is important for induction of LTP as well as activation of calcium-dependent transcription factors. S1R also modulates several GPCRs, which can influence several physiological processes including monoamine neurotransmission. Moreover, S1R activation promotes synaptic plasticity and neuronal survival through upregulation of BDNF expression and secretion as well as direct stimulation of TrkB receptors. S1R monomers are shown with the crystal structure adapted from [@B164], but S1R ligand-dependent oligomerization/monomerization may confer specificity in its diverse actions.](fnins-13-00862-g001){#F1} Molecular Analysis of S1R {#S3} ========================= Sigma-1 receptor is a promising therapeutic target in the treatment of neurodegenerative diseases as it stabilizes the function of several intracellular systems through its role as a chaperone when activated by a variety of ligands with neuroprotective properties. Despite S1R pluripotency, relatively little is known regarding the mechanisms of receptor functioning and regulation at the molecular level. There is evidence that the structural organization of S1R and its conformational state are important determinants of S1R activity. However, the structural basis for S1R functionality remains poorly understood. Originally, S1R was characterized as a type 1 transmembrane protein with only one transmembrane domain ([@B48]). Hydrophobicity analysis confirmed a single-pass transmembrane topology of S1R ([@B74]), however, subsequent studies predicted a two transmembrane domain model of S1R topology ([@B6]; [@B140]). For example, Aydar and colleagues proposed two-transmembrane domains (TM1 a.a. 11--29 and TM2 a.a. 80--100) based on antibody staining experiments with expression of S1R fused to GFP at the N- or C-terminus in *Xenopus* oocytes ([@B6]). Immunolabeling of the GFP-tags was absent without membrane permeabilization, but was detected after permeabilization with 0.5% acetone, leading them to conclude that both the N- and the C-termini are located near the plasma membrane but within the cytoplasm. By contrast, the topological model proposed by Hayashi and Su situates S1R in ER membranes with both N- and C-terminal regions oriented to the ER lumen ([@B56]). This was based on immunocytochemical staining of endogenous S1R in CHO cells with antibodies targeting N- and C-termini. Similar to results of [@B6], no labeling was detected without permeabilization, suggesting that S1R is not in the plasma membrane. Permeabilization of plasma and ER membranes with CHAPS or Triton X-100 enabled staining for all antibodies with a distribution similar to the shape of the ER. When the plasma membrane was permeabilized with streptolysin-O, staining was present only for the antibody targeting the loop domain. The discrepancy between the topology models of [@B6] and [@B56] may have arisen from altered membrane insertion of GFP-fused S1R and/or cell type specific differences in S1R localization. For example, Hayashi and Su note that fusion of YFP to the C-terminus of S1R, but not the N-terminus, mirrors the distribution of endogenous S1R ([@B56]). The two-pass transmembrane model was widely accepted for a long time and has been used as a structural basis for molecular modeling and ligand docking studies ([@B86]). However, the crystal structure for human S1R was recently solved revealing a single transmembrane domain structure ([@B164]). According to this study, a short N-terminus faces the ER lumen while the C-terminal domain of protein is oriented to the cytosolic side ([@B164]). Adding another possible model of S1R topology to the mix, [@B110] fused ascorbate peroxidase 2 (APEX2) to the N- or C-terminus of S1R and used electron microscopy to visualize deposition of 3,3′-diaminobenzidine outside or inside of the ER of ND7/23 cells and dorsal root ganglion (DRG) neurons. This experiment indicated that the N-terminus of S1R faces the cytosol, whereas the C-terminus is located in the ER lumen. This is consistent with the one transmembrane model suggested by the crystal structure, but suggests the orientation of S1R positions the bulk of its structure in the lumen of the ER with only a short N-terminus facing the cytosol. All of the experiments probing the topology and orientation of S1R have caveats that are important to keep in mind including specificity of antibodies and membrane permeabilization as well as alterations to S1R from protein fusions and crystallization conditions. [@B164] determined the first crystal structure of full-length human S1R using X-ray analysis. They expressed FLAG-tagged S1R in Sf9 insect cells, purified it using the detergent lauryl maltose neopentyl glycol (LMNG) and crystallized it using the lipidic cubic phase (LCP) method ([@B16]). According to the crystal structure, S1R homomers consist of three protein subunits, with each protomer having one transmembrane domain ([Figure 2A](#F2){ref-type="fig"}). The transmembrane alpha-helices of trimers (encompassing amino acid residues 8 to 32) are separated from each other and located in the corners of the complex, while the C-terminal cytosolic domains of each protomer organize the trimeric structure and form highly conserved ligand-binding sites. The membrane proximal surface of each C-terminal domain is tightly associated with the cytosolic surface of the ER membrane. Each C-terminal domain contains a cupin-like β-barrel that can envelop a ligand ([Figure 2A](#F2){ref-type="fig"}). ![Structure of the ligand-binding site of S1R bound to agonist (+)-pentazocine. **(A)** The overall structure of a sigma-1 receptor subunit bound to (+)-pentazocine (PDB ID: 6DK1) based on ([@B163]). **(B)** A close up of the binding pocket showing the key amino acids involved in coordination of the ligand. (+)-pentazocine is shown in orange. Glu172 interacts with (+)-pentazocine's nitrogen atom (blue) and both Tyr103 and Asp126 facilitate this through creating hydrogen bonds with Glu172. Other amino acids including Val84, Trp89, Met93, Leu95, Tyr103, Leu105, Phe107, Ile124, and Trp164 help to form the primarily hydrophobic binding pocket and stabilize the ligand through Van der Waals interactions.](fnins-13-00862-g002){#F2} An accurate model of the S1R ligand binding pocket is necessary for rational drug design aimed at the targeted treatment of neurodegenerative diseases. Mapping of the S1R ligand binding site was carried out in a large number of studies using mutational analysis and photoreactive probe labeling ([@B203]; [@B18]; [@B143], [@B142]). The first identified amino acids that are important for ligand binding include Ser99, Tyr103, Leu105, and Leu106 ([@B203]). The results of these studies are highly consistent with the structural model of S1R obtained via X-ray crystal analysis. Sigma-1 receptor crystal structures harboring chemically distinct ligands (the high-affinity, selective S1R ligands PD144418 and 4-IBP) show that both ligands bind in similar positions, forming electrostatic interactions with the highly conserved amino acid residue Glu172 ([Figure 2B](#F2){ref-type="fig"}). The amino acid Asp126 which is also essential for ligand binding forms a hydrogen bond with Glu172 ([@B164]). With the exception of only two amino acid residues, the S1R active site is hydrophobic and is occluded from aqueous solution. Other amino residues involved in ligand coordination are as follows: Val84, Trp89, Met93, Leu95, Tyr103, Leu105, Phe107, Ile124, Trp164, and Leu182 ([Figure 2B](#F2){ref-type="fig"}). Additionally, Tyr103 creates a hydrogen bond with Glu172, which is apparently necessary for the formation of the binding pocket ([Figure 2B](#F2){ref-type="fig"}). Indeed, in earlier works a significant decrease in the binding activity of the mutant Tyr103Phe was shown ([@B203]). The highly occluded structure of the binding pocket raises the questions about the pathway of ligand entry and explains the very slow ligand binding kinetics. [@B163] conducted additional structural studies and molecular dynamics (MD) simulation experiments to reconstruct the ligand binding mechanism in detail. They solved crystal structures of S1R bound to the classical antagonists haloperidol and NE-100 as well as the agonist (+)-pentazocine. The obtained structures were highly similar to each other and did not differ significantly from the previously determined trimeric structures of S1R. They share a similar organization of the ligand-binding pocket. The overall conformations of S1R in complex with the antagonists or agonist remain almost identical with the exception of a difference in the position of (+)-pentazocine in the ligand binding site. On the basis of structural data and MD simulations the authors suggest that agonist binding leads to conformational changes of S1R compared to the unliganded form of receptor or antagonist bound S1R. MD simulations were used to characterize conformational rearrangements occurring during ligand association. Ligand association was characterized as a two-step process: (1) the hydrogen bonds between Trp136 and Ala161 break, leading to receptor "lid" opening, and (2) the interior of the receptor separates, exposing the binding pocket and allowing ligand entry. [@B163] provided valuable insights on the structural basis for a ligand binding mechanism and describe the potential conformational differences between agonist and antagonist bound S1R. However, they do not explain the functional role of agonist/antagonist actions and the physiological relevance of agonist-induced structural rearrangements of S1R. The relationship between the ligand-receptor association and the subsequent biological response remains unclear. Extensive evidence indicates that S1R exists in multiple oligomeric states ([@B19]; [@B40]; [@B118]) and recent studies suggest that ligand binding induces a shift in the oligomeric state of S1R ([@B40]; [@B118]; [@B61]), which could in turn lead to the various functional responses. For example, high-molecular weight forms of S1R have been detected in rat liver microsomal membranes using radioactive photosensitive labels ([@B143]). Oligomeric forms of S1R corresponding to hexamers, tetramers, octamers were also identified by size-exclusion chromatography (SEC) ([@B40]). Analysis by SDS-PAGE after chemical crosslinking of individual oligomeric forms of MBP fused S1R has also confirmed the presence of monomers, tetramers and high molecular weight S1R oligomers ([@B109]). Cell-based fluorescence resonance energy transfer (FRET) studies have also confirmed the existence of several oligomeric states ([@B118]) and revealed that agonists stabilize low-molecular-weight species, whereas antagonists favor oligomers. This model suggests that monomeric form of S1R is an "active" conformation involved in interactions with client proteins ([Figure 3](#F3){ref-type="fig"}). On the other hand, [@B40] found similar stabilizing effects of agonists and antagonists on S1R oligomeric states. However, the detergents used in the purification procedure of their experiments do not reflect the native lipid environment of membrane proteins, potentially limiting the validity of this finding. While the crystal structure of S1R has a trimeric fold, size-exclusion, cross-linking and multi-angle light scattering (SEC-MALS) experiments have revealed a wide range of oligomeric states from 140 up to 400 kDa ([@B164]). Thus, it remains unclear which oligomeric form(s) of S1R exists *in vivo* and which state(s) is(are) functionally active. ![Model of S1R oligomerization and its functional role. The model is based on [@B118] and [@B61]. On the left, a S1R trimer is shown. Agonists promote dissociation of S1R into monomers, which may redistribute to other subcellular compartments and associate with client proteins. By contrast, antagonists prevent such interactions by stabilizing S1R oligomerization. Ligands regulate the interactions of S1R with its protein partners. While oligomeric forms of S1R have a demonstrated ability to bind ligands, S1R monomers may lose this property. As several oligomeric forms of S1R have been reported, they may also have functional roles and oligomer-specific interaction partners.](fnins-13-00862-g003){#F3} Sigma-1 receptor oligomerization is disrupted by mutations in the GXXXG motif corresponding to amino acid residues 87--91 ([@B40]; [@B139]). The GXXXG motif was previously through to reside in the second transmembrane domain and mediate subunit association via transmembrane alpha helices. However, the crystal structure suggests that it forms a beta-hairpin structure inside the oligomerization interface ([@B163]). The distance between Cα atoms of Gly88 in each protomer is about 6 Å ([@B163]). Thus, mutations of this residue can sterically interfere with subunit association. [@B40] demonstrated that mutations within the GXXXG motif cause a shift toward the monomeric state of S1R. Interestingly, this is associated with a significant decrease in ligand binding, suggesting that ligand binding affinity may depend on S1R oligomerization processes. While oligomeric forms of S1R have a demonstrated ability to bind ligands, S1R monomers may lose this property ([@B40]). Mutations within the GXXXG motif also decrease S1R expression, which may indicate reduced stability of GXXXG mutants ([@B40]). The oligomerization interface was further characterized with crystallographic data, leading to identification of key amino acids involved in subunit interactions. The sequence of the oligomerization interface is highly conserved among species, confirming its physiological importance. The trimerization surface is formed largely by hydrophobic residues within the C-terminal cupin domain. For example, a Phe191 residue from each protomer forms inter-subunit contacts. There are the polar interactions between sidechains of Thr141, His54, and Glu55. The amino acids Trp81, Phe83, Met90, Ala92 and Leu111, His116, Arg119, Trp136, Ala161, Trp169, Asp188, Phe191, Ser192, Gln194, Asp195, and Thr198 are also engaged in formation of the trimerization interface ([@B164]). Despite the detailed characterization of oligomerization interface and comprehensive studies of ligand-induced structural rearrangements, the significance of S1R oligomer-monomer transitions in the regulation of S1R functions remains unknown. To correlate structural rearrangements observed *in vitro* with physiological responses, it has been proposed that the physiological significance of S1R oligomerization may be linked to the protein--protein interactions of S1R with its partners ([@B40]; [@B204]). For example, the monomeric form of S1R interacts with Nav1.5, acid-sensing channels and D1R ([@B17]; [@B133]; [@B8]). Two groups identified a direct and agonist-dependent interaction between S1R and the dopamine transporter (DAT), resulting in attenuated DA efflux and calcium signals evoked by methamphetamine ([@B61]; [@B160]). [@B60] suggested that agonists induce dissociation of S1R multimers into monomers which in turn interact with DAT. Mutational analyses have shown that the interaction site is located in the transmembrane domain of S1R. As mentioned above, initiation of S1R activity was also proposed to involve ligand-, calcium-, or ER stress-dependent dissociation of S1R from binding immunoglobulin protein (BiP), which is a chaperone located in the lumen of the ER ([@B56]). Recently, [@B204] used a novel bioluminescence resonance energy transfer (BRET) assay to study the ligand-mediated oligomerization of S1R. They revealed the distinct effects of agonists and antagonists on S1R homomerization, consistent with previous results ([@B61]). Interestingly, while the agonist pentazocine facilitated interaction of BiP and S1R, haloperidol induced the dissociation of S1R from BiP. Thus, S1R ligands may regulate the association between S1R and BiP through controlling S1R oligomerization and monomerization. This is likely also relevant to S1R interactions with its client proteins ([Figure 3](#F3){ref-type="fig"}). It would be interesting to know more clearly how S1R associates with various proteins located in the ER lumen, ER membrane, cytoplasm and plasma membrane and to resolve the conflicting models of S1R topology and orientation. Given the topology model proposed by [@B110], the bulk of S1R may face the ER lumen. This topology is consistent with the well-described interaction of S1R with BiP, but raises it questions about how S1R interacts with proteins in the cytosol with only a small cytosolic N-terminal tail. Perhaps S1R has two or more structural elements or configurations responsible for the binding of S1R to different proteins. The structural and biological mechanisms of such interactions remain to be fully elucidated. S1R as a Target for Treating Neurodegenerative Diseases {#S4} ======================================================= Many S1R agonists are anti-amnestic, synaptogenetic, and neuroprotective in conditions of neuronal stress ([@B5]; [@B59]; [@B154]; [@B11]). They also mitigate disease and symptoms in experimental models of ALS ([@B94]; [@B138]; [@B146]; [@B67]), Alzheimer's disease (AD) ([@B113]; [@B29]; [@B99]; [@B47]; [@B36]; [@B156]), Parkinson's disease (PD) ([@B30]; [@B31]) Huntington's disease (HD) ([@B169]; [@B35]; [@B12]; [@B34]; [@B155]; [@B83]) stroke ([@B3]; [@B154], [@B153]; [@B162]; [@B189]) and traumatic brain injury ([@B24]). By contrast, S1R deficiency exacerbates progression of neurological disorders ([@B109], [@B108]; [@B46]; [@B30]; [@B115]; [@B103]) as well as symptoms commonly associated with neurodegenerative diseases. For example, pharmacological inhibition of S1R leads to mushroom spine loss in hippocampal cultures ([@B156]) and this could be related to memory impairments from the anti-psychotic drug and S1R antagonist haloperidol (K~D~ for S1R ∼3 nM) ([@B1]). S1R knockout (KO) mice have several phenotypes resulting from neuronal dysfunction and late-onset neurodegeneration ([@B157]; [@B45]; [@B166]). These data collectively highlight the innate, neuroprotective properties of S1R activity. The following sections summarize genetic associations of S1R mutations/polymorphisms with neurodegenerative diseases, examples of neuroprotection in respective disorders by S1R agonists, and possible mechanisms of action. S1R in Amyotrophic Lateral Sclerosis/Frontotemporal Dementia (ALS/FTD) {#S5} ====================================================================== Amyotrophic lateral sclerosis is a fatal neurodegenerative disease featuring progressive weakness of skeletal muscles due to upper and/or lower motor neuron dysfunction and loss. Several recessive, loss-of-function mutations in S1R have been associated with ALS, distal hereditary motor neuropathy and/or FTD ([@B92]; [@B4]; [@B77]; [@B89]; [@B188]; [@B38]; [@B62]; [@B193]). S1R is highly expressed in motor neurons ([@B42]; [@B107]), suggesting a possible cell autonomous mechanism of motor neuron degeneration in these patients. Although S1R KO mice do not develop an overt ALS phenotype ([@B85]), they have deficits in locomotion and motor performance ([@B107]) related to muscle weakness, axonal degeneration, and motor neuron loss ([@B10]). KO of S1R accelerates the onset and progression of ALS in the SOD1^G93A^ mouse model of ALS ([@B108]), whereas the S1R agonists PRE-084 and SA4503 are neuroprotective and extend survival of SOD1^G93A^ mice ([@B94]; [@B138]). PRE-084 is also protective in wobbler mice, which develop spontaneous motor neuron degeneration ([@B146]). PRE-084 is a derivative of phencyclidine (PCP) with nanomolar affinity for S1R and negligible affinity for PCP receptors and GPCRs ([@B172]). SA4503 has low nanomolar affinity for S1R, low micromolar affinity for S2R, and little affinity for 36 other receptors, ion channels and second messenger systems ([@B98]). Treatment of SOD1^G93A^ mice with the S1R agonist pridopidine improves axonal transport (e.g., of BDNF, GDNF, and mitochondria) and BDNF secretion while attenuating atrophy of neuromuscular junctions, muscle fibers and motor neurons ([@B67]) (pharmacological properties summarized in HD section). Pridopidine treatment reduces the prevalence of SOD1 aggregates in spinal cords of SOD1^G93A^ mice ([@B67]). S1R activity may additionally protect motor neurons by reducing their excitability through facilitation of potassium channel activity ([@B106]). Motor neuron degeneration from the absence of S1R is associated with reduced contacts between mitochondria and ER, ER stress, calcium dysregulation ([@B10]) and this may help to explain pathology in ALS patients with mutations in S1R. Role of S1R in Parkinson's Disease (PD) {#S6} ======================================= Dopamine receptors play important roles in learning and memory, motivation and movement and S1R agonists modulate dopaminergic signaling through multiple mechanisms. This has primarily been studied in the context of psychostimulant research, but these results may be important for understanding regulation of dopamine neurotransmission and its dysregulation in HD and PD. S1R appears to differentially regulate dopamine D1 and D2 receptors, as S1R activation by cocaine inhibits D2R ([@B134]) and prevents histamine H3 receptor-dependent inhibition of the dopamine D1 receptor, stimulating Gs, recruitment of β-arrestin and phosphorylation of ERK1/2 ([@B123]). Although S1R activation does not affect basal dopamine neurotransmission, it attenuates methamphetamine-induced and DAT-dependent increases in firing of dopamine neurons ([@B160]). It also interacts directly with the DAT and attenuates calcium signals evoked by methamphetamine ([@B160]). As a result, S1R limits hyperactivity, motivated behavior and reinforcement from methamphetamine ([@B160]). Abnormalities in movement and cognition in PD result from degeneration of dopaminergic neurons projecting from the substantia nigra to the striatum. S1R is expressed in these neurons ([@B61]) and it can bidirectionally modulate NMDAR-dependent release of dopamine in striatal brain slice experiments ([@B37]). S1R may be decreased in striatal regions that are preferentially affected in PD ([@B116]), which could contribute to neuropathology as indicated by studies with S1R KO mice. Similar to PD patients, S1R KO mice have age-related deficits in motor behavior and degeneration of dopaminergic neurons ([@B61]). This appears to be related to aggregation and phosphorylation of α-synuclein which may be driven by phosphorylation of eIF2α from ER stress and proteasomal dysfunction ([@B61]). Pharmacological inhibition of ER stress prevented oligomerization of α-synuclein, dopaminergic neuron loss and motor impairments in S1R KO mice ([@B61]). Recent studies found that S1R agonists are protective in PD models. For example, chronic treatment with PRE-084 gradually improves PD-like motor deficits from unilateral intrastriatal 6-hydroxydopamine (6-OHDA) lesions (hemiparkinsonian model) when treatment onset was prompt ([@B30]). This treatment suppressed neuroinflammation while increasing levels of neurotrophic factors, monoamines (e.g., dopamine and serotonin), dopaminergic innervation of the striatum, and nigral neuron survival ([@B30]). Low dose pridopidine treatment (0.3 mg/kg) of unilaterally 6-OHDA-lesioned mice partially protected nigral dopaminergic cell bodies and increased dopaminergic fiber density in the motor striatum ([@B31]). This was associated with a gradual restoration of forelimb use (cylinder test, stepping test) and prevention of rotational bias toward the ipsilateral side ([@B31]). The delayed recovery of motor function corresponds roughly with the expected timeline of pridopidine-dependent dopaminergic axon sprouting ([@B31]). Treatment efficacy was absent in S1R KO mice, which had reduced loss of dopaminergic neurons in the substantia nigra pars compacta, but a greater loss of dopaminergic fibers in the striatum compared to wild-type mice ([@B31]). The increased vulnerability of S1R knockout mice to axonal degeneration in the nigrostriatal pathway could relate S1R's ability to promote growth and repair of neurites ([@B31]). The neuroresorative effects of pridopidine were associated with upregulation of neurotrophic factors (BDNF, GDNF, pERK1/2) and associated signaling in the striatum and substantia nigra as well as reduced microglial activation ([@B31]). Role of S1R in Huntington's Disease (HD) {#S7} ======================================== Huntington's disease (HD) patients suffer from psychiatric, motor and cognitive disturbances that gradually worsen, leading to dementia, cachexia and eventually death. HD is a dominantly inherited neurodegenerative disease resulting from a CAG trinucleotide repeat expansion in exon 1 of the *Huntingtin* gene (\>35 CAG repeats), leading to expression of mutant Huntingtin (mHtt) protein with an elongated polyglutamine tract. mHtt is broadly expressed throughout the body, but striatal MSNs are preferentially vulnerable in HD. The most significant contributions of mHtt to HD pathology remains a matter of debate and intense investigation. The CAG expansion compromises normal functions of Htt and disrupts cellular functioning through gain of mHtt function mechanisms ([@B66]; [@B78]), with possible toxic contributions from repeat-associated non-AUG translation ([@B9]). This results in oxidative damage, glial reactivity, altered intracellular signaling, metabolism and energy levels, impaired axonal transport, transcriptional dysregulation, aberrant calcium regulation associated with ER stress, synapse loss and excitotoxicity ([@B206]; [@B119]; [@B87]; [@B155]). Initial studies on the potential role of S1R in HD pathology were carried out with cellular models of HD. [@B65], found that expression of mHtt (N-terminal huntingtin fragment proteins with 120 polyQ repeats or full-length Htt protein with 75 repeats) downregulates S1R expression in neuronal PC6.3 cells. There were no differences in S1R expression in control cells expressing the N-terminal fragment of Htt with 18 polyQ repeats or wild-type Htt. Administration of the selective S1R agonist PRE-084 prevented mHtt-dependent downregulation of S1R, SOD1, SOD2, thioredoxin 2, and Bcl-XL in neuronal PC6.3 cells ([@B65]). However, S1R expression appears to be upregulated in the striatum of YAC128 HD mice at 12 months of age and in the striatum of patients with advanced HD, possibly as an effort to compensate for ER calcium dysregulation and stress ([@B155]). PRE-084 also decreased caspase-3 cleavage and oxidative stress and upregulated calpastatin, NF-κB-p65 levels and NF-κB signaling in mHtt expressing PC6.3 cells, enhancing their viability ([@B65]). [@B65] proposed that the neuroprotective properties of S1R activity involved modulation of the calpastatin/calpain system, increasing NF-κB signaling and thereby upregulating antioxidants and decreasing ROS levels. Another study demonstrated that large neuronal nuclear inclusions were strongly positive for S1R in human brains affected by polyglutamine diseases and intranuclear inclusion body disease ([@B114]). Also, S1R immunostaining colocalized with most intranuclear mHtt aggregates in HeLa cells expressing the N-terminal fragment of mHtt. Downregulation of S1R with antisense RNA increased the amount of mHtt aggregates in both the cytoplasm and nucleus. This was reproduced by treatment with the proteasome inhibitor epoxomicin. Moreover, proteasome activity was significantly lower following knockdown of S1R. Pridopidine's Mechanism of Action in the Treatment of HD {#S8} ======================================================== Clinical trials with pridopidine indicate that it has efficacy in treating motor symptoms of HD ([@B91]; [@B22]; [@B27]; [@B64]; [@B149]) and recent evidence suggests that the therapeutic effect of pridopidine involves S1R. Pridopidine was originally dubbed a "dopamine stabilizer" based on behavioral experiments showing that it can both decrease locomotion in hyperactive rodents (e.g., from [D]{.smallcaps}-amphetamine or MK-801) and increase locomotion in hypoactive rodents (e.g., animals that have habituated to their environment or co-treated with the VMAT inhibitor tetrabenazine) ([@B147]; [@B194]; [@B159]). Increased locomotion in hypoactive rodents may relate to pridopidine's ability to increase dopamine and norepinephrine in the cortex and subcortical areas ([@B147]), which may also explain increased firing in prefrontal pyramidal neurons ([@B41]). Pridopidine's ability to bidirectionally normalize activity levels may have particular utility in the treatment of HD in which patients develop hyperkinetic motor disturbances followed by hypoactivity later in disease. The mechanism of action was initially proposed to involve low-affinity/fast-off negative modulation of dopamine D2 receptors with a slight binding preference for the agonist binding site when the receptor is in the active, catalytic, high-affinity state ([@B136]; [@B152]; [@B194]), but the affinity of pridopidine for D2 receptors of dopamine is relatively low being in the micromolar range ([@B25]). Unlike classical D2 receptor antagonists, pridopidine does not induce hypoactivity or catalepsy ([@B195]). Pridopidine also has micomolar affinity for several additional GPCRs including adrenergic alpha 2A/C receptors, serotonergic 5HT1A and 5HT2A receptors, and histamine H3 receptors ([@B41]) and these interactions may influence levels of extracellular monoamines and glutamatergic neurotransmission ([@B194]). The effects of pridopidine in behavioral assays are not fully blocked in D2R receptor knockout mice ([@B175]), prompting further investigation into potential molecular targets of pridopidine. More recently, pridopidine was found to have a high affinity (kD = ∼80 nM) for S1R ([@B158]) and it primarily binds S1R rather than D2 receptors *in vivo* at behaviorally relevant doses ([@B159]), suggesting that S1R might mediate the therapeutic effects of pridopidine. Recent studies show that activation of S1R by pridopidine might be disease-modifying in HD. [@B169] found that pridopidine reduces motor symptoms of R6/2 mice, improving performance on the horizontal ladder task and open-field locomotor measurements when treatment was started presymptomatically (5--6 mg/kg via daily intraperitoneal injections) ([@B169]). Pridopidine also extends their lifespan ([@B169]). *In vivo* treatment also normalized striatal BDNF and DARPP32 levels ([@B169]; [@B35]; [@B34]; [@B83]), while decreasing the size and amount of mHtt aggregates ([@B169]). 150 μM pridopidine reduced apoptosis and restored pERK1/2 levels in a mouse striatal knock-in cellular HD model (STHdh111/111) and these effects were blocked by the S1R antagonist NE-100 ([@B169]). [@B155] found that low nanomolar concentrations of pridopidine and the structurally similar S1R agonist (+)3-PPP are neuroprotective in another cellular model of HD ([@B155]). Both compounds stabilized synaptic connections between cortical and striatal MSNs in primary corticostriatal co-cultures prepared from from YAC128 HD mouse pups. Deletion of S1R with Cas9 prevented pridopidine and 3-PPP from rescuing dendritic spine loss in HD MSNs. Interestingly, S1R deletion also resulted in significant spine loss in WT MSNs. This observation indicated an important role for S1R in maintaining MSN spine stability. A synaptoprotective action of pridopidine was further supported by a series of Ca^2+^ imaging experiments. Previous studies demonstrated that abnormal Ca^2+^ signaling in post-synaptic spines is responsible for their destabilization in HD MSNs ([@B199], [@B198]). Decreased ER Ca^2+^ levels due to mHtt-induced InsP~3~R1 hyperactivity ([@B179], [@B178]) increases neuronal store-operated calcium entry (nSOC) in HD MSNs to synaptotoxic levels ([@B200], [@B199], [@B198]). Pridopidine treatment of corticostriatal co-cultures from YAC128 mice prevented InsP~3~R1 hyperactivity, restored ER Ca^2+^ levels, and decreased nSOC in HD MSNs ([@B155]). Deletion of S1R WT MSNs resulted in depletion of ER Ca^2+^ content, suggesting that it might stabilize MSN spines through homeostatic control of ER Ca^2+^ levels ([@B155]). Deletion of S1R in HD MSNs prevented the normalization of ER Ca^2+^ by pridopidine ([@B155]). The selective S1R agonist PRE-084 also prevents dendritic spine loss in HD MSNs and this rescue is blunted by the S1R antagonist NE-100 ([@B12]). These findings suggest that in addition to the ability of pridopidine to mitigate motor symptoms of HD, it may also foster synaptic and neuronal viability via activation of S1R. Consistent with this, [@B26] found that pridopidine and 3-PPP protects mouse primary striatal and cortical neurons from expression of mHtt (22 vs. 58 CAG repeats with the first 586 amino acids of Htt), as measured by imaging nuclear condensation in apoptotic cells and neuronal morphology. Pridopidine also rescued HD patient-derived induced pluripotent stem cells ([@B26]). The S1R antagonist NE-100 or genetic ablation of S1R blocked the neuroprotective effects ([@B26]). Although BDNF was also protective and is upregulated by S1R stimulation, blockade of BDNF signaling with the TrkB receptor antagonist ANA-12 did not impede the neuroprotective effects of pridopidine ([@B26]). However, ANA-12, like NE-100, suppressed pridopidine's ability to prevent mitochondrial depolarization from mHtt, as measured using tetramethyl rhodamine methyl ester (TMRM) ([@B26]). These data indicate that S1R activation by pridopidine or 3-PPP is neuroprotective, but neuroprotection is not entirely mediated by BDNF signaling. Pridopidine activates neuroprotective pathways that are compromised in HD (e.g., BDNF and AKT pathways), improving behavioral and transcriptional deficits in mouse models of HD ([@B35]; [@B34]; [@B83]). Consistent with its ability to promote neuronal plasticity and survival, pridopidine upregulates expression BDNF, dopamine D1 receptor, AKT/PI3K and glucocorticoid pathway components and stimulates BDNF secretion in an S1R-dependent fashion ([@B35]; [@B83]). Microarray and qPCR studies showed that pridopidine upregulates several genes downstream of BDNF including EGR1, EGR2, KLF5, CDKN1A, Homer1a, and Arc ([@B35]; [@B83]). BDNF overexpression is sufficient to rescue many phenotypic characteristics of YAC128 HD mice (e.g., motor performance, cognitive deficits, synaptic density) ([@B202]), further suggesting that BDNF signaling could be an important contributor to neuroprotection following S1R activation. The idea that some of the beneficial effects of pridopidine in HD models can be mediated through BDNF signaling was supported by recent experimental evidence from [@B168]. Synaptic scaling was suppressed in YAC128 cultures, as determined by recording the amplitude and frequency of mEPSCs after blockade of activity-dependent neurotransmission with TTX. Synaptic scaling was restored in YAC128 neurons by pharmacological activation of S1R with pridopidine or 3-PPP through BDNF-TrkB signaling ([@B168]). Given that AKT is a potent pro-survival kinase, its upregulation may help promote neuronal resilience by phosphorylating apoptotic proteins (e.g., BAD and GSK3) and forkhead family transcription factors (e.g., FOXOs) ([@B35]; [@B83]). Also, the calcium regulating genes calbindin and Homer1a are downregulated in the striatum of Q175 and YAC128 HD mice and they are both upregulated by pridopidine treatment ([@B155]). These results indicate that when activated by pridopidine S1R acts on several transcriptional networks to foster neuronal function and survival in HD models. A recent study demonstrated that pridopidine improves motor performance in YAC128 HD mice as well as anxiety- and depressive-like phenotypes, but it was unable to prevent striatal and corpus callosum atrophy ([@B34]), indicating that S1R agonism is insufficient to completely mitigate HD neuropathology and complementary treatment strategies should be considered. While pridopidine might be insufficient to completely prevent HD progression, when taken together the data on its effects in HD models and in HD patients shows that it can mitigate symptoms and is likely has the capacity to modify disease. Role of S1R in Alzheimer's Disease (AD) {#S9} ======================================= Alzheimer's disease (AD) is the most pervasive cause of dementia in elderly people and it involves progressive impairment of memory and other cognitive faculties from damage to the hippocampus and other parts of the brain. Age is the main risk factor for the sporadic form of the disease. Early onset of AD is characterized by the development of the disease before the age of 65 and most of these cases result from autosomal dominant inherited mutations in amyloid precursor protein (APP), presenilin-1 (PSEN1) or presenilin-2 protein (PSEN2). Autosomal dominant inheritance accounts for about 1% of all cases of AD. When APP is cleaved by β- and γ-secretases, Aβ is formed with a length of 39 to 42 amino acid residues ([@B49]). Presenilins are part of the γ-secretase protease complex and are key catalytic subunits. In AD mutations in the *APP* and *PSEN1*, *PSEN2* genes promote the formation of an extracellular fragment of Aβ with a length of 42 amino acid residues (Aβ~42~), the accumulation of which contributes to the formation of amyloid oligomers. Recently, it has been demonstrated that Aβ is generated intracellularly at the MAM domain and may influence functioning of the ER, mitochondria, and MAM ([@B165]). Given this finding and the importance of S1R at MAM domains ([@B56]; [@B193]), it is perhaps not surprising that common S1R polymorphisms influence risk of developing AD ([@B187]; [@B97]; [@B28]). In fact, certain genetic combinations of S1R and apolipoprotein E (APOE) genotypes synergistically increase the risk of AD ([@B63]). Several S1R agonists have anti-amnestic properties, overcoming learning and memory impairments from amyloid-β toxicity or scopolamine ([@B99]). S1R agonists promote neurogenesis in the hippocampus ([@B125]) and they may mitigate memory impairment because they can stabilize mature, mushroom spines ([@B156]), which serve as sites of robust synaptic connections encoding lasting information ([@B13]; [@B57]). Mushroom spine loss may underlie memory defects in models of AD, as hippocampal neuron mushroom spines are lost *in vitro* and *in vivo* in both presenilin-1-M146V knock-in (PS1-KI) and APP knockin (APP-KI) models of AD ([@B173]; [@B212]). As postmortem and *in vivo* brain imaging studies have demonstrated a reduced density of S1R in the brains of patients with AD ([@B70]; [@B117]) and S1R knockdown destabilizes mushroom spines ([@B184]; [@B29]; [@B156]), downregulation of S1R may contribute to AD pathology. Consistent with this, knockout of S1R in APP~Swe~ AD mice increases oxidative stress within the hippocampus and exacerbates memory impairments ([@B99]; [@B103]). The novel positive S1R modulator (±)-2-(3-chlorophenyl)-3,3,5,5-tetramethyl-2-oxo-oxazaphosphinane (OZP002) ([@B42]; [@B100]; [@B184]) was also neuroprotective in pharmacological and genetic models of AD. It potentiated the antidepressant effect of the S1R agonist igmesine and prevented scopolamine-induced learning deficits in the Y maze test and passive avoidance test. Its effect was blocked by NE-100 or in S1R knockout mice ([@B104]). Treatment of hippocampal cultures with Aβ~42~ oligomers induces loss of mushroom spines ([@B148]; [@B212]) and Aβ~42~ accumulation in hippocampal cultures prepared from APP knock-in mice also causes mushroom spine loss ([@B212]). This is also observed *in vivo* ([@B212]). Pridopidine and 3-PPP prevent mushroom spine loss from both of these sources of Aβ toxicity in hippocampal neuronal cultures ([@B156]). Pridopidine treatment normalized synaptic functioning, preventing LTP deficits caused by Aβ~42~ oligomers ([@B156]). Pridopidine and 3-PPP also prevented mushroom spine loss in hippocampal cultures prepared from PS1-KI mice ([@B156]) that model familial AD ([@B44]). Importantly, oral treatment with pridopidine rescued mushroom spines *in vivo* in PS1-KI mice ([@B156]), suggesting this as a viable treatment strategy for memory deficits in familial AD. AF710B also stabilized mushroom spines *in vitro* in hippocampal cultures prepared from AD mice (PS1-KI and APP-KI models) ([@B29]). AF710B was found to potently and selectively stimulate the M1 muscarinic acetylcholine receptor (M1R) and S1R ([@B29]). AF710B binds to M1R outside of its orthosteric binding site, suggesting an allosteric mechanism of action ([@B29]). This is supported by data showing that 0.1 nM AF710B enhances the affinity and potency of the M1R agonist carbachol ([@B29]). The mechanism by which AF710B activates S1R is less clear, but the anti-amnestic properties of AF710B appear to require S1R, because the S1R antagonist NE-100 can suppress them ([@B29]). No significant binding was observed with other targets in a screen involving 83 GPCRs, ion channels and transporters ([@B29]). Treatment of 3xTg-AD mice with AF710B (10 μg/kg delivered by intraperitoneal injections daily for 2 months) reduced levels of BACE1, Aβ~1--42~, plaques, p25/CDK5, GSK-3β activity, Tau phosphorylation and memory deficits in the Morris water-maze ([@B29]). It was previously known that M1R activation improves cognition and reduces AD-like pathology in animal models ([@B15]; [@B112]), but the combined activity of AF710B at both M1R and S1R might make it particularly therapeutic. Indeed, *in vivo* treatment of McGill-R-Thy1-APP transgenic rats also reduced amyloid burden and inflammation while enhancing synaptogenesis and cognition ([@B29]). Another mixed muscarinic/σ1R agonist, ANAVEX2-73, was able to mitigate Aβ~25--35~-induced tau phosphorylation and Aβ~1--42~ seeding in mice ([@B84]) and may help to preserve cognition in preliminary clinical trials clinical studies with AD patients ([@B93]). In addition to S1R agonists, positive allosteric modulators that do not compete with the (+)-pentazocine binding site might have therapeutic value. For example, SKF-83959 shows promise in the 6-OHDA-induced rat model of Parkinson's disease ([@B214], [@B209]; [@B43]). Also, OZP002 attenuated learning deficits from scopolamine, ICV injection of amyloid Aβ~25--35~, or the APP~Swe~ transgene and protected against neurotoxicity associated with ICV injection of amyloid Aβ~25--35~ ([@B190]). Several selective allosteric modulators for S1R have been discovered (methylphenylpiracetam and SOMCL-668) (see [@B190] for a review) and it will be interesting to see whether they are efficacious in models of neurodegenerative diseases. Hinting at potential utility, SOMCL-668 enhanced (+)-SKF-10047-stimulated neurite growth and BDNF production in an S1R-dependent manner ([@B192]). Although both direct agonists and positive allosteric modulators may have therapeutic promise, it is unclear which are better candidates for clinical trials. Recent data suggests that S1R functionally interacts with presenilin 1 (PS1) and presenilin 2 (PS2), which are implicated in AD. Although cleaved PS1 is the catalytic subunit in the γ-secretase complex, the holoprotein version of PS1 functions as a passive leak channel in the ER membrane ([@B186]; [@B135]; [@B210]). Similarly, PS2 forms a calcium leak channel in the ER ([@B186]; [@B135]; [@B210]). Many familial AD-causing mutations in either PS1 or PS2 disrupt tonic Ca^2+^ release from the ER via PS1 and PS2 leak channels, increasing the concentration of calcium in the ER ([@B186]; [@B210]). Pridopidine promotes ER calcium homeostasis by decreasing luminal calcium levels in cultured hippocampal neurons from WT, PS1-KI and conditional presenilin double-knockout mice (PS1^flox/flox^/PS2^--/--^) infected with lenti-NLS-GFP as well as in neurons expression Cas9 and gRNA targeting the PS1 gene ([@B156]). However, this effect is lost in PS1^flox/flox^/PS2^--/--^ hippocampal neurons infected with lenti-NLS-GFP-Cre ([@B156]). Knockout of PS1, PS2, or PS1/2 causes mushroom spine loss in hippocampal neurons ([@B156]). Consistent with functional data, pridopidine can compensate for this and restore mushroom spine integrity in presenilin 1 or presenilin 2 knockout neurons, but not in PS1/2 double knockout neurons ([@B156]). Consistent with the spine loss phenotype in PS1 and/or PS2 KO neurons, inactivation of PS1 in the mouse forebrain causes mild cognitive deficits, whereas inactivation of both PS1 and PS2 severely impacts cognition and synaptic plasticity, leading to neurodegeneration ([@B174]). PS1 and PS2 both contribute to resting ER calcium homeostasis and form a redundant ER calcium leak pathway ([@B186]; [@B135]) that is required for pridopidine to decrease the concentration of ER calcium and stabilize mushroom spines ([@B156]). The reason for this is unclear, but S1R agonists might decrease ER calcium in hippocampal neurons by enhancing ER calcium leakage either by modulating leak activity of PS channels or increasing the prevalence of PS holoprotein in the ER membrane. The later possibility is intriguing because S1R might impede catalytic cleavage of PS1, reducing its incorporation into the γ-secretase complex. This could underlie the ability of S1R agonists to reduce Aβ~42~ accumulation and aggregation ([@B29]). However, more work is needed to characterize how S1R activity normalizes ER calcium homeostasis and mushroom spine prevalence while mitigating other hallmarks of AD neuropathology. When ER calcium levels are chronically elevated from mutations in presenilin 1, suppression of neuronal store-operated calcium entry (nSOC) leads to destabilization of hippocampal mushroom spines ([@B173]; [@B211]). Upregulating nSOC either pharmacologically or via overexpression of STIM2 or EB3 prevents PS1-KI mushroom spine loss ([@B173]; [@B211]; [@B145]). Consistent with pridopidine's ability to decrease ER calcium levels in hippocampal neurons, it increased nSOC in the spines of cultured PS1-KI neurons and activity of the nSOC pathway was required for the rescue of PS1-KI mushroom spines ([@B156]). This suggests that pridopidine rescues PS1-KI mushroom spines through decreasing ER calcium levels and thereby stimulating nSOC pathway activity. Highlighting the importance of the client protein milieu in determining the effects of S1R activity, S1R suppresses store-operated calcium entry (SOC) in other cell types outside of the hippocampus. In non-neuronal cells, treatment with S1R agonists or overexpression of S1R or suppresses SOC ([@B14]; [@B170]). This may involve S1R binding to STIM1 and disrupting the interaction of STIM1 and Orai1 ([@B170]). Additionally, in MSNs from YAC128 mice that model Huntington's disease pridopidine decreases supranormal nSOC, which is synaptotoxic to MSNs ([@B155]). This contrasts with data involving hippocampal neurons in which pridopidine decreased ER calcium levels and enhanced nSOC ([@B156]). This divergence indicates that the effect of S1R in a given cell type depends on the availability of S1R interaction partners. For instance, InsP~3~R1 constitutes the primary ER calcium release pathway in MSNs ([@B199]), whereas presenilins preferentially mediate leakage of ER calcium in hippocampal neurons ([@B186]; [@B135]; [@B210]). Additionally, although S1R suppresses STIM1-dependent SOC in HEK293 and CHO cells ([@B170]), STIM2 is the predominant regulator of nSOC in hippocampal neurons ([@B173]). S1R might not bind and sequester STIM2 the way it does with STIM1 or if S1R does bind STIM2, it might enhance or minimally effect STIM2-gated nSOC. Thus, regulation of synaptic plasticity by S1R is likely to be multifaceted and highly dependent on the cellular context. Conclusion {#S10} ========== Sigma-1 receptor is incredibly versatile in its ability to foster neuronal homeostasis in the context of several neurodegenerative disorders. Several S1R agonists are FDA-approved ([@B68]), such as fluvoxamine ([@B137]) and donepezil ([@B101]) and they may be repurposed for the treatment of several neurodegenerative diseases. Additional S1R agonists such as pridopidine have shown promising results in preclinical studies and in clinical trials. Author Contributions {#S11} ==================== All authors listed have made a substantial, direct and intellectual contribution to the work, and approved it for publication. Conflict of Interest Statement {#conf1} ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This research was supported by the National Institutes of Health (R01NS056224 and R01AG055577 to IB and F32NS093786 to DR), by the State grant 17.991.2017/4.6 to IB, by a grant from the RFBF according to the research project 18-34-00994 to NK, and by the Russian Scientific Fund grant 19-15-00184 to IB. IB holds the Carl J. and Hortense M. Thomsen Chair in Alzheimer's Disease Research. [^1]: Edited by: Tangui Maurice, INSERM U1198 Mécanismes Moléculaires dans les Démences Neurodégénératives, France [^2]: Reviewed by: Wladyslaw-Lason, Institute of Pharmacology, Polish Academy of Sciences, Poland; Tomohisa Mori, Hoshi University, Japan [^3]: This article was submitted to Neuropharmacology, a section of the journal Frontiers in Neuroscience

Introduction {#Sec1} ============ DNA-directed, rational antineoplastic agent development has made an extensive use of the acridine pharmacophore^[@CR1]^. Among many diverse families of acridine derivatives, development of a series of novel triazoloacridinones with potent antitumor activities has been started some time ago^[@CR2]^. The most active triazoloacridinone derivative obtained to date, 5-\[\[3-(dimethylamino)propyl\]amino\]-8-hydroxy-6H-v-triazolo\[4,5,1-de\]acridin-6-one, codenamed C-1305 (Fig. [1](#Fig1){ref-type="fig"}), showed high antitumor activity towards a wide range of different experimental tumors in vitro and in vivo, including both murine and human colon carcinomas, which in most cases was associated with complete tumor regression^[@CR3]^. Induction of apoptosis in human leukemia cells after uptake of C-1305 has also been demonstrated^[@CR4],[@CR5]^. Its interaction with several molecular targets^[@CR6],[@CR7]^, as well as its metabolism^[@CR8]--[@CR10]^, were extensively studied. For instance, this compound was shown to be a topoisomerase II poison, which stabilizes unusually toxic covalent complexes between DNA and the enzyme^[@CR11]^. It was also demonstrated that C-1305 is a viable double stranded DNA (dsDNA) intercalator, yet no sequence-specificity of this potential drug was revealed since UV, CD and ELD studies have shown that it intercalated into ctDNA, p(dAdT)~2~ and p(dGdC)~2~ polymers at the same ratio^[@CR12]^. Moreover, chemical probing with DEPC, combined with molecular modelling studies suggested that C-1305 is able to induce substantial distortions of a dsDNA duplex while intercalating into the GGG triplets, which is a unique effect among the known topoisomerase II inhibitors. Yet, this result was strongly dependent on the protonation of the drug and was observed only in pH \> 7.5, in which C-1305 is believed to exist mostly in a zwitterionic form with deprotonated 8-OH hydroxyl group and protonated tertiary nitrogen atom at the end of the sidechain^[@CR12]^ (Fig. [1](#Fig1){ref-type="fig"}). Nevertheless, no structural studies on the mode of intercalation of C-1305 into DNA were performed as it was merely assumed that the binding occurs from the minor groove of the helix. That statement was made on the basis of the structural resemblance of C-1305 to other known minor-groove intercalators^[@CR12]^. It might be very misleading since there are several known examples of other, structurally close acridine-based dsDNA intercalators, such as 9-amino-DACA and its derivatives^[@CR13],[@CR14]^, which bind from the major groove of the helix. Figure 1The structure of 5-\[\[3-(dimethylamino)propyl\]amino\]-8-hydroxy-6H-v-triazolo\[4,5,1-de\]acridin-6-one, codenamed C-1305, with a protonated terminal tertiary nitrogen atom. Rational design of new drugs requires the knowledge of their molecular mechanism of action. We believe that the complete understanding of the structural aspects of the dsDNA:C-1305 non-covalent adducts formation is crucial for further, biochemically-oriented studies and for the synthesis of new compounds, potentially exhibiting better pharmacological properties. Our previous studies on a C-1305 analogue---imidazoacridinone C-1311^[@CR15],[@CR16]^---have suggested that the apparent lack of sequence-specificity of this drug family might not be the case. Thus, we decided to revisit also this aspect using NMR spectroscopy and molecular dynamics (MD) modelling. Herein, we report the three-dimensional structures of the d(CGATATCG)~2~:C-1305 and d(CCCTAGGG)~2~:C-1305 1:1 mol/mol complexes, fully and unambiguously reconstructed on the basis of 2D NMR experiments and refined by molecular dynamics simulations restrained by NOESY-derived data. Our structures demonstrate that, for these DNA sequences, C-1305 intercalates exclusively at the central 5′-TA-3′/5′-TA-3′ dinucleotide step (from now on denoted in a shorthand notation TA/TA). Furthermore, we also present a detailed, NMR-driven study on the sequence-specificity of C-1305, providing a completely new insight into the drug's mode of binding into dsDNA. Results {#Sec2} ======= Oligonucleotide sequence selection {#Sec3} ---------------------------------- The investigation started with screening for an oligonucleotide sequence able to form a well-defined complex with C-1305. In the light of some previous results^[@CR12]^, the d(CCCGGG)~2~ duplex, from now on referred to as **D1**, was chosen assuming that upon addition of C-1305 it would produce a stable, strictly defined non-covalent complex. Such an adduct would possibly enable a spectroscopic confirmation of the previously reported, unique changes in DNA topology, caused by C-1305 intercalating inside a guanosine triplet^[@CR12]^. Unfortunately, the resulting d(CCCGGG)~2~:C-1305 complex turned out to be too weakly interacting and too labile to produce a set of good quality NMR spectra indispensable for structure determination studies (Supplementary Figure [S1](#MOESM3){ref-type="media"}). On the second approach, basing on some previous findings regarding sequence specificity of a closely related pro-drug C-1311^[@CR16]^, the study was focused on a GA/TC-containing palindromic octamer, expecting that it would be able to produce a symmetrical, DNA:drug 1:2 mol/mol complex. The choice fell on the d(CGATATCG)~2~ duplex, from now on referred to as **D2**, which is an extended version of the d(CGATCG)~2~ and d(CGTACG)~2~ duplexes---the standard palindromes widely used in the structural studies on the DNA:drug intercalation complexes^[@CR13],[@CR17]--[@CR19]^. The d(CGATATCG)~2~ duplex (Fig. [2](#Fig2){ref-type="fig"}a) indeed formed a very well-defined complex with triazoloacridinone C-1305 (Fig. [2](#Fig2){ref-type="fig"}b), yet---as 2D NMR data has later shown---with 1:1 mol/mol stoichiometry. Furtherly, basing on the sequence-specificity studies described further in the text, it was discovered that insertion of the 5′-TA-3′ dinucleotide step at the very centre of the d(CCCGGG)~2~ duplex radically enhanced its ability to bind C-1305. The resulting d(CCCTAGGG)~2~ duplex, later on referred to as **D3** (Fig. [2](#Fig2){ref-type="fig"}f), in the presence of C-1305 produced particularly well-defined non-covalent adduct, which also---proven by the 2D NMR data---displayed 1:1 mol/mol stoichiometry (Fig. [2](#Fig2){ref-type="fig"}g).Figure 2The optimization of the experimental conditions for the formation of the d(CGATATCG)~2~:C-1305 (**D2L**) and d(CCCTAGGG)~2~:C-1305 (**D3L**) complexes: (a) uncomplexed **D2** duplex, (b) **D2L**,1:1 mol/mol in conditions (I), c~DNA~ = 125 µM, (c) **D2L**, 1:1 mol/mol in conditions (I), c~DNA~ = 1.25 mM, (d) **D2L**, 1:1 mol/mol in conditions (II), c~DNA~ = 1.25 mM, (e) **D2L**, 1:1.25 mol/mol in conditions (II), c~DNA~ = 1.25 mM, (f) uncomplexed **D3** duplex, (g) **D3L**, 1:1.5 mol/mol in conditions (II), c~DNA~ = 1.25 mM. The imino proton resonances raised by the formation of the complexes are labelled with asterisk (\*). Reference NMR data {#Sec4} ------------------ Standard 2D NMR studies were conducted on the **D2** and **D3** octamers, consisting of NOESY, TOCSY and ^1^H-^13^C-HSQC experiments in conditions with a close-to-physiological pH value (I), namely: 10 mM phosphate buffer, pH = 7.5, 150 mM NaCl (H~2~O and D~2~O solutions). Assignment of the ^1^H and ^13^C resonances was carried out using standard procedures^[@CR20]--[@CR22]^. Reference ^1^H NMR data for triazoloacridinone C-1305 (NOESY, TOCSY) were also acquired in the aforementioned experimental conditions. Optimization of the experimental conditions {#Sec5} ------------------------------------------- Experimental conditions were optimized for the **D2** octamer and then applied to the rest of the examined duplexes. Low concentrations of the **D2** duplex (\~ 125 µM) were initially investigated in conditions (I). In such setup, d(CGATATCG)~2~ oligomer and 1 molar equivalent of C-1305 formed a non-covalent complex, exhibiting only one spectral form in the solution (Fig. [2](#Fig2){ref-type="fig"}b). High specificity of ligand binding was deduced upon the appearance of new, narrow ^1^H resonances corresponding to a single form of the complex. Signal integration revealed that around 50% of total DNA amount was complexed by C-1305. However, the total DNA concentration was too low to produce viable 2D NMR spectra. Surprisingly, increasing the concentration while maintaining DNA/ligand proportions (1.25 mM concentration of the DNA material, 1 molar equivalent of C-1305) resulted in a significant broadening of the proton resonances (Fig. [2](#Fig2){ref-type="fig"}c). Even more unexpectedly, higher concentrations also slightly shifted the system's equilibrium towards the complex dissociation. A detailed investigation (see [Supplementary Data](#MOESM3){ref-type="media"} for details) allowed to conclude that this behavior is related to the pronounced self-association of the ligand. Aggregated ligand molecules are unsuitable for intercalation, yet instead they interact non-specifically with DNA, leading to line broadening. This problem could ultimately be alleviated by alteration of the experimental conditions (see [Supplementary Data](#MOESM3){ref-type="media"}). The optimal experimental conditions for the d(CGATATCG)~2~:C-1305 complex formation were identified as (II): 2.5 mM cacodylate buffer, pH = 5.0, 10 mM NaCl. In conditions (II), an ^1^H NMR spectrum was obtained for the sample of 1.25 mM concentration of the DNA plus 1 molar equivalent of the ligand (Fig. [2](#Fig2){ref-type="fig"}d) in which the protons of the complex displayed almost the same chemical shifts as in the spectrum recorded for the sample in conditions (I) (Fig. [2](#Fig2){ref-type="fig"}b). Even the imino proton resonances, potentially vulnerable to the changes of the pH of the solution, have not altered their positions. Therefore, it was concluded that the structures of the both d(CGATATCG)~2~:C-1305 complexes, obtained in different (I and II) experimental conditions, were identical. Moreover, it was possible to add an extra 0.25 molar equivalent of the ligand, which shifted the equilibrium of the system towards the complex formation, without broadening the DNA resonances. In such a setup, the well-defined d(CGATATCG)~2~:C-1305 complex, from now on referred to as **D2L**, became a dominant spectral form in the solution (\~ 70% of total DNA amount, Fig. [2](#Fig2){ref-type="fig"}e). Interestingly, in the same setup, the d(CCCTAGGG)~2~:C-1305 complex, further on referred to as **D3L** (Fig. [2](#Fig2){ref-type="fig"}g), displayed higher fraction of DNA bound by ligand in comparison to **D2L**, unveiling that **D3** duplex exhibits higher affinity towards C-1305 than the **D2** duplex. In the end, optimization of the experimental conditions, enabling a significant upscale in total DNA and ligand concentrations, allowed to obtain good quality 2D NMR spectra for the **D2L** and **D3L** complexes. 2D NMR studies of the d(CGATATCG)2:C-1305 (D2L) complex {#Sec6} ------------------------------------------------------- Assignment of the ^1^H resonances to the d(CGATATCG)~2~:C-1305 complex was performed in a similar fashion as it was done for the non-complexed dsDNA and the free ligand. However, the NMR spectra were severely more complicated, since two distinctive forms of both DNA and C-1305 were present in solution in slow chemical exchange. Moreover, the intercalation of the ligand into DNA duplex broke the symmetry of the helix, which manifested in doubling of ^1^H resonances of most of the residues of the complex. As a result, two DNA strands of the complex needed to be traced separately and were named **D2L-A** and **D2L-B**, respectively (Fig. [3](#Fig3){ref-type="fig"}). Nevertheless, despite severe spectral crowding, it was possible to assign all the aromatic base protons of the complex, as well as most of the deoxyribose resonances, namely H1′, H2′, H2″ and H3′ (please consult Supplementary Table [S1](#MOESM3){ref-type="media"} for more detailed information). The H4′ and H5′/H5″ resonances were also partially assigned. Due to severe signal overlap in 2D NMR spectra, performing a complete assignment of these protons was not possible. Crucially, all proton resonances of the bound ligand were also unambiguously identified. The chemical shift changes of the DNA resonances upon complex formation are presented in Fig. [4](#Fig4){ref-type="fig"}a.Figure 3H6/H8-H1′ NOESY-walks for the d(CGATATCG)~2~:C-1305 complex, depicted as green (strand A) and blue (strand B) lines. Missing/extremely weak correlations are marked in circles. This figure was prepared with NMRFAM-SPARKY 1.3^[@CR38]^. Figure 4Absolute values of the chemical shift changes (Δδ, ppm) of the DNA resonances upon (**A**) d(CGATATCG)~2~:C-1305 (**D2L**) and (**B**) d(CCCTAGGG)2:C-1305 (**D3L**) complex formation. For a full list of δ and Δδ, please consult Supplementary Tables [S1](#MOESM3){ref-type="media"} and [S2](#MOESM3){ref-type="media"}. PyMOL, version 1.8^[@CR39]^ was used for visualization. The above described analysis enabled finding over a twenty DNA/ligand intermolecular NOEs, which are listed in Table [1](#Tab1){ref-type="table"}. Most of these NOEs were the contacts between the aromatic protons of the ligand and the protons of the T4 and A5 residues of the both DNA strands. This finding, along with the broken symmetry of the dsDNA octamer and the observed 1:1 dsDNA:ligand molar ratio, clearly suggested that the ligand should be located at the centre of the DNA duplex. The resulting conclusion was well supported by the fact that in the H6/H8-H1′ NOESY-walks (Fig. [3](#Fig3){ref-type="fig"}), traced for the both DNA strands of the complex, the correlations T4H1′/A5H8 were either not observed (chain **D2L-A**) or were extremely weak (chain **D2L-B**).Table 1Observed d(CGATATCG)~2~:C-1305 (**D2L**) and d(CCCTAGGG)~2~:C-1305 (**D3L**) intermolecular NOE contacts. The intensities were classified as weak/medium/strong on the basis of the integration of the respective crosspeaks in the 2D NOESY spectra (τ~m~ = 150 ms) of the complexes and due to the lack of an internal standard of the DNA/ligand relaxation.NoC-1305 protonDNA protonD2L (intensity)D3L (intensity)**NOE contacts involving aromatic protons of the ligand**1LH3B-A5H1′−+ (weak)2LH3B-A5H8+ (medium)+ (weak)3LH3B-T4H1′+ (weak)+ (medium)4LH3B-T4H2′\'+ (strong)+ (strong)5LH3B-T4H3′−+ (weak)6LH3B-T4H6+ (weak)+ (weak)7LH3B-T4CH~3~−+ (weak)8LH4A-A5H2−+ (weak)9LH4B-A5H4′−+ (weak)10LH4B-A5H8−+ (weak)11LH7A-A5H1′−+ (weak)12LH7A-A5H8+ (weak)+ (weak)13LH7A-T4H6+ (medium)+ (weak)14LH7A-T4CH~3~+ (medium)+ (weak)15LH7A-T4H1′+ (medium)+ (medium)16LH7A-T4H2′+ (strong)+ (strong)^a^17LH7A-T4H2′\'+ (strong)+ (strong)^a^18LH7A-T4H3′+ (weak)−19LH9A-T4CH~3~+ (strong)+ (strong)20LH10A-T4CH~3~+ (weak)+ (weak)**NOE contacts involving protons of the aliphatic sidechain of the ligand**21LH15A-A5H2+ (weak)+ (weak)22LH15B-A5H1′−+ (medium)23LH15B-A5H2+ (weak)+ (weak)24LH15B-A5H4′−+ (medium)25LH17A-A5H2+ (strong)+ (medium)26LH17B-A5H2+ (strong)+ (medium)27LH17B-A5H4′−+ (medium)28LH18A-A5H2+ (strong)+ (weak)29LH18B-A5H1′−+ (medium)30LH18B-A5H2+ (strong)+ (medium)31LH18A-A3H2+ (strong)^b^−32LH18B-A3H2+ (strong)^b^−^a^These crosspeaks were superimposed.^b^These crosspeaks were superimposed. It should be noted that the extensive set of the NOESY contacts observed between the **D2** octamer and the ligand was sufficient to qualitatively reconstruct the intercalation mode, even without the performed in silico refinements. The fact that the aliphatic protons of the ligand exhibited NOE contacts uniquely to the H2 protons of A3 and A5 of the both DNA strands unambiguously placed the aminoaliphatic sidechain inside the minor groove. Regarding the exact positioning of the aromatic moiety, one should notice that the LH3 proton belonging to the one of the peripheral aromatic rings of the ligand interacted uniquely with the protons of one DNA strand (strand **D2L-B**), while the LH7, LH9 and LH10 protons of the opposite aromatic ring gave NOE contacts almost uniquely to the other DNA strand (strand **D2L-A**). Moreover, especially close vicinity was observed between the LH3 and the T4H2″ proton of the strand **D2L-B** and between the LH9 and the T4H2″ proton of the strand **D2L-A**. Such a set of NOEs could be satisfied only while the ligand was placed approximately parallel to the neighboring base pairs and spanned the entire intercalation cavity from the vicinities of the T4 ribose moiety of the one strand to the corresponding ribose moiety of the other. 2D NMR studies of the d(CCCTAGGG)2:C-1305 (D3L) complex {#Sec7} ------------------------------------------------------- Assignment of the ^1^H resonances to the d(CCCTAGGG)~2~:C-1305 (**D3L**) complex was performed in a similar way as it was done for the **D2L** complex. However, in this scenario the **D3L** complex stood out as almost the exclusive (\> 90%) DNA spectral form present in the solution, which was quantified upon the comparison of the averaged cytidine H5--H6 and thymidine CH~3~--H6 NOESY cross-peak intensities of the complexed and ligand-free DNA. The intercalation of the ligand into the **D3** duplex once again broke the symmetry of the helix, which manifested in doubling of ^1^H resonances, mostly of the central residues of the complex. Hence, two DNA strands of the complex were *per analogiam* traced separately and named **D3L-A** and **D3L-B**, respectively. Nevertheless, despite the fact that the resonances of the free **D3** duplex were next to non-existent, severe spectral crowding was observed for the C1--C3 and G6--G8 nucleotides of the complex, yielding a more complicated spectral dataset in comparison to the **D2L** study. Still, it was possible to assign all the aromatic base protons of the complex, as well as all the H1′-H3′ deoxyribose resonances (Supplementary Table [S2](#MOESM3){ref-type="media"}). The H4′ and H5′/H5″ resonances were also partially assigned. The chemical shift changes of the DNA resonances upon complex formation were presented in Fig. [4](#Fig4){ref-type="fig"}b in a similar fashion as for the **D2L** complex. The analysis of the spectral dataset resulted in marking of almost a thirty recorded DNA/ligand intermolecular NOEs, which are listed in Table [1](#Tab1){ref-type="table"}. As it was in case of the **D2L** complex, most of the observed DNA/ligand NOEs were the correlations between the aromatic protons of the ligand and the protons of the T4 and A5 residues of the DNA. On the basis of the aforementioned NOEs, the ligand was unambiguously located---again---at the very centre of the DNA duplex, which was additionally supported by the **D3L-A** and **D3L-B** NOESY-walks. The minor-groove mode of intercalation of the ligand was deduced upon a very similar set of spectral evidences as observed for the **D2L** complex (Table [1](#Tab1){ref-type="table"}). For the **D3L** complex, it was also possible to assign most of the phosphorus resonances (Supplementary Table [S3](#MOESM3){ref-type="media"}) through their correlations with sugar protons measured in the HP-COSY spectrum. Most of the backbone phosphorus atoms displayed very limited chemical shift perturbations (\< 0.1 ppm) upon C-1305 binding. The only major exceptions from this behavior were provided by the phosphate groups of the two A5 residues, which were downfield shifted by 0.6 and 1.0 ppm in strands A and B, respectively. Such a pattern of chemical shift perturbations throughout the **D3**/**D3L** backbone was another very direct proof of the ligand's exclusive intercalation into the central T4-A5 dinucleotide step. Molecular modelling studies {#Sec8} --------------------------- In order to reveal the atomistic details of the d(CGATATCG)~2~:C-1305 and d(CCCTAGGG)~2~:C-1305 interactions, these systems were subjected to NMR-restrained explicit solvent molecular dynamics simulations. Basing on the pH of the solution and according to previous findings^[@CR12]^, it was concluded that both the L8-O oxygen atom and the tertiary nitrogen at the end of the ligand's sidechain (Fig. [1](#Fig1){ref-type="fig"}) should be protonated in the refined experimental conditions (II). Hence, two systems were prepared, containing either **D2** or **D3** dsDNA duplexes and C-1305 in a moderate proximity, solvated with water containing appropriate amounts of Na^+^ and Cl^−^. On the basis of Table [1](#Tab1){ref-type="table"}, thirteen (**D2L** complex) and eighteen (**D3L** complex) DNA/ligand distance restraints (DR) were defined, corresponding to the correlations involving the aromatic protons of C-1305. It was decided to not consider the NOEs involving the protons of the ligand's sidechain, as the preliminary simulations (Supplementary Tables [S6](#MOESM3){ref-type="media"} and [S7](#MOESM3){ref-type="media"}) have demonstrated that not all of them could be satisfied by only one, strictly defined orientation of the sidechain in case of either of the two studied complexes. The complete and equilibrated systems were then subjected to 1 ns-long restrained molecular dynamics (rMD) simulations. The resulting trajectories revealed that for the both considered sequences C-1305 molecule almost instantly intercalated between T4 and A5 from the minor groove of the DNA. The aliphatic sidechain was maintained in the minor groove, whereas the aromatic system of the ligand slightly protruded from the major groove. Such an arrangement was in perfect agreement with all of the enforced DR. The last frames from these trajectories served as a starting point for two independent 1 µs-long MD simulations for each system. In the first simulation (**D2L-MD1** and **D3L-MD1**), DNA/ligand distance restraints were maintained, while during the second one (**D2L-MD2** and **D3L-MD2**) the respective DR were dropped. All four simulated systems were structurally stable on a simulated timescale. Cluster analysis (Supplementary Figures [S4](#MOESM3){ref-type="media"} and [S5](#MOESM3){ref-type="media"}), performed on the 1 µs trajectories, revealed that the complexes **D2L-MD1** and **D2L-MD2** existed in a similar conformation, classified as the main structural cluster, for 90.6% and 72.1% of the simulation time, respectively (Fig. [5](#Fig5){ref-type="fig"}a,b), whereas the systems **D3L-MD1** and **D3L-MD2** exhibited next to identical conformation for 92.2% and 75.6% of the simulation time, respectively (Fig. [5](#Fig5){ref-type="fig"}c,d). The structural clusters extracted from NMR-restrained simulations were very well defined, with RMSD of 2.437 Ǻ and 1.652 Å for simulations **D2L-MD1** and **D3L-MD1**, respectively. Moreover, a noticeable fraction of the intercluster structural variability was due to the non-restrained terminal regions of the DNA and when only the central region was considered (residues 3--6 and the ligand), the RMSD decreased to 1.687 Å and 1.229 Å, respectively.Figure 5The most representative structures of the **D2L** and **D3L** complexes, obtained as centres of the dominant conformational clusters from the four molecular dynamics simulations: (**A**) **D2L-MD1** (restrained), (**B**) **D2L-MD2** (unrestrained), (**C**) **D3L-MD1** (restrained), (**D**) **D3L-MD2** (unrestrained). For more details on clustering, please consult [Supplementary Information](#MOESM3){ref-type="media"}. This figure was prepared using VMD 1.9.3^[@CR40]^, <https://www.ks.uiuc.edu/Research/vmd>. As it goes for the aromatic system of the ligand, its positioning within the intercalation sites was very well established by the NMR-derived distance restraints. In case of the both examined complexes, the C7--C10 region of the ligand was located between A-T4 and A-A5 nucleotides, whereas the C2--C5 region has settled within the B-T4 and B-A5 bases (Fig. [6](#Fig6){ref-type="fig"}a,b). Such a geometry was in perfect agreement with the NOE data reported in Table [1](#Tab1){ref-type="table"}, with just a few observed average NOE violations and no one greater than 0.35 Å (see Table [S8](#MOESM3){ref-type="media"} for a detailed comparison of MD-modelled and NOE-derived distances). Such an orientation of the ligand also enabled an occasional formation of the hydrogen bond between the L8-OH group of C-1305 and the A-A5O4′ oxygen atom of the respective deoxyribose moiety^[@CR12]^. This hydrogen bond was observed in the trajectories of all the conducted MD simulations. Nevertheless, the ligand was still able to slightly rotate within the intercalation sites.Figure 6A closer look on the ligand bound at the 5′-TA-3′/5′-TA-3′ intercalation site in (**A**) **D2L** and (**B**) **D3L** complex. Observed NOEs between the aromatic protons of C-1305 and the protons of the DNA are depicted as red, bidirectional arrows. The accompanying numbers correspond to the respective NOE contacts listed in Table [1](#Tab1){ref-type="table"}. This figure was prepared using VMD 1.9.3^[@CR40]^, <https://www.ks.uiuc.edu/Research/vmd>. In both simulated systems, the aliphatic sidechain of C-1305 was prone to periodical reorientation within the minor groove, visiting both ends of the DNA helix and reaching up to the third and sixth nucleotides in the sequence. This result explained the observation that the DNA/ligand NOEs involving the protons of the aminoalkyl sidechain couldn't be satisfied by a single structure in case of either **D2L** complex and **D3L** complex. For the three-dimensional models of the d(CGATATCG)~2~:C-1305 and d(CCCTAGGG)~2~:C-1305 complexes in PDB format, obtained as centres of the dominant conformational clusters from the restrained **D2L-MD1** and **D3L-MD1** trajectories, please consult [Supplementary Data](#MOESM3){ref-type="media"}. Sequence-specificity studies {#Sec9} ---------------------------- The originally selected d(CGATATCG)~2~ (**D2**) duplex contained 4 out of 10 possible dinucleotide steps: CG/CG, GA/TC, AT/AT and TA/TA. In the presence of the TA/TA step in the duplex, the alternative intercalation sites were utterly ignored by the ligand. This result was rather surprising, since, considering the previous studies on the closely related imidazoacridinone C-1311, a symmetrical intercalation into the GA/TC steps^[@CR15],[@CR16]^ was expected. While the experiments reported herein have proven otherwise, in the next step a systematic sequence-specificity study on the triazoloacridinone C-1305 has been initiated, in order to: (1) confirm the binding preference of C-1305 to the TA/TA step in other sequence contexts, (2) examine the CG/CG, GA/TC and AT/AT steps without the presence of the TA/TA step, and finally (3) examine the remaining 6 dinucleotide sequences. Thus, 11 additional dsDNA palindromic duplexes have been engineered in a way that they would contain all 10 possible dinucleotide steps with a varying 5′ and 3′ base-pair neighbourhood. The first four of the designed duplexes (labelled **D3**--**D6**, Table [2](#Tab2){ref-type="table"}) contained TA/TA or AT/AT step at their very centres with the 5′- or 3′-adjacent GGG/CCC triplets. The following seven palindromes (labelled **D7**--**D13**, Table [2](#Tab2){ref-type="table"}) contained TA/TA step not at the centre of their sequences or did not contain this step at all.Table 2The dinucleotide step preferences of C-1305. The entries in bold represent the intercalation sites unambiguously identified on the basis of the NMR spectra while the remaining ones were inferred through the joint analysis of the entire dataset as described in the "[Results](#Sec2){ref-type="sec"}" and "[Discussion](#Sec10){ref-type="sec"}".Oligonucleotide codenameSequenceRelative affinityMolar ratio of the dsDNA:ligand complexDinucleotide steps binding C-1305Dinucleotide steps missed by C-1305D1CCCGGGWeak1:1CG/CGCC/GGD2CGATATCGStrong1:1**TA/TACG/CG, GA/TC, AT/AT**D3CCCTAGGGVery strong1:1**TA/TACC/GG, CT/AG**D4GGGTACCCStrong1:1**TA/TAGG/CC, GT/AC**D5CCCATGGGMedium1:1TG/CACC/GG, AT/ATD6GGGATCCCNoneNone**NoneGG/CC, GA/TC, AT/AT**D7GTACGTACStrong1:2TA/TAGT/AC, CG/CGD8CTAGCTAGStrong1:2TA/TACT/AG, GC/GCD9GAACGTTCVery weak1:1CG/CGGA/TC, AA/TT, AC/GTD10CGATGCATCGMedium-strong1:1 or 1:2TG/CACG/CG, GA/TC, AT/AT, GC/GCD11CGTAGCTACGStrong1:2TA/TACG/CG, GT/AC, AG/CT, GC/GCD12CTGACGTCAGMedium1:2TG/CACT/AG, GA/TC, AC/GT, CG/CGD13CTAGCGCTAGMedium-strong1:2TA/TACT/AG, AG/CT, GC/GC, CG/CG The results of the observations on a total of 13 palindromic DNA duplexes interacting with triazoloacridinone C-1305 are listed in Table [2](#Tab2){ref-type="table"}. All the sequences were examined by the means of ^1^H NMR in the previously described, refined experimental conditions (II), in the presence of 1 molar equivalent of the ligand. The ligand intercalation propensity to the different duplexes was estimated based on the ratio of the free and ligand-bound DNA in each sample. Considering the duplexes formed by sequences **D3**--**D6** (Table [2](#Tab2){ref-type="table"}), in case of **D3** and **D4** the ligand intercalated exclusively into the central TA/TA step, producing well-resolved NMR spectra similar to the ones recorded for **D2** (Fig. [7](#Fig7){ref-type="fig"}a--f). In a striking contrast, C-1305 exhibited no intercalation into the sequence **D6** at all; only non-specific interactions were observed (Fig. [7](#Fig7){ref-type="fig"}i,j). This finding excluded the GG/CC, GA/TC and AT/AT steps from being even low affinity binding sites for C-1305. The elimination of these three steps from a set of the possible binding sites of C-1305 was crucial for the identification of the most plausible binding site(s) in the rest of the studied sequences.Figure 7NMR spectra recorded for **D2**--**D6** and **D9** duplexes free in solution (a, c, e, g, i, k) and interacting with the triazoloacridinone C-1305 (b, d, f, h, j, l). In every case, the same concentration of duplex DNA of 0.125 mM was used and the drug was added to achieve 1:1 molar ratio. The duplex **D5** interacted with C-1305 with moderate affinity, yet more than one form of the resulting complex was present in the system (Fig. [7](#Fig7){ref-type="fig"}g,h). The coexistence of multiple complexed species disabled the possibility of identification of the binding site(s) by 1D NMR. However, after the exclusion of the GG/CC and AT/AT steps (see duplex **D6**) as possible intercalation sites, only the TG/CA step remained as the plausible binding site of C-1305 in case of the duplex **D5**. Yet, this interaction was noticeably weaker than for the TA/TA step. It should be mentioned that the unique intercalation into the TG/CA base step proposed for **D5** is not at odds with the observed coexistence of multiple bound forms. That is due to the fact that the TG/CA step is not present at the centre of the duplex and thus (1) the two different orientations of the ligand ('up' and 'down') are no longer equivalent and (2) the coexistence of both 1:1 and 1:2 DNA:ligand mol/mol complex stoichiometries is possible. Placing the TG/CA step at the centre of a duplex was not attempted, as this study was limited solely to the palindromic sequences. Whether the GT/AC and AG/CT steps constitute viable intercalation sites remains unsettled for now as they were present in sequences containing the dominating TA/TA site. At the very beginning of this study, the duplex **D1** (Table [2](#Tab2){ref-type="table"}) exhibited a weak interaction with the ligand, yet it was impossible to identify the binding site of C-1305 due to the severely broadened ^1^H resonances (Supplementary Figure [S1](#MOESM3){ref-type="media"}). On the basis of the aforementioned **D6**-based GG/CC exclusion, the CG/CG doublet was proposed as a binding site of the ligand in case of **D1**. However, the affinity of C-1305 towards this site cannot be high due to the following observations made for duplex **D9**. The **D9** palindrome contained CG/CG step at the very centre of the sequence, yet it exhibited a barely detectable interaction with triazoloacridinone C-1305 (Fig. [7](#Fig7){ref-type="fig"}k,l). As the GA/TC step was previously excluded (see **D6**), in case of **D9** the ligand could have bound to the AA/TT, AC/GT and CG/CG steps. Considering the results obtained for **D1**, the CG/CG step seems to be the suspect, yet this conclusion is unsure. Nevertheless, it is certain that the triazoloacridinone C-1305 does not exhibit considerably high affinity to the AA/TT, AC/GT and CG/CG doublets. The remaining three, still unsettled dinucleotide steps are the AG/CT, GT/AC and GC/GC. They are present, individually or in-pairs, in the sequences **D7**, **D8** and **D10**--**D13** (Table [2](#Tab2){ref-type="table"}, Supplementary Figure [S6](#MOESM3){ref-type="media"}). In all of these cases, the binding of the ligand to any of those three doublets was not observed. It must be noted, though, that each one of these six duplexes contained the notably preferred TA/TA or TG/CA steps. Therefore, the resulting conclusion is that the binding of C-1305 to the AG/CT, GT/AC and GC/GC doublets cannot be excluded, yet the ligand's affinity to any of those three steps has to be incomparably lower than to the TA/TA and TG/CA sequences. Discussion {#Sec10} ========== On the basis of the collected experimental data, we have confirmed that triazoloacridinone C-1305 binds into double stranded DNA via intercalation through the minor groove of the helix. Extensive NMR and MD studies on the d(CGATATCG)~2~:C-1305 (**D2L**) and d(CCCTAGGG)~2~:C-1305 (**D3L**) complexes have revealed that the ligand binds at the very centres of the hosting octamers. Detailed analysis has shown that in both cases the aromatic system of C-1305 slightly protrudes from the major groove, whereas the aliphatic sidechain is located within the minor groove of a DNA duplex. Such an alignment of the ligand was unambiguously defined by two sets of host/guest intermolecular NOEs (Table [1](#Tab1){ref-type="table"}). For instance, for both systems we could observe NOE contacts between the aromatic protons of C-1305 and the T4 methyl groups inside the major groove, whereas the aliphatic sidechain of the ligand exhibited NOE contacts uniquely to the DNA protons located in the minor groove. All the remaining intermolecular NOEs, listed in Table [1](#Tab1){ref-type="table"}, have also strongly supported our resulting conclusion. Given the established minor-groove-binding mode and the asymmetry of the aromatic region of the ligand, one should notice that C-1305 molecule can enter a given intercalation site assuming one of the two possible orientations of the ring system ("up" or "down"). This should result in two spectral forms of the DNA:ligand adduct per intercalation site. Thus, one should ask the question: why only a single form of the complex was observed for the **D2L** and **D3L** systems? The reason was the fact that the studied triazoloacridinone has intercalated into the very *centres* of the *palindromic* octamers. In such cases, two possible orientations of the ligand's ring system within a binding site became equivalent. In any other case, the systems' equilibrium would get more complex, especially if the intercalation of C-1305 would have occurred at multiple, various dinucleotide steps. As a result, we would record barely readable ^1^H NMR spectra consisting of many, overlapping resonances, possibly broadened due to non-specific interactions with the self-associated ligand molecules. The observed selective binding of C-1305 at the TA/TA step of the originally selected d(CGATATCG)~2~ (**D2**) octamer was a rather unexpected result. We thus continued the research to check whether this was a harbinger of a sequence-specificity of C-1305 to be discovered. Hence, we have examined the interactions of C-1305 with 11 additional short dsDNA palindromes containing all 10 possible dinucleotide steps in varying sequence contexts. This was done in order to provide as deep insight into the potential sequence-specificity of triazoloacridinone C-1305 as possible. Basing on all the results gathered, we were able to define the C-1305's dinucleotide step preference in the following order: TA/TA \> TG/CA ≫ CG/CG. On the other hand, three other dinucleotide steps: GG/CC, GA/TC and AT/AT were definitively ruled out from the family of the possible binding sites of the triazoloacridinone C-1305. For the remaining four steps: AA/TT, GT/AC, AG/CT and GC/GC, the intercalation was not observed, yet could not be unambiguously excluded based on the available data. It should be noted that all the dinucleotide steps we have observed intercalation into, namely TA/TA, TG/CA and CG/CG, are the 5′-pyrimidine-purine-3′ sequences. Based on the order of affinity towards these steps it can be concluded that triazoloacridinone C-1305 clearly prefers thymidine at the 5′ side of the binding cavity. The exclusion of the GG/CC step was surprising in the light of the previously reported effective intercalation of C-1305 into the G-triplet-rich oligonucleotides^[@CR12]^. However, we believe that the previously reported binding of C-1305 into such sequences may be related to the GGG-containing DNA fragments assuming other-than-duplex conformations in a solution. For instance, a G-triplet-rich DNA is likely to form the G-quadruplexes, which were previously demonstrated to be able to effectively bind molecules structurally similar to the triazoloacridinone C-1305. The GGG-containing duplexes used in our study were confirmed in situ to be present uniquely in the duplex form in our refined experimental conditions, while no such test was performed in the previous report^[@CR12]^. On the basis of the interactions of C-1305 with the **D2**, **D3** and **D4** duplexes, which contained the AT\*AT/AT\*AT, CT\*AG/CT\*AG and GT\*AC/GT\*AC tetranucleotide steps, respectively (the asterisk stands for the intercalation site), we have observed that while the triazoloacridinone C-1305′s sequence-specificity can be interpreted in terms of dinucleotide steps, the strength of the interaction depends also on the 5′ and 3′ base pairs adjacent to the binding cavity. Previous studies have reported that intercalative binding can induce structural rearrangements in the base pairs separated by one nucleotide from the direct binding site, thus confirming they can play a role in ligand recognition by the DNA^[@CR23],[@CR24]^. Our study revealed that in case of the CT\*AG/CT\*AG sequence (incorporated in the **D3** duplex) the intercalation of the triazoloacridinone C-1305 was so effective that the system's equilibrium could be almost completely shifted towards the complex formation, while the narrowness of the ^1^H NMR resonances was still maintained. This was not possible in case of the AT\*AT/AT\*AT (**D2**) and GT\*AC/GT\*AC (**D4**) binding regions. The influence of the neighboring base pairs on the C-1305 affinity to the TA/TA step could be explained by the combination of the two effects: (1) the stacking interactions between the 5′- and 3′-adjacent base pairs and the distorted dinucleotide binding cavities, (2) the interactions between the ligand sidechain and the 5′- and 3′-adjacent base pairs. While sidechain/minor-groove interactions were indeed clearly observed during our stereostructural studies, they appear to be rather similar in both the **D2L** and **D3L** complexes. In both cases, the terminal methyl groups of the side chain locate themselves in proximity of the base pairs flanking the intercalation site, and likely undergo regular reorientations from one side of the duplex to the other. The energy of these interactions might still be different in the **D2L** and **D3L** complexes (due to slightly different geometries and charge distributions of the minor grooves), but our current data does not provide definite answers in this matter. Likewise, to reliably evaluate the degree to which the stacking interactions throughout the duplex are differently affected in **D2L** and **D3L**, additional computational studies would be required. Nevertheless, a glance at Fig. [4](#Fig4){ref-type="fig"} allows to conclude that the chemical shift perturbations in **D3L** appear to be mostly confined to the central TA/TA step, while in **D2L** they spread considerably further throughout the duplex. Such a pattern could suggest that only negligible structural changes occur in the neighboring base pairs in **D3L**, while in the case of **D2L** ligand binding induces some geometric rearrangements (likely resulting in stacking distortions) further away from the binding site. Such a comparably less strained bound state for **D3L** might provide a partial explanation for the higher affinity of C-1305 towards **D3**. The observation of a marked selectivity of C-1305 for TA/TA dinucleotide steps raises the question of the energetic basis for such a high preference towards this specific site. Our NMR-derived atomistic model of the C-1305 molecule bound within a TA/TA dinucleotide step does not reveal any specific interactions (apart from the sequence-independent transient hydrogen bonding between the ligand's OH group and a deoxyribose moiety). Thus, the reason for the selective intercalation into the TA/TA steps must reside in a favorable energetic balance from some other sources (stacking, hydration etc.). While considering such a balance for the intercalation process at a given site, one has to consider not only the favorable contributions from the newly formed DNA--ligand interactions, but also the energetic penalties for the interactions lost when the two base pairs are separated, i.e., loss of stacking energy between the two base pairs, disruption of specific hydration patterns, etc. While the part of the equation dealing with the DNA--drug interactions is of course ligand-dependent, the penalty for the separation of a given dinucleotide step is, on the other hand, an intrinsic property of the dsDNA itself and, as such, has been extensively studied. Multiple studies aimed at the understanding of the DNA thermodynamics have formulated the findings in terms of intrinsic stacking energies at different dinucleotide steps, as gathered for example by SantaLucia^[@CR25]^. However, perhaps the most direct estimation of a dinucleotide step separation energy comes from the study of the stacked-unstacked equilibria in the nicked DNA molecules of different sequences^[@CR26]^. Interestingly, the energetic penalty of the separation of the TA/TA step is the lowest one among all the possible dinucleotide steps according to both SantaLucia^[@CR25]^ and Protozanova et al.^[@CR26]^. Thus, it may appear that for the intercalators with the relatively low differences in the dsDNA:ligand interaction energies at different dinucleotide steps, the TA/TA step should stand out as the default binding site. Whether the TA/TA preference of C-1305 is actually decided by this mechanism is difficult to prove with the current data and some additional research is needed, aiming at evaluation of dsDNA:C-1305 stacking interactions at various dinucleotide steps. Nevertheless, an additional hint in favor of such an interpretation is provided by the fact that TG/CA step, the other site preferred by C-1305, is also the second most easily separated one according to Protozanova et al.^[@CR26]^. Regardless whether our hypothesis is true, the striking difference in the ligand's affinity to the TA/TA and AT/AT steps is evident (stacking energies of 0.2 kcal/mol and 1.3 kcal/mol, respectively^[@CR26]^). Such a preference was actually observed for several other intercalators, also some structurally unrelated to C-1305. For instance, intercalation into the very centre of the d(CGTACG)~2~ duplex was previously reported for ACRAMTU, another acridine-based dsDNA topoisomerase II poison^[@CR27]^. The Authors established the TA/TA, CG/CG and GA/TC as the preferred binding sites of the drug, while the compound intercalated most effectively into the d(TATATATA)~2~ octamer. In a striking contrast, the d(AGGGCCCT)~2~ and d(AAAATTTT)~2~ duplexes were proven to be the disfavored recipients of ACRAMTU^[@CR27]^. In another study^[@CR28]^, the d(CCGGTACCGG)~2~ and d(CCGGATCCGG)~2~ DNA duplexes were crystalized in the presence of λ-\[Ru(phen)~2~dppz\]^2+^. In case of the d(CCGGTACCGG)~2~ sequence deep intercalation into the TA/TA step was observed, while the inverted AT/AT step, present in the second duplex, was utterly ignored^[@CR28]^. The Authors of this report explained this selectivity by excellent ligand-DNA shape-complementarity at the TA/TA step. Recently, Cu-Oda was demonstrated to be able to distinguish between the TA/TA and AT/AT steps. The copper complex was proven to effectively intercalate into the TA/TA step, whereas it induced the transition from B-DNA to Z-DNA while bound to the AT/AT dinucleotide sequence^[@CR29]^. The intercalation of a drug molecule into dsDNA is a complex process involving a delicate balance between the energetic contributions from stacking interactions lost and formed, as well as from specific hydrogen bonding interactions. The net effect of all these factors dictates the sequence specificity of a given intercalator. For the triazoloacridinone C-1305 studied here we have found a marked preference for the interaction with TA/TA step, characterized by very low intrinsic stacking energy. As the interaction lacks specific hydrogen bonding, it seems reasonable to assume that this property has a substantial contribution into determining the binding specificity. Such an interpretation might hold true for other intercalators, selecting the TA/TA step without forming specific hydrogen bonding at this site. Materials and methods {#Sec11} ===================== NMR sample preparation {#Sec12} ---------------------- All the chemically synthesized oligonucleotides used in this study were purchased from Sigma-Aldrich. Before the NMR measurements, each purchased oligo was further purified by precipitation from acetone containing 1% NaClO~4~. This process served to remove an impurity giving rise to very strong signals in the proton NMR spectra at around 1.28, 1.99, 3.21 and 8.60 ppm (triethylamine-acetate, used by the oligo supplier during HPLC purification). Thanks to their palindromic nature, each oligo used in this study is fully self-complementary and thus the preparation of the duplex samples consisted simply of dissolving the purified material in an appropriate buffer. A range of buffer compositions (buffering agent, pH, NaCl concentration) were tested throughout this study in a pursuit to optimize the quality of the measured NMR spectra. The two most relevant conditions (see "[Results](#Sec2){ref-type="sec"}") were: (I) 10 mM phosphate buffer of pH 7.5, containing 150 mM NaCl and (II) 2.5 mM cacodylate buffer of pH 5.0, containing 10 mM NaCl. To prepare the samples of the dsDNA:C1305 intercalation complexes, the ligand was added to the pre-mixed NMR sample from a concentrated stock solution in water, to reach to duplex:ligand molar ratio of 0.5, 1.0, 1.25 or 2.0, depending on the sample. NMR spectra {#Sec13} ----------- All NMR spectra were collected using a 700 MHz Bruker Avance III HD spectrometer, equipped with a QCI CryoProbe. After two 8 bp DNA duplexes forming a single, well-defined complexes with C1305 were identified (see "[Results](#Sec2){ref-type="sec"}"), a set of 2D spectra was recorded for resonance assignment for each of the free duplexes. It comprised the NOESY (150 ms mixing time) and TOCSY (60 ms spin-lock time) spectra measured at 5 °C in 90% H~2~O/10% D~2~O, as well as, the NOESY (150 and 400 mixing time), HC-HSQC, HP-COSY and DQF-COSY spectra acquired in 100% D~2~O at 25 °C. These spectra were recorded in conditions (I). The resonance assignment process itself was performed using standard approaches^[@CR21]^. In order to obtain the reference data for the ligand, NOESY (150 mixing time) and TOCSY (60 ms spin-lock time) spectra were recorded for this compound at 5 °C in 90% H~2~O/10% D~2~O in conditions (I) and (II). For the assignment of the dsDNA--C1305 complexes and for the identification of the DNA--ligand cross-peaks, the NOESY (150 ms mixing time), TOCSY (60 ms spin-lock time) and HC-HSQC spectra were recorded for the complexes at 5 °C, in both 90% H~2~O/10% D~2~O and 100% D~2~O in conditions (II). HP-COSY spectra were also recorded for the two complexes, but only in the case of **D3L** a good quality spectrum was observed as for **D2L** the resonances turned out to be too broad to provide enough signal. The sets of spectra originally collected to assign the free duplexes were recorded in somewhat different experimental conditions (I) than the spectra of the dsDNA:C-1305 complexes (II). Thus, in order to extract the most meaningful chemical shift differences due to complex formation, the following procedures were applied to reassign the free duplexes in conditions (II). The **D2** duplex, which coexisted in a significant quantity with the **D2L** complex, was reassigned directly from the spectra recorded for the complex. For the **D3** duplex, additional NOESY spectra were recorded at 5 °C in conditions (II), in both 90% H~2~O/10% D~2~O and 100% D~2~O. Using these spectra, the assignments made previously in conditions (I) were easily transferable to conditions (II). Fragments of TOCSY and NOESY spectra, displaying correlations essential for assignment of ligand's resonances and identification of intermolecular DNA:C-1305 interactions in **D2L** and **D3L** complexes have been presented in Supplementary Information as Figures [S7](#MOESM3){ref-type="media"}--[S14](#MOESM3){ref-type="media"}. Molecular modelling {#Sec14} ------------------- Molecular dynamics (MD) simulations were performed for the d(CGATATCG)~2~:C-1305 and d(CCCTAGGG)~2~:C-1305 intercalation complexes explicitly solvated in cubic boxes, with 4,497 and 4,502 TIP3P water molecules, respectively, at 0.15 M concentration of NaCl. The force field parameters for the DNA octamers were taken from the latest iteration of CHARMM36 nucleic acid force field^[@CR30]^. The parameters for C-1305 were taken from the latest version of CHARMM36 Generalized Force Field (CGenFF)^[@CR31]^, whereas the partial atomic charges of the ligand were calculated ab initio using GAUSSIAN09 software on the MP2/6-31G\* level of theory^[@CR32]^. All energy minimizations and MD simulations were carried out using GROMACS 2018.2^[@CR33]^. All the MD simulations were conducted using the leapfrog scheme with a time step of 2 fs. The particle mesh Ewald technique with a cutoff of 1 nm and grid spacing of approx. 0.1 nm was used to evaluate electrostatic forces^[@CR34]^. The van der Waals interactions were calculated using Lennard--Jones potential with a cutoff of 1 nm. The simulations were conducted at a constant temperature of 278 K and at a constant pressure of 1 bar, using the weak coupling method^[@CR35]^. Preparation of the molecular models of the dsDNA:ligand complexes was identical for the both studied sequences and was executed as follows. After obtaining an initial B-form of DNA structure from X3DNA 2.3^[@CR36]^, one C-1305 molecule with a protonated tertiary nitrogen at the end of the sidechain was placed in a moderate proximity of a DNA duplex using VMD software. After the appropriate energy minimization and 100 ns of MD-based initial equilibration with position restraints set on DNA and ligand molecule, the system was simulated for 1 ns. During this run, 13 (**D2L** complex) and 18 (**D3L** complex) distance restraints corresponding to the NOE contacts between the aromatic protons of the ligand and the protons of the DNA were applied (Table [1](#Tab1){ref-type="table"}). This was done using the GROMACS implementation of the restraining potential which adds a quadratic penalty to the potential when a distance exceeds a lower or upper threshold. The same force constants of 1,000 kJ mol^−1^ nm^−2^ were used for all restrained distances. The above described simulation resulted in a DR-driven intercalation of C-1305 molecule into the 5′-TA-3′/5′-TA-3′ site from the minor groove of the DNA duplex. Afterwards, the final frame from the resulting trajectory was extracted. This frame was set as a starting point for the two independent 1 µs-long MD simulations described below. Each of them was preceded by 100 ns of further equilibration with position restraints set on DNA and ligand molecules. During the first run, designated as **D2L-MD1** and **D3L-MD1** in the main text, respectively, the systems were subjects to the aforementioned 13 (**D2L**) and 18 (**D3L**) distance restraints derived from the NOESY experiment of τ~m~ = 150 ms and another 14 (**D2L**) and 16 (**D3L**) distance restraints, strengthening the hydrogen bonds in Watson--Crick base pairs. It was decided to stabilize all the base pairs except the terminal GC pairs, since the G8 imino proton resonance was not observed in the ^1^H NMR spectra of the both **D2L** and **D3L** complexes. During the second run, designated as **D2L-MD2** and **D3L-MD2** in the main text, the systems were subjects only to the 14/16 distance restraints strengthening the hydrogen bonds in Watson--Crick base pairs. This was done in a similar way as it was described above for the **D2L-MD1** and **D3L-MD1** simulations. For more details on distance restraints, please consult [Supplementary Data](#MOESM3){ref-type="media"} (Supplementary Tables [S4](#MOESM3){ref-type="media"} and [S5](#MOESM3){ref-type="media"}). Cluster analysis was performed using the Daura method^[@CR37]^. Supplementary information ========================= {#Sec15} Supplementary Information 1. Supplementary Information 2. Supplementary Information 3. **Publisher\'s note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Tomasz Laskowski and Witold Andrałojć. Supplementary information ========================= is available for this paper at 10.1038/s41598-020-68609-8. This work was funded by a grant from Polish National Science Centre no. 2017/01/X/NZ7/00752. Jakub Grynda received a financial doctoral scholarship from Polish National Science Centre no. 2018/28/T/ST5/00165. This research was supported in part by PLGrid infrastructure. T.L., W.A. and J.M. conceived the experiments, T.L., W.A., J.G. and Z.G. conducted the experiments, T.L., W.A. and P.G. analyzed the results. T.L. and W.A. wrote the main manuscript and prepared Figs. [1](#Fig1){ref-type="fig"}, [2](#Fig2){ref-type="fig"}, [3](#Fig3){ref-type="fig"}, [4](#Fig4){ref-type="fig"}, [5](#Fig5){ref-type="fig"}, [6](#Fig6){ref-type="fig"}, [7](#Fig7){ref-type="fig"}. All authors have reviewed the manuscript. Most of the data generated or analyzed during this study are included in this published article (and its Supplementary Information files). The remaining datasets generated and analyzed during the current study, i.e. full NMR spectra and MD trajectories, are available from the corresponding author on reasonable request. The authors declare no competing interests.

Introduction {#s1} ============ RNA viruses possess an extraordinary capacity to adapt to changing environments due to their high mutation rate [@ppat.1000502-Holland1]. Recent epidemics of West Nile virus (WNV, *Flaviviridae*, *Flavivirus*) and chikungunya virus (CHIKV, *Togaviridae*, *Alphavirus*) were driven by relatively small genetic changes that increased the efficiency with which the viruses were transmitted by mosquito vectors (*Culex spp.* and *Aedes albopictus*, respectively) [@ppat.1000502-Ebel1]--[@ppat.1000502-Tsetsarkin1]. In the case of WNV, a single conservative amino acid substitution is responsible for increased transmission efficiency at early timepoints after the mosquito acquires an infectious bloodmeal [@ppat.1000502-Ebel1],[@ppat.1000502-Moudy1]. The change in transmission efficiency has led to the complete displacement of the parental WNV genotype by the derived strain, demonstrating the power of small genetic changes to profoundly influence arbovirus transmission patterns and the importance of mosquitoes in shaping their populations [@ppat.1000502-Snapinn1]. Studies examining the population genetics of WNV have shown that WNV consists of a genetically diverse population within hosts, and that infections within mosquitoes are more genetically diverse than are those within birds [@ppat.1000502-Jerzak1]. The underlying mechanism for the increased genetic diversity in mosquitoes, however, is poorly understood. Eukaryotic organisms generally possess pathways that respond to pathogen-associated molecular patterns (PAMPs) that, when activated, lead to suppression or elimination of the pathogen. Double-stranded RNA (dsRNA) is one such PAMP that is a powerful trigger of innate antiviral responses. Invertebrates, including mosquitoes and other dipterans, respond to virus infections through RNA interference (RNAi) [@ppat.1000502-Aliyari1]--[@ppat.1000502-Zambon1]. In RNAi, dsRNA molecules are identified, processed and ultimately used to guide sequence-specific degradation of homologous RNA sequences. Because many RNA viruses, including WNV and other arboviruses, have complex secondary structures within their genomes and form dsRNA intermediates during replication, their genomes may be susceptible to RNAi-based degradation. Indeed, several studies have demonstrated that the RNAi pathway can function to limit viral infections in members of the diptera [@ppat.1000502-Aliyari1]--[@ppat.1000502-Wang1]. Virus-derived small interfering RNAs (viRNAs) have been detected in *Aedes aegypti* mosquitoes infected with the arboviruses dengue virus (DENV) and Sindbis virus (SINV), and in *Drosophila* infected with flock house virus (FHV) [@ppat.1000502-Aliyari1],[@ppat.1000502-Myles1],[@ppat.1000502-SanchezVargas1]. In addition, suppression of RNAi machinery in *Ae. aegypti* resulted in transient increases in DENV and SINV titers [@ppat.1000502-SanchezVargas1],[@ppat.1000502-Campbell1]. Similar studies in RNAi deficient *Anopheles gambiae* showed increased viral dissemination rates and titers of intra-thoracically inoculated O\'nyong-nyong virus [@ppat.1000502-Keene1]. An expanding body of literature thus suggests that RNAi functions as an innate antivirus response in mosquitoes. Interpreting this literature, however, has been problematic. First, the systems studied have generally not included ecologically relevant virus-vector pairs (but with the notable exception of Sanchez-Vargas et al. [@ppat.1000502-SanchezVargas1]). For example, neither *Aedes aegypti* nor *Drosophila* are vectors of SINV or FHV, respectively. Second, although mosquitoes acquire infection orally during bloodfeeding, many published studies have examined intra-thoracically inoculated mosquitoes. Mosquito midgut epithelial cells are the first cells to become infected and midgut tissues present important infection and escape barriers that appear to control the vector competence of mosquitoes for arboviruses [@ppat.1000502-Bennett1]--[@ppat.1000502-Myles2]. Consequently, surprisingly little is known about the innate antivirus response mounted by ecologically important vector mosquitoes in physiologically relevant tissues. Viruses possess counter-measures to escape host antiviral responses. One such counter-measure may be their characteristically high mutation rates. RNAi is highly sequence specific and single nucleotide mismatches between the guide and target sequences can drastically reduce or abolish silencing effects [@ppat.1000502-Pusch1],[@ppat.1000502-Westerhout1]. The sequence specificity of the RNAi response may therefore influence virus genetics. For example, artificial induction of the RNAi pathway using virus-specific siRNAs has been shown to result in (a) the accumulation of mutations in the targeted regions and (b) viral escape in poliovirus (*Picornaviridae*; *Enterovirus*), hepatitis C virus (*Flaviviridae*; *Hepacivirus*), LaCrosse virus (*Bunyaviridae*; *Bunyavirus*) and HIV-1 [@ppat.1000502-terBrake1]--[@ppat.1000502-Gitlin1]. Whether this occurs spontaneously following oral infection of competent vector mosquitoes, however, is less clear. Since WNV genetic diversity is increased within mosquitoes, we hypothesized that RNAi might lead to increased mutational diversity by creating an intracellular environment in which mutant genomes are less likely to be degraded by viRNAs deriving from common WNV genome sequences. In particular, we used a deep sequencing approach (technology reviewed extensively in [@ppat.1000502-Shendure1]) to determine whether siRNAs mapping to the WNV genome are produced in mosquito midguts in response to infection. Using these data, we then determined whether particular genome regions are consistently targeted and whether differential targeting of the genome leads to increases in genetic diversity in intensely targeted regions. Results {#s2} ======= Mosquito infection {#s2a} ------------------ To ensure that only WNV-positive midguts were included in the infected group, and that all putatively negative mosquito midguts were indeed negative, midgut RNAs were screened for the presence of WNV RNA by RT-PCR. At both timepoints sampled, 12 of the 15 midguts analyzed were positive, and all control mosquito midguts were negative. Characterization of viRNAs from *Cx. p. quinquefasciatus* midguts {#s2b} ----------------------------------------------------------------- To define the population of small RNAs in mosquito midguts, small RNA (sRNA) libraries from WNV-infected and uninfected midguts after a 7 or 14 day extrinsic incubation period (EIP) were subjected to high-throughput sequencing. The total number of sRNA reads obtained from each of the four libraries sequenced ranged from 3.7 to 4.9 million ([Table 1](#ppat-1000502-t001){ref-type="table"}). To determine the presence of viRNAs, sRNA reads were aligned to the WNV genome. There were 2,544 (1,701 unique) and 4,419 (2,629 unique) reads that aligned to the WNV genome following a 7 or 14 day EIP, respectively, where unique reads represent individual viRNAs that code for a specific nucleotide sequence. As expected, these reads had a mean and mode length of 21 nt ([Figure 1](#ppat-1000502-g001){ref-type="fig"}, [Table 1](#ppat-1000502-t001){ref-type="table"}). In addition, the average quality scores \[Illumina scores, scaled from 1 (minimum) to 40 (maximum)\] were extremely high (90% of the reads \>36) and greater than 80% of these reads perfectly aligned with the viral genome ([Table 1](#ppat-1000502-t001){ref-type="table"}). A small population of sRNAs from the uninfected mosquito midguts aligned to the WNV genome. A high proportion of these reads required 2 mismatches in order to align to the WNV genome; 50% and 86% at 7 and 14 days post bloodmeal, respectively, ([Table 1](#ppat-1000502-t001){ref-type="table"}). {#ppat-1000502-g001} 10.1371/journal.ppat.1000502.t001 ###### Characteristics of small RNA libraries sequenced. {#ppat-1000502-t001-1} Mosquito sRNA library sequenced: Total \# reads (×10^6^) Reads Aligning to WNV genome: Number of mismatches required for alignment n (%): ---------------------------------- ------------------------- ------------------------------- ---------------------------------------------------- ------ ------------ ---------- --------- WNV (−) 7 dpi 4.9 18 (14) 20.6 37.8 7 (39) 2 (11) 9 (50) WNV (−) 14 dpi 3.9 65 (21) 20.5 38.9 8 (12) 1 (2) 56 (86) WNV(+) 7 dpi 4.7 2,544 (1,701) 21.1 38.5 2,104 (83) 343 (13) 97 (4) WNV(+) 14 dpi 3.7 4,419 (2,629) 21.0 38.8 3,754 (84) 513 (12) 152 (3) The orientation of the WNV strand targeted by viRNAs was determined. At both a 7 and 14 day EIP, 74% of all viRNAs were derived from the positive sense viral genome. Subsequently, genomic equivalents for both the positive and negative strands were determined by two-step RT-PCR. 81% and 80% of the total WNV RNA detected was from the positive strand at 7 and 14 days, respectively. viRNA distribution and abundance {#s2c} -------------------------------- The percent coverage of the WNV genome by viRNA was determined. 81.75% and 91.88% of the genome was targeted by at least one viRNA in the WNV infected samples following a 7 or 14 day EIP, respectively. To assess positional and regional differences in the intensity of viRNA targeting of the WNV genome in mosquito midguts, the frequency of viRNA reads mapping to each nucleotide in the WNV genome was computed ([Figure 2A](#ppat-1000502-g002){ref-type="fig"}). Inspection of these results revealed (A) an asymmetric distribution of viRNAs across the genome with some regions being highly targeted and others weakly or not targeted and (B) that peaks in the frequency distribution of hits along the genome at 7 days were also apparent at 14 days, although peaks at 14 days tended to be higher, with some exceptions. {#ppat-1000502-g002} The most intensely targeted portion of the WNV genome, at both 7 and 14 days, was an approximately 200 nt region of the Capsid (C) coding sequence. Genome position 176 within this region was targeted by 64 and 120 viRNAs at 7 and 14 days, which was the highest peak in either dataset. Examination of the orientation of viRNAs targeting this region revealed roughly equal targeting of both positive and negative strand WNV genomes ([Figure 2C](#ppat-1000502-g002){ref-type="fig"}). A similar examination of a portion of the 3′UTR that possesses well-characterized secondary structures revealed less intense targeting, and a more pronounced bias toward viRNAs targeting the positive WNV genome ([Figure 2D](#ppat-1000502-g002){ref-type="fig"}). We then examined viRNAs that were apparently abundantly produced in mosquito midguts. At both 7 and 14 days the abundance of each unique read was calculated which ranged from 1 to 23 ([Table 1](#ppat-1000502-t001){ref-type="table"}). In order to assess whether the observed distribution of read abundance at both 7 and 14 days was significantly different than would be expected by chance, we conducted a permutation analysis. For each time point, n = 2544 (day 7) or n = 4419 (day 14) genomic positions were randomly sampled with replacement. The sampling procedure was repeated 100,000 times to obtain a theoretical distribution of read abundances, and this distribution was compared to the observed distribution by the Kolmorogov-Smirnov test ([Figure 3](#ppat-1000502-g003){ref-type="fig"}). The observed abundance distribution was significantly different from the null expectation (P\<0.001 at 7 and 14 days). In addition, the intensity of viRNA targeting of each nucleotide of the genome was highly correlated at 7 and 14 days ([Figure 4A](#ppat-1000502-g004){ref-type="fig"}). Further, a subset of viRNAs was detected at both sampled timepoints. The abundance of these "common" reads at 7 and 14 days was also highly correlated ([Figure 4B](#ppat-1000502-g004){ref-type="fig"}). {#ppat-1000502-g003} {#ppat-1000502-g004} RNAi influence on virus diversification {#s2d} --------------------------------------- The RNAi pathway is highly sequence specific and mutations in the target dsRNA sequence can have profound effects on its efficiency. Therefore we sought to determine whether highly targeted regions would be more likely to contain mutations compared with relatively weakly targeted regions. Initially, we used imperfect reads (i.e. reads that required one or two mismatches in order to be aligned to the WNV genome) as a proxy for viral mutant sequences, hypothesizing that these reads derived from mutant genomes and had been processed by the host cell RNAi machinery. Putative mutant basecalls with quality scores of less than 30 were discarded. Thus, the probability that basecalls included as mutations in this analysis are read errors was less than 0.0010. At both timepoints sampled, mismatched nucleotides within imperfect reads were covered by a significantly higher number of viRNAs compared to nucleotides that had only perfect reads and were covered by at least a single viRNA ([Table 2](#ppat-1000502-t002){ref-type="table"}, T-test P\<0.0001). 10.1371/journal.ppat.1000502.t002 ###### Increased viRNA coverage of WNV genome at imperfectly aligned positions. {#ppat-1000502-t002-2} Match type: Mean viRNA coverage per nt (Std. Dev., n) ------------------------------------------ ------------------------------------------- -------------------- Perfect 5.80 (7.00, 8,707) 8.91 (9.96, 9,712) Imperfect 10.72 (10.76, 300) 15.40 (14.32, 420) P-value[\*](#nt101){ref-type="table-fn"} 6.14×10^−14^ 1.63×10^−18^ **\*:** T-test for independent samples with unequal variances. Next, we tested the association between viRNA coverage and mutation frequency directly. To accomplish this, we identified regions in NS5 and the 3′-UTR of the WNV genome that had varying degrees of viRNA coverage and sampled WNV genomes from the same RNA specimens used to generate sRNA libraries. viRNA coverage per nucleotide along the entire genome was used to determine the frequency distribution of number of viRNA hits per nucleotide position. Positions were then ranked into quartiles and the interquartile range combined. At 7 days there were three mutations identified in NS5 and no mutations in the 3′-UTR. These mutations were evenly distributed across the three frequency distribution classifications. In WNV genomes sampled at 14 days, mutations were detected in both NS5 and 3′-UTR with increasing viRNA coverage associated with increasing mutation frequency. (Chi-squared test for trend, p = 0.0393) ([Figure 5](#ppat-1000502-g005){ref-type="fig"}). {#ppat-1000502-g005} Discussion {#s3} ========== We used massively parallel sequencing to comprehensively analyze viRNA from the midgut of *Cx. p. quinquefasciatus*. Mosquitoes were sampled after a 7 or 14 day EIP in order to assess differential production and positioning of viRNAs aligned to the WNV genome. At both sampling points, 80% of midguts were infected, indicating that the *Cx. p. quinquefasciatus* used in our studies were highly susceptible to infection by WNV. WNV-positive midgut RNA samples from each were then pooled in order to obtain sufficient RNA concentrations for sRNA library construction. In addition, pooling RNA allowed us to look at a generalized response to WNV within midguts. Additional studies will be required to characterize individual mosquito RNAi responses to WNV. Between two and five thousand high-quality reads from infected midguts mapped to the WNV genome at both sampling points ([Table 1](#ppat-1000502-t001){ref-type="table"}). The average read length was approximately 21 nt, and the vast majority were between 20 and 22 nt, suggesting that reads mapping to the WNV genome were mainly viRNAs ([Figure 1](#ppat-1000502-g001){ref-type="fig"}). Furthermore, the majority of these reads matched perfectly to the sequence of the infectious clone-derived WNV used to infect the mosquitoes. In contrast, a very small population (\<70 reads) of siRNAs mapping to the WNV genome were identified from uninfected control mosquitoes. These reads were similar in length and quality to those from infected midguts, but tended to require more mismatches to align to WNV ([Table 1](#ppat-1000502-t001){ref-type="table"}). The origin of these siRNAs is unknown, although the presence of insect-specific flaviviruses such as *Culex* flavivirus and T\'Ho virus has been reported in wild *Culex spp.* populations and could have led to the observed data [@ppat.1000502-FarfanAle1]--[@ppat.1000502-MoralesBetoulle1]. In addition, flavivirus-like genetic material, termed cell-fusing agent virus, has been identified from the genomes of *Culex spp.* populations in Puerto Rico [@ppat.1000502-Cook1]. To eliminate the possibility that these viruses were the source of the siRNAs in our uninfected control mosquitoes, three pools of ten male and female *Cx. p. quinquefasciatus* from our colony were screened for the presence of flavivirus RNA by one-step RT-PCR using flavivirus specific primers [@ppat.1000502-FarfanAle1]. By this approach, we were unable to detect the presence of flavivirus RNA in our colony mosquitoes, suggesting that these siRNAs were not derived from a contaminating insect-specific flavivirus (data not shown). Further, alignment of our sRNA libraries against multiple irrelevant *Flavivirus* genomes and one *Alphavirus* genome produced very few reads mapping to anything other than WNV ([Text S1](#ppat.1000502.s004){ref-type="supplementary-material"}, [Table S2](#ppat.1000502.s003){ref-type="supplementary-material"}). Alternatively, there may have been cross contamination between the infected and uninfected samples during preparation of the sRNA libraries. However, this seems unlikely considering that the majority of the siRNAs identified in the uninfected control groups contained mismatches (61% and 88% at 7 and 14 days post bloodmeal, respectively) while the viRNAs retrieved from the infected samples were predominantly perfect matches (83% and 84% at 7 and 14 days post infection, respectively) with the WNV genome ([Table 1](#ppat-1000502-t001){ref-type="table"}). Finally, because the genome of *Culex pipiens* has not been completed we cannot rule out the possibility that these sequences were derived from the *Culex* genome. In any case, examination of basic quality metrics on read data from all four sRNA libraries allows us to conclude that the viRNA sequences obtained from infected midguts are of generally high quality and are products of the RNAi pathway. We observed viRNAs targeting both the positive and negative sense WNV genome. *Flaviviruses* have highly structured genomes with many secondary structures and replicate in an asymmetric manner through dsRNA replicative intermediates. While dsRNA clearly activates the RNAi pathway, it has not been clear whether replicative intermediate or stem-loop and other RNA secondary structures are mainly responsible for activation. Studies with (+) ssRNA plant viruses have demonstrated that either may predominate, depending on the virus-host system studied. For example, infection of two different plant species with *Cymbidium ringspot tombusvirus* resulted in the production of viRNAs almost exclusively from the positive-strand. These results suggest that imperfect duplexes in the secondary structure of the (+) ssRNA genome serve as targets for Dicer (DCR) cleavage [@ppat.1000502-Ho1],[@ppat.1000502-Molnar1]. Conversely, when analyzing the viRNA profile of *Brassica juncea* infected with *Turnip mosaic potyvirus*, viRNAs originated from both the positive- and negative-strands and were present in almost equal proportions suggesting that DCR targets dsRNA replicative intermediates in this system [@ppat.1000502-Ho1]. We found that RNAi targeting of the negative sense genome was proportional to the amount of negative sense genome in the infected midguts, which was also observed in the FHV-drosophila system [@ppat.1000502-Aliyari1]. In addition, a highly structured portion of the 3′UTR was not highly targeted in our data. These results differ from those recently reported by Myles and others whom reported a strong positive-strand bias in *Aedes aegypti* inoculated with SINV [@ppat.1000502-Myles1]. The reasons for this are likely to be complex, and possibly related to differences in the virus-vector system, route of infection and/or source of sRNAs (midguts vs. whole body), among others. In any case, our data establish that the RNAi pathway in WNV-infected *Culex* midguts clearly targets dsRNA replicative intermediates and most likely secondary structures within the positive-sense genomic RNA as well. Computing the number of viRNA reads targeting each nucleotide of the WNV genome allowed us to determine whether particular regions are more intensely targeted by the RNAi response than others. Inspection of [Figure 2](#ppat-1000502-g002){ref-type="fig"} clearly shows that this is true. Further, regions targeted at 7 days were also targeted at 14 days: peaks in the figure tend to coincide with one another. Indeed, the per-nucleotide intensity of viRNA coverage along the genome at 7 and 14 days is highly correlated ([Figure 4A](#ppat-1000502-g004){ref-type="fig"}). This suggests that particular regions of the genome are more accessible to the RNAi pathway than others. Supporting this observation, some peaks are extremely high, while others are quite small. Notably, the most intense targeting of WNV by viRNAs at both timepoints is a region in the 5′ aspect of the C coding region ([Figure 2C](#ppat-1000502-g002){ref-type="fig"}). The reasons for this are unclear at present, but may be related to inefficient replication initiation, whereby a stalled replicase complex may leave dsRNA targets available for cleavage. In fact, this has been described in *Drosophila* using FHV [@ppat.1000502-Aliyari1]. Interestingly, the extreme 3′ portion of the 3′UTR, which has well characterized stem-loop structures, is not highly targeted ([Figure 2D](#ppat-1000502-g002){ref-type="fig"}). Upon further analysis of the distribution of viRNA aligning to the WNV genome a small number of sequences obtained from both infected libraries were detected very frequently. The possibility that this observation was due to a sampling artifact was ruled out by permutation analysis. This suggests that detecting any read greater than three times was unlikely to occur by chance ([Figure 3](#ppat-1000502-g003){ref-type="fig"}). Therefore, we considered any viRNA with abundance greater than three to be highly abundant. We also observed that the abundance of read sequences detected at both 7 and 14 days (i.e. "common reads") were highly correlated ([Figure 4B](#ppat-1000502-g004){ref-type="fig"}). Collectively, these data demonstrate that there are 'hotspots' within the WNV genome targeted by the RNAi pathway in the midgut of *Cx. p. quinquefasciatus* and suggest a stereotypical RNAi response to WNV infection that is characterized by the production of a small subset of common viRNA molecules. WNV populations within hosts are highly genetically diverse, with greater levels of population diversity (i.e. more mutations) found in mosquitoes compared to birds [@ppat.1000502-Jerzak1]. The mechanism(s) that give rise to this increase in genetic diversity within mosquitoes, however, remain poorly characterized. We hypothesized that the sequence specificity of the RNAi response in mosquitoes might increase viral genetic diversity. Mutant genomes would be favored because they would be less susceptible to degradation by the RISC loaded with wild-type, un-mutated viRNAs, i.e., they could escape degradation by mutation. We observed two lines of evidence in support of this. First, viRNAs that were imperfect matches to the input WNV genome tended to occur on nucleotides that were significantly more highly targeted than un-mutated nucleotides, excluding uncovered positions ([Table 2](#ppat-1000502-t002){ref-type="table"}). Second, when viral genetic diversity was independently assessed using viral RNA from the midguts analyzed, it was found that after fourteen days extrinsic incubation, regions that were more intensely targeted by viRNAs had greater genetic diversity than regions that were weakly targeted ([Figure 5](#ppat-1000502-g005){ref-type="fig"}). To ascertain whether the observed results were a result of differences in selective constraints across the genome, we analyzed the selective pressures influencing the WNV populations within the mosquito midgut ([Text S1](#ppat.1000502.s004){ref-type="supplementary-material"}, [Figure S1](#ppat.1000502.s001){ref-type="supplementary-material"}, [Table S1](#ppat.1000502.s002){ref-type="supplementary-material"}). These analyses failed to detect significant correlation between intrahost mutational diversity (estimated from sRNA reads mapping to the WNV genome) and dN/dS or interhost genetic diversity. Thus, our data support the hypothesis that intense targeting of WNV by the RNAi pathway in mosquitoes might result in the observed increases in genetic diversity in mosquitoes relative to birds. Indeed, this phenomenon has been observed by other investigators. Multiple studies have demonstrated that artificial induction of the RNAi pathway with viral-specific siRNAs can drive viral evolution [@ppat.1000502-Westerhout1],[@ppat.1000502-Konishi1],[@ppat.1000502-Gitlin1]. For example, exposure of HIV-1 to viral-specific siRNAs resulted in the accumulation of mutations and viral escape mutants. When the viral escape mutants were sequenced it was found that they contained a mutation distribution similar to sequence variants of naturally circulating HIV-1 [@ppat.1000502-terBrake1]. It is notable that even with the power constraints implicit in our strategy for independently assessing genetic diversity in our samples (a relatively small number of nucleotides were sequenced) we observed a significant association between viRNA coverage and viral genetic diversity. High-throughput virus genome sequencing would undoubtedly enhance our ability to characterize the WNV genotypes in our samples. Nonetheless, our results on viral genetic diversity describe for the first time a direct correlation between the RNAi pathway and viral evolution under natural conditions. Further, they suggest a mechanism for the increased fitness observed in highly genetically diverse WNV populations *in vitro* [@ppat.1000502-Ciota1] and may in part explain how WNV can persist *in vivo* in mosquitoes despite the presence of an apparently robust RNAi response. A genetically diverse virus population, whether acquired during bloodfeeding or arising through mutation *in situ*, may present a more complex target for the RNAi response. Materials and Methods {#s4} ===================== Virus {#s4a} ----- WNV used in these experiments was generated from an infectious cDNA clone derived from the NY99 strain as previously described [@ppat.1000502-Shi1]. Virus was produced in baby hamster kidney (BHK) cells and used without subsequent passage. WNV obtained in this manner is highly genetically homogeneous [@ppat.1000502-Jerzak1], and well characterized phenotypically [@ppat.1000502-Jerzak2]. Mosquitoes {#s4b} ---------- The mosquitoes used for these experiments were obtained from our *Culex pipiens quinquefasciatus* colony. Mosquitoes were housed in environmental chamber at constant temperature of 27°C with a 16∶8 light∶dark photoperiod for the duration of these experiments. Artificial bloodmeals containing defibrinated goose blood alone (controls) or with 2×10^8^ plaque-forming units (pfu)/ml of WNV were offered to adult female *Cx. quinquefasciatus* five to seven days post-eclosion using a Hemotek (Accrington, UK) membrane feeding apparatus. Mosquitoes were then cold anesthetized and engorged individuals reserved and held for either a 7 or 14 day EIP under the above standard conditions. After the 7 and 14 day EIPs, mosquitoes were cold-anesthetized and fifteen mosquitoes from each group were dissected. Midguts were isolated, washed four times in phosphate buffered saline (PBS) and placed in 300 µl lysis buffer (mirVana, Ambion, Austin, TX). Forceps were flame sterilized between each dissection. RNA extractions and one-step reverse transcriptase polymerase chain reaction (RT-PCR) {#s4c} ------------------------------------------------------------------------------------- RNA from individual midguts was extracted using the mirVana miRNA Isolation Kit according to the manufacturer\'s instructions. The midgut RNA was subsequently analyzed by one-step RT-PCR for the presence of WNV RNA using the SuperScript® One-Step RT-PCR with Platinum® Taq (Invitrogen, Carlsbad, CA) according to standard methods. WNV-positive midgut RNAs, were pooled. An equivalent number of uninfected midgut RNAs were pooled in our control groups. RNA was then precipitated with ethanol, resuspended in 100 ul H~2~O, and quantity and integrity determined on an Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). Preparation of small RNA populations and high-throughput sequencing {#s4d} ------------------------------------------------------------------- Approximately 6--7 µg of total RNA from pooled WNV-positive and negative specimens were size fractionated on a 15% TBE/ urea polyacrylamide gel and small RNA (sRNA) populations (18--30 nt) recovered. 5′ and 3′ sequencing adapters were then ligated to sRNAs and reverse transcription and PCR amplification performed according the manufacturer\'s instructions (Illumina, San Diego, CA). The resulting libraries were sequenced at the National Center for Genomic Resources (Santa Fe, NM) using an Illumina Cluster Station and Genome Analyzer. Assembly and analysis of sRNA libraries {#s4e} --------------------------------------- Reads from small RNA libraries were trimmed of adapter sequences and aligned to the WNV infectious clone reference genome using the Short Oligonucleotide Alignment Package v.1 (SOAP) (<http://soap.genomics.org.cn/>) with seed size of eight, and a maximum of two mismatches allowed. Further trimming and gaps were not permitted. Additional analyses were performed using R, Microsoft Excel and GraphPad. Determination of WNV genetic diversity {#s4f} -------------------------------------- Intra-population genetic diversity was determined according to methods described elsewhere [@ppat.1000502-Jerzak1]. For these studies, 160 nt (10776--10936) and 454 nt (7800--8254) of the 3′-UTR and NS5 of the WNV genome, respectively, were chosen based on inspection of frequency distributions of the number of viRNA "hits" per nucleotide. Regions were chosen that had highly variable viRNA coverage. Viral RNAs extracted from WNV-infected pools were reverse transcribed using the AccuScript High Fidelity 1^st^ Strand cDNA Synthesis Kit (Stratagene, LaJolla, CA). The fragments were PCR amplified using high fidelity *Pfu*Ultra polymerase (Stratagene) using the following parameters: 94°C for 30 s, 60°C for 30 s and 72°C for 45 s repeated an additional 39 cycles followed by a 72°C final extension for 6 min. The primers used for both the RT and PCR reactions were WNV 10,524 F 5′- CGCCACCGGAAGTTGAGTAGAC -3′ and WNV 3′ End 5′- AGATCCTGTGTTCTCGCACCACCA - 3′ or WNV 7670 F 5′-GACTAAAAAGAGGTGGGGCAAAAG -3′ and WNV 8335 R 5′-GAAGCTCGACTCACCCAATACAT -3′. Amplicons were gel purified and cloned into pCR Script Amp^+^ vector (Stratagene). Clones were sequenced using the M13 Rev primer at the University of New Mexico\'s School of Medicine DNA Services laboratory. Sequences were aligned and analyzed for genetic diversity using DNAStar\'s SeqMan program. Quantitative reverse transcriptase polymerase chain reaction (Q-RT-PCR) {#s4g} ----------------------------------------------------------------------- To determine the amount of positive and negative strand WNV genome in mosquito midguts at each sampling point, equal concentrations of RNA from WNV-infected midguts was reverse transcribed with strand specific primers using the AccuScript High Fidelity 1^st^ Strand cDNA Synthesis Kit (Stratagene). WNV RNA of each polarity was then quantified using real-time quantitative (TaqMan) RT-PCR as described elsewhere [@ppat.1000502-Lanciotti1]. Supporting Information {#s5} ====================== ###### Intrahost mutational diversity and dN/dS, Pi and viRNA coverage. (1.73 MB TIF) ###### Click here for additional data file. ###### Genetic diversity and selection in cloned WNV. \* An alignment of 38 WNV genomes obtained from the North American epidemic was used to generate interhost diversity values. \# Undefined. Only nonsynonymous mutations were present in the alignment, yielding a denominator of zero for dN/dS calculation. (0.01 MB PDF) ###### Click here for additional data file. ###### Alignment of sRNA reads against viruses not used in these studies. \* For all four sRNA libraries generated in these studies (WNV-infected and uninfected). (0.01 MB PDF) ###### Click here for additional data file. ###### Supplemental text. (0.04 MB PDF) ###### Click here for additional data file. The authors thank Ivy Brown, Kelly Fitzpatrick and Robert Nofchissey for technical assistance. A portion of this research was conducted using the Boston University Linux Cluster for Genetic Analysis (LinGA). Sanger sequencing was performed at the UNM DNA research facility. The authors have declared that no competing interests exist. This work was supported in part by funds from the National Institute of Allergy and Infectious Diseases, National Institutes of Health, under grant AI067380, the Ruth L. Kirschstein National Research Service Award under grant T32 AI007538-11, and by the University of New Mexico School of Medicine, Department of Pathology. LinGA (Boston University) is funded by the NIH NCRR (National Center for Research Resources) Shared Instrumentation grant 1S10RR163736-01A1). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Conceived and designed the experiments: DEB GDE. Performed the experiments: DEB GDE. Analyzed the data: DEB JEB GDE. Contributed reagents/materials/analysis tools: JEB GDE. Wrote the paper: DEB GDE.

###### Strengths and limitations of this study - This study will provide insight into the substance-use screener(s) that works best to identify illicit drug use and prescription-drug misuse during pregnancy, using hair and urinalysis for biochemical verification of long-term and short-term substance use in a convenience sample of 500 pregnant women. - The study will provide evidence of screener usefulness and acceptability in prenatal clinic settings that could inform US Preventive Services Task Force recommendations for substance-use screening during pregnancy. - The study uses electronic medical records to capture prescribed drugs and birth-outcome data of enrolled participants to assess for associations between substance use in pregnancy and adverse birth outcomes. - A limitation of this study is the reliance on a convenience sample from two urban clinics rather than a national sample. - Findings from this study will not be generalisable to pregnant adolescents who were not included in our study sample. Introduction {#s1} ============ Abuse of prescription and illicit drugs in pregnancy is a growing cause of maternal and neonatal morbidity and mortality in the USA. According to data from the 2012 and 2013 US National Survey on Drug Use and Health, the rate of current illicit drug use (including non-medical use of prescription drugs) in pregnant adolescents and women was 14.6% among adolescents (aged 15--17 years), 8.6% among young adults (aged 18--25 years) and 3.2% among adults (aged 26--44 years).[@R1] The consequences of this problem include spontaneous abortions, stillbirths, low birth weight, prematurity, neonatal abstinence syndrome and congenital malformations.[@R2] Given the relatively high frequency of provider--patient contact during the prenatal period, obstetrical care providers have the unique opportunity to identify substance abuse in pregnancy. Furthermore, for pregnant women from socioeconomically disadvantaged groups, obstetricians often serve as primary care physicians and typically are the only contact these women have with the healthcare system.[@R3] Prenatal screening for drug use is an important way to identify drug abuse in pregnancy, as strongly recommended by the American Congress of Obstetricians and Gynecologists (ACOG).[@R4] But, while validated alcohol and tobacco screeners have been recommended by the US Preventive Services Task Force, there is currently no universally recommended validated screening tool for identifying illicit drug use in pregnancy. Currently, three separate validated tools exist that screen for use of more than one substance among pregnant women: The Alcohol, Smoking, and Substance Involvement Screening Test (ASSIST); the 4 P's Plus; and the Substance Use Risk Profile--Pregnancy (SURP-P).[@R5] ASSIST has been validated across several populations, but it has not yet been formally validated with pregnant women.[@R5] A modified ASSIST, with items on tobacco and alcohol use removed, was incorporated by National Institute on Drug Abuse (NIDA) to their Quick Screen as a follow-up to the four-question prescreener; this is referred to as the NIDA Quick Screen/ASSIST. The 4 P's Plus was designed to identify drug use in pregnancy and has been validated with pregnant women.[@R7] The 4 P's Plus is brief but is associated with a licensing fee which may be a hindrance to widespread use. The SURP-P is a validated scale composed of three questions that can differentiate between populations of pregnant women at low risk or high risk for substance use.[@R8] The SURP-P is a simple and flexible tool for identifying possible substance use in pregnancy; however, a further screen is required for identifying those who would require treatment. To bridge this gap and identify the most universally valid and reliable screening tool for drug abuse in pregnancy, this study aims to compare and validate three existing substance-use screeners---4 P's Plus, NIDA Quick Screen/ASSIST and the SURP-P scale---among a cross-section of 500 pregnant women presenting to two obstetrics clinics in Baltimore, Maryland, USA. The overarching goal of this effort is to determine which screening tool is most effective in identifying prescription-drug abuse and illicit drug use among pregnant women and acceptable among patients and clinicians so that evidence-based guidance may be offered. Methods/design {#s2} ============== Specific aims {#s2a} ------------- Specific aims of this study are to: (a) conduct validity analyses to determine sensitivity, specificity, usability (test--retest reliability) and how each scale compares with the others and to the gold standard of urine and hair drug testing in identifying prescription and illicit drug use; (b) determine the impact of clinic population variables (age, race, trimester of pregnancy) on validity of the three substance-use screeners; and (c) assess birth outcomes (birth weight, gestational age, head circumference and neonatal intensive care unit (NICU) admissions) associated with the most widely used prescription drug and multidrug exposure. Study design {#s2b} ------------ This study is a cross-sectional study that evaluates the sensitivity, specificity and usability of existing substance-use screeners. We chose this study design following an extensive search of the literature, an overall assessment of feasibility and consultation with stakeholders (eg, clinicians, pregnant women and substance-use researchers). We believe that a cross-sectional study such as ours is appropriate for the evaluation of the accuracy and reliability of these screeners. We were also aided by the knowledge that the prevalence of substance use in pregnancy is high.[@R1] This implies that we are likely to obtain good sensitivity and specificity estimates, with narrow CIs, in a cross-sectional design which is favourable in terms of cost and feasibility. Setting {#s2c} ------- The study is being implemented at two urban obstetrics clinics which serve diverse populations of pregnant women. The study plans to recruit 500 participants to complete a demographic questionnaire, followed by a randomised order of the NIDA Quick Screen/NIDA-modified ASSIST, 4 P's Plus and SURP-P. Participants are recruited during their regularly scheduled prenatal appointment, then contacted again 1 week later by telephone to re-administer the screeners. Participants consent to multidrug urine testing, hair drug testing and access to prescription-drug and birth-outcome data from electronic medical records (EMR). ### Recruitment sites {#s2c1} We are recruiting participants from two obstetrics outpatient clinics from January 2017 to January 2018. Currently, all obstetrical patients are screened for use of drugs, alcohol and tobacco at their first prenatal visit by medical staff. Additionally, all new obstetrical patients receive an in-depth evaluation by a social worker which includes a more detailed assessment of both substance use and mental health disorder history. In the first clinic, which is the larger of the two clinics, most patients (97%) are publicly insured with medical assistance and are over the age of 20 (80%). This clinic's population is primarily African-American and low-income, all of whom undergo urine toxicological screening for substance-use identification. Based on preliminary data obtained from the clinic, about 950 individual obstetrical patients are cared for at this clinic annually. In the second (smaller) clinic, approximately 500 pregnant women are cared for annually. Most patients (87%) have commercial insurance, and 13% have either medical assistance or Medicare. Most are over the age of 20 years (90%). Due to varying insurance coverage for urine toxicology screens, patients in this office do not universally undergo urine toxicology screening but all are screened for drug use using various interview techniques by their obstetrical care providers at their first prenatal visit. Based on historical data, we expect about 500 individual obstetrical patients to be cared for in this clinic across all trimesters of pregnancy in the 1 year of study recruitment. Across both study sites, our source population covers a diverse set of participants and captures pregnant women across all socioeconomic categories, insurance types, ethnicities and drug-use patterns. This ensures that our study results are generalisable to most populations of pregnant women. Study population {#s2d} ---------------- In the first clinic, of the estimated 950 individual obstetrical patients cared for at this clinic annually, we anticipated approaching 403 (50%), and expected 322 (80%) or more to agree to participate in this study. In the second clinic, of the approximately 500 pregnant women cared for annually, we expect at least 450 (90%) to meet eligibility criteria. We anticipate approaching 225 pregnant women (50%) and expect 180 (80%) or more to agree to participate in this study. Expected participation percentages are based on a similar grant-funded study that recruited pregnant smokers from the same population and required consent for urine testing (cotinine) and birth-data abstraction from EMR. Participant eligibility criteria include the following: (a) currently pregnant (predetermined by clinic staff), (b) age 18 or older, (c) able to speak and understand English sufficiently to provide informed consent and (d) natural hair length at least 3 cm to allow for substance-use testing. If eligibility criteria are met, research staff then obtain informed consent and medical releases for urine collection, hair drug testing and prescription-drug and birth-outcome data abstraction from the EMR. Ethics and dissemination {#s2e} ------------------------ All participants are required to give their informed consent prior to any study procedure. All research staff complete ethics training via the Collaborative Institutional Training Initiative annually. Study procedures {#s2f} ---------------- ### Approach {#s2f1} All patients entering the clinics for prenatal appointments are approached by research staff at check-in and asked to read a brief description of the study to determine their interest in participating (excluding those previously approached). Research staff keep track of which patients have been approached already to avoid repetitive recruitment efforts. The study description includes a section requesting basic demographic information (if they would allow its use for anonymous, grouped analysis) and at the bottom asks potential participants to note their interest and return to clinic staff. There are check boxes for 'not interested' (with additional space beneath for noting reasons for lack of interest) and 'interested in learning more'. Patients who are not interested in the study are not to be contacted further; however, the basic demographic information provided is used for comparative analyses with study participants to assess for selection bias. If a patient expresses interest, the research staff approaches her as she waits for her prenatal appointment either on the same day or at a future prenatal appointment. ### Recruitment {#s2f2} At the enrolment visit, the staff escorts potential participants from the waiting area to a private room, further describes the study and determines whether potential participants meet all eligibility criteria. If eligibility criteria are met, the staff obtains informed consent and Health Insurance Portability and Accountability Act (HIPAA) authorisation (for urine collection, hair drug testing and prescription-drug and birth-outcome data abstraction from the EMR). Women who refuse to participate are thanked for their time, and no further contact is made. The research visit takes 20--30 min. Enrolled participants are compensated for their time using a reloadable gift card for their time. The typical patient wait time to see medical staff at each clinic is from 30 minutes to 1 hour, so data collection does not typically interfere with medical visits. See [figure 1](#F1){ref-type="fig"} for study procedures. {#F1} ### Self-report measures {#s2f3} Participants complete a demographic questionnaire. Afterwards, the NIDA Quick Screen/NIDA-modified ASSIST, 4 P's Plus and SURP-P surveys are administered on a Wi-Fi enabled iPad Pro through SurveyMonkey (ie, online survey software). See [table 1](#T1){ref-type="table"} for description of surveys and the timing of administration during the study. These surveys are assigned to participants in a random sequence; this randomisation service is provided by SurveyMonkey. The questions are read aloud by the interviewer and entered directly into SurveyMonkey so that electronic submission is instantaneous, and data can be obtained by the research team at any time. ###### Study Instruments Instrument Description/construct Use in study --------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ----------------------------- Demographic questionnaire 20-Item questionnaire that collects demographic and general information such as age, marital status, education, employment status, ethnicity and reproductive history. Enrolment NIDA Quick Screen/ASSIST 9-Item combined NIDA Quick Screen and modified-ASSIST to screen for tobacco, alcohol and illicit drugs. Enrolment, 1-week follow-up 4 P's Plus 4-Item screener for alcohol and general substance use. Enrolment, 1-week follow-up SURP-P 3-Item screener for alcohol and substances Enrolment, 1-week follow-up ASSIST, Alcohol, Smoking, and Substance Involvement Screening Test; NIDA, National Institute on Drug Abuse; SURP-P, Substance Use Risk Profile-Pregnancy. ### Biochemical measures {#s2f4} Participants are asked to consent that urine collected for their prenatal appointment that day is also tested for various drugs by research staff ([table 2](#T2){ref-type="table"}). If sufficient urine is unavailable for testing, participants are given bottled water and asked to provide another sample prior to leaving the clinic. Participants must also consent to hair testing which involves the cutting of approximately 100 strands of hair from the crown of the head (or other body hair if head hair is unavailable). Samples are then shipped to an external laboratory on the same day for drug testing using mass spectrometry. ###### Drug detection windows and cut-offs for urine and hair testing Drug class Detection window Confirmation cut-off ---------------------------------------- ----------------- ------------------ ---------------------- URINE Cocaine (COC) 2--4 Days 300 ng/mL Marijuana (THC) 15--30 Days 50 ng/mL Opiates (OPI) 2--4 Days 2000 ng/mL Amphetamines (AMP) 2--4 Days 1000 ng/mL Methamphetamines (mAMP) 3--5 Days 1000 ng/mL Phencyclidine (PCP) 7--14 Days 25 ng/mL Benzodiazepines (BZO) 3--7 Days 300 ng/mL Barbiturates (BAR) 4--7 Days 300 ng/mL Methadone (MTD) 3--5 Days 300 ng/mL Tricyclic antidepressants (TCA) 1000 ng/mL Oxycodone 2--4 Days 100 ng/mL Propoxyphene 1--2 Days 300 ng/mL Buprenorphine (BUP; Suboxone, Subutex) 2--3 Days 10 ng/mL HAIR Marijuana (THC) Up to 90 days Amphetamines (AMP) Up to 90 days Cocaine (COC) Up to 90 days Opiates (OPI) Up to 90 days Phencyclidine (PCP) Up to 90 days All women who screen positive on either biological multidrug test or any one of the screeners are contacted immediately (for urine and screener results) or within 72 hours (for hair results) to detail the results of her test, encourage the participant to talk with her physician about her substance use and offer her referrals to community resources for treatment that mirror what is currently given to patients by medical staff in each clinic. They are encouraged to speak with the on-site clinic social worker who can provide further support. ### Birth outcome measures {#s2f5} Birth outcome data, including miscarriage, stillbirth, birth weight, gestational age, head circumference and NICU admissions, as well as a list of drugs prescribed during pregnancy and their dosage are collected by research staff via the EMR and entered into SurveyMonkey. ### Participant follow-up {#s2f6} After completion of this research visit, participants are contacted once more by telephone 1 week after completing the surveys to complete the three screeners again to assess test--retest reliability. The average time commitment for the call is about 10--15 minutes, and on completion, \$25 is loaded onto the reloadable gift card provided the week prior. ### Pilot study {#s2f7} To examine the recruitment process and determine acceptability from the target population of substance-using pregnant women prior to the start of the study, we conducted a 1-month pilot study. Each step of the recruitment process was reviewed to determine where improvements could be made. We recruited 21 participants from each site for a total of 42 participants ([table 3](#T3){ref-type="table"}). Mean age (SD) of participants was 30.1 years (5.64). By race/ethnicity, 11 participants (26.2%) were white, 25 (59.5%) were black/African-American, 4 (9.5%) were Asian, 1 (2.4%) was Hispanic and 1 (2.4%) was of other race. About 24.4% tested positive for illicit drugs on urine testing, 22% tested positive on hair sample testing. Seven (7) participants (16.7%) were lost to follow-up. ###### Pilot study participant characteristics Characteristics, n=42 Clinical site -------------------------------------- --------------- -------------- -------------- Number of participants 21 21 42 Participant age in years, mean (SD) 27.10 (5.09) 33.05 (4.54) 30.07 (5.64) Ethnicity, n (%)  African-American/black 18 (85.7) 7 (33.3) 25 (59.5)  Asian 0 (0.0) 4 (19.0) 4 (9.5)  Caucasian/white 2 (9.5) 9 (42.9) 11 (26.2)  Hispanic, Latino or Chicano 0 (0.0) 1 (4.8) 1 (2.4)  Some other group 1 (4.8) 0 (0.0) 1 (2.4) Trimester, n (%)  First 2 (9.5) 2 (9.5) 4 (9.5)  Second 6 (28.6) 6 (28.6) 12 (28.6)  Third 13 (61.9) 13 (61.9) 26 (61.9) Urine results, n (%)  Negative for all substances 15 (71.4) 16 (80.0) 31 (75.6)  Positive for at least one substance 6 (28.6) 4 (20.0) 10 (24.4) Hair results, n (%)  Negative for all substances 12 (60.0) 18 (85.7) 30 (73.2)  Positive for at least one substance 7 (35.0) 2 (9.5) 9 (22.0)  Invalid 1 (5.0) 1 (4.8) 2 (4.9) Study disposition, n (%)  Study completes 19 (90.48) 16 (76.2) 35 (83.3)  Lost to follow-up 2 (9.52) 5 (23.8) 7 (16.7) Results from the pilot study confirmed the feasibility of this study. Eligibility criteria did not appear too restrictive, given the eligibility rate of 78% (although slightly lower than anticipated; [figure 2](#F2){ref-type="fig"}). Overall, there was good comprehension of surveys, a low refusal rate for hair sampling (1 refusal/95 approached, 1.1%) and high study enrolment ([figure 2](#F2){ref-type="fig"}). The recruitment process, on an average, took 40 minutes. {#F2} ### Power and sample size {#s2f8} The sample size of 500 participants was chosen based on power analyses for the primary study questions. Based on a one-sample binomial approach, with a sample size of 500 participants, we can be 95% confident that the false-negative rate in the population is under 10% (assuming no more than 35 individuals test positive in the biologic drug tests without a positive survey screener result). Similarly, we can be 95% confident that the false-negative rate in the population is under 5% (assuming no more than 15 individuals test positive in the urine drug test without a positive survey screen result in the study). According to McNemar's test, if at least 15% of the study participants have disagreement between any pair of survey results, 500 is a sufficient sample size to determine significant disagreement. After a preliminary sample size of 500 was chosen, a power analysis was conducted to determine the detectable differences in age, race and trimester with a sample size of 500. The power of the test of proportions was calculated based on the difference in the proportion of false negatives in each age group, race and trimester of pregnancy. Assuming recruitment of an equal number of women aged 18--25 years and women 26 and older, and that their respective positive screener results are 20% and 10%, then the power to detect that difference is 0.88. If the respective screener results are 15% and 20%, then the power is much lower (0.31). If we further assume that recruitment of 23% white women and 77% non-white women and that white women have a false-negative rate of 5% and non-white women have a false-negative rate of 15%, then the power is high (0.87). Similarly, if we assume recruitment of an equal number of women in each of the three trimesters of pregnancy and that women in one trimester have a false-negative rate of 20% while women in another trimester have a false-negative rate of 35%, then the power is high (0.87). Analysis {#s2g} -------- For each screener, reliability and validity (convergent/discriminant validity) will be assessed, including calculating correlation coefficients between each pair of screeners and between each screener and the appropriate biologic drug tests. Test--retest reliability analysis will be conducted by examining the results of repeated screener administrations within 1 week of original screener administrations for consistency via correlation analysis. The sensitivity and specificity of each instrument will be calculated, presented and interpreted. Each survey instrument will be compared with the gold standard (hair and urine sample drug testing) by comparing the false-negative rates to a predetermined limit of acceptability. If the upper one-sided 95% binomial CI around the false-negative rate in the sample is less than that limit, then the survey instrument is considered acceptable. The 4 P's Plus and SURP-P survey screeners will be compared with both urine and hair testing in the assessment of their sensitivity and specificity in relation to short-term and long-term drug use, respectively. The NIDA Quick Screen/NIDA-modified ASSIST screener will be compared with urine testing, while particular questions from the screener regarding long-term drug use will be compared with hair testing. Furthermore, we will assess if there are differences in the validity of each screener by age, race and trimester. The false-negative rate for each screener will be presented by age, race and trimester. A two-sided test of proportions will be conducted to test for significant differences in false-negative rates between age, race and trimester for each screener. Χ^2^ tests (or Fisher's exact tests if subgroup sizes are small) may be conducted to determine whether the distribution of responses on each survey instrument is similar for age, race and trimester. To examine differences in screener validity by age, race and trimester, logistic regression models will be fitted to the data. To separately analyse differences in probability of false-positive results and false-negative results on each survey, data will be stratified by screener and screener result (positive or negative) for a total of six models. In each model, the dependent variable will be coded 1 for invalid screener result (false negative or false positive) and 0 for valid screener result (true negative or true positive). Independent variables for age, race and trimester will be added to the models to test whether they have a significant effect on the probability of an invalid screener result. Two-way interaction terms will be included in the model if they are found to have significant effects. In order to stratify results by trimester, if trimester or any two-way interaction term including trimester is a significant effect in the models for any of the screeners, probabilities of false-positive/false-negative result will be presented separately by each trimester. Finally, the prevalence of prescription and illicit drug use will be calculated based on hair test results and self-report. Prevalence of multidrug exposure will also be calculated. An analysis of variance model will be fitted to the data with a fixed effect for drug use (negative, positive, positive for multidrug exposure) to test for significant differences in birth weight, gestational age and head circumference based on participant hair drug test results. Significant differences will be noted and discussed. The relative risk of NICU admission, stillbirth and miscarriage will be examined. A risk ratio will be calculated and will quantify the percentage difference in these three variables between those with positive hair drug tests versus negative drug test. The risk ratio takes on values between zero and infinity. A risk ratio of one means that there is no difference in NICU admissions, stillbirth or miscarriage between the participants' biologic drug test results. A risk ratio very small (close to zero) or very large means a large difference between NICU admissions, stillbirth or miscarriage based on the hair drug test results. Approximate 95% CIs for the relative risk will be calculated. The same relative risk ratios and 95% CIs will be calculated for a positive biologic drug test for multidrug exposure versus positive for a single-drug exposure. Further, relative risk ratios will be computed with 95% CIs stratified by trimester. Discussion {#s3} ========== Our ongoing research has five aspects of significance. First, the importance of screening pregnant women and the public-health impact of the current research is tied directly to the negative health consequences associated with illicit-drug and prescription- drug use during pregnancy. Second, it uses both urine and hair testing to enable us to examine past 90-day substance-use history with precision. Hair analysis provides nearly twice the number of positives due to its longer detection window, but often cannot capture very recent use. Urinalysis supplements hair analysis to allow for the most comprehensive validation of screeners possible. Third, the study compares three screeners acknowledged by the WHO to screen for multiple substances to each other and to the biological screeners (gold standard). This is the first study to conduct a direct, head-to-head comparison of multiple screening tools for prescription-drug and illicit-drug use among pregnant women, while also using biologic measures as a gold standard against which to compare. Fourth, the study uses EMR to capture prescribed drugs and birth outcome data of enrolled participants. The ability to access a participant's prescription-drug orders enables better tracking and distinction between prescription-drug use and abuse, while birth outcome data allows for determination of associations between specific drug use and birth outcomes. Fifth, the study has the potential to shift clinical practice towards universal standardised substance-use screening. Despite the significant contributions of this work, it is not without limitations. Though the study will enrol a large sample of pregnant women, it is a convenience sample from two prenatal clinics in an urban area. We have attempted to increase generalisability by enrolling women from two clinics with different population characteristics: one clinic serves low-income, Medicaid-eligible, primarily African-American women and the other serves privately insured, primarily white women. Second, there is a possibility of selection bias. Incentive may be more appealing to those who have lower socioeconomic status, individuals with more time may be those willing to take the study and pregnant women who use substances may not want to participate. For the latter point, we have obtained a certificate of confidentiality and ensured participants that their data will not be shared with anyone including clinic staff. Finally, our study is limited to adults. Though our initial protocol included adolescents, the institutional review board did not allow for 'no-benefit' studies enrolling pregnant adolescents. This is an important area for further exploration, given that pregnant adolescents report higher substance-use rates than pregnant adults in national surveys. The primary innovation of this project is that it may provide a final evidence-based recommendation for the tool(s) best suited to screen for illicit and prescription drugs among a diverse sample of pregnant women. The provision of this evidence-based guidance to clinicians is a concrete application of findings that is rare in public health research. We recognise that screening is a first step; also important is the need for a public health focus on treatment of substance use during pregnancy to enhance the odds of a successful pregnancy outcome. Barriers to treatment that are imperative to address are the potential legal repercussions of identifying substance use during pregnancy that exist in some states[@R9] and unintentional breach of confidentiality.[@R10] There is a strong need for a re-examination of state policies so that women are not punished for having a treatment need. Substance use during pregnancy, and specifically prescription-drug and illicit-drug use, are high-priority topics for the Centers for Disease Control and Prevention, WHO, ACOG, Substance Abuse and Mental Health Services Administration, NIDA and the National Institutes of Health. Universal screening has the potential to greatly enhance maternal and infant health outcomes and reduce healthcare costs. Specifically, the current research supports the following Healthy People 2020 public health goals and objectives which include reducing maternal illness and complications due to pregnancy; increasing the proportion of pregnant women who receive adequate prenatal care' increasing abstinence from alcohol, cigarettes and illicit drugs among pregnant women; and increasing the proportion of women delivering a live birth who received preconception care services and practised key recommended preconception health behaviours. This research addresses an important problem by identifying a valid substance-use screening instrument for illicit and prescription drugs among pregnant women that is accurate, brief and acceptable to both patients and healthcare providers in a primary care setting. Identifying and validating one instrument that functions the closest to the 'gold standard' of biologic testing (ie, urine and hair) and disseminating this information widely will increase the likelihood that primary care clinics nationwide may adopt a quick and easy screener universally. We may find that one instrument does not stand out but that each has its distinct advantages and disadvantages; in this case, the performance of each measure will be detailed with recommendations for which screener may work the best with a given population. Supplementary Material ====================== ###### Reviewer comments ###### Author\'s manuscript **Contributors:** VC-C conceived the study. VC-C, EAO, ENP, KT, BK and KM participated in the drafting of the manuscript and each approved the final draft. **Funding:** Research reported in this publication was supported by the National Institute On Drug Abuse of the National Institutes of Health under Award Number R01DA041328. **Disclaimer:** The content is solely the responsibility of the authors and does not represent the official views of the National Institutes of Health. **Competing interests:** None declared. **Patient consent:** Obtained. **Ethics approval:** This study is approved by the institutional review boards of the University of Maryland (HP-00072042), Baltimore; and Battelle Memorial Institute (0619-100106433). All participants are required to give their informed consent prior to any study procedure. **Provenance and peer review:** Not commissioned; externally peer reviewed.

HCM is an inherited cardiovascular disease primarily caused by mutations in genes encoding proteins in the sarcomere, the contractile apparatus of cardiac myocytes. HCM is characterized by increased heart mass and abnormal cardiac function with susceptibility to arrhythmias and sudden cardiac death. Histological manifestations of the disease include cardiac myocyte hypertrophy, myocardial fibrosis, extracellular matrix disorganization, and myocyte disarray. While many affected individuals are asymptomatic and remain undiagnosed, HCM is the most frequent cause of sudden death in young athletes (Seidman & Seidman, [@b15]; Maron & Maron, [@b12]). To date, 13 genes containing over 900 distinct mutations have been identified as genetic causes of HCM. Most of these genes encode for proteins found within the thick and thin filaments of sarcomeres, such as β-myosin heavy chain (*MYH7*) and troponin T (*TTNT2*). Mutations in *MYH7* increase force generation and actin-myosin sliding velocity within sarcomeres. These findings indicate that genetic mutations in HCM patients are the primary cause of cardiac hypertrophy (Wang *et al*, [@b16]). Numerous animal models have been generated to investigate HCM (Maass & Leinwand, [@b10]), and much focus has been given to an R403Q mutation in *MYH7*, which causes an especially severe clinical phenotype (Seidman & Seidman, [@b15]). While the various animal models of R403Q highlight different aspects of HCM, they share common traits of HCM including cardiac hypertrophy and fibrosis (Maass & Leinwand, [@b10]). An interesting, and poorly understood characteristic of hypertrophic cardiomyopathy, as opposed to other types of cardiomyopathies, is that systemic metabolic alterations occur secondary to the cardiomyopathy (Maron & Maron, [@b12]). This is recapitulated in the R403Q model used in the study published by Magida and Leinwand ([@b11]). Clinical studies have revealed that HCM patients harboring mutations in sarcomeric genes present with deficient cardiac energetics (Crilley *et al*, [@b6]). In the present study, the authors demonstrate that the R403Q HCM mouse model has diminished cardiac ATP levels and impaired lipid utilization in the heart, assessed by decreased cardiac triglycerides and fatty acid content, and decreased expression of fatty acid translocase (CD36), lipoprotein lipase (LPL), and very low density lipoprotein receptor (VLDLR). Notably, they observed an approximate two-fold reduction in active CD36 protein at the plasma membrane, coupled with a similarly decreased level of nonesterified fatty acid (NEFA) released from VLDL by the left ventricle. The authors suggest that this decreased lipid uptake in the heart leads to the observed lipid elevation in the plasma, ultimately resulting in hepatic lipid accumulation and pathologically enhanced gluconeogenesis. The authors propose that this elevation in hepatic glucose production creates a vicious cycle between the heart and the liver in which the spillover of VLDL triglyceride and oleic acid from the heart insults the liver via elevated protein kinase C signaling. The liver responds by increasing blood glucose levels leading to exacerbation of the primary cardiac disease (summarized in Fig [1](#fig01){ref-type="fig"}). Importantly, features of the diseased state can be rescued either by restoring the energetic deficit at the level of the cardiomyocyte via AMPK agonism, or by blocking the deleterious elevation in hepatic glucose output using the phosphoenolpyruvate carboxykinase (PEPCK) inhibitor 3-MPA (Magida & Leinwand, [@b11]). > These findings raise the interesting concept that the lack of use of a specific metabolic substrate by one tissue directly affects another {#fig01} These findings raise the interesting concept that the lack of use of a specific metabolic substrate by one tissue directly affects another, perhaps revealing an inter-tissue homeostatic feedback mechanism. Namely, that the heart signals to the liver to elevate glucose production by selectively excluding uptake and use of oleic acid and triglyceride in VLDL particles. Indeed, an emerging theme in homeostatic feedback is the recognition of metabolites as signaling effectors between tissues as means of physiologic integration within the body \[see (Blad *et al*, [@b2]; Liu *et al*, [@b9]; Roberts *et al*, [@b14]) for examples\]. However, in the setting of an HCM genotype, the current work suggests this relationship is injurious. Many metabolic diseases, such as diabetes and obesity, are ultimately detrimental to cardiac function, but the reverse has yet to be investigated. There is a clear relationship between cardiac metabolism and cardiac function, but diminished cardiac function, *per se*, has thus far not been reported to negatively influence systemic metabolism. There is a clear link between liver dysfunction, specifically non-alcoholic fatty liver disease, and cardiac dysfunction (Bhatia *et al*, [@b1]), but new evidence reported in this issue of *EMBO Molecular Medicine* suggests the reverse is also true. While the link between cardiac dysfunction, specifically the alteration of cardiac metabolism, and deregulated hepatic lipid metabolism is interesting, the mechanisms regulating this crosstalk are not resolved by the work of Magida and Leinwand ([@b11]). Further studies are required to clarify whether HCM-induced metabolic abnormalities are the primary cause of liver dysfunction. It remains unclear whether hepatic lipid accumulation in this mouse model results from decreased fatty acid uptake in the heart alone. Certainly, the relationship between the heart and liver is not monogamous, and other tissues such as skeletal muscle, pancreas, and adipose are likely to be directly affected by elevated circulating oleic acid and VLDL triglyceride. Indeed it is likely that lipid uptake, utilization, or storage in each of these tissues contributes to the metabolic phenotype described by Magida and Leinwand ([@b11]) and would be influenced by systemic agonism of AMPK. Further, PEPCK inhibition not only affects glucose production by the liver, kidney, and intestine, but also glyceroneogenesis in adipocytes. Additionally, it would be interesting to know if other sarcomeric mutations also decrease liver function in end-stage disease, and if so, if a similar mechanism is involved. Other aspects of HCM can also be explored in the R403Q HCM mouse model within the framework of metabolic abnormalities. For example, what role does calcium homeostasis play in the development of cardiac and metabolic dysfunction? Calcium is an important regulator of energy metabolism and calcium levels and homeostasis are altered in human HCM patients (Wang *et al*, [@b16]). Perhaps restoring calcium homeostasis in the heart could restore metabolism in this mouse as well? Moreover, what is the basis for the phenotypic gender differences in HCM? Is there likely a protective role for estrogen at the level of cardiac energetics as well as liver metabolism in the HCM patient? Estrogen certainly has a role both as it relates to AMPK and hepatic lipid metabolism (D\'Eon *et al*, [@b7]; Bryzgalova *et al*, [@b5]), properties which could be therapeutically exploited. > Certainly, the relationship between the heart and liver is not monogamous The studies of Magida and Leinwand ([@b11]) add to a growing number of examples in which the heart modulates energy homeostasis and metabolism in non-cardiac tissues. In this regard, the cardiac natriuretic peptides, ANP and BNP, have been shown to improve metabolic parameters by inducing the "browning" of white adipocytes (Bordicchia *et al*, [@b4]). While the thermogenic action by ANP is restricted to human, but not rodent adipocytes (Bordicchia *et al*, [@b4]), ANP was shown to induce gluconeogenesis in rat hepatocytes (Rashed *et al*, [@b13]). Therefore, it is curious that ANP expression is dramatically enhanced in HCM, but this mechanism for hepatic glucose output was left unexplored in these studies. Similarly, elevated expression of the Mediator subunit MED13 in the heart confers metabolic benefits in mice. MED13 is negatively regulated by a cardiac specific microRNA, miR-208, which plays a key role in cardiac hypertrophy (Grueter *et al*, [@b8]). Whether the miR-208/MED13 axis influences the metabolic consequences associated with HCM is an interesting question for the future. Perhaps a miR-208 inhibitor can remedy the metabolic defects observed in HCM by activating cardiac MED13, thus enhancing systemic metabolism, and reversing or preventing liver steatosis. In summary, the work of Magida and Leinwand ([@b11]) highlights the inextricable relationship between sarcomeric structural integrity and metabolically-derived energy at the organismal level, and opens up many more avenues for future investigation. Conflict of interest ==================== The authors declare that they have no conflict of interest. [^1]: See also: JA Magida & LA Leinwand (April 2014)

Introduction {#s1} ============ In the last decades, primate populations have suffered great demographic declines [@pone.0071527-Walsh1], [@pone.0071527-Campbell1]. These declines are due to several reasons, all having human activities and/or infectious disease epidemics as their core basis. However, little is known about how these threats translate into actual decrease in population size. Poaching, pet trade, slash-and-burn agriculture, deforestation associated with logging and agricultural activities, large-scale agricultural plantations, and other threats explain the biodiversity loss and fragmentation of several primate habitats worldwide [@pone.0071527-Gippoliti1], [@pone.0071527-S1]. On the large scale distribution patterns of species are shaped by environmental and historical constraints [@pone.0071527-Lehman1], [@pone.0071527-Kamilar1], [@pone.0071527-Harcourt1]. On the small scale behavioural characteristics including territoriality, location of nesting sites, predation, and competition for food or mates determine where a species is found [@pone.0071527-Nkurunungi1]. Today, human disturbance, quantifiable by population density, socio-economic and cultural factors, and the extent of roads and highways [@pone.0071527-Fa1], [@pone.0071527-Yackulic1], [@pone.0071527-Vanthomme1], is one of the major determinants of wildlife distributions [@pone.0071527-Paudel1], including chimpanzees [@pone.0071527-Torres1], [@pone.0071527-Junker1]. Primate distributions in Africa have been greatly affected by the expansion of road networks [@pone.0071527-Blom1], [@pone.0071527-Bald1], not only providing access to settlers but facilitating illegal hunting and logging [@pone.0071527-Vanthomme1], [@pone.0071527-Hashimoto1], [@pone.0071527-Devos1]. In West and Central Africa, hunting is one of the greatest threats due to the dependence of local populations on bushmeat, for subsistence and for commerce [@pone.0071527-Oates1], [@pone.0071527-Linder1], [@pone.0071527-Kuehl1]. Rivers can act as natural barriers shaping primate distribution patterns [@pone.0071527-Harcourt1], [@pone.0071527-Inogwabini1], while at the same time allow for an easy transport of bushmeat [@pone.0071527-Fa1]. Many studies have analyzed primate distributions with respect to different levels of human disturbance [@pone.0071527-Oates1], [@pone.0071527-Tutin1], [@pone.0071527-Hall1], [@pone.0071527-Pusey1], but few have provided a detailed quantification of the relationship [@pone.0071527-Torres1], [@pone.0071527-Junker1], [@pone.0071527-Lahm1], [@pone.0071527-Stokes1]. The impact of human activities on chimpanzee populations has been evaluated over large areas [@pone.0071527-Junker1], [@pone.0071527-Stokes1], but there are few quantitative studies that have been conducted at a small geographic scale [@pone.0071527-Vanthomme1], [@pone.0071527-Torres1], [@pone.0071527-Kuehl1]. The Western chimpanzee *Pan troglodytes verus* has been listed as Endangered on the IUCN Red List since 1988 [@pone.0071527-IUCN1]. *Pan t. verus* has, nonetheless, undergone a considerable population reduction over the last 20 to 30 years [@pone.0071527-Boesch1]. Its range encompasses nine West African countries, although it is already considered rare or close to extinction in four of them: Burkina-Faso, Ghana, Guinea-Bissau, and Senegal [@pone.0071527-Kormos1]. Junker et al. [@pone.0071527-Junker1] carried out a meta-analysis for eight taxa of African great apes that assessed continent-wide suitable environmental conditions and how they had changed over 20 years. They found that the western chimpanzee had suffered a decline of 11% in the area of suitable environmental conditions since 1992. The Cantanhez National Park in Guinea-Bissau has suffered the same loss of chimpanzee habitat (11%) since 1986 (surveys in 1986, 1994 and 2003; [@pone.0071527-Torres1]). Population estimates for this subspecies range from 21,300 to 55,600 individuals [@pone.0071527-Kormos1], with 600--1000 individuals in Guinea-Bissau [@pone.0071527-Gippoliti1]. Questionnaire surveys suggest that the range of chimpanzees is restricted by humans [@pone.0071527-Brugiere1]. In Guinea-Bissau, the highest human population densities are found in the north of the country (<http://www.bestcountryreports.com/Population_Map_Guinea-Bissau.php>) and suitable habitat for *P. t. verus* is found only in the south [@pone.0071527-Torres1]; in the south-west, in the region of Tombali, including the Cantanhez Forest and Cacine Basin, and in the region of Quinara, particularly in Lagoas de Cufada Natural Park (LCNP), and in the east in Boé ([Figure 1](#pone-0071527-g001){ref-type="fig"}) [@pone.0071527-Gippoliti2], [@pone.0071527-Sousa1]. {#pone-0071527-g001} The first report on the status of *P .t. verus* in this country dates from 1940 (Monard 1940 *in* [@pone.0071527-Gippoliti3]) and only recently have studies provided updated assessments [@pone.0071527-Kamilar1], [@pone.0071527-Casanova1], [@pone.0071527-Sousa2], which, however, have focused only on a small area of the country or have employed a less robust study design [@pone.0071527-Brugiere1], [@pone.0071527-Sousa1], [@pone.0071527-Sousa2] than is recommended [@pone.0071527-Buckland1]. For LCNP, the evidence available before 2008 suggests that chimpanzees occurred in the forests surrounding 23 villages, while in other areas they were scarce and seriously threatened by deforestation, poaching and increasing human populations, in others there were no signs at all of their presence [@pone.0071527-Casanova2]. No reliable estimates of the current population size and density of *P.t verus* in this important protected area were available. In this paper, we provide robust density and population size estimates for the western chimpanzee population in LCNP using a distance-sampling approach [@pone.0071527-Buckland1], [@pone.0071527-Buckland2]. We also assess patterns of chimpanzee occurrence inside LCNP in relation to landscape-scale covariates of human disturbance, such as roads, rivers, and settlements. Methods {#s2} ======= Ethics Statement {#s2a} ---------------- All research was conducted under permissions from *Instituto da Biodiversidade e Áreas Protegidas* (IBAP), Guinea-Bissau. No animals were captured or handled during this study. Study Site {#s2b} ---------- Guinea-Bissau is a small (36,125 km^2^) West African country ([Figure 1](#pone-0071527-g001){ref-type="fig"}) with relatively flat topography [@pone.0071527-Gippoliti1]. Patches of primary forest remain in the north-west and south-west of the country, in the regions of Cacheu, Quinara, and Tombali [@pone.0071527-Gippoliti1]. Landscape satellite images have shown, however, that dense canopy forests continue to decline in extent and number, being replaced by open canopy forests and savannah-woodland [@pone.0071527-Oom1], [@pone.0071527-FAO1] ([Table S1](#pone.0071527.s003){ref-type="supplementary-material"}). The dense canopy forests are being replaced by subsistence farming of rice, sugarcane and maize, and cashew plantations [@pone.0071527-Oom1]. LCNP is located in the region of Quinara, in southern Guinea-Bissau, between 11°34′ and 11°51′ N and 14°49′ and 15°16′ W ([Figure 1](#pone-0071527-g001){ref-type="fig"}) [@pone.0071527-Catarino1], [@pone.0071527-Catarino2]. The climate is characterized by an average annual temperature of 26°C and an average annual rainfall of 2200 mm, with a pronounced rainy season from June to October/November [@pone.0071527-Catarino3]. LCNP is an internationally recognized Ramsar site and covers an area of 890 km^2^ [@pone.0071527-Sousa1]. It is managed by a governmental organization, the *Instituto da Biodiversidade e das Áreas Protegidas* (IBAP). IBAP faces severe funding and personnel limitations despite multiple threats to the park\'s integrity. Different ethnic groups (around 11,000 people) live in the park, relying extensively on natural resources for their survival. The villages are close to roads or water sources (rivers or lagoons). The villagers\' livelihoods depend on subsistence farming, and bushmeat hunting is common. Chimpanzee Nest Surveys {#s2c} ----------------------- As chimpanzees in LCNP are not habituated to humans and, as such, very elusive (only 10 encounters were recorded in 2011), we relied on nest surveys for estimating population densities [@pone.0071527-Hashimoto1], [@pone.0071527-Tutin1], [@pone.0071527-Ghiglieri1], [@pone.0071527-Furuichi1]. Nest counts are a useful surrogate for estimating ape densities and monitoring their populations over time [@pone.0071527-Khl1], [@pone.0071527-Buckland3]. Most studies recommend using line transect surveys [@pone.0071527-Blom1], [@pone.0071527-Devos1], [@pone.0071527-Buij1], [@pone.0071527-FleuryBrugiere1], during which all the nests visible from the transect line are counted either individually [@pone.0071527-Hashimoto1], [@pone.0071527-Ghiglieri1] or in groups [@pone.0071527-Tutin1], or both [@pone.0071527-Furuichi1]. Two nest count techniques are recommended: Standing-Crop Nest Counts (SCNC) and Marked Nest Counts (MNC). SCNC consists of only a single visit to all transects, counting all nests irrespective of their age class, whereas MNC consists of counting only nests built between successive visits to the same transect, with all nests removed in the first visit, within an interval short enough to guarantee that no new nests will disappear between repeated visits. SCNC is logistically easier [@pone.0071527-Blom1], [@pone.0071527-Tutin1], [@pone.0071527-FleuryBrugiere1], [@pone.0071527-Plumptre1], although it requires independent estimates of rates of nest production and nest decay. MNC, on the other hand, is the only alternative when nest decay rate is lacking [@pone.0071527-Hashimoto1], [@pone.0071527-Devos1], [@pone.0071527-Furuichi1], [@pone.0071527-Plumptre1]. The pros and cons of these methods are well described in the literature [@pone.0071527-Devos1], [@pone.0071527-Khl1]. Here, we used a hybrid approach, depending on survey year, sampling unit, and nest count method. We estimated (1) densities of chimpanzee nests for 2010 and 2011 using line transect surveys and SCNC, (2) chimpanzee density for 2010 using strip transect surveys and MNC, (3) nest decay rate in 2010 for LCNP, and (4) subsequently, based on line transect sampling and SCNC, the chimpanzee density for 2011 using our estimate of nest decay rate and published information on nest production rate. Following the recommendations for an adequate study design [@pone.0071527-Buckland1], [@pone.0071527-Buckland2], 11 linear transects (each one 3 km long) were superimposed over LCNP as a grid of equally-spaced (5×6 km) parallel lines ([Figure 1](#pone-0071527-g001){ref-type="fig"}). Our design adhered to the assumptions underlying distance sampling [@pone.0071527-Buckland1]. All field work was conducted by JC, during 10 months in the dry seasons of 2010 and 2011. Habitat types along each line transect were classified based on canopy coverage as either dense canopy forest, open canopy forest, or savannah-woodland [@pone.0071527-Catarino1], [@pone.0071527-Catarino3] and geo-referenced to calculate the exact proportion of each in these sampling units. Line transects were visited four and five times, respectively, in 2010 and 2011, at biweekly intervals. The transects were walked at a steady speed of about 1 km/h [@pone.0071527-Furuichi1], [@pone.0071527-Plumptre1]. Only data on individual nests were collected, considering that nest groups were hard to identify as nests showed the highest aggregation in dense canopy forests. Whenever a nest was found, the perpendicular distance from the transect line to the nest was measured with a tape measure or range finder (Bushnell Yardage Pro Sport 450). To avoid double counting, nest trees were tagged with a rope. Nests observed during the 2011 survey were geo-referenced using a Global Positioning System (GPS Garmin 60) (only one reading was taken when there were several nests close to each other in the same tree). Analyses {#s3} ======== Distance Sampling Nest Surveys {#s3a} ------------------------------ We used the software DISTANCE 6.0 [@pone.0071527-Thomas1] to estimate nest encounter rate, the nest detection function, and the densities of nests and chimpanzees. We first explored the distance data in histograms, considering different cut-off points and fitting a half-normal model without adjustment terms to get a first feel for the shape of the detection function and to assess the best truncation distance (*w*). Some data truncation is recommended to avoid problems fitting the tail of the distribution, and 5% has been recommended as a plausible omnibus value for *w* [@pone.0071527-Buckland1]. Subsequently, we considered a range of other models implemented in DISTANCE to assess which model provided the best fit to the data. The best model was selected using Akaike\'s Information Criterion (AIC), and by evaluating the goodness of fit of the models based on the standard chi-square, the Kolmogorov-Smirnov and the Cramér-von Mises tests [@pone.0071527-Buckland1]. Estimation of Nest Densities from Line Transects using SCNC {#s3b} ----------------------------------------------------------- Only nest data from the first visit to any particular transect in each year were used to obtain the nest detection function by habitat. We opted not to include data from subsequent visits for nest density estimation because, because even under the assumption that during biweekly intervals no newly built nest will disappear, old nests were detected during repeated visits [@pone.0071527-Furuichi1]. Nest data were examined following the procedures described above for model selection and model evaluation. First, we estimated nest densities for each habitat. Nest density was estimated using the conventional distance sampling estimator as where *n* represents the number of the detected nests, is the estimated probability density function of detected nests evaluated at distance 0 and *L* is the total length of transects [@pone.0071527-Buckland1]. Global nest density was obtained as a weighted average of habitat specific estimates, with weights given by habitat area. Estimation of Chimpanzee Density and Nest Decay Rate from Strip Transects using MNC {#s3c} ----------------------------------------------------------------------------------- The linear transects were also regarded as constituting a grid of randomly positioned strips. Unlike line transect sampling, standard strip transect sampling assumes that all objects (either individuals or indirect evidence of their occurrence) within a distance along transects are detected, and providing a large enough sample size an unbiased estimator of density and precise estimates of abundance can be obtained [@pone.0071527-Blom1], [@pone.0071527-Buckland4], [@pone.0071527-Ogutu1]. To maximize the likelihood of detecting all nests within distance *s*, we used the 2010 dataset considering only nests from the second visit onwards to find the distance *s* for which we could consider that all new nests were detected, i.e., that would allow us to define sensible strip transects. The width of strip transects (*s*) was defined by the distance over which the shoulder of the detection function extended. We considered habitat-specific strip transects, given that the width over which it is reasonable to assume that all nests are detected was expected to be habitat dependent (compare also [@pone.0071527-Chapman1]). Following data exploration as described above, we estimated chimpanzee density by habitat, and then global density weighted by habitat, as where *n* represents the number of new nests detected within the strip transect from the second visit onwards, *L* is the total strip transect length, *s* is the width of the strip transect (taken from the shoulder of the detection function as described above), *t* is the number of days elapsed between the first and last survey, and is the daily nest production rate [@pone.0071527-Buckland1]. As an estimate of nest production rate for our study area or Guinea-Bissau is lacking, we used a published estimate of 1.143 nests built per animal per day (%CV = 3.51) from Taï National Park, Ivory Coast [@pone.0071527-Kouakou1]. 95% confidence intervals for nest encounter rates and density estimates were calculated in R version 2.15.3 [@pone.0071527-R1] using a nonparametric bootstrap procedure (999 resamples). Using the above 2010 estimates of nest density and chimpanzee density, we subsequently obtained an estimate of nest decay rate by rearranging the following equation and solving it for where denotes the estimate of nest density, is the nest decay rate (days) and is the nest production rate per day [@pone.0071527-Buckland1]. Nest decay rate was calculated for 2010. The variance for the decay rate estimator can be approximated via the delta method [@pone.0071527-Powell1] as where *CV* represents the coefficient of variation of the corresponding estimate, i.e., the standard error of the estimate divided by the estimate. Estimation of Chimpanzee Densities from Line Transects using SCNC {#s3d} ----------------------------------------------------------------- Based on the estimated rate of nest decay and again using the estimate of the daily rate of nest production from Taï National Park, we were able to apply the SCNC technique [@pone.0071527-Plumptre2] to estimate a habitat-weighted average of chimpanzee densities using [equation 3](#pone.0071527.e008){ref-type="disp-formula"} for the 2011 data. Relationship between Nest Distribution and Landscape-Scale Covariates {#s3e} --------------------------------------------------------------------- Nests were used as an indirect measure of the presence of chimpanzees [@pone.0071527-Torres1]. A randomization test was performed using the package *COIN* in R version 2.15.3 [@pone.0071527-R1] to assess relationships between the spatial distribution of chimpanzee nests and a set of landscape-scale variables which can be regarded as proxies for human disturbance: principal rivers, roads (including main and secondary roads), and human settlements. To determine whether nests were distributed in a non-random fashion with respect to these variables we compared the mean distances between nest locations and each environmental feature to mean distances generated in the same way based on random locations of 214 (number of independent nest locations in the data) nests within transects. This procedure was repeated 1000 times and statistical significance was assessed by recording the number of times the mean value from random locations was lower than the observed value for nest locations [@pone.0071527-Manly1]. We used the Geographic Information System (GIS) ARCMAP 9.3 package to calculate the shortest straight-line distance between each nest and a given environmental feature. All spatial layers were projected into Universe Transverse Mercator (UTM) Zone 28N. Environmental digital data were made available through the CARBOVEG project (<http://carboveg-gb.dpp.pt/>) or taken from Amaro [@pone.0071527-Amaro1]. To ensure that sampling of random points (*n* = 214) fell within the area surveyed, a buffer was constructed along both sides of the transects and limited by the maximum distance at which a nest was observed from the transect line surveys (84 m). In addition, to investigate the distribution of habitats in relation to the environmental features considered, we also plotted the measured distances grouped by habitat type and tested for statistical differences. Results {#s4} ======= Chimpanzee Nest Surveys {#s4a} ----------------------- Survey effort for SCNC was 67.2 km, whereas 235.2 km were walked for MNC. Line transects were composed mostly of savannah-woodland (46.81%), followed by dense canopy forest (26.28%), open canopy forest (9.97%), agricultural areas (10.08%), herbaceous savannah (5.35%), rivers or lagoons (1.23%) and human settlements (0.28%) ([Figure S1](#pone.0071527.s001){ref-type="supplementary-material"}). These relative proportions of habitat types in LCNP constitute a good representation of their occurrence countrywide ([Table S1](#pone.0071527.s003){ref-type="supplementary-material"}). In 2010, 211 nests were detected, 182 of them during the first visit. A total of 248 nests were recorded in 2011, 117 of those during the first visit. The highest count during the first visit is the natural consequence of nest accumulation over time. Once corrected for estimated habitat specific detectability (see results below) nests were found mostly in dense forest (71.55 %), and fewer nests were recorded in the two habitats with less canopy coverage: savannah-woodland (17.98%) and open forest (10.48 %). Distances at which nests were detected from the line transect differed significantly among habitat types (Kruskal-Wallis test, 40.82, *df* = 2, *p*\<0.001) ([Figure S2](#pone.0071527.s002){ref-type="supplementary-material"}), being greatest in savannah and shortest in dense forest. Estimation of Nest Densities from Line Transects using SCNC {#s4b} ----------------------------------------------------------- Truncating the data at 42 meters, a uniform model with a cosine adjustment provided the best fit for the 2010 dataset (AIC = 0.77; the reported values of AIC correspond to the comparison with the second best model unless otherwise noted). Open forests showed the highest nest encounter rate followed by dense forests and savannahs ([Table 1](#pone-0071527-t001){ref-type="table"}). Habitat-specific nest density estimates were substantially higher for open and dense canopy forests compared to savannahs ([Table 1](#pone-0071527-t001){ref-type="table"}). For 2010, the global nest density estimate for LCNP was 167.97 nests per km^2^ (95% CI 55.61--507.34). 10.1371/journal.pone.0071527.t001 ###### Comparison of nest count, nest encounter rate (nests/km) and nest density estimates (nests/km^2^) between 2010 and 2011 of the chimpanzees in Lagoas de Cufada Natural Park. {#pone-0071527-t001-1} Year Survey habitat No. of nests Nests/km \[95% CI\][a](#nt101){ref-type="table-fn"} Nests/km^2^ \[95% CI\][a](#nt101){ref-type="table-fn"} \% CV[b](#nt102){ref-type="table-fn"} ------ ---------------------------------------- ------------------------------------- ----------------------------------------------------- -------------------------------------------------------- --------------------------------------- 2010 Global[c](#nt103){ref-type="table-fn"} 167.97 \[55.61--507.34\] 44.21 DF 65[\*](#nt104){ref-type="table-fn"} 7.37 \[1.77--30.64\] 229.68 \[55.22--955.30\] 66.17 OF 67[\*](#nt104){ref-type="table-fn"} 20.03 \[4.88--82.26\] 364.37 \[90.07--1,474.10\] 54.36 SAV 36[\*](#nt104){ref-type="table-fn"} 2.29 \[0.81--6.44\] 27.28 \[9.70--76.72\] 48.20 2011 Global[c](#nt103){ref-type="table-fn"} 75.56 \[27.21--209.86\] 42.22 DF 72[\*](#nt104){ref-type="table-fn"} 8.16 \[3.36--19.81\] 233.21 \[96.02--566.39\] 38.87 OF 17[\*](#nt104){ref-type="table-fn"} 5.08 \[1.01--25.59\] 129.79 \[26.39--638.39\] 63.52 SAV 13[\*](#nt104){ref-type="table-fn"} 0.83 \[0.19--3.52\] 11.82 \[2.77--50.36\] 71.23 CI, confidence interval. Coefficient of variation. Average nest density weighted by habitat. \*Distance truncated at 42 m (2010) and 35 m (2011). DF- dense canopy forests; OF- open canopy forests; SAV- savannah-woodlands. Applying a truncation distance of 35 m, a uniform model with a cosine adjustment (AIC = 3.06) best fitted the 2011 data. The number of nests observed in dense forests was similar to 2010, whereas far fewer nests were observed in open forests and savannahs, leading to much lower estimates of nest encounter rate for these habitats compared to the previous year ([Table 1](#pone-0071527-t001){ref-type="table"}). Nest density was highest in dense forests, followed by open forests and savannahs, resulting in a global nest density estimate of 75.56 nests per km^2^ (95% CI 27.21--209.86) estimated for the entire park, less than half of the previous year\'s estimate ([Table 1](#pone-0071527-t001){ref-type="table"}). Note, however, the considerable overlap in the 95% CIs, suggesting that this reduction is not statistically significant. Estimation of Chimpanzee Density and Nest Decay Rate from Strip Transects using MNC {#s4c} ----------------------------------------------------------------------------------- Based on AIC, a half-normal model with cosine adjustment showed the best fit to the 2010 data after truncation at 44 m (AIC = 2.08). Based on the detection function, strip transects of different width were obtained for each habitat ([Table S2](#pone.0071527.s004){ref-type="supplementary-material"}). Habitat-specific density estimates were similar in dense and open forests ([Table S2](#pone.0071527.s004){ref-type="supplementary-material"}). The park-wide weighted average estimate was 0.50 nest builders per km^2^ (95% CI 0.18--1.39), corresponding to about 311 chimpanzees for LCNP. Using these estimates of chimpanzee densities and the nest densities estimated above for 2010 ([Table 1](#pone-0071527-t001){ref-type="table"}), nest decay rate was estimated to be 293.9 days (%CV = 58.80). Estimation of Chimpanzee Densities from Line Transects using SCNC {#s4d} ----------------------------------------------------------------- Using the above estimated rate of nest decay, chimpanzee density estimates for 2011 were highest in dense forests and lower in savannahs ([Figure 2](#pone-0071527-g002){ref-type="fig"}, [Table 2](#pone-0071527-t002){ref-type="table"}). The habitat-weighted average for LCNP was 0.22 nest builders per km^2^ (95% CI 0.08--0.62), corresponding to 137 chimpanzees. {#pone-0071527-g002} 10.1371/journal.pone.0071527.t002 ###### Chimpanzee density estimates (builders/km^2^) for each habitat and for the Lagoas de Cufada Natural Park obtained in 2011 based on Standing-Crop Nest Counts, using our estimated rate of nest decay. {#pone-0071527-t002-2} Survey habitat Density (builders/km^2^) 95% CI[a](#nt106){ref-type="table-fn"} (builders/km^2^) %CV[b](#nt107){ref-type="table-fn"} ---------------------------------------- --------------------------------------- --------------------------------------------------------- ------------------------------------- Global[c](#nt108){ref-type="table-fn"} 0.22[\*](#nt109){ref-type="table-fn"} 0.08--0.62 42.22 Dense canopy forests 0.69[\*](#nt109){ref-type="table-fn"} 0.28--1.67 38.87 Open canopy forests 0.37[\*](#nt109){ref-type="table-fn"} 0.08--1.90 64.00 Savannah-woodlands 0.03[\*](#nt109){ref-type="table-fn"} 0.01--0.15 71.23 Confidence interval Coefficient of variation. Average nest density weighted by habitat. \*results using the nest decay rate of 293.9 days (%CV = 58.80). It is important to note that the greatest chimpanzee density was estimated for the least available habitat type ([Figure 2](#pone-0071527-g002){ref-type="fig"}). Relationship between Nest Distribution and Landscape-Scale Covariates {#s4e} --------------------------------------------------------------------- The randomization test showed that the spatial distribution of chimpanzee nests differs significantly from a random pattern for all three environmental variables examined. Chimpanzees prefer to build their nests farther away from roads (*Z* = 9.55, *p*\<0.001), settlements (*Z* = 7.60, *p*\<0.001), and rivers (*Z* = −5.81, *p*\<0.001) than would be expected by chance. On average nests were observed farther away from settlements (4.13 km, 95% CI 3.88--4.37), than from roads (2.58 km, 95% CI 2.37--2.80) or rivers (1.28 km, 95% CI 1.10--1.45) ([Figure 3](#pone-0071527-g003){ref-type="fig"}). {#pone-0071527-g003} The distribution of habitats differed significantly in relation to the environmental features (Kruskal-Wallis test: rivers 10.55, *df* = 2, *p*\<0.05; roads 124.29, *df* = 2, *p*\<0.001; settlements 56.89, *df* = 2, *p*\<0.001), whereby habitats with a lower tree canopy cover (open forests and savannahs) were found closer to all landscape variables, contrasting with the large distances obtained for dense forests ([Figure 4](#pone-0071527-g004){ref-type="fig"}). {#pone-0071527-g004} Discussion {#s5} ========== Chimpanzee Population Density and Size {#s5a} -------------------------------------- We presented chimpanzee density estimates at LCNP for 2010 (0.22 nest builders per km^2^, 95% CI 0.08--0.62) and 2011 (0.50 nest builders per km^2^, 95% CI 0.18--1.39). While it is true that the 95% confidence intervals overlap considerably, hence not suggesting a large population change, we believe the apparent doubling of the point estimates is an artefact (i.e. a consequence of the large CV\'s associated with these estimates) and these numbers provide nonetheless a good comparison for future studies in this region. By comparison with previous studies for Guinea-Bissau we report the lowest chimpanzee density estimate ([Table S3](#pone.0071527.s005){ref-type="supplementary-material"}). There are three published studies for this country that provide density estimates based on distance sampling, however, just one of them employed a random sampling of line transects [@pone.0071527-Sousa3] and the remainder used abandoned trails or trails used by locals for access to crops or for hunting [@pone.0071527-Sousa2], [@pone.0071527-Sousa4]. Our estimates of chimpanzee densities for each habitat type show an inverse relationship with habitat availability ([Figure 2](#pone-0071527-g002){ref-type="fig"}), which highlights the preference for building nests and the suitable nesting conditions offered by dense canopy forests. The present results confirm previous studies [@pone.0071527-Sousa2], [@pone.0071527-Sousa3], [@pone.0071527-Sousa4] in demonstrating that tree canopy cover plays an important role in habitat choice for nest building in chimpanzees from Guinea-Bissau, in contrast to what has been found at other sites [@pone.0071527-Tutin1], [@pone.0071527-Pruetz1], [@pone.0071527-HernnandezAguilar1], [@pone.0071527-Boyer1]. As an adaptation when dealing with declines of their preferred habitat, chimpanzees evidently opt to nest in savannah-woodlands [@pone.0071527-Tutin1], [@pone.0071527-Marchesi1]. LCNP chimpanzees also use open canopy forests for nest building, which taken together underscores the importance of considering all habitat types for estimating chimpanzee densities, and also with respect to conservation efforts. Method selection is a compromise between sound and well-established methodologies and the available resources and personnel [@pone.0071527-Marshall1]. Population size estimates of primates rely on certain assumptions, which vary depending on the different methodologies available [@pone.0071527-Khl1], [@pone.0071527-Marshall1], [@pone.0071527-Whitesides1], [@pone.0071527-Boyko1]. Several studies applied nest count methods ([Table S3](#pone.0071527.s005){ref-type="supplementary-material"}). Nest production rates are usually taken from long-term monitoring of habituated chimpanzees due to the difficulty of observing wild chimpanzees [@pone.0071527-Khl1], [@pone.0071527-Plumptre2]. Despite the differences found in this variable between sites and seasons, many studies used non-site specific information (see [Table S3](#pone.0071527.s005){ref-type="supplementary-material"}). Our estimate of nest decay corresponds only to the dry season and with 293.9 days (%CV = 58.80) for 2010 was close to those reported from other locations within the western chimpanzee\'s range ([Table S4](#pone.0071527.s006){ref-type="supplementary-material"}). Further studies are required during the rainy season to compare robust estimates of the life span of nests from LCNP with those from other sites. Although we applied SCNC to estimate chimpanzee densities among habitats the overlap of the confidence intervals indicates low power to detect changes ([Figure 2](#pone-0071527-g002){ref-type="fig"}, [Table 2](#pone-0071527-t002){ref-type="table"}). In future research we suggest that the decay rate presented here be incorporated when using SCNC. However, as decay rate depends on unmeasured covariates that may vary temporally and spatially [@pone.0071527-Kouakou1], [@pone.0071527-Buij2], to avoid bias and obtain an accurate population size estimate, we recommend a new estimate under actual survey conditions [@pone.0071527-Marques1] by monitoring the decay of new nests during successive visits. Other techniques have emerged that deal with the time consuming process of monitoring the decay of a large and diverse sample of nests to obtain accurate estimates of life span of nests [@pone.0071527-Laing1]. For example, assuming a Markov chain for the state of a nest, with an absorbing state which represents nest disappearance, van Schaik et al. (1995) were able to estimate the time a nest takes to disappear based on the observation of nests (and their corresponding state) over time. For more details see [@pone.0071527-Khl1], [@pone.0071527-Buij2], [@pone.0071527-vanSchaik1]. Current work in progress based on our data set uses both state space models (D. L. Borchers, pers. comm.) and hidden Markov Models (R. Langrock, pers. comm.) to simultaneously estimate nest decay rate and abundance. Chimpanzee Distribution in Relation to Human Disturbance {#s5b} -------------------------------------------------------- Even though chimpanzees reportedly show a certain ability to coexist with humans [@pone.0071527-Hockings1], places they consider safe for nest building have previously been shown to be distant from human settlements, roads, and rivers [@pone.0071527-Vanthomme1], [@pone.0071527-Kuehl1], [@pone.0071527-Stokes1], further pointing towards a perhaps not surprising negative influence of human disturbance on chimpanzee distribution. A recent meta-analysis by Junker et al. (2012) showed that measures of human impact such as proximity to settlements make a large contribution to the loss of suitable ecological conditions for chimpanzees. Historically most of the chimpanzee populations in Guinea-Bissau had human settlements within their range, and hence people regularly come into contact with chimpanzees on roads (main and secondary roads), in cultivated areas, and around the edges of forest fragments [@pone.0071527-Hockings2]. Roads have been shown to be prejudicial for chimpanzee populations as they facilitate poaching and illegal hunting, and indirectly boost illegal logging [@pone.0071527-Hashimoto1], [@pone.0071527-Devos1], as has also been reported for other primates and other taxa such as ungulates, rodents and carnivores [@pone.0071527-Fa1], [@pone.0071527-Yackulic1]. Cashew nuts are Guinea-Bissau\'s principal cash crop, representing 90% of the country\'s exports since 2000 [@pone.0071527-Bald1], [@pone.0071527-UNEP1]. Most of the roads and settlements in LCNP are surrounded by extensive cashew plantations. Replacement of native forest by these monocultures reduces the availability of those trees that have canopies suitable for chimpanzees to build their nests. The cashew pulp is widely appreciated by many taxa, and some farmers reported that chimpanzees sometimes split branches while trying to reach the fruit at greater heights, leading to irreversible damages of trees and often resulting in chimpanzee-human conflict (S. Camará, pers. comm.). This study coincided with the period of cashew harvesting (March to late June), when the number of people inside the park, as well as road traffic, usually increases. Unlike the park residents, in general, these temporary harvest workers show little awareness with respect to the conservation of park biodiversity, compromising and undermining the conservation efforts by guards and residents. The distributions of several forest-dwelling primate taxa in west and central Africa have been shown to be limited by rivers; larger rivers have a greater barrier effect on species distribution of forest taxa than smaller rivers, as observed for the Congo River and the rivers bounding the Dahomey Gap [@pone.0071527-Harcourt1]. The Dahomey Gap, a dry savanna corridor interrupting the West African rainforest, has been a barrier for primate species either by its aridity or by its flanking rivers, the Volta and Niger [@pone.0071527-Harcourt1]. The main rivers surrounding Lake Tumba, Congo, have also acted as a barrier, influencing the distribution of bonobos and chimpanzees [@pone.0071527-Inogwabini1]. LCNP is delimited by two main rivers, the Corubal in the north and the Buba River in the south, which limits the chimpanzee distribution north and south of the protected area ([Figure 1](#pone-0071527-g001){ref-type="fig"}). People living in remote areas of LCNP with limited road access use navigable rivers as transportation routes, which could have the same negative impact on chimpanzees as roads. Methodological Implications {#s5c} --------------------------- Chimpanzee populations worldwide are declining at alarming rates and an immediate reclassification of chimpanzees to a status of "critically endangered" has been recommended (e.g. [@pone.0071527-Walsh1]). In light of such declines there is an urgent need to standardise appropriate designs and methodologies for long-term monitoring if the conservation of remaining chimpanzee populations is a priority for biodiversity management [@pone.0071527-Morgan1]. In this context, it is essential to consider the bias associated with a certain survey methodology, as well as its efficiency and cost-effectiveness [@pone.0071527-Ogutu1]. How can we reliably detect population declines within and between protected areas? What is the best way to provide baseline information for long-term population monitoring? In this regard it is crucial to stress that using trails or reconnaissance surveys might result in biased density estimates, compared to line transect surveys based on randomly placed transects, which, although more labour-intensive and expensive, should be the method of choice as they provide unbiased and potentially more accurate population estimates [@pone.0071527-Buckland1]. SCNC have been a viable and economical way to detect population declines, and procedures of monitoring programs and assessment of human impacts are performed using MNC surveys. The Ape Populations, Environments, and Surveys (A.P.E.S) Database aims to compile existing great ape survey data and make density and distribution data accessible to the scientific community (<http://apes.eva.mpg.de> ). Our data will be made available in this database to help incentivize more standardized monitoring efforts and enable comparisons between different study sites [@pone.0071527-Campbell1], [@pone.0071527-Junker1]. Final Considerations {#s5d} -------------------- Long-term population monitoring in LCNP, an important refuge for coastal populations of the western chimpanzee, would be highly desirable and may be achieved by investing in local training and capacity building. In general, human communities need to be included in conservation management, for instance by employing local people as park guards or tourist guides, to ensure effective long-term conservation [@pone.0071527-Kormos1]. As a mitigation measure to minimize human-chimpanzee conflict it would be desirable to concentrate crops, including future cashew plantations, in zones that are already disturbed and where environmentally sustainable practices could be implemented [@pone.0071527-Vanthomme1]. We also recommend an effective control of illegal hunting by strengthening and enforcing the existing law, which forbids poaching (Decree No. 21/1980). As ours, several other studies have shown the importance of protected areas for the preservation of stable primate populations. As there is evidence, however, that primates continue to use resources outside protected areas, recent studies advocate a landscape-scale conservation approach that takes into account the ecological requirements of species at larger spatial scales [@pone.0071527-Stokes1], [@pone.0071527-Butsic1], [@pone.0071527-Jones1]. Finally, our study contributes to our understanding of ecological patterns and how chimpanzees are influenced by human disturbance. In this regard it is, however, important to keep in mind that the chimpanzee-human relationship is complex, and present-day distribution patterns may not be explained alone by currently measurable variables as they may in part also reflect species adaptive responses to historical human activities. Supporting Information {#s6} ====================== ###### **Relative proportions of habitat type found along each line transect.** (TIFF) ###### Click here for additional data file. ###### **Boxplots showing the distances at which nests were detected from the line transects.** The data from 2010 and 2011 surveys were combined, and distances truncated at \>60 m. (TIF) ###### Click here for additional data file. ###### Deforestation rate in Guinea-Bissau based on Landsat satellite imagery from 1990 to 2007 \[data from Oom et al. [@pone.0071527-Oom1]\]. (DOCX) ###### Click here for additional data file. ###### Chimpanzee density estimates (builders/km^2^) for each habitat and for the *Lagoas de Cufada* Natural Park obtained in 2010 based on marked-nest counts, using strip transect surveys. (DOCX) ###### Click here for additional data file. ###### Estimates of chimpanzee densities (chimpanzees/km^2^) and population size reported for several study sites based on *nest count methods*. Estimates of chimpanzee densities from Guinea-Bissau are shown in italics. (DOCX) ###### Click here for additional data file. ###### Life span of nests from several study sites, including our estimate of nest decay for *Lagoas de Cufada* Natural Park. (DOCX) ###### Click here for additional data file. We are grateful to the administration and staff of the Instituto da Biodiversidade e das Áreas Protegidas (IBAP) and Lagoas de Cufada Natural Park for research permissions, logistical support, and guidance. Thanks to Honório Pereira, Benjamim Indec, Idrissa Camará, Musa Mané, Umaru Candê, Bacari Sanhá, Agostinho N\'fanda, Bafode Mané and Abu Dabô, and the villagers, especially Sajdo Camará, for their logistical support and assistance with data collection. A special thanks to *projecto Dari*, especially C. Casanova for logistical help. We also thank A. Rylands, C. Meyer, H. Kuehl, and one anonymous reviewer for helpful comments. Land cover data were generously provided by L. Catarino, J.B. Melo and A. Pinto from *Instituto de Investigação Ciêntifica Tropical* (Lisbon, Portugal). [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JSC. Performed the experiments: JSC. Analyzed the data: JSC TAM LV. Contributed reagents/materials/analysis tools: JSC TAM LV. Wrote the paper: JSC TAM LV.

{#sp1 .293} {#sp2 .294}

K. I. Assaf, J. Holub, E. Bernhardt, J. M. Oliva-Enrich, M. I. Fernández Pérez, M. Canle, J. A. Santaballa, J. Fanfrlík, D. Hnyk, W. M. Nau, *ChemPhysChem* **2020**, *21*, 971. Boron cluster chemistry is dominated by icosahedrally shaped cages (Figure [1](#cphc201901225-fig-0001){ref-type="fig"}), which can be exemplified by *closo*‐B~12~H~12~ ^2−^. Larger clusters can be obtained formally by their mutual fusion. The *closo,closo*‐\[B~21~H~18~\]^−^ ion, **B21**, is an example of shared icosahedral moieties with three joint vertices.^[\[]{.ul}1[\]]{.ul}^ The COSAN ion (CObalt SANdwich, Co(C~2~B~9~H~11~)~2~ ^−^) represents another way of fusion, i. e., via a single vertex.^[\[]{.ul}2[\]]{.ul}^ **B21** exhibits a low chemical reactivity, which resists most reaction conditions except for fluorination.^[\[]{.ul}3[\]]{.ul}^ However, complexation of the potassium salt of **B21** with β‐ and γ‐cyclodextrin (CDs, naturally occurring macrocycles that are water‐soluble and formed by oligomerization of seven or eight glucopyranoside units, respectively)[4](#cphc201901225-bib-0004){ref-type="ref"}, [5](#cphc201901225-bib-0005){ref-type="ref"}, [6](#cphc201901225-bib-0006){ref-type="ref"} has already been observed in the gas phase.[7](#cphc201901225-bib-0007){ref-type="ref"} Quantum‐chemical computations in the gas phase suggested dihydrogen bonds between the B−H vertices and the polar hydroxyl groups of CDs.[7](#cphc201901225-bib-0007){ref-type="ref"} Since the interaction of boron clusters with CDs in the aqueous solution phase might open novel avenues towards interactions with other water‐soluble biomolecules, we now extended our studies of **B21** to the solution phase.[8](#cphc201901225-bib-0008){ref-type="ref"}, [9](#cphc201901225-bib-0009){ref-type="ref"} {#cphc201901225-fig-0001} High‐affinity (micromolar) binding in aqueous solution has recently been documented between various COSANs and CDs.[10](#cphc201901225-bib-0010){ref-type="ref"} However, the *closo*‐B~12~H~12~ ^2−^ cluster, the smaller building block of **B21**, only exhibits a lower affinity (millimolar) to γ‐CD.[11](#cphc201901225-bib-0011){ref-type="ref"} The association constant was considerably increased by halogenation, up to an affinity of 10^6^ M^−1^ for B~12~Br~12~ ^2−^ with γ‐CD.[11](#cphc201901225-bib-0011){ref-type="ref"} The complexation of *closo*‐B~12~X~12~ ^2−^ to the hydrophobic inner cavity of CDs appears counterintuitive at first glance, due to the doubly negative charge of these clusters and their high water solubility (hydrophilicity). The high affinity of such large anionic clusters, not only borate clusters but also others such as polyoxometalates[12](#cphc201901225-bib-0012){ref-type="ref"}, [13](#cphc201901225-bib-0013){ref-type="ref"}, [14](#cphc201901225-bib-0014){ref-type="ref"}, [15](#cphc201901225-bib-0015){ref-type="ref"}, [16](#cphc201901225-bib-0016){ref-type="ref"} or octahedral metal clusters,[13](#cphc201901225-bib-0013){ref-type="ref"}, [17](#cphc201901225-bib-0017){ref-type="ref"} to hydrophobic cavities as well as to neutral surfaces, membranes, and proteins has been described as the chaotropic effect[11](#cphc201901225-bib-0011){ref-type="ref"}, [18](#cphc201901225-bib-0018){ref-type="ref"}, [19](#cphc201901225-bib-0019){ref-type="ref"} that presents a topic of considerable current interest. Accordingly, the propensity of anionic clusters to associate to hydrophobic cavities has been attributed to (*i*) the comparative ease of desolvation of the large clusters in combination with (*ii*) the stabilization of the resulting assemblies by dispersion interactions, promoted by the high polarizability of these large anions. In this study, we carried out an experimental and computational investigation of the complex formation between **B21** and CDs in aqueous solution in order to get a deeper insight to binding of icosahedron‐based elliptic boron clusters and compare them with their closest relatives, B~12~X~12~ ^2−^ and COSANs, see Figure [1](#cphc201901225-fig-0001){ref-type="fig"}. Through this comparison, we tried to advance structure‐affinity relationships in the emerging field of chaotropic anion recognition. We prepared **B21** according to literature procedures.[1](#cphc201901225-bib-0001){ref-type="ref"}, [7](#cphc201901225-bib-0007){ref-type="ref"} The formation of host‐guest complexes between **B21** (as sodium salt) and CDs was first studied by ^1^H NMR spectroscopy. No significant changes in the ^1^H NMR spectra were observed with α‐CD, which is expected due to its small cavity size. In contrast, large complexation‐induced chemical shifts were obtained with β‐CD and γ‐CD, see Figure [2](#cphc201901225-fig-0002){ref-type="fig"}, which indicated the formation of host‐guest complexes in aqueous solution, qualitatively in agreement with the gas‐phase results.[7](#cphc201901225-bib-0007){ref-type="ref"} In particular, we observed pronounced complexation‐induced chemical shifts for the H3 and H5 protons, which are located inside the cavity, signaling the formation of deep inclusion complexes, in analogy to the complexation of perhalogenated *closo*‐B~12~X~12~ ^2−^ anions with CDs.[11](#cphc201901225-bib-0011){ref-type="ref"}, [20](#cphc201901225-bib-0020){ref-type="ref"} The fact that the two diasterotopic H6 protons of the larger CDs split into an AB system upon complexation can be accounted for by a hindered rotation of the CH~2~OH groups upon complexation. The larger shifts obtained upon complexation of **B21** with the largest investigated macrocycle, γ‐CD, indicated the formation of a tightly packed complex. The potassium salt of **B21** provided identical NMR results, which demonstrated that the counter‐ion plays no major role in the complexation in water. In contrast to the large ^1^H NMR shifts of the H3 and H5 CD protons, the ^11^B chemical shifts[21](#cphc201901225-bib-0021){ref-type="ref"} of **B21** in the complex showed comparably less diagnostic changes with respect to uncomplexed **B21**, see Figure S1 in the Supporting Information. {#cphc201901225-fig-0002} Isothermal titration calorimetry (ITC) was used to determine the absolute binding constants and to analyze the complexation thermochemistry (Figure [3](#cphc201901225-fig-0003){ref-type="fig"}); access to this technique was essential, because the binding turned out to be too strong to employ NMR titrations for accurate determination of the binding constants (*K* ~a~). An ITC dilution titration of **B21** (1 mM) was performed first; no significant heat effect was observed, which excluded a competitive enthalpic effect due to micellization/de‐aggregation events (reported for *ortho*‐COSANs),[22](#cphc201901225-bib-0022){ref-type="ref"}, [23](#cphc201901225-bib-0023){ref-type="ref"}, [24](#cphc201901225-bib-0024){ref-type="ref"} and the subsequent host‐guest titrations were performed at much lower **B21** concentrations (0.1 mM). The ITC titration profiles were consistent with a 1 : 1 binding stoichiometry, consistent with the mass‐spectrometric study, in which also no 2 : 1 host‐guest complexes had been observed.[7](#cphc201901225-bib-0007){ref-type="ref"} The *K* ~a~ values obtained by ITC for the complexation of **B21** (as sodium salt) with β‐CD and γ‐CD were found to be (1.3±0.1) ×10^5^ and (1.8±0.5) ×10^6^ M^−1^, respectively. These values are on the same order of magnitude as those measured for *meta*‐COSAN,[10](#cphc201901225-bib-0010){ref-type="ref"} twice higher than those obtained for the most tightly binding single icosahedron, i. e., B~12~Br~12~,^2−\[11\]^ and they significantly exceed the affinities of other "mononuclear" borate clusters or carboranes.[25](#cphc201901225-bib-0025){ref-type="ref"} The same trend applies for the B~12~H~12~ ^2−^ cluster, whose association constant with γ‐CD is even smaller, by three orders of magnitude.[11](#cphc201901225-bib-0011){ref-type="ref"} The ITC data showed that the complexation is enthalpically driven. The large Δ*H* values are compensated by large entropic penalties, which are jointly in line with the thermochemical fingerprint of the chaotropic effect.[11](#cphc201901225-bib-0011){ref-type="ref"}, [13](#cphc201901225-bib-0013){ref-type="ref"}, [15](#cphc201901225-bib-0015){ref-type="ref"}, [16](#cphc201901225-bib-0016){ref-type="ref"}, [17](#cphc201901225-bib-0017){ref-type="ref"}, [19](#cphc201901225-bib-0019){ref-type="ref"}, [20](#cphc201901225-bib-0020){ref-type="ref"}, [25](#cphc201901225-bib-0025){ref-type="ref"}, [26](#cphc201901225-bib-0026){ref-type="ref"} {#cphc201901225-fig-0003} We also used a computational model to gain deeper insight into the binding of **B21** to γ‐CD and expanded this investigation, for comparison, to other compounds with icosahedral γ‐CD binding motifs, specifically *ortho*‐COSAN, *meta*‐COSAN (*ortho* and *meta* define the positions of the respective C atoms in the COSAN framework),[10](#cphc201901225-bib-0010){ref-type="ref"} B~12~H~11~NH~3~ ^−^, B~12~H~12~ ^2−^, B~12~Cl~12~ ^2−^, B~12~Br~12~ ^2−^, and B~12~I~12~ ^2−^. For all these compounds, association constants (*K* ~a~) are known under the same conditions (as sodium salts, in H~2~O) and a predominant 1 : 1 binding stoichiometry can be assumed.[11](#cphc201901225-bib-0011){ref-type="ref"}, [20](#cphc201901225-bib-0020){ref-type="ref"} Very important to consider, boron clusters as well as other large anions have been found to act as so‐called superchaotropic anions, that is, they display a special hydration behavior,[11](#cphc201901225-bib-0011){ref-type="ref"}, [18](#cphc201901225-bib-0018){ref-type="ref"}, [19](#cphc201901225-bib-0019){ref-type="ref"} which can be classically described as a "water‐structure breaking" effect or in terms of a weak hydration. Accordingly, it was imperative to consider, in addition to gas‐phase interactions energies, the free energies of hydration of all involved species. The environment was described by using implicit solvent models (COSMO and SMD); only the sodium counterion was treated as a by‐stander in the host‐guest complexation process, in line with the observed absence of a counter‐ion effect when changing to potassium as cation. The known crystal structure of B~12~Br~12~ ^2−^ with γ‐CD was employed to build starting geometries for our quantum mechanics‐based molecular dynamic/quenching (MD/Q) simulations.[11](#cphc201901225-bib-0011){ref-type="ref"} The most favorable structures from the MD/Q simulations are shown in Figure [4](#cphc201901225-fig-0004){ref-type="fig"}, and the computed binding 'free' energies (Δ*G*′) are summarized in Table [1](#cphc201901225-tbl-0001){ref-type="table"}. Note that **B21** is almost fully immersed inside γ‐CD (Figure [4](#cphc201901225-fig-0004){ref-type="fig"}A), in line with the observed 1 : 1 complexation pattern. The Δ*G*′ values were decomposed into three terms, i. e., the gas‐phase interaction energy (Δ*E*), the change of solvation free energy upon complex formation (ΔΔ*G* ~solv~), and the change of the conformational 'free' energy upon complex formation (Δ*G*′~conf~), see Ref. \[27\] for details. {#cphc201901225-fig-0004} ###### Experimental association constant (*K* ~a~) of boron clusters with γ‐CD, polarizability (*α* in Å^3^), experimental binding free energy (Δ*G* ^0^), computed binding 'free' energy (Δ*G* ^'^), interaction energy (Δ*E*), change of hydration free energy upon complex formation (ΔΔ*G* ^hydr^), hydration free energies of the clusters \[Δ*G* ^hydr^(cluster)) as well as of the γ‐CD ⋅ cluster complexes (Δ*G* ^hydr^(complex)\], and conformational distortion 'free' energy (ΔG'~conf~). All energies given in kcal mol^−1^. The non‐tabulated hydration free energy of γ‐CD amounts to −69.2 kcal mol^−1^. Cluster *K* ~a~ \[10^3^ M^−1^\] *α* ^\[a\]^ Δ*G* ^0^ Δ*G*′ Δ*E* ΔΔ*G* ^hydr^ Δ*G* ^hydr^(cluster) Δ*G* ^hydr^(complex) Δ*G*′~conf~ -------------------------- ------------------------- ------------- ---------- ------- ------- -------------- ---------------------- ---------------------- ------------- B~21~H~18~ ^−^ (**B21)** 1800±200 36.7 −8.5 −32.5 −42.7 9.2 −23.1 −83.1 1.0 *meta*‐COSAN^−^ 3000^\[b\]^ 39.3 −8.8 −28.9 −39.9 10.5 −24.6 −83.3 0.5 *ortho*‐COSAN^−^ 191^\[b\]^ 39.7 −7.2 −22.5 −40.1 17.2 −33.4 −85.4 0.3 B~12~H~11~NH~3~ ^−^ 1.7^\[c\]^ 24.5 −4.4 −9.5 −44.8 33.6 −63.7 −99.3 1.7 di‐anions                   ------------------ ------------ ------------- ------ ------- ------- ------ -------- -------- ----- B~12~H~12~ ^2−^ 2.0^\[c\]^ 25.7^\[c\]^ −4.5 −6.1 −75.6 67.0 −163.6 −165.8 2.6 B~12~Cl~12~ ^2−^ 17^\[c\]^ 46.7^\[c\]^ −5.8 −15.7 −58.9 41.9 −127.6 −154.9 1.3 B~12~Br~12~ ^2−^ 960^\[c\]^ 58.2^\[c\]^ −8.2 −27.1 −60.6 32.2 −116.8 −153.8 1.3 B~12~I~12~ ^2−^ 67^\[c\]^ 84.1^\[c\]^ −6.6 −15.1 −62.3 44.2 −127.3 −152.3 3.0 \[a\] Calculated in the gas phase by using the B3LYP/aug‐cc‐pvdz method in the Gaussian 09 software. \[b\] Taken from ref. \[10\]. \[c\] Taken from ref. \[11\]. Wiley‐VCH Verlag GmbH & Co. KGaA Expectedly, there is no absolute agreement between the experimental and calculated data (columns with Δ*G* ~0~ and Δ*G*′ in Table [1](#cphc201901225-tbl-0001){ref-type="table"}). Among other reasons, the COSMO continuum solvation model has not been optimized for macromolecular species with concave interiors. Secondly, although high‐energy cavity water contributes comparably less to the driving force of host‐guest complexation for large cavities (such as that of γ‐CD) than for smaller ones,[28](#cphc201901225-bib-0028){ref-type="ref"} an energetic offset will nevertheless result from this neglect. Third, the SMD continuum solvation model (employed for the free borate clusters) has not been optimized for large polarizable ions and does not incorporate boron‐specific coordinative bonding effects of the H~2~O−H−B type, which have been previously considered in the hydration of borate clusters.[29](#cphc201901225-bib-0029){ref-type="ref"} Nevertheless, regardless of the absolute variation, the Δ*G*′ values show a very good linear correlation with the experimentally determined binding free energies (Δ*G* ^0^ values, *R* ^2^=0.94, *n*=8, see Chart 1), which demonstrates that the computational model picks up essential contributions to the driving force of the host‐guest complexation process. Besides the correctly reproduced overall trend, the model predicts salient experimental details, namely (*i*) the strongest binding (most negative Δ*G*′ values) for **B21** and *meta*‐COSAN, (*ii*) the significantly stronger binding of *meta*‐COSAN than *ortho*‐COSAN,[10](#cphc201901225-bib-0010){ref-type="ref"} and (*iii*) the maximal binding for X=Br in the B~12~X~12~ ^2−^ series.  In searching (*i*) for the driving forces responsible for the high affinities and (*ii*) for the intermolecular interactions underlying the observed correlation, we could disregard three factors under discussion. First, *host* desolvation effects, related to emptying the cavity of γ‐CD upon guest binding, should be comparable along the series; they can be presumed to remain constant and cannot account for the large variations with cluster type. Second, the conformational free energies upon complex formation (Δ*G*′~conf~) are quite small (0.3 to 3.0 kcal mol^−1^) and cannot account for the much larger variations in Δ*G*′.[30](#cphc201901225-bib-0030){ref-type="ref"} Third, dihydrogen bonding does not dominate the host‐guest complexation process in water. The gas‐phase geometry optimizations of the host‐guest complexes do, indeed, predict the formation of numerous dihydrogen bonds between the protiated clusters and γ‐CD, in which the partially negative H atoms of the borate clusters interact with the partially positive H3 and H5 hydrogens of γ‐CD, see Figure [5](#cphc201901225-fig-0005){ref-type="fig"}.[31](#cphc201901225-bib-0031){ref-type="ref"} However, although such coordinative bonds may be very important in the gas‐phase[7](#cphc201901225-bib-0007){ref-type="ref"} and structure‐determining in the solid‐state,[32](#cphc201901225-bib-0032){ref-type="ref"}, [33](#cphc201901225-bib-0033){ref-type="ref"} they are not expected to present a sizable driving force in aqueous solution, because the same type of bonds will be formed with the protic O−H hydrogens of bulk water molecules, and these H~2~O−H−B bonds are likely stronger than those inside CDs. Indeed, despite the assigned dihydrogen bonds, **B21** and *meta*‐COSAN did not have highly negative Δ*E* values, which indicated that these dihydrogen bonds are not very strong, and are not the major driving force for complexation. It is worth noting that COSANs can exist as three rotamers (*cisoid*, *gauche*, and *transoid*) that differ in dipole moment, and among which the *cisoid* one is presumed to dominate in polar media.[10](#cphc201901225-bib-0010){ref-type="ref"}, [22](#cphc201901225-bib-0022){ref-type="ref"}, [34](#cphc201901225-bib-0034){ref-type="ref"}, [35](#cphc201901225-bib-0035){ref-type="ref"} {#cphc201901225-fig-0005} We therefore returned to the two more likely contributors, dispersion interactions, which should be governed by the polarizabilities of the clusters (see Table [1](#cphc201901225-tbl-0001){ref-type="table"}), and *guest* desolvation effects for the uncomplexed borate cluster. Desolvation effects should be absent in the gas phase, such that the Δ*E* values should mainly report on dispersions interactions as well as other bonding effects. Indeed, the equally large positional isomers *meta*‐COSAN and *ortho*‐COSAN do not differ significantly in polarizability (Table [1](#cphc201901225-tbl-0001){ref-type="table"}) and show the same Δ*E* values, within ±0.1 kcal mol^−1^. Moreover, the trend in Δ*E* values for B~12~I~12~ ^2−^\>B~12~Br~12~ ^2−^\>B~12~Cl~12~ ^2−^ coincides with the polarizability trend of these globular halogenated clusters (Table [1](#cphc201901225-tbl-0001){ref-type="table"}).[11](#cphc201901225-bib-0011){ref-type="ref"}, [19](#cphc201901225-bib-0019){ref-type="ref"} The Δ*E* values for B~12~H~12~ ^2−^ is very high (negative) despite its lower polarizability, which can be rationalized by the electrostatically driven dihydrogen bonding interactions in the gas phase; they do not contribute to the driving force in solution, because the same or stronger interactions apply in aqueous bulk (see above). However, although contributions arising from dispersion interactions must contribute, they do not overall correlate well with the experimental binding energies, such that our final analysis narrowed down on differential dehydration as main dominator to the driving force of host‐guest complexation, an effect which was previously postulated[11](#cphc201901225-bib-0011){ref-type="ref"}, [19](#cphc201901225-bib-0019){ref-type="ref"} but which has never been quantitatively inspected for large cluster anions up to now. Towards this end, we further dissected the calculated differential desolvation term into the hydration free energies of the clusters and those of the γ‐CD ⋅ cluster complexes. For the associated discussion, we furthermore focused on two subsets, for which the hydration free energies of the γ‐CD ⋅ cluster complexes were very similar: the *meta*‐COSAN/**B21**/*ortho*‐COSAN mono‐anion subset and the B~12~Cl~12~ ^2−^, B~12~Br~12~ ^2−^, and B~12~I~12~ ^2−^ di‐anion subset. The reason why the hydration free energies are similar within the two subsets is due to the fact that the encapsulated clusters are shielded from bulk water, although there remains a sizable difference between the mono‐anion and di‐anions subsets as a consequence of the incomplete charge screening by the macrocycle. Within these subsets, the differential desolvation energies can be directly attributed to variations of the clusters themselves. For the mono‐anion subset with *meta*‐COSAN, **B21**, and *ortho*‐COSAN, where Δ*G* ^hydr^(complex) is similar, within −83 to −85 kcal mol^−1^, the Δ*G* ^hydr^(cluster) values show a much larger spread from −23 to −33 kcal mol^−1^. Within this subset, the lower (less negative) calculated (and experimental) binding free energy of *ortho*‐COSAN can be traced back to its much larger (more negative) hydration free energy. *Ortho*‐COSAN (in its *cisoid* conformation) has a more than twice larger value of the dipole moment than *meta*‐COSAN (directions of these vectors are identical).^10^ *Ortho*‐COSAN is therefore better electrostatically hydrated than *meta*‐COSAN and the hydration shell is more difficult to strip off in the binding process, which leads to its weaker binding to γ‐CD. Since the dipole moment of the *D* ~3h~‐symmetrical **B21** is zero on account of symmetry, we employed also the molecular electrostatic potential surfaces (MEPs) of **B21** and COSANs to pin‐point the source of the low desolvation of **B21**. Indeed, **B21** and *meta*‐COSAN have a lower magnitude of the MEP than *ortho*‐COSAN (−74.8, −79.1 and −87.7 kcal mol^−1^, respectively, see also Table S1). The di‐anion subset includes B~12~I~12~ ^2−^, B~12~Br~12~ ^2−^, and B~12~Cl~12~ ^2−^, where Δ*G* ^hydr^(complex) is much larger but also similar, within −152 to −155 kcal mol^−1^, but where Δ*G* ^hydr^(cluster) shows a much more pronounced spread from −117 to −128 kcal mol^−1^. In this set of three, the brominated cluster B~12~Br~12~ ^2−^ stands out experimentally and theoretically, because it has the highest affinity but the lowest (least negative) hydration free energy, which -- in relative terms -- facilitates its binding to γ‐CD. The irregularity in the calculated hydration free energies (maximum for B~12~Br~12~ ^2−^) is likely due to the fact that the hydration of the largest cluster, B~12~I~12~ ^2−^, is disfavored due to its lower charge density but favored due to its very high polarizability. Note, in this context, that the polarizability of any hydrated species is an important determinant of its solubility, which accounts, for example, for the larger aqueous solubility of xenon than neon.[36](#cphc201901225-bib-0036){ref-type="ref"} To sum up, we reported an exceptional thermodynamic signature for the formation of the host‐guest inclusion complex between *closo,closo*‐\[B~21~H~18~\]^−^ (**B21**) and γ‐CD; the corresponding affinity reaches a micromolar value, 1.8×10^6^ M^−1^. Although the original quantum‐chemical analysis indicated that *closo,closo*‐\[B~21~H~18~\]^−^ interacts with γ‐CD *via* formation of numerous dihydrogen bonds in the gas phase, a detailed analysis in comparison with other boron clusters revealed that, in addition to dispersion interactions, the desolvation of the anionic clusters governs the trend in affinities of the different boron clusters with γ‐CD in aqueous solution. Accordingly, the exceptional affinity of **B21** to γ‐CD is enabled by a very small desolvation penalty, thereby qualifying macropolyhedral boron hydrides as new and impressive examples of superchaotropic anions. Experimental Section {#cphc201901225-sec-0002} ==================== Experimental and computational details are given in the Supporting Information. Conflict of interest {#cphc201901225-sec-0004} ==================== The authors declare no conflict of interest. Supporting information ====================== As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors. ###### Supplementary ###### Click here for additional data file. K.I.A. and W.M.N. are grateful to the DFG for grant NA‐686/8 within the priority program SPP 1807 "Control of London Dispersion Interactions in Molecular Chemistry". J.H., J.F., and D.H. thank the Czech Science Foundation (grant number 17‐08045S) and M.I.F.P., M.C.L, and J.A.S.L. thank the regional government Xunta de Galicia for financial support (Project Grupo Potencial Crecemento ‐GPC‐ ED431B 2017/59). J.M.O.‐E. acknowledges financial support from the Spanish MICINN through project CTQ2018‐094644‐B‐C22.

{#sp1 .69}

1. Introduction {#sec1-materials-12-01741} =============== The prosperity of a human society is mainly dependent on the supply of sustainable energy. Because of inevitable and imminent shortages of fossil fuels together with global warming, we should look for affordable and clean energy resources. Among renewable energy resources, solar energy is considered a promising way to meet future energy demand. Since the dye-sensitized solar cell (DSSC) was invented in 1991 \[[@B1-materials-12-01741]\], numerous efforts on materials and device architecture have been made to achieve high performance and long-term stability \[[@B2-materials-12-01741],[@B3-materials-12-01741],[@B4-materials-12-01741]\]. In DSSCs, the chromophore responsible for capturing solar light and producing photocurrent is a critical component to enhance their power conversion efficiency (PCE). For example, Ru-complexes have attained remarkable efficiencies over 11% due to their full absorption window from the visible to the near-infrared (NIR) region \[[@B5-materials-12-01741],[@B6-materials-12-01741]\]. However, conventional Ru-complexes have significant drawbacks, such as scarcity, complicated synthesis, and toxicity \[[@B2-materials-12-01741]\]. Metal-free organic chromophores have been investigated as alternatives because of their many synthetic protocols, high molar extinction coefficients, and reduced toxicity issues \[[@B7-materials-12-01741],[@B8-materials-12-01741],[@B9-materials-12-01741]\]. Among these organic chromophores, a donor-π bridge-acceptor (D-π-A) architecture has been mostly adopted, and a structural modification can tune their photophysical properties to the architecture. Triarylamine-based moieties have been popularly selected as the donor unit because of their expedient synthesis and strong electron donating ability. Cyanoacrylic acid has been extensively used as the acceptor unit due to its direct synthesis from aldehydes and anchoring ability to TiO~2~. Methine or aromatic units have been introduced as π-spacers. However, the selection of an appropriate π-spacer is still of paramount importance in modulating the photophysical properties and enhancing the light-harvesting ability in the visible region. In particular, a planar π-spacer can increase electron delocalization and thus facilitate the electron transfer from the donor to acceptor for higher efficiency. Thiophene and its derivatives have been widely used as π-bridges because of their low steric hindrance between adjacent molecules and a strong conjugation effect. Also, alkyne units have been used to maintain the planar structure of the D-π-A architecture while increasing the distance between D and A moieties. Recently, ethynyl-thiophene and ethynyl-thienothiophene spacers have been explored theoretically and experimentally. Wang's group employed thienothiophene and bisthienothiophene linkers to achieve a high molar extinction coefficient of the metal-free organic sensitizers \[[@B10-materials-12-01741]\]. Recently, we also reported the facile synthesis of two organic dyes featuring ethynyl-thiophene from commercially available starting materials \[[@B11-materials-12-01741],[@B12-materials-12-01741]\]. Zhao et al. introduced ethynyl-thiophene or 2-ethynyl-6-methylthieno\[3,4-b\]thiophene to zinc-porphyrin dyes to enhance the light-harvesting capability and prevent the intermolecular aggregation \[[@B13-materials-12-01741]\]. Fernandes et al. synthesized three push-pull organic dyes featuring 2,2'-bithiophene, ethynyl-2,2'-bithiophene or ethynyl-thieno\[3,2-b\]thiophene spacers \[[@B14-materials-12-01741]\]. In this study, we report two organic dyes, coded as **CSD-03** and **CSD-04** ([Scheme 1](#materials-12-01741-sch001){ref-type="scheme"}), which can be synthesized in three steps from commercially available thienothiophene derivatives. The derivative has an n-hexyl side chain in order to improve the solubility of the final compounds. Importantly, the acetylene linker has been included to promote the planarity between the D and A units and thus increase electron delocalization. We also investigate the effect of the number of thienothiophene units on the optical, redox, and photovoltaic properties. 2. Results and Discussion {#sec2-materials-12-01741} ========================= 2.1. Synthesis {#sec2dot1-materials-12-01741} -------------- The two compounds, **CSD-03** and **CSD-04**, were directly synthesized in a three-step protocol ([Scheme 1](#materials-12-01741-sch001){ref-type="scheme"}). The Vilsmeier--Hack reaction of commercially available thienothiophene derivatives yielded compounds **1** and **2** as the aldehyde. The Sonogashira coupling reactions of compound **1** and **2** with 4-ethynyl-N,N-dimethylbenzenamine provided compounds **3** and **4**, respectively. Subsequently, compounds **3** and **4** were condensed with cyanoacetic acid to obtain **CSD-03** and **CSD-04**, respectively, under Knoevenagel condensation conditions. The two final dyes were fully characterized by ^1^H nuclear magnetic resonance (NMR) spectroscopy, ^13^C NMR spectroscopy, heteronuclear single quantum coherence (HSQC) spectroscopy, Fourier-transform infrared (FTIR) spectroscopy, and high-resolution mass spectrometry (HR-MS) in [Figures S1 and S2](#app1-materials-12-01741){ref-type="app"}. 2.2. Characterization {#sec2dot2-materials-12-01741} --------------------- The UV-visible spectra of the two compounds in dimethylformamide (DMF) and dye-grafted mesoporous TiO~2~ films are shown in [Figure 1](#materials-12-01741-f001){ref-type="fig"}a,b, respectively. In [Figure 1](#materials-12-01741-f001){ref-type="fig"}a, two broad absorption bands of each chromophore were observed between 300 nm and 600 nm. The absorption band at lower wavelengths is attributed to localized aromatic π--π\* transition, while that at higher wavelengths is ascribed to the intramolecular charge transfer (ICT) from donor to acceptor. The addition of another thienothiophene unit resulted in red-shift in the maximum absorption wavelength (λ~max~) from 419 nm to 455 nm, presumably a result of elongating π-conjugation. The molar extinction coefficients of **CSD-03** and **CSD-04** in DMF were 46,700 M^−1^ cm^−1^ and 50,400 M^−1^ cm^−1^, respectively. Their coefficients are higher than those of the ruthenium-complex dyes N3 (13,900 M^−1^ cm^−1^) \[[@B15-materials-12-01741]\] and N719 (14,000 M^−1^ cm^−1^) \[[@B16-materials-12-01741]\]. The optically determined bandgaps were 2.48 eV and 2.25 eV for **CSD-03** and **CSD-04**, respectively. The incorporation of n-hexyl to thienothiophene made the absorption spectra red-shifted and broadened when compared to dye **6** in Reference \[[@B14-materials-12-01741]\] because of the weaker intermolecular π--π stacking \[[@B17-materials-12-01741]\]. Compared with the spectra in DMF, the ICT absorption band of the two dyes on TiO~2~ films exhibits significant spectral broadening because of J-aggregation of chromophore molecules and the interaction of the anchoring moieties with the TiO~2~ surface ([Figure 1](#materials-12-01741-f001){ref-type="fig"}b). The electrochemical behavior of the two dyes was investigated by square wave voltammetry (SWV), as shown in [Figure 2](#materials-12-01741-f002){ref-type="fig"}. Both dyes exhibited a single oxidation wave and a single reduction wave. The addition of the second thienothiophene unit results in more positive reduction potential, ensuring a narrow fundamental energy gap, which correlates to the narrowing of the optical bandgap. According to the energy level of *Fc*/*Fc^+^* redox couple (−4.8 eV under vacuum level \[[@B18-materials-12-01741]\]), the estimated ionization potentials (IPs) and electron affinities (EAs) are summarized in [Table 1](#materials-12-01741-t001){ref-type="table"}. The IPs are slightly lower than the energy level of the *I*^−^/*I*~3~^−^ redox couple (−4.92 eV), confirming that the regeneration of the two dyes is energetically favorable. The EAs are almost 1 eV above the TiO~2~ conduction band (-4.26 eV), indicating efficient electron transfer from the lowest unoccupied molecular orbital (LUMO) of the organic dye to the conduction band of nanocrystalline TiO~2~. 2.3. Theoretical Calculations {#sec2dot3-materials-12-01741} ----------------------------- We performed computational calculations based on density functional theory (DFT) to recognize electron distribution in the frontier molecular orbitals. [Figure 3](#materials-12-01741-f003){ref-type="fig"}a shows the ground-state optimized geometries of the **CSD-03** and **CSD-04** chromophores. They are fully conjugated throughout the D, π-spacer, and A moieties. The ethynyl-thienothiophene linker allows for maintaining a coplanar molecular structure, which can facilitate the ICT and suppress rotational disorder, even if the n-hexyl chain is on the periphery of the thienothiophene. Interestingly, the ground-state structure of **CSD-04** has a twist angle of 13.4° between two thienothiophene moieties, and thus **CSD-03** has a more planar geometry due to small torsion angles. In [Figure 3](#materials-12-01741-f003){ref-type="fig"}b, the electrons are distributed from the donor to the π-bridge for both chromophores at the highest occupied molecular orbital (HOMO) level, while the excited electrons are localized on the cyanoacrylic acid at the LUMO level. It is beneficial to the efficient charge transfer from the dye to TiO~2~ upon irradiation. The addition of another thienothiophene unit drives up both HOMO and LUMO levels, in particular, due to a stronger influence on the HOMO. It should be noted that the larger difference between *I*~3~^−^/*I*^−^ potential and the HOMO level allows more efficient dye regeneration because of preventing the geminate recombination between oxidized chromophore molecules and existing electrons in the TiO~2~ under solar illumination. 2.4. Dye-Sensitized Solar Cells {#sec2dot4-materials-12-01741} ------------------------------- [Figure 4](#materials-12-01741-f004){ref-type="fig"}a displays J-V characteristics of translucent DSSCs (5 μm-thick TiO~2~) sensitized with **CSD-03** and **CSD-04** without any co-adsorbent. The photovoltaic parameters are summarized in [Table 2](#materials-12-01741-t002){ref-type="table"}. Adding the second thienothiophene unit resulted in higher *J~sc~* (from 12.55 mA/cm^2^ to 14.85 mA/cm^2^) due to stronger ICT absorption and higher extinction coefficient in the visible region but lower *V~oc~* (from 0.627 V to 0.587 V) due to the up-shift of the HOMO energy level. The photovoltaic properties depending on dye-soaking time were also measured and summarized in [Table 2](#materials-12-01741-t002){ref-type="table"}. Interestingly, when compared to **N719**, extended dye-soaking time of **CSD-03** and **CSD-04** resulted in lowering DSSC performance because of the faster aggregation of organic dyes on the TiO~2~ surface. It should be noted that the maximum efficiencies of both **CSD-03** (5.46%) and **CSD-04** (5.20%) are close to that of **N719** (5.92%) in the translucent DSSCs. [Figure 4](#materials-12-01741-f004){ref-type="fig"}b exhibits incident photon-to-current efficiency (IPCE) spectra of the semitransparent DSSCs. **CSD-04** provides broader IPCE spectrum than **CSD-03**, and thus this agrees with the enhancement of *J~sc~*. However, because of lower *V~oc~* and fill factor (FF), **CSD-04** unexpectedly exhibited lower power-conversion-efficiency (***η***) than CSD-03. FF is mainly affected by many factors, such as the internal resistance in the DSSC relating on the charge transportation at the electrolyte/chromophore/TiO~2~ interface, the charge transfer process at Pt counter electrode and diffusion in the electrolyte and the conductance of transparent electrodes \[[@B19-materials-12-01741]\]. Also, the dye-containing n-hexyl side chain had better DSSC performance, and IPCE than its counterpart of the nonalkyl chain (dye **6** in Reference \[[@B14-materials-12-01741]\]) because of its enhanced solubility for dye loading and suppressed π--π stacking. Electrochemical impedance spectroscopy (EIS) can offer a deeper understanding of interfacial charge-transfer processes \[[@B20-materials-12-01741]\]. [Figure 4](#materials-12-01741-f004){ref-type="fig"}c shows the Nyquist plots of the DSSCs in the dark. The EIS parameters fitted by an equivalent circuit are summarized in [Table 3](#materials-12-01741-t003){ref-type="table"}. The radius of the semicircle in the middle frequency region increases in the order of **CSD-03** (84.46 Ω) \< **CSD-04** (111.20 Ω), yielding recombination resistance (*R~rec~*). Large *R~rec~* helps the charge recombination at the electrolyte/chromophore/TiO~2~ interface to be hampered. Unfortunately, **CSD-04** had higher transport resistance (*R~tr~*), shorter electron lifetime (*τ~e~*), and lower charge collection efficiency (*η~cc~*) than **CSD-03** and consequently provided the lower power-conversion-efficiency (***η***) of the DSSC. 3. Materials and Methods {#sec3-materials-12-01741} ======================== 3.1. General {#sec3dot1-materials-12-01741} ------------ ^1^H and ^13^C NMR spectra were recorded on a Bruker 500 MHz instrument (Billerica, MA, USA) with tetramethylsilane (TMS) as an internal standard. MS data were obtained using either a JEOL-700 MStation (EI, Akishima, Tokyo, Japan) or a Thermo Scientific LTQ Orbitrap XL (Waltham, MA, USA) with negative ion mode nanoelectrospray ionization (nESI^−^). Infrared (IR) spectra were recorded on a Perkin-Elmer FTIR (Waltham, MA, USA). UV-visible spectra were obtained using a JASCO UV/Vis/NIR spectrophotometer (Tokyo, Japan). Optical bandgaps (*E~g~*) were determined using the edge of the longest wavelength absorption (λ) using *E~g~* = 1240/λ. A three-electrode system consisting of a Pt working electrode, a Pt wire counter electrode, and an Ag wire pseudoreference electrode was used for cyclic voltammetry measurements using a CHI 440A electrochemical analyzer (Austin, TX, USA). The redox potential of the dyes was measured in DMF containing 0.1 M tetrabutylammonium hexafluorophosphate, and the potential was calibrated against the ferrocene/ferrocenium (*Fc*/*Fc*^+^). Before recording the electrochemical data, the solutions were carefully purged with nitrogen gas. 3.2. Syntheses {#sec3dot2-materials-12-01741} -------------- *Synthesis of 5-bromo-6-hexylthieno\[3,2-b\]thiophene-2-carbaldehyde (***1***).* 2-bromo-3-hexylthieno \[3,2-b\]thiophene (1.35 g, 4.44 mmol) and phosphorus oxychloride (2.04 g, 13.4 mmol) were dissolved in DMF (16 mL). The solution was stirred for 30 min at 85 °C and then cooled to room temperature, quenched with H~2~O (20 mL). Dichloromethane (DCM) (200 mL) and H~2~O (200 mL) were added, and the organic layers were separated, dried over MgSO~4~ and filtered. The solvent was evaporated, and the residue was purified by silica gel column chromatography (Hex/EA = 20:1) to give **1** as a dark red oil (0.87 g, 59%). ^1^HNMR (400 MHz, CDCl~3~) δ 0.88 (t, J = 6.78 Hz, 3 H) 1.23--1.45 (m, 4 H) 1.48--1.59 (m, 2 H) 1.71--1.83 (m, 2 H) 3.32 (t, J = 7.53 Hz, 2 H) 8.04 (s, 1 H) 9.56 (s, 1 H) ppm; MS m/z(EI^+^) calculated for C~13~H~15~^81^BrOS~2~ 331.97 found 332.00. *Synthesis of 5-bromo-6,6′-dihexyl-2,2′-bithieno\[3,2-b\]thiophene-5′-carbaldehyde (***2***)*. 5-bromo-6,6′- dihexyl-2,2′-bithieno\[3,2-b\]thiophene (4.24 g, 8.07 mmol) and phosphorus oxychloride (24.7 g, 161 mmol) were dissolved in DMF (40 mL). The solution was stirred for 20 h at 95 °C and then cooled to room temperature, quenched with H~2~O (20 mL). Dichloromethane (MC) (500 mL) and H~2~O (400 mL) were added. The organic layer was separated, dried over MgSO~4~ and filtered. The solvent was evaporated and the residue was purified by silica gel column chromatography (Hex/EA = 10:1) to give **2** as a yellow solid (0.8 g, 18%). ^1^H (400 MHz, CDCl~3~) δ 10.05, 7.32, 7.30, 3.06 (t), 2.75--2.67 (m), 1.79 (q), 1.74--1.66 (m), 1.41--1.29 (m), 0.93--0.86 (m) ppm; MS m/z(EI^+^) calculated for C~25~H~29~^81^BrOS~4~ 554.03 found 554.10. *Synthesis of 5-((4-(dimethylamino)phenyl)ethynyl)-6-hexylthieno\[3,2-b\]thiophene-2-carbaldehyde (***3***).* A stirred mixture of 4-ethynyl-N,N-dimethylbenzenamine(0.46 g, 3.2 mmol), **1** (1.00 g, 3.02 mmol), copper(I) iodide (0.012 g, 0.060 mmol), Triphenylphosphine (0.01 6g, 0.060 mmol) and bis(triphenylphosphine)palladium(II) dichloride (0.11 g, 0.039 mmol) in trimethylamine (30 mL) was refluxed for 3 h. After cooling the solution, MC (200 mL) and H~2~O (200 mL) were added. The organic layer was separated, dried over MgSO~4~ and filtered. The solvent was evaporated and the residue was purified by silica gel column chromatography (Hex/EA = 7:1) to give **3** as an orange solid (1.1 g, 92%). ^1^H NMR (400 MHz, CDCl~3~) δ 9.93(s,1H), 7.81(s,1H),6.67(d, J = 8.9 Hz, 2H), 6.61 (d, J = 8.9 Hz, 2H), 3.01 (s, 6H), 2.91 (t, 2H), 1.79 (p, J = 7.6 Hz, 2H), 1.46 -- 1.26 (m, 7H), 0.89 (t, J = 7.1 Hz, 3H) ppm; MS m/z(EI^+^) calculated for C~23~H~25~NOS~2~ 395.14 found 394.70. *Synthesis of 5′-((4-(dimethylamino)phenyl)ethynyl)-6,6′-dihexyl-\[2,2′-bithieno\[3,2-b\]thiophene\]-5-carbaldehyde (***4***).* A stirred mixture of 4-ethynyl-N,N-dimethylbenzenamine (0.154 g, 1.06 mmol), **3** (0.56 g, 1.01 mmol), copper(I) iodide (7.7 mg, 0.041 mmol), triphenylphosphine (5.3 mg, 0.020 mmol), bis(triphenylphosphine)palladium(II) dichloride (35.5 mg, 0.0506 mmol) in triethylamine (30 mL) was refluxed for 6 h. After cooling the solution, MC (200 mL) and H~2~O (200 mL) were added. The organic layers were separated, dried over MgSO~4~ and filtered. The solvent was evaporated and the residue was purified by silica gel column chromatography (Hex/EA = 7:1) to give **4** as a yellow solid (0.31 g, 50%).^1^H NMR (400 MHz, DMSO-d~6~) δ 0.83 (t, J = 7.03 Hz, 3 H) 1.23--1.34 (m, 9 H) 2.90 (t, J = 7.28 Hz, 2 H) 2.98 (s, 6 H) 6.74 (m, J = 8.53 Hz, 2 H) 7.39 (m, J = 8.53 Hz, 2 H) 8.28 (s, 1 H) 8.57 (s, 1 H) ppm; MS m/z(EI^+^) calculated for C~35~H~39~NOS~4~ 617.96 found 617.50. *Synthesis of (E)-2-cyano-3-(5-((4-(dimethylamino)phenyl)ethynyl)-6-hexylthieno\[3,2-b\]thiophen-2-yl) acrylic acid (**CSD-03**).* A mixture of **2** (1.1 g, 2.8 mmol) and cyanoacetic acid (0.50 g, 5.6 mmol) was dissolved in CH~3~CN (50 mL) containing piperidine (1.42 g, 16.7 mmol). The solution was refluxed for 3 h. After cooling the solution, it was poured into HCl (2 M, 200 mL) and MC (500 mL). The organic layer was separated, washed with water, dried over MgSO~4~, and filtered. Then the solvent was removed in vacuo to afford **CSD-03** as a dark red solid (0.3 g, 23%). ^1^H NMR (500 MHz, DMSO-d~6~) δ 8.47 (s, 1H), 8.20 (s, 1H), 7.39--7.37 (m, 2H), 6.74--6.72 (m, 2H), 2.97 (s, 6H), 2.89 (t, J = 7.4 Hz, 2H), 1.76--1.70 (m, 2H), 1.34 -- 1.22 (m, 6H), 0.83 (t, J = 7.2 Hz, 3H) ppm.; ^13^C(126 MHz, DMSO-d~6~) δ 163.8, 151.0, 145.9, 138.7, 138.0, 137,2, 132.9, 131.8, 117.6, 112.3, 107.6, 101.7, 100.0, 81.0, 40.1, 31.2, 28.6, 28.5, 28.2, 22.4, 14.4 ppm; FT-IR (ATR) υ 2920, 2852, 2181, 1672, 1603, 1568, 1533, 1470, 1418, 1368, 13.12, 1244, 1192, 1148, 1125, 945, 832, 812, 758, 745, 608, 584 cm^−1^; HRMS m/z (FAB^+^) calculated for C~26~H~26~N~2~O~2~S~2~ \[M\]^+^ 462.1436, found 462.1439. *Synthesis of (E)-2-cyano-3-(5-((4-(dimethylamino)phenyl)ethynyl)-6,6′-dihexyl-2,2′-bithieno\[3,2-b\] thiophen-5′-yl)acrylic acid (**CSD-04**).* A mixture of **4** and cyanoacetic acid (0.0173 g, 0.204 mmol) was dissolved in CH~3~CN (40 mL) containing piperidine (0.052 g, 0.061 mmol). The solution was refluxed for 5 h. After cooling the solution, it was poured into HCl (2 M, 200 mL) and MC (500 mL). The organic layer was separated and washed with water, dried over MgSO~4~, and filtered. Then, the solvent was evaporated in vacuo to afford **CSD-04** as a red solid (0.056 g, 80%). ^1^H NMR (500 MHz, DMSO-d~6~) δ 8.22 (s, 1H), 7.79 (s, 1H), 7.76 (s, 1H), 7.45 (d, J = 9.0 Hz, 2H), 7.01 (s, 1H), 6.72 (d, J = 9.0 Hz, 2H), 2.96 (s, 6H), 2.89--2.86 (m, 4H), 1.69--1.63 (m, 12H), 0.86--0.83 (m, 6H) ppm.; ^13^C(151 MHz, DMSO-d~6~) δ 168.6, 150.4, 145.5, 141.8, 141.7, 140.6,139.5, 138.6, 138.3, 137.9, 136.9, 134.7, 133.0, 132.2, 125.6, 121.3, 119.2, 118.1, 117.0, 112.0, 104.6, 40.1, 31.0, 30.9, 29.1, 28.8, 28.4, 28.4, 27.9, 27.5, 22.0, 21.6, 14.0, 13.9, 13.9 ppm; FT-IR (ATR) υ 2911, 2854, 2202, 1763, 1603, 1521, 1458, 1427, 1363, 1312, 1283, 1204, 1182, 1105, 1057, 945, 810 cm^−1^; HRMS m/z (FAB^+^) calculated for C~38~H~41~N~2~O~2~S~4~ \[M+H\]^+^ 685.2051, found 685.0078. 3.3. Computational Details {#sec3dot3-materials-12-01741} -------------------------- Theoretical calculations were performed with the Gaussian 09 \[[@B21-materials-12-01741]\]. The ground-state geometries were fully optimized using the DFT method with Becke's three-parameter hybrid exchange functionals and the Lee-Yang-Parr correlation functional (B3LYP) and 6-311G (d, p) basis set for all atoms without additional diffused function. Vibrational frequencies were also calculated to confirm that the optimized geometry would correspond to the lowest point of the respective potential energy surface. 3.4. DSSC Fabrication and Photovoltaic Measurements {#sec3dot4-materials-12-01741} --------------------------------------------------- The acetylene-black TiO~2~ paste \[[@B22-materials-12-01741]\] was screen-printed onto fluorine-doped SnO~2~ (FTO) conducting glass (TEC 8, Pilkington, 2.2-mm-thick, sheet resistance = 8 Ω/sq). The printed films were placed in a muffle furnace and slowly heated to 300 °C over a 30 min period, heated at 300 °C for 1 h, heated to 575 °C for 1 h, and then cooled to room temperature after 3 h. The final thickness of the sintered films was about 5 μm, and their active area was 0.2025 cm^2^ (i.e., 0.45 cm×0.45 cm). The prepared TiO~2~ photoanodes were dipped in a 0.04 M TiCl~4~ solution at 75 °C for 30 min, rinsed several times with deionized water and ethanol, and then annealed at 500 °C for 30 min on a hot plate. The photoanodes were treated with the O~2~ plasma for 10 min, and then immersed in 0.1 M HNO~3~ solution for 30 min to facilitate dye adsorption. The final ones were submerged in a 0.5 mM dye-containing ethanol solution (**N719**, **CSD-03**, and **CSD-04**) for different soaking time. After two holes were drilled in the FTO-coated glass, chloroplatinic acid (H~2~PtCl~6~) solution in 2-propanol was spin-coated and annealed at 525 ºC for 1 h in the muffle furnace to prepare the Pt counter electrode. Both the dye-grafted photoanode and the counter electrode were assembled with 25-μm-thick Surlyn (Solaronix, Aubonne, Switzerland). A liquid electrolyte (Iodolyte AN-50, Solaronix) was injected through the holes in the backside of the counter electrode. The photovoltaic performance of the masked devices was investigated by using a solar cell I-V measurement system (K3000 LAB, McScience, Suwon, Korea) under the standard AM 1.5G spectrum (100 mW cm^−2^). Short-circuit photocurrent density (*J*~sc~), open-circuit voltage (*V*~oc~), fill factor (*FF*), and power conversion efficiency (*η*) were obtained simultaneously. The rectangular mask size (e.g., 0.6 cm × 0.6 cm) was somewhat larger than the active area of the photoanodes because this is the preferred configuration to capture both direct and diffuse lights and to minimize internal reflection \[[@B23-materials-12-01741]\]. The incident photon-to-current conversion efficiency (IPCE) was obtained using a spectra IPCE measurement system (K3100, McScience, Korea). EIS experiments were performed with cells biased to *V*~oc~ in the dark using a frequency response analyzer (Solartron 1260, AMETEK, Leicester, UK). A sinusoidal potential modulation of 10 mV was applied over a frequency range from 0.1 Hz to 100 kHz. The obtained EIS spectra were fitted using the equivalent circuit built in the ZView software (Version 3.5f, Scribner Associates Inc., Southern Pines, NC, USA). 4. Conclusions {#sec4-materials-12-01741} ============== In this article, we report the facile synthesis of the two organic dyes featuring an ethynyl--thienothiophene linker with an n-hexyl chain, which will add impetus to develop lower cost photosensitizers for the DSSCs. UV-vis spectroscopy and square wave voltammetry have shown that adding the second thienothiophene allowed for narrowing the bandgap of the molecule and ensuring better light harvesting in the DSSCs. Unexpectedly, the extended π-conjugation negatively affects the structural planarity and HOMO level of **CSD-04**. Therefore, the photovoltaic performance of **CSD-03** (5.45 ± 0.03%) is better than that of **CSD-04** (5.07 ± 0.18%). The following are available online at <https://www.mdpi.com/1996-1944/12/11/1741/s1>, Figure S1: (a) FTIR, (b) ^1^H NMR, (c) ^13^C NMR, and (d) HSQC spectra of CSD-03, Figure S2: (a) FTIR, (b) ^1^H NMR, (c) ^13^C NMR, and (d) HSQC spectra of CSD-04. ###### Click here for additional data file. J.H. conceived and designed the experiments; D.-S.L. synthesized the molecules; K.-W.P. conducted DFT/TDDFT calculations; A.A.W. performed electrochemical experiments and analyzed the data; D.-S.L. and K.-W.P. performed photovoltaic experiments. D.-S.L. and K.-W.P. prepared the manuscript; A.A.W. and J.H. revised the manuscript. This work was supported by the National Research Foundation of Korea (NRF) Grant funded by the Ministry of Science and ICT for First-Mover Program for Accelerating Disruptive Technology Development (NRF-2018M3C1B9088457) and the Chung-Ang University Research Grants in 2015. The authors declare no conflict of interest. Figures, Scheme and Tables ========================== {#materials-12-01741-sch001} {#materials-12-01741-f001} {#materials-12-01741-f002} {#materials-12-01741-f003} {#materials-12-01741-f004} materials-12-01741-t001_Table 1 ###### Summary of optical and electrochemical properties of the as-synthesized dyes. ----------------------------------------------------------------------------------------------------------------- Dyes *λ~max~*/*λ~onset~*\ *E~g~ (optical)*\ *E~ox~*\ *E~red~*\ *IP*\ *EA*\ *E~g~ (e-chem)*\ (nm) (eV) (V) (V) (eV) (eV) (eV) ------------ ---------------------- ------------------- ---------- ----------- ------- ------- ------------------ **CSD-03** 419/499 2.48 0.52 −1.35 −5.32 −3.45 1.87 **CSD-04** 455/550 2.25 0.53 −1.32 −5.34 −3.48 1.85 ----------------------------------------------------------------------------------------------------------------- materials-12-01741-t002_Table 2 ###### Masked DSSC (5 μm-thick TiO~2~) performance parameters of **CSD-03**, **CSD-04**, and **N719**. Dyes (Soaking Time) *V~oc~* (V) *J~sc~* (mA/cm^2^) *FF* *η* (%) --------------------- ---------------- -------------------- ------------ ------------- **CSD-03** (3 h) 0.627 ± 0.013 12.55 ± 0.18 69.4 ± 1.1 5.46 ± 0.03 **CSD-03** (6 h) 0.619 ± 0.022 12.64 ± 0.68 69.1 ± 0.2 5.41 ± 0.08 **CSD-03** (12 h) 0.6111 ± 0.004 11.15 ± 0.51 68.7 ± 0.3 4.69 ± 0.27 **CSD-04** (3 h) 0.587 ± 0.004 14.85 ± 0.29 58.2 ± 1.4 5.07 ± 0.18 **CSD-04** (6 h) 0.576 ± 0.009 13.86 ± 0.35 65.2 ± 1.0 5.20 ± 0.03 **CSD-04** (12 h) 0.587 ± 0.016 12.974 ± 0.42 64.1 ± 0.9 4.88 ± 0.01 **N719** (3 h) 0.608 ± 0.003 14.74 ± 0.47 62.1 ± 0.1 5.57 ± 0.21 **N719** (24 h) 0.671 ± 0.001 14.95 ± 0.39 59.1 ± 0.6 5.92 ± 0.01 materials-12-01741-t003_Table 3 ###### Electrochemical impedance spectroscopy (EIS) fitting parameters of the DSSCs (5 μm-thick TiO~2~). Dyes *R~tr~* (Ω) *R~rec~* (Ω) *C~μ~* (mF) *τ~e~* (ms) *η~cc~* (%) ------------------ ------------- -------------- ------------- ------------- ------------- **CSD-03** (3 h) 5.68 84.46 0.08 6.34 93.70 **CSD-04** (3 h) 11.99 111.20 0.05 5.65 90.27 [^1]: These authors contributed equally to this work.

INTRODUCTION {#sec1-ZoolRes-40-3-219} ============ Major depressive disorder (MDD) is a debilitating psychiatric mood disorder that affects millions of individuals globally ([@B8-ZoolRes-40-3-219]). Our understanding of the biological basis of MDD is poor, and current treatments are ineffective in a significant proportion of cases. This likely relates to the lack of human and non-human primate research models compared with the dominant rodent models of depression, which possess translational limitations due to limited homologies with humans. Therefore, a more homologous primate model of depression is needed to advance our understanding of the pathophysiological mechanisms underlying depression and to provide a sound basis for conducting pre-clinical therapeutic trials. Social stress plays a major role in the pathogenesis of depression ([@B17-ZoolRes-40-3-219]). In human research, depressive patients, especially women, are more likely to experience depression after prolonged stress ([@B26-ZoolRes-40-3-219]). The diagnosis of depression in humans is based on various scales. In line with the DSM-V (Diagnostic and Statistical Manual of Mental Disorders, fifth edition) diagnostic criteria, MDD is characterized by five (or more) of the following symptoms: depressed mood, loss of interest, weight change, sleep disturbance, psychomotor agitation or retardation, loss of energy, feelings of worthlessness, difficulty concentrating, and recurrent thoughts of death. These symptoms must persist for at least two weeks and must include depressed mood and/or loss of interest ([@B9-ZoolRes-40-3-219]). In determination of primate depressive behavior, the most reliable method is through behavioral phenotypes. Because of the inability to transfer pressure after being the subject of aggression, depressed female cynomolgus monkeys (*Macaca fascicularis*) can experience long-term social pressure. [@B27-ZoolRes-40-3-219] indicated that socially-subordinate female cynomolgus monkeys, who are likely weak competitors in social environments due to long-term attack and suppression, can exhibit similar pathogenetic processes as depression in humans. Female cynomolgus monkeys have several behavioral and physiological characteristics in common with human depression. They tend to spend more time alone than their dominant counterparts, exhibit greater vigilance, display slumped or collapsed body posture, show diminished interest in feeding and sex, and subdued communication and reciprocal grooming with others for at least two weeks ([@B27-ZoolRes-40-3-219]). Huddling, which is defined as a slumped body posture with the head at or below the shoulders during the awake state (i.e., when the monkey's eyes are opened) accompanied by a relative lack in responsiveness to environmental stimuli, is used as a behavioral indicator of depression and is a core posture reflecting depressed mood in monkeys ([@B7-ZoolRes-40-3-219]). Based on these well-established criteria, we successfully constructed a naturally occurring depression model in macaques ([@B32-ZoolRes-40-3-219]). Because of the similarities in complex behavioral and psychological processes between macaques and humans, the development of naturally occurring animal models of disease, whether physical or psychological, is a valuable approach for translational research between human studies and induced primate models, especially for depression ([@B2-ZoolRes-40-3-219]). The ratio of the index to ring finger length is a commonly used measure. The male ring finger is longer than the index finger, whereas the index finger of the female is basically equal to that of the ring finger. The second-to-fourth (2D:4D) digit ratio was first proposed by [@B21-ZoolRes-40-3-219] to predict estrogen and sperm number and prenatal sex hormones. Under regulation of the Hox gene, hormone levels during embryonic development, especially the level of androgen, affect finger development ([@B10-ZoolRes-40-3-219]). The 2D:4D ratio is present in a human embryo by the seventh week of pregnancy ([@B19-ZoolRes-40-3-219]). Additionally, evidence suggests that in both males and females, the digit ratio can act as an indicator of the level of testosterone the developing fetus was exposed to, making it a useful indirect measure of organizational prenatal hormone exposure. Some research indicates that the 2D:4D ratio may also be an indicator of perinatal androgen action, whereby lower digit ratios predict greater androgen sensitivity ([@B28-ZoolRes-40-3-219]). Thus, given that depression is a strongly sexually dimorphic trait, it is reasonable to expect that the 2D:4D ratio may be related to depression ([@B6-ZoolRes-40-3-219]). Furthermore, previous investigations have confirmed that the 2D:4D ratio is associated with human behavior, such as aggression, cooperation, left-handedness, and human disease, including breast cancer, dyslexia, infertility, myocardial infarction, and autism ([@B25-ZoolRes-40-3-219]). For instance, [@B1-ZoolRes-40-3-219] reported that the 2D:4D ratio is related to individual aggressiveness, confidence, and competitive ability. Furthermore, other developmental disorders, emotional behaviors, and negative and affective symptoms in schizophrenia are reported to be related to the 2D:4D ratio ([@B3-ZoolRes-40-3-219]). Previous studies have also shown that individuals with a higher 2D:4D ratio are more likely to suffer from depression. For instance, [@B28-ZoolRes-40-3-219] found that a higher digit ratio is correlated with higher depression scores in females, but not males. To date, however, results have been inconsistent. For example, using a large sample comprised of 298 college students (149 males and 149 females), [@B1-ZoolRes-40-3-219] found that more feminine ratios were associated with higher depression in men, but found no correlation between the 2D:4D ratio and depression in women. [@B1-ZoolRes-40-3-219] produced the unusual finding of no sex differences in depression, although digit ratios did differ in the expected direction. As such, it remains unclear to what degree depression and the 2D:4D ratio, both characterized by marked sex differences, are related ([@B28-ZoolRes-40-3-219]). Females are more susceptible to depression in social groups and depression is approximately twice more common among women than men ([@B29-ZoolRes-40-3-219]). Therefore, we chose female cynomolgus monkeys as our research targets. Based on our previous method ([@B32-ZoolRes-40-3-219]), a total of 15 depressed female monkeys were selected across 52 enclosures, with 16 healthy subjects selected as controls from the original population (*n*=6 012). A strict radiographic procedure was used to measure index and ring finger lengths in both forelimb hands. This investigation aimed to provide evidence of a relationship between the digit ratio and depression status in the cynomolgus monkey. MATERIALS AND METHODS {#sec2-ZoolRes-40-3-219} ===================== Ethics statement {#sec2dot1-ZoolRes-40-3-219} ---------------- Behavioral data acquisition was observational under normal circumstances and did not involve physical manipulation of the subjects or changes to their environment or diet. Animal care and housing procedures followed Chinese regulatory requirements and the Association for Assessment and Accreditation of Laboratory Animal Care International. In brief, complete animal husbandry and veterinary care were provided daily. Animals were fed a nutritious standardized diet, supplemented daily with fresh fruits and vegetables. Animals had unrestricted access to potable water and their enclosures were cleaned each day. Animals were observed daily by trained care-takers. Any observed abnormality, disease, or injury was reported to the veterinary staff for diagnosis and treatment; this veterinary support was documented in both hard copy and electronic formats. In addition, this study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the Institute of Neuroscience of Chongqing Medical University (Approval No.: 20100031). Prior to implementation, the experimental protocol was approved by the Committee on the Ethics of Animal Experiments at Chongqing Medical University and was in accordance with state regulations. Observation site {#sec2dot2-ZoolRes-40-3-219} ---------------- The *M. fascicularis* Feeding Base of Zhongke Experimental Animal Co., Ltd. is in Suzhou, China, at E31º07′03″ to E31º07′06″, N120º19′08″ to N120º19′15″. The company imported the *M. fascicularis* subjects from Guangdong Province, China and from Vietnam in 1990, from which they established a domestication and breeding base for these monkeys. Subjects {#sec2dot3-ZoolRes-40-3-219} -------- We scanned a total population of 6 012 adult female cynomolgus monkeys across all 52 enclosures. Depressive behavior was identified using the operational definition according to Shively's criteria: slumped or collapsed body posture ([Figure 1](#ZoolRes-40-3-219-f001){ref-type="fig"}), diminished interest in feeding and sex, and diminished communication and reciprocal grooming with others ([@B32-ZoolRes-40-3-219]). Sixteen healthy adult female *M. fascicularis* subjects (aged 9--13 years) were randomly selected from the pool of 6 012 monkeys. A total of 15 depressed female monkeys (aged 10--12 years) were selected from the 52 enclosures based on the above-mentioned depression phenotypes lasting for at least two weeks. All subjects were reared in socially-stable colonies with negligible rates of conflict ([@B30-ZoolRes-40-3-219]). Staff veterinarians ruled out disease in the subjects. Each colony was housed in an indoor free enclosure measuring 8.0 m×3.0 m×3.0 m (L×W×H) with continuous daylight exposure. Every colony was composed of two males, 16--22 adult females, and their offspring of less than six months of age. To reflect wild populations, the male:female ratio was maintained at 1:(7--10). {#ZoolRes-40-3-219-f001} Behavioral recording methods and scored behavioral items are described in our previous work ([@B33-ZoolRes-40-3-219]). The duration and frequency of the core depressive behavior "Huddling" ([Figure 1](#ZoolRes-40-3-219-f001){ref-type="fig"}) and 58 other behaviors in 12 categories (Ingestive, Thermoregulatory, Rutting and estrous, Mating, Resting, Parental, Amicable, Conflict, Vigilance, Communication, Locomotive, Miscellaneous behaviors) were video recorded by three well-trained observers blind to the behavioral definition using NOLDUS Observer XT software (v10.0, Noldus Information Technology, Leesburg, PA, USA) during two consecutive weeks with four phases per day (A1 0900--0930 h; A2 0930--1000 h; P1 1500--1530 h; P2 1530--1600 h). Behavioral data were coded as duration (in seconds) and frequency (in count) for each discreet behavioral item per 30 min observational phase and presented as means±*SD*. Finger length measurement with radiography {#sec2dot4-ZoolRes-40-3-219} ------------------------------------------ We measured index and ring finger lengths from the right and left forelimb hands in both groups after the last day of behavioral data acquisition using a digital radiography unit with a flat-panel digital detector (PLX8200, Perlove, Nanjing, China) ([@B16-ZoolRes-40-3-219]). The digital detector was exposed to X-ray at 60 kVp, with an approximate detector-to-tube distance of 1 m. Exposure times were no greater than 0.1 s, resulting in 4.0 mA exposure. Three qualified staff performed this process in cooperation: one undertook anesthesia and hand-position adjustment, one operated the machine to acquire finger length, and one recorded the data and was blind to the experiment. Ketamine (10 mL/kg) anesthetic was administered intramuscularly (I.M.) in the distal hind limb at 0900 h ([@B23-ZoolRes-40-3-219]). Approximately 10 min after ketamine injection, the finger length ratio was measured at an accuracy of 0.01 mm using X-ray imaging ([@B4-ZoolRes-40-3-219]) ([Figure 2](#ZoolRes-40-3-219-f002){ref-type="fig"}). {#ZoolRes-40-3-219-f002} Statistical methods {#sec2dot5-ZoolRes-40-3-219} ------------------- To assess the behavioral differences between depressed subjects and healthy controls, Student's *t*-test was performed if the data were normally distributed; otherwise, the Mann-Whitney *U* test was applied. As there were 58 behavioral items in total, Bonferroni correction was used to reduce type I errors for multiple comparisons. Primate fingers are termed 1 through 5 from the thumb to the smallest digit, respectively. In this investigation, the 2D:4D digit ratio was used ([@B18-ZoolRes-40-3-219]). To minimize error, all finger length data were presented as means±*SD*. Finger length and digit ratio comparisons between the two groups were conducted using Student's *t*-test. *P* values of less than 0.05 were deemed significant for all data. Data management and statistical analysis were performed using SPSS 21.0. RESULTS {#sec3-ZoolRes-40-3-219} ======= Differences in behavioral phenotype between depressed and control monkeys {#sec3dot1-ZoolRes-40-3-219} ------------------------------------------------------------------------- In terms of the duration of behavior, depressed monkeys spent more time "Huddling" (*P*\<0.001), in which the target displayed a fetal-like, self-enclosed posture with a lowered head during the awake state (i.e., when the monkey's eyes were opened). The increased duration of "Huddling" indicated that the depressed subjects recursively stayed in the depressed mood condition. For Ingestive behavior, the control subjects preferred to "feed while sitting" (*P*\<0.05), whereas the depressed subjects displayed more vigilance and preferred "feeding while hanging" to avoid potential threats and attacks. They also spent more time "licking food residue off the cage" (*P*\<0.05) due to the pressure of other monkeys during normal food intake. In accordance with the depression criteria, depressed monkeys also spent less time drinking. The impact of pressure on both groups was further reflected in resting behavior, with the depressed group more reluctant to "hang on iron chain rest" (*P*\<0.001) compared to the control group, but more willing to choose a remote area for resting, namely, "hanging on skylight rest" (*P*\<0.001). Regarding locomotive behavior, depressed individuals exhibited less vitality in walking and standing and spent relatively less time performing "quadrupedal walking on floor" (*P*\<0.001), "walking on iron chain" (*P*\<0.001), and "standing" (*P*\<0.05). Indicative of a friendly relationship among others, the control group received more amicable grooming (i.e., "being groomed" (*P*\<0.001)) and groomed others more often (i.e., "mutual grooming" (*P*\<0.001)) than exhibited by the depressed group, suggesting reduced interaction in depression. Furthermore, there was a reduction in the duration of "nursing" (*P*\<0.05) parental behavior. For mating behavior, which is a sign of sexual interest, the duration of "copulation" (*P*\<0.05) was higher in the control monkeys in comparison to the depressed females, with a significant difference in the frequency of mating behavior also observed ([Table 2](#ZoolRes-40-3-219-t002){ref-type="table"}). In the end, except the frequency of behaviors matched with the duration of behaviors, a lower frequency of miscellaneous behavior (i.e., "rub hand back and forth" (*P*\<0.001)) existed in depressed individuals, suggesting that depressed monkeys may be less imposing and lacking in confidence. ###### Duration of behaviors observed in depressed and control subjects Behavior Duration --------------------- ------------------------------------ ----------------- ------------------ ------------ -------------------- Behavioral category Behavior Depressed group Control subjects *P* value Adjusted *P* value Core behavior Huddling\*\*\* 365.44±514.14 159.00±355.08 *P*\<0.001 *P*\<0.001 Ingestive behavior Drinking\*\*\* 2.51±13.09 5.92±21.59 *P*\<0.001 *P*\<0.001 Feeding while hanging\* 6.16±33.19 3.56±24.14 *P*=0.0004 *P*=0.0243 Lick food residue off cage\* 1.29±13.78 0.44±3.24 *P*=0.0007 *P*=0.0419 Feeding while sitting\*\*\* 29.73±89.06 63.29±171.94 *P*\<0.001 *P*\<0.001 Amicable behavior Mutual grooming\*\*\* 51.96±153.61 114.13±226.53 *P*\<0.001 *P*\<0.001 Being groomed\*\*\* 51.75±120.17 97.94±198.63 *P*\<0.001 *P*\<0.001 Parental behavior Nursing\* 7.33±36.69 11.75±57.42 *P*=0.0006 *P*=0.0334 Locomotive behavior Quadrupedal walking on floor\*\*\* 62.82±58.99 82.03±75.68 *P*\<0.001 *P*\<0.001 Walking along iron chain\*\*\* 0.13±1.26 0.85±5.62 *P*\<0.001 *P*\<0.001 Standing\* 9.75±25.35 13.94±29.95 *P*=0.0006 *P*=0.0378 Resting behavior Hanging on iron chain rest\*\*\* 5.85±71.70 45.35±176.59 *P*\<0.001 *P*\<0.001 Hanging on skylight rest\*\*\* 23.18±143.93 8.88±81.59 *P*\<0.001 *P*\<0.001 Data are means±*SD*. Bonferroni correction was used to reduce type I errors for multiple comparisons. Mann-Whitney *U* test, *\**: *P*\<0.05; *\*\**: *P*\<0.01; *\*\*\**: *P*\<0.001*.* ###### Frequency of behaviors observed in depressed and control subjects Behavior Frequency ------------------------ ---------------------------------- ----------------- ------------------ ------------ -------------------- Behavioral category Behavior Depressed group Control subjects *P* value Adjusted *P* value Core behavior Huddling\*\*\* 1.53±2.17 0.51±1.25 *P*\<0.001 *P*\<0.001 Ingestive behavior Feeding while hanging\*\*\* 0.11±0.63 0.05±0.28 *P*\<0.001 *P*\<0.001 Feeding while sitting\*\* 0.85±2.15 1.11±2.47 *P*\<0.001 *P*\<0.01 Drinking\*\*\* 0.20±0.86 0.51±1.59 *P*\<0.001 *P*\<0.001 Amicable behavior Mutual grooming\*\*\* 0.74±1.83 1.33±2.40 *P*\<0.001 *P*\<0.001 Mating behavior Copulation\* 0.08±0.35 0.12±0.53 *P*=0.0007 *P*=0.0418 Parental behavior Nursing\* 0.28±1.01 0.40±1.44 *P*=0.0002 *P*=0.0129 Resting behavior Hanging on iron chain rest\*\*\* 0.02±0.22 0.23±0.80 *P*\<0.001 *P*\<0.001 Hanging on skylight rest\*\*\* 0.29±1.63 0.10±0.62 *P*\<0.001 *P*\<0.001 Locomotive behavior Walking along iron chain\*\*\* 0.03±0.21 0.10±0.53 *P*\<0.001 *P*\<0.001 Miscellaneous behavior Rub hand back and forth\*\*\* 0.02±0.15 0.06±0.43 *P*\<0.001 *P*\<0.001 Data are means±*SD*. Bonferroni correction was used to reduce type I errors for multiple comparisons. Mann-Whitney *U* test, *\**: *P*\<0.05; *\*\**: *P*\<0.01; *\*\*\**: *P*\<0.001*.* Finger length data {#sec3dot2-ZoolRes-40-3-219} ------------------ Finger length was measured in the 16 control and 15 depressed animals ([Table 3](#ZoolRes-40-3-219-t003){ref-type="table"}). For both the left and right hand, the ring finger was significantly longer in the control group than that in the depressed group. ###### Digit length in right and left forelimbs of depressed and control subjects (cm) Left/Right Digit^†^ Control (*n*=16) Depressed (*n*=15) ------------ ---------- ------------------ -------------------- ------- ------ Left 2 29.76 2.46 29.28 2.62 4\* 37.75 2.60 35.45 3.17 Right 2 30.25 2.67 29.04 2.78 4\* 37.14 2.64 35.09 3.24 ^†^: Fingers are numbered 1 to 5 from thumb to smallest digit. Data are means±*SD*. Student's *t*-test, \*: *P*\<0.05. Digit ratio comparison {#sec3dot3-ZoolRes-40-3-219} ---------------------- The digit ratio was significantly lower in the control group than in the depressed monkeys ( [Table 4](#ZoolRes-40-3-219-t004){ref-type="table"}), including the 2D:4D ratio in the right and left forelimb hands of depressed and control subjects. ###### 2D:4D ratio in the right and left forelimb hands of depressed and control subjects +-------+-------+---------+-----------+-----------+----+------+------+-----------+ | Left/ | Digit | Control | Depressed | *P* value | | | | | | | | | | | | | | | | Right | Ratio | | | | | | | | +=======+=======+=========+===========+===========+====+======+======+===========+ | Left | 2D:4D | 16 | 0.79 | 0.04 | 15 | 0.83 | 0.03 | 0.002\*\* | +-------+-------+---------+-----------+-----------+----+------+------+-----------+ | Right | 2D:4D | 16 | 0.81 | 0.03 | 15 | 0.84 | 0.05 | 0.0372\* | +-------+-------+---------+-----------+-----------+----+------+------+-----------+ Data are means±*SD*. Student's *t*-test, *\**: *P*\<0.05; *\*\**: *P*\<0.01; *\*\*\**: *P*\<0.001*.* DISCUSSION {#sec4-ZoolRes-40-3-219} ========== With respect to primate finger length investigations, the 2D:4D ratio is strongly related to social behavior and physical aggression. [@B23-ZoolRes-40-3-219] reported that a low 2D:4D ratio is associated with more competitive social systems, which is in accordance with our previous observation ([@B34-ZoolRes-40-3-219]) that depressed monkeys face greater competition for social resources--including feeding opportunities, comfortable resting places, and mating opportunities--and display significant deficits in social interactions. Thus, the cynomolgus monkey population is a suitable choice to study the relationship between the 2D:4D ratio and depression. In the present study, we employed a reliable naturally occurring primate depression model to identify depressed cynomolgus monkeys. Our data disclosed that the ring fingers in both the left and right forelimb hands were longer in healthy female monkeys than in depressed female monkeys. In terms of the finger length data, we found that the digit ratios were significantly higher in depressed monkeys, including the 2D:4D ratios in the left and right forelimb hands. Depression can occur due to long-term social pressures, especially in females ([@B26-ZoolRes-40-3-219]). In a competitive environment, the dorsal anterior cingulate cortex (dACC) region in the brain controls support, emotion regulation, conflict monitoring, and behavioral inhibition ([@B5-ZoolRes-40-3-219]). [@B11-ZoolRes-40-3-219] revealed a significant positive correlation between the 2D:4D ratio and gray matter volume of the dACC in women but not in men. Interestingly, maturation of the dACC influences the development of MDD ([@B13-ZoolRes-40-3-219]). The critical hippocampal brain area, which is strongly associated with the pathogenesis of depression, is also related to the 2D:4D ratio ([@B15-ZoolRes-40-3-219]). These studies provide a possible intrinsic link in the brain tissue between 2D:4D ratio and depression. Previous investigations have shown that the digit ratio persists in a stable range during embryogenesis and increases in accordance with personal growth ([@B10-ZoolRes-40-3-219]; [@B12-ZoolRes-40-3-219]). Interestingly, [@B31-ZoolRes-40-3-219] used the 2D:4D ratio to reflect the degree of prenatal androgen exposure in humans. However, a growing body of evidence indicates that the 2D:4D ratio is unrelated to adult sex hormone (e.g., estrogen and androgen) concentrations ([@B22-ZoolRes-40-3-219]). The 2D:4D ratio appears to be relatively stable, although it does increase somewhat throughout childhood ([@B29-ZoolRes-40-3-219]). Thus, there is a general consensus that the 2D:4D ratio is a relatively stable biomarker for the balance between fetal testosterone (FT) and fetal estrogen (FE), with low FT and high FE linked to high 2D:4D ([@B20-ZoolRes-40-3-219]). Based on these findings, we hypothesize that the 2D and 4D finger length ratios are primarily determined by prenatal sex hormone exposure, and that the effects of this prenatal hormone on the 2D:4D ratio are not presented as estrogen or androgen concentrations differences in a later period. However, the prenatal sex hormon affect the subject\'s neural development and biochemistry ([@B14-ZoolRes-40-3-219]). Those with longer finger lengths tend to possess poor aggression tendencies and emotion regulation function. This is also consistent with our findings. In a competitive environment, high 2D:4D female individuals exhibited a high correlation with depression. Thus, our findings provide a novel way in which to select depressed monkeys according to comparison of the 2D:4D ratio. Future work should examine the relationship between the 2D:4D ratio and the severity of depression in larger samples that report a wider range of depression symptoms. Measurement of the 2D:4D ratio may provide a predictive tool for the diagnosis of depression and strong support for indications of depression risk to proceed early intervention. CONCLUSIONS {#sec5-ZoolRes-40-3-219} =========== Most previous primate digit ratio studies have been examined in regard to social behaviors and rank. However, few have investigated the relationship between digit ratios and depression in primates. This is the first study to reveal significant differences in finger lengths and digit ratios between depressed monkeys and healthy controls. We discovered that depressed monkeys presented with shorter 4th fingers and a higher 2D:4D ratio in both forelimbs. These metrics show promise as gross biological indicators to facilitate screening for depressed monkeys in large population-based studies. However, whether this conclusion can be applied to screen for human depression requires further investigation. We thank the Zhongke Experimental Animal Co. Ltd., China for providing the facilities to conduct the field observations. We thank the scientific editors at Impactys (www.impactys.com) for editing and proofreading this manuscript. All authors declare no conflicts of interest. \>L.W. and P.X. jointly conceived and designed the study; J.G.Z., L.B., and L.W. collected the data; Q.Y.W., C.H.Q., and L.Y.L. analyzed the data; L.W., G.W., and P.X. drafted the original manuscript; G.W., Y.X., and L.B. supervised the data analysis and edited the manuscript for intellectual content and style. All authors read and approved the final version of the manuscript. [^1]: Authors contributed equally to this work

{#sp1 .319}

Introduction {#Sec1} ============ Progranulin (PGRN), which is a 68.5-kDa glycoprotein containing 7.5 tandem repeats of a cysteine-rich motif, is proteolytically processed into 6-kDa granulins (GRNs) \[[@CR1]\]. PGRN is involved in multiple physiological functions and various disease processes, including autoimmune disorders, tumorigenesis, and insulin resistance \[[@CR2]--[@CR4]\]. In the human brain, haploinsufficiency of PGRN is one of the major factors causing frontotemporal lobar degeneration (FTLD) \[[@CR5],[@CR6]\], which is characterized by ubiquitinated cytoplasmic inclusions containing TAR DNA binding protein 43 (TDP-43) \[[@CR7]\]. However, the mechanism by which impaired production of PGRN causes the formation of TDP-43 inclusions in the cytoplasm is not yet fully understood. We previously demonstrated that PGRN expression increases after traumatic brain injury (TBI) in mice, and the major source of PGRN is CD68-positive activated microglia \[[@CR8]\]. PGRN deficiency induces exacerbated inflammatory responses that are associated with increased lysosomal biogenesis in activated microglia after TBI. Further, the expression of cytotoxic factors is increased in relation to increased lysosomal biogenesis, and neuronal damage is exacerbated in PGRN-deficient mice \[[@CR9]\]. Because microglial activation characterized by morphologic and phenotypic changes also occurs with aging \[[@CR10]\], increased lysosomal biogenesis and cytotoxic factor production may occur with aging in activated microglia in PGRN-deficient mice. Wils et al. suggested that PGRN-deficient mice have a disrupted autophagy-lysosomal pathway with alterations in lysosomal homeostasis \[[@CR11]\]. Patients with a homozygous mutation in the *GRN* gene present with neuronal ceroid lipofuscinosis (NCL), a group of neurodegenerative lysosomal storage disorders \[[@CR12]\]. Additionally, Götzl et al. (2014) recently revealed that FTLD patients due to PGRN deficiency have NCL-like pathology \[[@CR13]\]. These reports suggest that disorders with aging from PGRN deficiency occur associated with lysosomal dysfunction. The NCLs display a relatively uniform phenotype in the central nervous system that is characterized by a dramatic loss of cortical neurons, pronounced gliosis, and accumulation of lysosomal autofluorescent lipopigments \[[@CR14]\]. NCL model animals show a particular vulnerability early during disease progression in the sensory thalamocortical pathways, i.e., projections from the ventral posteromedial nucleus/ventral posterolateral nucleus (VPM/VPL) of the thalamus to the primary somatosensory barrelfield cortex (S1BF) \[[@CR15]\]. Localized aggregation of activated microglia and astrocytes followed by neuronal cell loss occur in this area \[[@CR16]--[@CR18]\]. Impaired myelination also occurs as a feature of some NCL models such as mice deficient in palmitoyl-protein thioesterase 1 (*Ppt1*), ceroid-lipofuscinosis, neuronal 5 (*Cln5*), and ceroid-lipofuscinosis, neuronal 8 (*Cln8*) genes \[[@CR19],[@CR20]\]. However, characterization of the brain pathology regarding NCL in the VPM/VPL has yet to be performed in PGRN-deficient mice. In the present study, the effects of PGRN deficiency on age-associated lysosomal biogenesis and function, inflammatory responses, and TDP-43 aggregates in the cytoplasm were investigated. Further, the brain pathology regarding NCL was also characterized in aged PGRN-deficient mice. Materials and methods {#Sec2} ===================== Animals {#Sec3} ------- In this study, 10- and 90-week-old female wild-type (WT) and PGRN-deficient (KO) C57BL/6J mice were used. They were produced from breeding heterozygous pairs in our laboratory and were genotyped using previously described protocols \[[@CR21]\]. Mice were maintained under controlled light (lights on, 07:00--19:00), temperature (23 ± 1°C), and humidity (55 ± 10%), and were given free access to food and water. All animal experiments were performed in accordance with the Guide for the Care and Use of Laboratory Animals of the University of Tokyo and were approved by the Institutional Animal Care and Use Committee of the University of Tokyo (Permit Number: P12-651). Antibodies {#Sec4} ---------- The primary antibodies from the following sources were used for immunohistochemistry: mouse monoclonal anti-ubiquitin antibody (MAB1510), rabbit polyclonal anti-microtuble-associated protein 2 (MAP2) antibody (AB5622), and rat monoclonal anti-CD11b antibody (MAB1387Z) from Millipore (Billerica, MA, USA); rabbit polyclonal anti-glial fibrillary acidic protein (GFAP) antibody (Z0334) and rabbit polyclonal anti-myelin basic protein (MBP) antibody (A0623) from Dako (Glostrup, Denmark); rabbit polyclonal anti-TDP-43 antibody (10782-2-AP) from Proteintech (Chicago, IL, USA); rabbit polyclonal anti-phospho-TDP-43 (Ser409/410) antibody (TIP-PTD-P02) from CosmoBio (Tokyo, Japan); rabbit polyclonal anti-ionized calcium-binding adapter-1 (Iba1) antibody (019-19741) from Wako chemicals (Osaka, Japan); rat monoclonal anti-CD68 antibody (MCA1957GA) from AbD serotec (Oxford, UK); mouse monoclonal anti-MAP2 antibody (M1406) from Sigma (St. Louis, MO, USA); guinea pig polyclonal anti-p62 antibody (03-GP62-C) from American research products (Waltham, MA, USA); rat monoclonal anti-lysosome associated membrane protein 1 (Lamp1) antibody (553792) from BD pharmingen (Franklin Lakes, NJ, USA). Subsequently, the secondary antibodies from following sources were used: biotin-conjugated goat anti-rabbit IgG antibody and biotin-conjugated goat anti-guinea pig IgG from Vector Laboratories (Burlingame, CA, USA); biotin-conjugated donkey anti-rat IgG from Jackson ImmunoResearch Laboratories (West Grove, PA, USA); Alexa Fluor 488 goat anti-mouse IgG antibody, Alexa Fluor 594 goat anti-mouse IgG antibody, Alexa Fluor 594 goat anti-rabbit IgG antibody, Alexa Fluor 594 goat anti-rat IgG antibody, and Alexa Fluor 488 goat anti-guinea pig IgG antibody from Invitrogen (Carlsbad, CA, USA). Traumatic brain injury (TBI) {#Sec5} ---------------------------- TBI was placed in the cerebral cortex of 10- and 90-week-old WT and KO mice as described previously \[[@CR8]\]. Briefly, mice were anesthetized by intraperitoneal administration of xylazine hydrochloride (8 mg/kg) and chloral hydrate (300 mg/kg) dissolved in sterile saline and subsequently placed in a stereotaxic apparatus (Narishige, Tokyo, Japan). The right cerebral cortex (1.3 mm lateral to the midline, 1 to 3 mm posterior to the bregma, 2 mm in depth) was exposed and injured with a stainless steel cannula with a 0.5-mm outer diameter. Brain samples were taken 4 days after TBI under xylazine hydrochloride and chloral hydrate anesthesia or by decapitation, and all efforts were made to minimize suffering. Tissue preparation for histological analyses {#Sec6} -------------------------------------------- For immunohistochemical studies, 10- and 90-week-old WT and KO mice (n = 5 for each group) were used. Mice were transcardially perfused with saline followed by 4% paraformaldehyde in PBS (pH 7.2) under xylazine hydrochloride and chloral hydrate anesthesia. Brains were removed from the skull, postfixed in 4% paraformaldehyde overnight, immersed in 10% sucrose in PBS for 24 h, 20% sucrose for 24 h, and 30% sucrose for 48 h for cryoprotection, and cut into 30-μm sections on a cryostat. Sections were preserved in PBS with 0.01% sodium azide. Brain sections from bregma −1.46 to −1.82 were used for the analysis. In the present study, we focused on age-related phenotypical changes in the brain due to PGRN deficiency, and analyzed the uninjured side of the brain unless otherwise mentioned. Immunochromogenic staining {#Sec7} -------------------------- Immunochromogenic staining for Iba1, CD68, Lamp1, GFAP, p62, phospho-TDP-43 (pTDP-43) and MBP was carried out on free-floating sections. In brief, sections were incubated in 0.3% hydrogen peroxide for 30 min, blocked in blocking solution (Block Ace, Snow Brand Milk Products, Sapporo, Japan) at room temperature (RT) for 2 h, and exposed to primary antibody (1:500) at 4°C for 60 h. Sections were incubated with biotinylated secondary antibody (1:500) at RT for 2 h, amplified with avidin-biotin complex (Vector Laboratories), and visualized with 3-3′-diaminobenzidine (Sigma) as the chromogen for 1.5 min. All washes were done in 0.03% PBST, incubations with primary and secondary antibodies were done in 0.3% PBST containing 1% BSA, incubations with the avidin-biotin complex were done in 0.3% PBST. Lamp1-, p62-, and pTDP-43-stained sections were counterstained with cresyl violet solution (0.2% cresyl violet, 0.1% acetic acid in distilled water (DW)). Sections were mounted, dehydrated, and coverslipped with Multi Mount 220 (Matsunami Glass, Osaka, Japan). The sections were then observed under a BX-53 microscope equipped with a DP-73 digital microscope camera (Olympus, Tokyo, Japan). To measure the immunoreactive (IR) areas, the number of pixels per image with an intensity above the predetermined threshold level was quantified, and the total was calculated using NIH ImageJ software (National Institutes of Health, Bethesda, MD, USA) as the percentage area density, which is defined as the number of pixels divided by the total number of pixels in the imaged field. To determine the number of Iba1-IR cells, cells with a clearly outlined, stained soma were counted. Images on the medial and lateral region along the wound site (injured side) and the comparable contralateral side (uninjured side) were used for the analysis of the cerebral cortex. On the other hand, four images of the left VPM/VPL without overlapping were captured and used for the analysis of VPM/VPL, and the values of the four images were averaged. For analysis of the number of Iba1-IR cells in layers IV of the S1BF, an image on the left S1BF was used. A rectangular box (400 × 150 μm) was localized to layers IV of the S1BF, and the number of Iba1-IR cells was counted. To measure fractal dimension (FD) of Iba1-IR cells, ten images of Iba1-IR cells of the left cerebral cortex and VPM/VPL were used for analysis. FD was calculated by the software (PopImaging v5.00, the trial version, <http://www.dbkids.co.jp>), and the value of the ten images was averaged. Immunofluorescent staining {#Sec8} -------------------------- Immunofluorescent staining for MAP2, GFAP, MBP, CD68, p62, ubiquitin, and TDP-43 was carried out on free-floating sections. To improve specific fluorescent signals and reduce autofluorescence, antigen retrieval and Sudan black B treatment was performed as described previously \[[@CR22]\]. Briefly, sections were incubated in 10 mM citrate buffer (pH 6.0) at 90°C for 30 min, followed by incubation in 0.1% Sudan black B (Wako Pure Chemicals) in 70% ethanol for 20 min at RT. The sections were blocked in blocking solution (Block Ace, Snow Brand Milk Products) at RT for 2 h and exposed to primary antibody at 4°C for 60 h. Sections were incubated with Alexa Fluor 488- or 594-conjugated secondary antibody (1:500) at RT for 2 h. Nuclei were stained with Hoechst33258 (0.1 μg/ml; Polysciences Inc., Warrington, PA) at RT for 30 min. All sections were mounted on glass slides and coverslipped with fluoromount (Diagnostic BioSystems, Pleasanton, CA). To analyze the localization of p62, the sections were examined under an LSM 510 laser scanning confocal microscope (Carl Zeiss, Oberkochen, Germany). For triple staining for p62, ubiquitin, and Hoechst33258; p62, TDP-43, and Hoechst33258, the sections were examined under an LSM 5 PASCAL laser scanning confocal microscope (Carl Zeiss). Real-time PCR {#Sec9} ------------- The left thalamus was dissociated from 90-week-old WT (n = 6) and KO mice (n = 5) using tweezers with a fine tip. Total RNA was isolated from the samples with TRIzol (Invitrogen), and first-strand cDNA was synthesized with SuperScriptII reverse transcriptase (Invitrogen). PCR reactions for cathepsin D (*Ctsd*), V-type proton ATPase subunit d 2 (*Atp6v0d2*), transcription factor EB *(Tfeb)*, macrophage expressed gene 1 (*Mpeg1*)*,* cytochrome b-245 light chain (*Cyba*), cytochrome b-245 heavy chain (*Cybb*), complement C4 (*C4*), tumor necrosis factor α (*TNF-α*), lipocalin 2 (*Lcn2*), *p62*, and hypoxanthine phosphoribosyltransferase (*Hprt*; an internal standard) were performed with Thunderbird SYBR qPCR MIX (Toyobo, Osaka, Japan), a LightCycler (Roche, Mannheim, Germany), and each primer set listed in Additional file [1](#MOESM1){ref-type="media"}: Table S1. The PCR reactions for *Hprt* were carried out as follows: 1 min at 95°C for the initial denaturation, followed by 40 cycles of amplification at 95°C for 15 s, 67°C for 20 s, and 72°C for 15 s. For other primers, the PCR conditions were: 1 min at 95°C for the initial denaturation, followed by 40 cycles of amplification at 95°C for 15 s and 60°C for 60 s. The genes of interest were normalized to the expression level of *Hprt*. Assessment of viable neurons {#Sec10} ---------------------------- Viable neurons were visualized using Nissl staining. For Nissl staining, the brain sections were mounted on slides and dried at RT. The slides were incubated with Nissl staining solution (0.2% cresyl violet, 0.1% acetic acid in DW) for 30 min. The slides were washed with DW, dehydrated, and coverslipped with Multi Mount 220 (Matsunami Glass). The sections were then observed under a BX-53 microscope equipped with a DP-73 digital microscope camera (Olympus). For analysis of the number of viable neurons in layers IV, V, and VI of the S1BF, an image on the left S1BF was used. A rectangular box (400 × 150 μm) was localized to layers IV, V, and VI of the S1BF, and the number of viable neurons was counted. For analysis in the VPM/VPL, four images without overlapping were captured, and the number of viable neurons was counted in each image and averaged. A viable neuron was defined as a neuron not exhibiting cytoplasmic shrinkage or nuclear pyknosis. Lipofuscin analysis {#Sec11} ------------------- For autofluorescence analysis, brain sections were mounted on slides, dried at RT, and coverslipped with fluoromount (Diagnostic BioSystems). The sections were then observed under a BX-53 microscope equipped with a DP-73 digital microscope camera (Olympus). The CA3 area was isolated from the hippocampal image using digital image manipulation software (GIMP v2.8.0, Open source software, <http://www.gimp.org/>) and used for analysis. Four images were captured to avoid overlap with each other on the VPM/VPL. To measure the autofluorescent areas, the number of pixels per image with an intensity above the predetermined threshold level was quantified, and the total was calculated using NIH ImageJ software as the percentage area density, which is defined as the number of pixels divided by the total number of pixels in the imaged field. Further, the brown-yellow pigment in the cytoplasm was evaluated with hematoxylin staining, PAS reactivity after diastase treatment, reducing substances by Schmorl staining, and sudanophilia by Oil-red-O or Sudan black B staining. Assessment of myelination {#Sec12} ------------------------- Myelination was visualized using avidin-biotin-peroxidase complex method with diaminobenzidine as the chromogen as described in the immunochromogenic staining. The sections were observed under a BX-53 microscope equipped with a DP-73 digital microscope camera (Olympus). To analyze myelination in the temporal association cortex (TeA) and ectorhinal cortex (Ect), an image was captured on the left TeA and Ect, and the image was binarized using digital image manipulation software (GIMP v2.8.0). After that, the distance from the terminal of MBP-IR area to the rim of the cortex in the TeA and Ect was investigated using digital image manipulation software (GIMP v2.8.0). Statistical analyses {#Sec13} -------------------- The data were analyzed with an unpaired *t*-test or two-way ANOVA followed by a Tukey-Kramer test. Data are expressed as the mean ± SEM. Differences were considered statistically significant when P \< 0.05. Results {#Sec14} ======= PGRN deficiency leads to enhanced CD68 expression in activated microglia {#Sec15} ------------------------------------------------------------------------ We first performed immunohistochemistry for Iba1, a marker for microglia, and compared the morphology and number of Iba1-IR cells between 10- and 90-week-old WT and KO mice (Figure [1](#Fig1){ref-type="fig"}). Conspicuous accumulation of Iba1-positive cells was observed in the thalamus, especially in the VPM/VPL, on the uninjured side of 90-week-old KO mice (Figure [1](#Fig1){ref-type="fig"}A). We therefore focused on both the thalamus and the cerebral cortex in this study. High-power images revealed that Iba1-IR cells in 10-week-old mice were characterized by a small soma and numerous thin, branched processes (ramified), whereas those in 90-week-old mice had a larger soma with fewer and shorter processes in both WT and KO mice (Figure [1](#Fig1){ref-type="fig"}A), suggesting that microglia were activated with aging. Indeed, FD, which is the index for the complexity of the morphology, of Iba1-IR cells in WT and KO mice was decreased with aging both in the cortex and VPM/VPL, while FD was not different between 90-week-old WT and KO mice both in the cortex and VPM/VPL (Figure [1](#Fig1){ref-type="fig"}B and C). On the other hand, in the cortex, the number of Iba1-IR cells was not different among the groups (Figure [1](#Fig1){ref-type="fig"}D). In the VPM/VPL, the number of Iba1-IR cells was not different between 10-week-old WT and KO mice, but significantly larger in KO mice than WT mice at 90 weeks (Figure [1](#Fig1){ref-type="fig"}E).Figure 1**Increased Iba1-IR cells in the VPM/VPL of aged PGRN-deficient mice.** ***A***, Iba1-immunostained images of the cerebral cortex (upper images) and the VPM/VPL (lower images) of 10- and 90-week-old wild-type (WT) and PGRN-deficient (KO) mice. The magnified view is shown on the lower-left corner of each image. Low-power field: scale bar = 200 μm. High-power field: scale bar = 25 μm. ***B***, ***C***, The fractal dimension (FD) of Iba1-IR cells in the cerebral cortex **(** ***B*** **)** and VPM/VPL **(** ***C*** **)** of 10- and 90-week-old WT and KO mice. The figure in a box shows the magnified view of the corresponding figure. The data are presented as the change relative to the number in 10-week-old WT mice. Values are the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). ***D***, ***E***, The number of Iba1-IR cells in the cerebral cortex **(** ***D*** **)** and VPM/VPL **(** ***E*** **)** of 10- and 90-week-old WT and KO mice. The data are presented as the change relative to the number in 10-week-old WT mice. Values are the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). Immunohistochemical images for CD68, a member of the Lamp family expressed in microglia, in the cerebral cortex and VPM/VPL are shown in Figure [2](#Fig2){ref-type="fig"}A. The CD68-IR area was not significantly different between 10-week-old WT and KO mice, but significantly larger in 90-week-old KO mice than WT mice in both the cortex and VPM/VPL (Figure [2](#Fig2){ref-type="fig"}C and D). Additionally, CD68-IR cells in the VPM/VPL in PGRN-deficient mice were round or amoeboid cells (Figure [2](#Fig2){ref-type="fig"}B). These results suggest that PGRN deficiency enhances CD68 expression in activated microglia in both the cerebral cortex and VPM/VPL of the aged brain, although microgliosis with aging was more prominent in the VPM/VPL.Figure 2**Increased CD68 immunoreactivity in the cerebral cortex and VPM/VPL of aged PGRN-deficient mice.** ***A***, CD68-immunostained images of the cerebral cortex (upper images) and VPM/VPL (lower images) of 10- and 90-week-old wild-type (WT) and PGRN-deficient (KO) mice (scale bar = 200 μm). ***B***, High-power image of CD68 immunostaining of the VPM/VPL of 90-week-old KO mice (scale bar = 10 μm). ***C***, ***D***, CD68-positive area in the cerebral cortex **(** ***C*** **)** and VPM/VPL **(** ***D*** **)** of 10- and 90-week-old WT and KO mice. The data are presented as the change relative to the level in 10-week-old WT mice. Values are the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). PGRN deficiency leads to an increase in lysosomal biogenesis {#Sec16} ------------------------------------------------------------ We then assessed lysosomal biogenesis using Lamp1 (lysosome marker) immunostaining in the layer V of the cerebral cortex and the VPM/VPL of the thalamus (Figure [3](#Fig3){ref-type="fig"}). The Lamp1-IR area was significantly increased with aging in both the layer V and VPM/VPL. Lamp1 immunoreactivity in the layer V was not different between WT and KO mice at 10 weeks, but was significantly higher in KO mice than WT mice at 90 weeks of age (Figure [3](#Fig3){ref-type="fig"}A, B). Similarly, Lamp1 immunoreactivity in the VPM/VPL was not different between WT and KO mice at 10 weeks, but was significantly higher in KO mice than WT mice at 90 weeks of age (Figure [3](#Fig3){ref-type="fig"}A, C). Further, we compared gene expressions of the lysosomal enzyme (*Ctsd*), protein (*Atp6v0d2*) and the master regulator of lysosomal gene expression (*Tfeb*) in the thalamus between 90-week-old WT and KO mice. These gene expressions in KO mice were significantly increased compared to expression in WT mice (Figure [3](#Fig3){ref-type="fig"}D).Figure 3**Increased lysosomal biogenesis in the cerebral cortex and VPM/VPL of aged PGRN-deficient mice.** ***A***, Lamp1-immunostained images of the layer V of the cerebral cortex (upper images) and VPM/VPL (lower images) of 10- and 90-week-old wild-type (WT) and PGRN-deficient (KO) mice (scale bar = 20 μm). ***B***, ***C***, Lamp1-positive area in the layer V of the cerebral cortex **(** ***B*** **)** and VPM/VPL **(** ***C*** **)** of 10- and 90-week-old WT and KO mice. The data are presented as the change relative to the level in 10-week-old WT mice. Values are the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). ***D***, Gene expressions of *Ctsd, Atp6v0d2, and Tfeb* in 90-week-old WT and KO mice. The data are presented as the change relative to the level in WT mice. Values are the mean ± SEM (n = 6 for WT; n = 5 for KO, \*P \< 0.05, unpaired t-test). PGRN deficiency leads to an increase in astrogliosis {#Sec17} ---------------------------------------------------- Next, we performed immunohistochemistry for GFAP, a marker of astroglia (Figure [4](#Fig4){ref-type="fig"}). In both the cerebral cortex and VPM/VPL, the GFAP-IR area was significantly increased with aging. Although the GFAP-IR area was not different between WT and KO mice at 10 weeks, the GFAP-IR area was significantly larger in KO than WT mice at 90 weeks of age in both areas (Figure [4](#Fig4){ref-type="fig"}A, C, and D). As shown in Figure [4](#Fig4){ref-type="fig"}B, GFAP-IR astrocytes in the VPM/VPL of 90-week-old WT mice showed variable degrees of cellular hypertrophy, suggesting mild to moderate reactive astrogliosis. On the other hand, GFAP-IR astrocytes in 90-week-old KO mice had thicker processes, suggesting severe diffuse reactive astrogliosis \[[@CR23]\].Figure 4**Increased GFAP immunoreactivity in the cerebral cortex and VPM/VPL of aged PGRN-deficient mice.** ***A***, GFAP-immunostained images of the cerebral cortex (upper images) and VPM/VPL (lower images) of 10- and 90-week-old wild-type (WT) and PGRN-deficient (KO) mice (scale bar = 200 μm). ***B***, High-power GFAP-immunostained images of the VPM/VPL (scale bar: 20 μm) of 90-week-old WT and KO mice. ***C***, ***D***, GFAP-positive area in the cerebral cortex **(** ***C*** **)** and VPM/VPL **(** ***D*** **)** of 10- and 90-week-old WT and KO mice. The data are presented as the change relative to the level of 10-week-old WT mice. Values are the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). PGRN deficiency results in neuronal loss in the VPM/VPL and increased microgliosis in the S1BF layer IV {#Sec18} ------------------------------------------------------------------------------------------------------- We concentrated our analysis on the VPM/VPL and the S1BF. Neurons in the VPM/VPL project to the S1BF, and both of these structures show particular vulnerability in previously reported NCL mouse models \[[@CR15]\]. We performed Nissl staining to compare the number of viable neurons in the VPM/VPL in 10- and 90-week-old WT and KO mice. The number of viable neurons was not different among 10-week-old WT, 10-week-old KO, and 90-week-old WT mice, but significantly decreased in 90-week-old KO mice (Figure [5](#Fig5){ref-type="fig"}A and B). Next, we examined gene expression of glial cell-derived cytotoxic factors, i.e., *Mpeg1*, *Cyba*, *Cybb*, *C4*, *TNF-α*, and *Lcn2* in the left thalamus in 90-week-old mice. The expression of all of these genes was significantly higher in KO mice than in WT mice (Figure [5](#Fig5){ref-type="fig"}C). On the other hand, neuronal loss was not detected in 90-week-old KO mice in layers IV, V, and VI of the S1BF (data not shown). However, the number of Iba1-positive cells in layer IV of the S1BF, which receives projections from the VPM/VPL, was significantly increased only in 90-week-old KO mice (Figure [5](#Fig5){ref-type="fig"}D and E). These results suggest that aged PGRN-deficient mice show particular vulnerability in the VPM/VPL at least partially due to increased expression of cytotoxic factors by glial cells.Figure 5**Neuronal loss and increased expression of cytotoxic factors in the VPM/VPL of aged PGRN-deficient mice.** ***A***, Nissl-stained images of the VPM/VPL of 10- and 90-week-old WT and KO mice (scale bar = 50 μm). ***B***, The number of Nissl-positive cells in the VPM/VPL of 10- and 90-week-old WT and KO mice. The data are presented as the change relative to the number of 10-week-old WT mice. Values are presented as the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). ***C***, Comparison of gene expression for *Mpeg1*, *C4, Cyba*, *Cybb*, *TNF-α*, and *Lcn2* between 90-week-old WT and KO mice in the thalamus. The data are presented as the change relative to the level in WT mice. Values are the mean ± SEM (n = 6 for WT; n = 5 for KO, \*P \< 0.05, unpaired t-test). ***D***, Iba1-immunostained images in the layer IV of the S1BF of 10- and 90-week-old WT and KO mice (scale bar = 100 μm). ***E***, The number of Iba1-IR cells in the VPM/VPL of 10- and 90-week-old WT and KO mice. The data are presented as the change relative to the number of 10-week-old WT mice. Values are presented as the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). PGRN deficiency results in impaired cellular degradation in lysosomes {#Sec19} --------------------------------------------------------------------- To investigate cellular degradation in lysosomes, we performed immunohistochemistry for p62, which is selectively degraded by the autophagy-lysosomal pathway, in 10 and 90-week-old WT and KO mice. Considerable p62-positive aggregates in the VPM/VPL were observed only in 90-week-old KO mice (Figure [6](#Fig6){ref-type="fig"}A). The p62-IR area in the VPM/VPL of 90-week-old mice was significantly larger in KO mice than WT mice (Figure [6](#Fig6){ref-type="fig"}B), although gene expression for p62 was not different between 90-week-old WT and KO mice (Figure [6](#Fig6){ref-type="fig"}C). Because a high-power image suggested that p62-positive inclusions were present in both neuronal and glial cells (Figure [6](#Fig6){ref-type="fig"}D), we next investigated what types of cells in the VPM/VPL accumulated p62 using double immunostaining for p62 and MAP2 (neuronal marker), GFAP (astrocyte marker), MBP (oligodendrocyte marker), and CD68 (microglia marker; mainly localized to late endosomes and lysosomes). As shown in Figure [6](#Fig6){ref-type="fig"}E, p62 was occasionally colocalized with MAP2, GFAP, MBP, and CD68. In addition, some p62-positive inclusions were also positive for ubiquitin (Figure [6](#Fig6){ref-type="fig"}F). These results suggest that PGRN deficiency leads to disruption of the autophagy-lysosomal system in neuronal and glial cells.Figure 6**Excessive p62 accumulation in the VPM/VPL of PGRN-deficient mice with aging.** ***A***, p62-immunostained images with Nissl counterstaining in the VPM/VPL of 10- and 90-week-old WT and KO mice (scale bar = 100 μm). ***B***, ***C***, p62-positive area **(** ***B*** **)** and p62 gene expression **(** ***C*** **)** in the VPM/VPL of 90-week-old WT and KO mice. The data are presented as the change relative to the level of WT mice. Values are the mean ± SEM **(** ***B***: n = 5 for each group; ***C***: n = 6 for WT, n = 5 for KO, \*P \< 0.05, unpaired t-test). ***D***, High-power p62-immunostained image with Nissl counterstaining (scale bar = 20 μm). ***E***, Double-stained images of p62 (green) with MAP2 (red), GFAP (red), MBP (red), and CD68 (red) in the VPM/VPL of 90-week-old KO mice (scale bar = 10 μm). ***F***, The triple-stained image of p62 (green), ubiquitin (red) and Hoechst33258 (blue) in the VPM/VPL of 90-week-old KO mice (scale bar = 10 μm). PGRN deficiency leads to TDP-43 aggregation in the cytoplasm of neurons and lipofuscin accumulation in microglia {#Sec20} ---------------------------------------------------------------------------------------------------------------- We next performed TDP-43 immunostaining in the VPM/VPL in 10 and 90-week-old WT and KO mice. TDP-43 immunoreactivity was decreased in 90-week-old KO mice compared with other groups (Figure [7](#Fig7){ref-type="fig"}A), probably due to neuronal loss (see Figure [5](#Fig5){ref-type="fig"}A). Triple staining for TDP-43, p62, and Hoechst33258 showed that the TDP-43 aggregate colocalized with p62 in the cytoplasm was observed only in aged KO mice (Figure [7](#Fig7){ref-type="fig"}B). Triple staining for TDP-43, MAP2, and Hoechst33258 further showed that the TDP-43 aggregate was present in the cytoplasm of neurons in aged KO mice (Figure [7](#Fig7){ref-type="fig"}C). Additionally, pTDP-43-IR aggregate was observed in aged KO mice (Figure [7](#Fig7){ref-type="fig"}D).Figure 7**TDP-43 aggregation in the cytoplasm of neurons in the VPM/VPL of aged PGRN-deficient mice.** ***A***, TDP-43-immunostained images in the VPM/VPL of 10- and 90-week-old WT and KO mice (scale bar = 500 μm). The area enclosed by dashed line indicates the VPM/VPL. ***B***, The triple-stained image of TDP-43 (red), p62 (green), and Hoechst33258 (blue) in the VPM/VPL of 90-week-old KO mice (scale bar = 50 μm). ***C***, The triple-stained image of TDP-43 (red), MAP2 (green), and Hoechst33258 (blue) in the VPM/VPL of 90-week-old KO mice (scale bar = 20 μm). ***D***, Phospho-TDP-43 (pTDP-43)-immunostained image with Nissl counterstaining in the VPM/VPL of 90-week-old KO mice (scale bar = 10 μm). To estimate lipofuscin accumulation, autofluorescence in the cerebral cortex, hippocampus, and thalamus was evaluated. Autofluorescence was not detected in 10-week-old WT and KO mice (data not shown). On the other hand, at 90-week-old, autofluorescence was similar in the cerebral cortex and the CA3 of the hippocampus between WT and KO mice, whereas it was significantly increased in KO mice compared with WT mice in the VPM/VPL (Figure [8](#Fig8){ref-type="fig"}A, B). Lipofuscin represents an intracellular lipopigment of lysosomal origin and appears in unstained sections as brown-yellow autofluorescent intracellular granules that are positive for periodic acid-Schiff (PAS), Schmorl, and Oil-red-O staining \[[@CR24]\]. A brown-yellow pigment was observed in the cytoplasm (Figure [8](#Fig8){ref-type="fig"}C), and reactivity for Schmorl (Figure [8](#Fig8){ref-type="fig"}D), PAS (Figure [8](#Fig8){ref-type="fig"}E), and Oil-red-O (Figure [8](#Fig8){ref-type="fig"}F) was positive in the VPM/VPL in 90-week-old KO mice. Additionally, double staining with Sudan black B and CD11b (microglia marker; localized to the plasma membrane) revealed that lipofuscin was accumulated in microglia in the VPM/VPL of KO mice (Figure [8](#Fig8){ref-type="fig"}G).Figure 8**Increased lipofuscin accumulation in the VPM/VPL of aged PGRN-deficient mice.** ***A***, Autofluorescence images of the cerebral cortex, hippocampus, and VPM/VPL of 90-week-old wild-type (WT) and PGRN-deficient (KO) mice (scale bar = 200 μm). ***B***, Autofluorescence (AF) area of the cerebral cortex, CA3 of the hippocampus, and VPM/VPL of 90-week-old WT and KO mice. Values are the mean ± SEM (n = 5 for each group, \*P \< 0.05, unpaired t-test). ***C***, ***D***, ***E***, ***F***, Hematoxylin-stained **(** ***C*** **)**, Schmorl-stained **(** ***D*** **)**, periodic acid-Schiff (PAS)-stained **(** ***E*** **)**, and Oil-red-O-stained **(** ***F*** **)** images of the VPM/VPL of 90-week-old KO mice (scale bar = 20 μm). ***G***, Images double-stained with Sudan black B and CD11b in the VPM/VPL of 90-week-old KO mice (scale bar = 10 μm). Black and white arrows indicate Sudan black B-positive and CD11b-positive cells, respectively. PGRN deficiency results in impaired myelination in the cerebral cortex {#Sec21} ---------------------------------------------------------------------- NCL mouse models such as *Ppt1*-, *Cln5*-, and *Cln8*-deficient mice exhibit a loss of myelin as one of the typical disease features \[[@CR19],[@CR20]\]. We therefore performed immunohistochemical staining for MBP, a major protein constituent of the myelin sheath that surrounds axons and is synthesized by oligodendrocytes \[[@CR25]\], using 10 and 90-week-old WT and KO mice. Compared to 90-week-old WT mice, MBP immunoreactivity in cerebral cortex, especially on the lateral side including the TeA and Ect, was relatively disorganized in 90-week-old KO mice (Figure [9](#Fig9){ref-type="fig"}A). The distance from the terminal of MBP-IR area to the rim of the cortex in the TeA and Ect was lessened with aging in WT mice, suggesting myelin was formed with aging. On the other hand, in KO mice, the distance was not different with aging, suggesting myelin formation was disturbed (Figure [9](#Fig9){ref-type="fig"}B).Figure 9**Disrupted myelination in the cerebral cortex of aged PGRN-deficient mice.** ***A***, Binary MBP-immunostained images in the TeA and Ect of 10- and 90-week-old WT and KO mice (scale bar = 200 μm). Double dashed arrows indicate the distance from the terminal of MBP-IR area to the rim of the cortex. ***B***, The distance from the terminal of MBP-IR area to the rim of the cortex in the TeA and Ect of 10- and 90-week-old WT and KO mice. The data are presented as the change relative to the level in 10-week-old WT mice. Values are the mean ± SEM (n = 5 for each group). Values having a different superscript are significantly different (P \< 0.05, Tukey-Kramer test). Discussion {#Sec22} ========== The present study demonstrated that PGRN deficiency leads to enhanced lysosomal biogenesis in the cerebral cortex and thalamus, especially in the VPM/VPL, with aging. Additionally, we showed impaired lysosomal function and the TDP-43 aggregate in the cytoplasm of neurons in the VPM/VPL in aged PGRN-deficient mice. Further, aged PGRN-deficient mice developed NCL-like pathology. Thus, although the possibility that contusion in the unilateral cerebral cortex might affect the severity of histological changes and the levels of gene expressions in the contralateral cortex and the thalamus could not be ruled out, substantial pathological changes were observed in the aged brain due to PGRN deficiency. In our previous study, we showed that although the total number of Iba1-IR microglia was not different between WT and KO mice, CD68 immunoreactivity was significantly higher in young (8- or 9-week-old) KO than WT mice that underwent TBI \[[@CR8]\]. In addition, we have shown that PGRN deficiency leads to decreased mammalian target of rapamycin complex 1 (mTORC1) activity with a resultant increase in lysosomal biogenesis in activated microglia after TBI \[[@CR9]\]. Using aged (90-week-old) mice, the present study showed that both the Lamp1 and CD68 immunoreactivity in the cerebral cortex were significantly higher in KO than in WT mice, even in the absence of TBI, whereas the number of Iba1-IR microglia was not different between the two genotypes. Thus, we made similar observations in the TBI and aging models, both of which are associated with microglial activation and microgliosis. PGRN is therefore likely to be primarily implicated in lysosomal biogenesis that leads to microglial activation and microgliosis. In the present study, we showed that aged PGRN-deficient mice presented with increased accumulation of p62 and ubiquitin in the VPM/VPL, where the main pathological changes have been observed in reported NCL model mice \[[@CR15]--[@CR18]\]. p62/ubiquitin inclusions have been observed when the autophagy-lysosomal system is disrupted \[[@CR26]\]. Therefore, aged PGRN-deficient mice may have an impaired autophagy-lysosomal system. Since p62 inclusions were observed in neuronal and glial cells, PGRN may play a role in maintaining lysosomal function during aging in both neuronal and glial cells. Disruption of both the ubiquitin-proteasome system and the autophagy-lysosomal system is one of the characteristic features of FTLD with TDP-43 inclusions in the cytoplasm of neurons \[[@CR27]\], and impaired production of PGRN is involved in this pathology \[[@CR5]--[@CR7]\]. Previous studies reported increased TDP-43 phosphorylation in PGRN-deficient mice \[[@CR11],[@CR13],[@CR28]\]. In the present study, TDP-43 colocalized with p62 in the cytoplasm of neurons as well as the phosphorylated TDP-43 aggregate in the VPM/VPL were only observed in aged PGRN-deficient mice. Previous studies have shown that abnormal degradation by the autophagy-lysosomal pathway affects TDP-43 aggregation in the cytoplasm \[[@CR29]--[@CR32]\]. Taken together, these observations suggest that PGRN deficiency may be involved in generating TDP-43 aggregation in the cytoplasm associated with impaired cellular degradation by lysosomes. We characterized the brain pathology related to NCL in aged PGRN-deficient mice in the present study. PGRN deficiency resulted in neuronal loss, increased gliosis, lipofuscin accumulation in the VPM/VPL, and defective myelination in the cerebral cortex. These pathological changes in aged PGRN-deficient mice are partially consistent with previous studies \[[@CR11],[@CR13],[@CR33]\], though the brain regions examined were different. Analyses using NCL model mice in the previous studies revealed that glial activation precedes neuronal loss in the VPM/VPL \[[@CR17],[@CR34]\]. Therefore, the particular vulnerability of the VPM/VPL in PGRN-deficient mice might be attributed to cytotoxicity of glial cells. Since expression of genes for glial cell-derived neurotoxic factors was increased in the VPM/VPL of aged PGRN-deficient mice, PGRN deficiency is likely to make glial cells cytotoxic, thereby leading to enhanced neuroinflammation with aging, though further analysis on the vulnerability of the sensory thalamocortical pathways in PGRN-deficient mice might be needed. In addition, lipofuscin was accumulated in microglia of aged PGRN-deficient mice. Lipofuscin is thought to be an end product of cellular protein metabolism, especially protein catabolism \[[@CR24]\]. Since macrophage in PGRN-deficient mice displayed enhanced rates of apoptotic-cell phagocytosis \[[@CR35]\], microglia in PGRN-deficient mice might be more phagocytically active and thereby overloaded with waste products with a resultant increase in lipofuscin deposits. In the present study, defective myelination and enhanced astrogliosis occurred in the cerebral cortex of aged PGRN-deficient mice, but neuronal loss was not observed in the cerebral cortex. Similar observations were also made in 3-month-old mice lacking Cln1, a mutation in which causes classic infantile NCL \[[@CR19]\]. Because astrocytes are implicated in myelination in the brain \[[@CR36],[@CR37]\], PGRN deficiency might lead to impaired astrocyte function and consequently induce defective myelination, though further studies are necessary to elucidate myelination deficits due to PGRN deficiency. These observations indicate that our PGRN-deficient mice may be another useful experimental model for NCL. Recently, it is reported that PGRN-deficient mice recapitulate pathobiomedical features of NCL and FTLD, e.g. increased lysosomal proteins and TDP-43 phosphorylation, and that FTLD patients due to PGRN deficiency have NCL-like pathology \[[@CR13]\]. The present study also revealed that aged PGRN-deficient mice have characteristic features for both FTLD and NCL. Since lysosomal functions are involved in the pathogenesis of these diseases \[[@CR13],[@CR38],[@CR39]\], increased lysosomal biogenesis and lysosomal dysfunction as observed in PGRN-deficient mice in the present study may be common processes in the pathogenic mechanism of these diseases. We have previously suggested that PGRN is localized in lysosomes and may be involved in the activation of mTORC1, which prevents translocation of TFEB to the nucleus by phosphorylation, thereby decreasing lysosomal biogenesis \[[@CR9]\]. Further studies on the molecular mechanism underlying the actions of PGRN are needed to elucidate pathophysiological changes of the aged brain due to PGRN deficiency. Conclusions {#Sec23} =========== The present study demonstrated that aged PGRN-deficient mice show exacerbated neuronal loss, lipofuscin accumulation, gliosis, and increased lysosomal protein and gene expressions in the VPM/VPL of the thalamus, suggesting that aged PGRN-deficient mice develop NCL-like pathology. In addition, the presence of p62 aggregates in the VPM/VPL of aged PGRN-deficient mice suggests that TDP-43 aggregates observed in the cytoplasm of neurons in this brain region are associated with impaired cellular degradation by lysosomal dysfunction. To our knowledge, this is the first report showing NCL-like pathology and TDP-43 aggregates in the VPM/VPL of aged PGRN-deficient mice. Clarifying the mechanisms by which PGRN regulates lysosomal function may be crucial to elucidate pathoetiology of brain disorders resulting from PGRN deficiency. Additional file {#Sec24} =============== Additional file 1: Table S1The primer list used in the present study. **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** YT designed the present study, carried out the experiments and analyzed the data. YT also drafted the initial manuscript. JKC carried out lipofuscin analysis, microscopic observation and helped to draft the manuscript. TM and KY participated in the design of the study and helped to draft the manuscript. MN supervised the present study, and worked up the manuscript. All the authors read and approved the final manuscript. **Authors' information** YT is a JSPS postdoctoral research fellow, Department of Neuropathology and Cell Biology, Tokyo Metropolitan Institute of Medical Science. JKC is an assistant Professor, Department of Veterinary Pathology, The University of Tokyo. TM is an assistant Professor, Department of Veterinary Physiology, The University of Tokyo. KY is an associate Professor, Department of Veterinary Physiology, The University of Tokyo. MN is a Professor, Department of Veterinary Physiology, The University of Tokyo. We are grateful to Sei Sato, Kazunori Sato, and Atsuhiro Tsutiya in the Graduate School of Life Sciences, Toyo University, Japan for technical help using the confocal microscope. This work was supported in part by JSPS KAKENHI Grant Number 23228004 to MN and 24-7548 to YT.

1. Introduction {#sec1-molecules-21-00436} =============== The use of inhibitors to protect steel from corrosion, mainly in acidic solutions, is one of the most practical methods \[[@B1-molecules-21-00436],[@B2-molecules-21-00436],[@B3-molecules-21-00436]\]. Acids such as HCl, \[[@B4-molecules-21-00436],[@B5-molecules-21-00436],[@B6-molecules-21-00436]\] HNO~3~, and H~2~SO~4~ \[[@B7-molecules-21-00436]\] are usually used for the elimination of metal scale and rust in many manufacturing processes. Because of their aggressive effect on the metal surface, inhibitors are comprehensively used to decrease the corrosion of the metal \[[@B8-molecules-21-00436],[@B9-molecules-21-00436]\]. Organic compounds, especially those composed of nitrogen, oxygen and sulfur atoms, represent the majority of inhibitors used in industry \[[@B10-molecules-21-00436],[@B11-molecules-21-00436],[@B12-molecules-21-00436],[@B13-molecules-21-00436]\]. Inhibitors which contain multiple (double or triple) bonds play an important role in the adsorption and the interaction of these compounds onto the metal surfaces \[[@B2-molecules-21-00436]\]. The adsorption of the organic molecules \[[@B14-molecules-21-00436]\], the charge density \[[@B15-molecules-21-00436],[@B16-molecules-21-00436]\], the mode of adsorption \[[@B17-molecules-21-00436],[@B18-molecules-21-00436]\], the exposal area \[[@B19-molecules-21-00436]\] of the organic inhibitors and the molecular weight could all affect the efficiency of an inhibitor \[[@B20-molecules-21-00436]\]. Organic compounds generally inhibit the corrosion of the metal by forming a film on the metal surface. The degree of inhibition depends on the chemical composition, the structure and the ability of the inhibitor to adsorb on the metal surface. The adsorption of these compounds is influenced by the their electronic structure, steric factors, aromatic character and electron density of the functional groups such as --NH, --N=N--, --C=N, --CHO, R--OH, *etc.* located at the donor site such \[[@B20-molecules-21-00436],[@B21-molecules-21-00436],[@B22-molecules-21-00436],[@B23-molecules-21-00436],[@B24-molecules-21-00436],[@B25-molecules-21-00436]\]. Recently Shukla *et al.* \[[@B26-molecules-21-00436]\] reported the hexahydro-*s*-triazine derivatives **1a**--**e** ([Figure 1](#molecules-21-00436-f001){ref-type="fig"}) as corrosion inhibitors of steel in 1 N HCl and the inhibition effect depend on the electronic nature of the functional groups present in the compounds. The work we report here describes a one step and an easy method for the synthesis of three *sym*-trisubstituted 1,3,5-triazine derivatives T3Q, T3AMP**y** and T3EA with different terminal side chains on the triazine ring ([Scheme 1](#molecules-21-00436-sch001){ref-type="scheme"}) and their application as promising novel organic corrosion inhibitors for steel in HCl solutions. 2. Results and Discussion {#sec2-molecules-21-00436} ========================= 2.1. Synthesis of sym-Trisubstituted-s-triazines ***4***, ***6***, and ***8*** {#sec2dot1-molecules-21-00436} ------------------------------------------------------------------------------ The three inhibitors T3Q (**4**), T3EA (**6**), and T3AMPy (**8**) were prepared by the reaction of nucleophiles **3**, **5**, or **7** with cyanuric chloride (**2**) at 0 °C for 2 h and then the temperature of the reaction mixtures was increased to 80 °C and kept at this temperature for 24 h to afford the target products in 83%--86% yield as shown in [Scheme 1](#molecules-21-00436-sch001){ref-type="scheme"}. All the spectral data were in good agreement with their proposed structures and with the reported data for compounds **6** and **8** (Experimental Section). As an example the obtained ^1^H-NMR spectrum ([Figure S1](#app1-molecules-21-00436){ref-type="app"}) of the new compound 2,4,6-tris(quinolin-8-yloxy)-1,3,5-triazine (T3Q, **4**), showed doublet of doublets at δ 8.81, 8.34, and 7.81 ppm, corresponding to the three protons Ha, Hc, and Hd, respectively ([Scheme 1](#molecules-21-00436-sch001){ref-type="scheme"}). The other three protons Hb, He, and Hf appear at δ 7.46--7.55 as a multiplet peak. The ^13^C-NMR of compound **4** showed nine carbon signals in the aromatic region at δ 120.9, 122.1, 126.3, 126.2, 128.9, 136.0, 139.9, 147.0, 150.5, corresponding to the quinolone moiety and a signal at δ 173.6 related to the triazine moiety, respectively. 2.2. Potentiodynamic Polarization Measurements {#sec2dot2-molecules-21-00436} ---------------------------------------------- The effect of the concentration of **4**, **6** and **8** on the polarization curves of steel in acidic chloride solutions is shown in [Figure 2](#molecules-21-00436-f002){ref-type="fig"}, [Figure 3](#molecules-21-00436-f003){ref-type="fig"} and [Figure 4](#molecules-21-00436-f004){ref-type="fig"}, respectively. These figures show that the addition of **4**, **6**, or **8** to the acidic chloride solution is accompanied by a lowering of the current densities in both directions of the polarization curves. Accordingly, the compounds **4**, **6** and **8** can be labeled as mixed type inhibitors that decrease the anodic and cathodic reactions \[[@B27-molecules-21-00436]\]. All the estimated electrochemical parameters for the compounds **4**, **6**, and **8** are quoted in [Table 1](#molecules-21-00436-t001){ref-type="table"}. The data clearly shows a dependence of the current density *(Icorr)* on the concentrations of **4**, **6** and **8**. It is obvious that *Icorr* values decreased with increasing inhibitor concentration and the corrosion protection occurred via blocking adsorption of T3Q, T3AMPy and T3EA on the active sites of steel, which enhances the corrosion protection performance of T3Q, T3AMPy and T3EA. The results can be explained on the basis of adsorption of T3Q, T3AMPy and T3EA molecules on the cathodic active sites, which retards the reduction of the H^+^ ion and the corresponding hydrogen evolution by increasing the energy barrier for proton discharge \[[@B28-molecules-21-00436],[@B29-molecules-21-00436]\]. In addition, the anodic dissolution of steel was decreased by the addition of the adsorbed molecules (T3Q, T3AMPy and T3EA) on the anodic active sites. The adsorption of T3Q, T3AMPy and T3EA molecules on the steel surface led to formation of a protective layer and causes a shift in *Ecorr* to more noble \[[@B2-molecules-21-00436],[@B30-molecules-21-00436]\]. Then, we can conclude that the additions of T3Q, T3AMPy and T3EA to the acidic chloride solution caused an enhancement in the corrosion protection performance of steel. The percentage inhibition efficiency (IE, *%*) can be calculated from: $$IE\% = \frac{Icorr\left( b \right) - Icorr\left( i \right)}{Icorr\left( b \right)} \times 100$$ where *Icorr (i)* is the corrosion current density in the inhibited and *Icorr (b)* is the corrosion current density in the uninhibited solution. The *IE* values were calculated from Equation (4) and presented in [Table 1](#molecules-21-00436-t001){ref-type="table"} for T3Q and T3AMPy. As observed from [Table 1](#molecules-21-00436-t001){ref-type="table"}, the *IE* is highly dependent on the concentration of the tested materials and increased with increasing concentration of T3Q, T3AMPy and T3EA. The maximum inhibition efficiency (98% for T3Q, 96% for T3AMPy and 85% for T3EA) was attained at a concentration of 250 ppm. 2.3. EIS Studies {#sec2dot3-molecules-21-00436} ---------------- The influence of T3Q, T3AMPy and T3EA concentrations on the Nyquist plots of steel in acidic chloride solution is shown in [Figure 5](#molecules-21-00436-f005){ref-type="fig"}, [Figure 6](#molecules-21-00436-f006){ref-type="fig"} and [Figure 7](#molecules-21-00436-f007){ref-type="fig"}, respectively. A capacitive loop is clearly shown in [Figure 5](#molecules-21-00436-f005){ref-type="fig"}, [Figure 6](#molecules-21-00436-f006){ref-type="fig"} and [Figure 7](#molecules-21-00436-f007){ref-type="fig"}, where the larger the size of the capacitive loops the higher the concentration of T3Q, T3AMPy and T3EA. *EIS* data were fitted by an equivalent circuit (*EC*) comprised of solution resistance (*Rs*), charge transfer resistance (*Rct*) in parallel with double layer capacitance (*Cdl*) as shown in [Figure 8](#molecules-21-00436-f008){ref-type="fig"}. All the impedance parameters for T3Q, T3AMPy and T3EA were estimated and are quoted in [Table 1](#molecules-21-00436-t001){ref-type="table"}. The percentage inhibition efficiency (*IE*, *%*) as follows: $$IE\% = \frac{Rct\left( i \right) - Rct\left( b \right)}{Rct\left( i \right)} \times 100$$ where *Rct(i)* and *Rct(b)* are the charge transfer resistances in the inhibited and uninhibited solution, respectively. It is clear that the *IE%* values are strongly dependent upon T3Q, T3AMPy and T3EA concentration and increased with increasing concentration of the tested materials. It seems that the surface of coverage of steel increased with increasing concentration due to an increase in the amount of adsorbed of T3Q, T3AMPY and T3EA on the steel surface, which led to the increase in *IE*. The results are comparable to those obtained from previous studies \[[@B31-molecules-21-00436],[@B32-molecules-21-00436]\]. On the other hand, the value of *Cdl* decreases with the increasing in T3Q, T3AMPy and T3EA concentration. The adsorption of T3Q, T3AMPy and T3EA can be explained by the gradual replacement of water molecules with high dielectric constant by T3Q, T3AMPy and T3EA molecules with low dielectric constant. As immersion time and concentration increased, the thickness of the adsorbed layer increased leading to a decrease in the electrical capacitance \[[@B33-molecules-21-00436]\]. The results of increasing of *Rct* and the decreasing of *Cdl* values suggests that T3Q, T3AMPy and T3EA blocking the active sites on the steel surface via adsorption \[[@B34-molecules-21-00436]\], which could be confirmed that a protective layer formed at the steel/solution, which enhances the corrosion protection performance of steel \[[@B35-molecules-21-00436]\]. 2.4. Adsorption Isotherm {#sec2dot4-molecules-21-00436} ------------------------ The replacement of the pre-adsorbed water molecules by adsorption of the tested materials can be attributed to the interaction energy between the tested materials and the exposed surface. Investigating the different types of adsorption isotherms provide an explanation about the interaction between the steel surface and the T3Q, T3AMPy and T3EA. The experimental data was found to fit well with the Langmuir adsorption isotherm among the various adsorption isotherms. The Langmuir adsorption isotherm related the surface coverage (θ) and the concentration of inhibitor (*Cinh*) as follows \[[@B36-molecules-21-00436]\]: where *C~(inh)~* is inhibitor concentration and *Kads* is the equilibrium constant for the adsorption process. The linear relation obtained by plotting *C~(inh)~/θ* *versus* *C~(inh)~* is shown in [Figure 9](#molecules-21-00436-f009){ref-type="fig"}a--c for T3Q, T3AMPy and T3EA, respectively. The regression coefficients were also calculated and found to be 0.999 for T3Q and T3AMPy and 0.99 for T3EA. The calculated value of the slope and the R2 suggests that the adsorption of T3Q, T3AMPy and T3EA on the steel surface follows the Langmuir adsorption isotherm. The standard free energy of adsorption (*ΔG°ads*) and the adsorption constant (*Kads*) are related by the following equation: There are two types of adsorption, the first is the physical adsorption and the second one is the chemical adsorption. The former one is predominating when the values of *ΔGads* around −20 kJ·mol^−1^ or lower due to the electrostatic interactions between charged T3Q and TA3MPy and charged steel. The latter process is predominant when the value of *ΔG°ads* around −40 kJ·mol^−1^ or higher through formation of coordinate bond. The estimated values of *ΔG°ads* are −35.65 and −35.85 kJ·mol^−1^ for T3Q and T3AMPy, respectively. In case of T3EA it was found to be −31.87 kJ·mol^−1^. The results suggest that the adsorption mechanism of T3Q, T3AMPy and T3EA on steel surface occurs via physical and chemical adsorption (comprehensive adsorption) \[[@B37-molecules-21-00436]\]. The physical adsorption of the triazine derivatives occurred via electrostatic interactions between the protonated tested materials and the negatively charged steel surface resulted from the adsorption of Cl−anions as shown in [Figure 10](#molecules-21-00436-f010){ref-type="fig"}a. Nitrogen atoms of the triazine ring and hydrazine have unshared electron pairs, which are shared with the empty d-orbital of iron atoms on the steel surface and enhanced the chemical adsorption as shown in [Figure 10](#molecules-21-00436-f010){ref-type="fig"}b. In addition, electron donor−acceptor interactions may also arise between the π-electrons of the imine (C=N) groups of the 1,3,5 triazine ring and the empty d-orbital of iron atoms as shown in [Figure 10](#molecules-21-00436-f010){ref-type="fig"}c. The spontaneity and strong adsoption of T3Q and T3AMPy on steel surface can be accounted the high negative values of *ΔG°ads* \[[@B35-molecules-21-00436],[@B38-molecules-21-00436]\]. 3. Experimental Section {#sec3-molecules-21-00436} ======================= 3.1. General Information {#sec3dot1-molecules-21-00436} ------------------------ The solvents used were of HPLC reagent grade. Cyanuric chloride, ethanolamine, 2-picolyl-amine, and 8-hydroxyquinoline were purchased from Sigma-Aldrich (Sigma-Aldrich Chemie GmbH, 82024 Taufkirchen, Germany Melting points were determined with a Mel-Temp apparatus (Sigma-Aldrich Chemie GmbH, 82024 Taufkirchen, Germany)and are uncorrected. Magnetic resonance spectra (^1^H-NMR and ^13^C-NMR) were recorded on a JEOL 400 MHz spectrometer (JEOL, Ltd., Tokyo, Japan). Chemical shift values are reported in δ units (ppm). Elemental analyses were performed on a mod. 2400 elemental analyzer (PerkinElmer, Inc.940 Winter Street, Waltham, MA, USA), and the values found were within ±0.3% of the theoretical values. The compounds were named using ChemDraw Ultra version 14 (Cambridge Soft Corporation, Cambridge Park Dr, Cambridge, MA, USA). The chemical composition, the method of electrode preparation, the reference and the counter electrode are the same as used previously in our studies \[[@B39-molecules-21-00436]\]. 3.2. General Method for the Synthesis of 1,3,5-Triazine Derivatives {#sec3dot2-molecules-21-00436} ------------------------------------------------------------------- A solution of cyanuric chloride (**2**, 1.84 g, 10.0 mmol) in dioxane (50 mL) was slowly added to a solution of the nucleophile \[8-hydroxyquinoline (**3**, 4.78 g, 33.0 mmol), ethanolamine (**5**, 2.01 g, 33.0 mmol), or 2-(aminomethyl)pyridine (**7**, 3.57 g, 33.0 mmol)\] in dioxane (50 mL) at 0 °C. Anhydrous potassium carbonate (9.7 g, 70.0 mmol) was added to the reaction mixture at the same temperature. After complete addition of the K~2~CO~3~ the reaction mixture was stirred for 24 h at 80 °C. After cooling to room temperature, the reaction mixture was filtered and washed with hot dioxane. The solvent was removed under reduced pressure, and the residue was extracted with dichloromethane (2 × 50 mL, in case of preparation of **8** followed washing with H~2~O (20 mL). The organic extracts were combined and dried with anhydrous MgSO~4~. The pure product was obtained by recrystallization from dichloromethane--hexane (1:2) to afford the product **8** in pure state. In case of **4**, the product was obtained after evaporation of dioxane and the residue washed with water and the solid product recrystallized from ethyl acetate. In case of **6**, the product was obtained after evaporation of dioxane at room temperature to get a white solid product. *2,4,6-Tris(Quinolin-8-yloxy)-1,3,5-triazine* (T3Q, **4**). White solid mp 268--271 °C; yield 86%; ^1^H-NMR (DMSO-*d*~6~, [Figure S1](#app1-molecules-21-00436){ref-type="app"}): δ 7.46--7.55 (m, 9H, Ar; b,e,f); 7.81 (dd, *J* = 2.4, 2.0 Hz, 3H, Ar,d), 8.34 (dd, *J* = 1.6 Hz, 3H, Ar,*c*), 8.81(dd, *J* = 1.6, 1.6 Hz, 3H, Ar, a) ppm; ^13^C-NMR (DMSO-*d*~6~, [Figure S1](#app1-molecules-21-00436){ref-type="app"}): δ 120.9, 122.1, 126.3, 126.2, 128.9, 136.0, 139.9, 147.0, 150.5, 173.6 ppm; Anal. Calc. for (C~30~H~18~N~6~O~3~; 510.51): C, 70.58; H, 3.55; N, 16.46; found C, 70.79; H, 3.64; N, 16.29. *2,2′,2″-\[(1,3,5-Triazine-2,4,6-triyl)tris(azanediyl)\]tris(ethan-1-ol)* (T3EA, **6**). White solid mp 99 °C; yield 83% \[[@B40-molecules-21-00436]\] mp 97--99 °C, 81.5% yield\]; FT-IR: 3415 (OH), 3340 (NH), 1585--1478 (C=N) cm^−1^; ^1^H-NMR (DMSO-*d*~6~, [Figure S2](#app1-molecules-21-00436){ref-type="app"}): δ 3.25 (brs, 6H, --NCH~2~--), 3.45 (brs, 6H, --CH~2~O), 4.57 (brs, 3H, 3OH), 6.35-6.55 (brt, 3H, NH) ppm; ^13^C-NMR (DMSO-*d*~6~, [Figure S2](#app1-molecules-21-00436){ref-type="app"}): δ 42.7 (--NCH~2~), 60.3 (--CH~2~O--), 165.7 (C=N) ppm. MS: *m*/*z* = 259.47 \[M\]^+^ ([Figure S3](#app1-molecules-21-00436){ref-type="app"}). *N^2^,N^4^,N^6^-Tris(Pyridin-2-ylmethyl)-1,3,5-triazine-2,4,6-triamine* (T3AMPy, **8**). Off white solid mp 76--8 °C; yield 85% \[[@B41-molecules-21-00436],[@B42-molecules-21-00436]\], yield 83%\]; ^1^H-NMR (CDCl~3~): δ 4.46 (br, d, *J* = 41 Hz, 6H), 7.27--7.09 (br, m, 9H), 7.63 (br, d, *J* = 50 Hz, 3H), 8.43 (br, d, *J* = 17 Hz, 3H) ppm; ^13^C-NMR (CDCl~3~): δ 45.9 (--CH~2~--), 121.8, 121.9, 136.5, 148.9, 160.0, 166.1 (C=N) ppm. 3.3. Electrochemical Measurements {#sec3dot3-molecules-21-00436} --------------------------------- Polarization curves and EIS data were conducted using a Solartron 1470E (multichannel system, (Solartron, Armstrong Mall, Farnborough, Hampshire, UK)) with the Solartron 1455A as FRA. The polarization curves were recorded with a sweep rate of 1 mV/s. EIS measurements were measured in the frequency range of 0.01--10 kHz. Pt sheet was used as counter electrode while a calomel electrode was used as arefernce electrode. The working electrode was cut from steel rod with the following composition (wt: 0.14% C, 0.57% Mn, 0.21% P, 0.15% S, 0.37% Si, 0.06% V, 0.03% Ni, 0.03% Cr and Fe balance. The test solution used in all experiments was 1M HCl containing different concentrations of the investigated inhibitors (50--250 ppm). 4. Conclusions {#sec4-molecules-21-00436} ============== The electrochemical results revealed that T3Q and TA3MPy have excellent corrosion protection performance towards the corrosion of steel in acidic chloride solution. The protection of steel occurs via adsorption of T3Q, T3AMPy and T3EA and blocking of the active sites on the steel surface. The electrochemical results revealed that the inhibition occurs by suppressing the anodic and cathodic reactions. The calculated values of *IE* from the *EIS* method follow the same trend as those obtained from the polarization results. The results obtained indicated that the more nitrogen atoms in the terminal groups of the inhibitor, the better the corrosion protection performance, so T3Q and T3AMP**y** gave better protection for steel than T3EA, which has an oxygen atom in its terminal side chain of the triazine moiety. Finally, the triazine moiety is not the only factor responsible for the inhibitory action, but also the structure of the side chain attached has a major effect for the corrosion inhibition. The authors thank the Deanship of Scientific Research at King Saud University for funding this work through Prolific Research Group Program (PRG-1437-33; Saudi Arabia). **Sample Availability:** Samples of the prepared compounds **4**, **6**, and **8** are available from the authors. ###### Click here for additional data file. Supplementary materials can be accessed at: <http://www.mdpi.com/1420-3049/21/4/436/s1>. The main part of the work was carried out by Kholod A. Dahlous, with the direct supervision of Ayman El-Faham and Zeid A. AL Othman. Conceptually the work was designed by Ayman El-Faham, Gamal A. El-Mahdy, and Hamad A. Al-Lohedan. The corrosion mearuments was carried by Gamal A. El-Mahdy. All authors read and approved the final manuscript. The authors declare no conflict of interest. Figures, Scheme and Table ========================= ![Structures and names of compounds **1a**--**e** as corrosion inhibitors \[[@B26-molecules-21-00436]\].](molecules-21-00436-g001){#molecules-21-00436-f001} {#molecules-21-00436-sch001} {#molecules-21-00436-f002} {#molecules-21-00436-f003} {#molecules-21-00436-f004} {#molecules-21-00436-f005} {#molecules-21-00436-f006} {#molecules-21-00436-f007} {#molecules-21-00436-f008} {#molecules-21-00436-f009} {#molecules-21-00436-f010} molecules-21-00436-t001_Table 1 ###### The effect of T3Q, T3EA and T3AMPY concentrations on the IE% values for steel in acidic chloride solutions using electrochemical methods. Inhibitor Polarization Method EIS Method ---------------- --------------------- ------------ --------- --------- ------ ---------- ------ ------ ---------- Blank 69 120 −0.3955 839 \_\_\_\_ 1.80 334 \_\_\_\_ **T3Q (4)** 50 55 131 −0.3337 55 93.4 28.5 109 93.6 150 50 187 −0.3271 53 93.6 30 108 94.0 250 77 103 −0.3463 14 98.8 100 90 98.2 **T3EA (6)** 50 98 631 −0.3315 557 33.6 2.8 198 35.6 150 79 413 −0.3357 299 64.3 5.2 153 65.3 250 63 174 −0.3352 125 85.1 13 125 86.1 **T3AMPy (8)** 50 56 170 −0.3324 63 92.4 24 112 92.5 150 53 105 −0.3382 36 95.7 46 105 96.0 250 50 112 −0.3348 34 96.9 47.5 104 96.2

Introduction ============ Papillomaviruses are small double-stranded DNA viruses that infect the squamous epithelia (skin and internal mucosae) of both animals and humans ([@bau031-B1]). Papillomaviruses are diverse and species-specific. The human papillomavirus (HPV) genome is composed of six early (E1, E2, E4, E5, E6 and E7) Open Reading Frames (ORFs), two late (L1 and L2) ORFs and a non-coding long control region ([@bau031-B2]). More than 170 HPV types have been characterized to date. Of those, nearly 40 are transmitted through sexual contact, infecting the anogenital region and the oropharynx. Among these, ∼20 are designated as 'oncogenic high risk' because they have been linked to cervical, anal, vulvar, vaginal, penile and oropharyngeal cancers ([@bau031-B3]). Worldwide, \>5% of all new cancers are attributed to high-risk HPV infections ([@bau031-B4]). HPV, Hepatitis B virus, Hepatitis C virus and *Helicobacter pylori* were responsible for 1.9 million cancer cases worldwide in 2008 ([@bau031-B5]). HPV is the cause of virtually all cases of cervical cancer, the third most common female cancer globally ([@bau031-B6]). The high-risk type HPV16 alone is responsible for 50% of cervical cancers and high-grade cervical intraepithelial lesions ([@bau031-B7]). Furthermore, in the developing world where HPV disease burden is the greatest, cervical carcinoma is the leading cause of cancer mortality among women. Two prophylactic HPV L1 VLP (virus-like particle) vaccines have been developed to provide protection against infection for at least 5 years and reduce the risk of cervical cancer ([@bau031-B8]). Gardasil, an HPV quadrivalent recombinant vaccine that is a mixture of VLPs derived from the L1 capsid proteins of HPV types 6, 11, 16 and 18, was approved by the US Food and Drug Administration in 2006 ([@bau031-B9]). Cervarix, a prophylactic vaccine composed of a mixture of VLPs derived from the L1 capsid proteins of HPV types 16 and 18 has been shown to be 100% effective in preventing HPV strains 16 and 18 ([@bau031-B10]). T cells naturally eliminate the majority of HPV infections by recognizing epitopes displayed on the virally altered epithelium. Therapeutic HPV vaccines aim to treat established HPV infections and HPV-associated malignancies by targeting non-structural oncogenic proteins E6 and E7 ([@bau031-B11]). A recent vulvar intraepithelial neoplasia clinical trial showed promising results. Kenter *et al.* ([@bau031-B12]) reported a vaccination using synthetic long peptides spanning the complete sequence of the HPV16 E6 and E7 oncoproteins and a conventional adjuvant-induced clinical responses and relief of symptoms in 60% of the patients with high-grade vulval intraepithelial neoplastic disease. However, most trials of therapeutic HPV vaccines have yielded disappointing clinical results ([@bau031-B11]). Available immunological data related to HPV vary in granularity, quality and complexity and are stored in various formats, from publications, technical reports and general-purpose databases. The challenge is to collect the data scattered in these resources, clean, annotate, store and analyse them to facilitate meaningful knowledge discovery. To bridge the gap between data and knowledge, we developed a framework called KB-builder to streamline the development and deployment of web-accessible immunological knowledge bases. The KB-builder framework is generic and can be applied to any immunological sequence data set. We developed the Human Papillomavirus T cell Antigen Database (HPVdb) using KB-builder to support the discovery of T cell-based HPV vaccine targets and reported it at the Association for Computing Machinery - Conference on Bioinformatics, Computational Biology and Biomedical Informatics (ACM-BCB) 2013 ([@bau031-B13]). We updated and improved HPVdb by expanding the data sets and integrating more advanced analysis tools such as antigen sequence variability analysis tool based on Shannon entropy calculation and conservation analysis of T cell epitopes and human leukocyte antigen (HLA) ligands. HPVdb integrates curated data and information with tailored analysis tools to facilitate data mining and to aid rational vaccine design by discovery of vaccine targets. It is publicly available at <http://cvc.dfci.harvard.edu/hpv/>. Material and Method =================== KB-builder ---------- The framework, KB-builder, consists of seven major functional modules, each facilitating a specific aspect of the database construction process. The input to the framework is data scattered in primary databases and scientific literatures. As shown in [Figure 1](#bau031-F1){ref-type="fig"}, the modules enable automated data collection and integration, semi-automated data cleaning and annotation, automated data storage and retrieval, fast deployment of basic computational tools, development and integration of advanced tools for an in-depth analysis of various structural and functional properties associated with immune responses and vaccine development, definition of workflows to answer specific research questions and semi-automated update and maintenance. The KB-builder framework helps set up a web-accessible knowledgebase and the corresponding analysis pipeline within a short period (typically within 1--2 weeks), given a set of annotated genetic or protein sequences. In addition to the HPVdb, using KB-builder, several other prototype knowledge bases have been built including TANTIGEN: Tumor T cell Antigen Database (cvc.dfci.harvard.edu/tadb), FLAVIdB: Flavivirus Antigen Database ([@bau031-B14]) and FLUKB: Flu Virus Antigen Database (research4.dfci.harvard.edu/cvc/flukb/). A workflow is an automated process that takes a request from the user, performs complex analysis by combining data and tools preselected for common questions and produces a comprehensive report ([@bau031-B15]). Several workflows have been implemented in FLAVIdB and FLUKB to answer various research questions, such as the identification of broadly protective viral vaccine targets ([@bau031-B14]). The web interface of these online knowledge bases uses a set of graphical user interface forms with a combination of Perl, PHP, CGI and C background programs. Development was carried out in the CentOS 4.5 Linux environment. Figure 1.Schematic overview of the KB-builder framework. Data collection --------------- Eight HPV proteins are included in the HPVdb, namely, E1, E2, E4, E5, E6, E7, L1 and L2. Eighteen HPV genotypes (16, 31, 33, 35, 52, 58, 18, 39, 45, 59, 68, 26, 51, 82, 73, 53, 56 and 66) of five HPV species (HPV16, HPV18, HPV26, HPV34 and HPV53) were annotated as the high-risk group in HPVdb. Eighteen HPV genotypes (11, 44, 55, 40, 91, 32, 42, 54, 61, 72, 81, 83, 84, 62, 87, 89, 71 and 64) of six HPV species (HPV6, HPV7, HPV32, HPV54, HPV61, HPV71 and one unclassified species) were classified as the low-risk group. Names, full names, virus genotypes, UniProt review status, sequence status and primary amino acid sequences of HPV antigens from the aforementioned 36 genotypes were collected from the UniProtKB ([@bau031-B16]). Primary amino acid sequences of HPV were collected via database search against the UniProtKB using the NCBI taxonomic identifiers of the 18 HPV genotypes, for example, 333 760 is the taxonomic identifier of organism 'HPV type 16'. HPV T cell antigens were collected based on experimentally characterized T cell epitopes and/or HLA ligands. The criteria for the selection of HPV T cell antigens require that the antigen must be presented via one or more HLA alleles or be recognized by T cells. If the peptides were able to stimulate T cell function, they are called 'T cell epitopes'. If the peptides had only been tested for their binding affinities to HLA molecules and not for T cell reactivity, these peptides are called 'HLA ligands'. Collection of T cell epitopes and HLA ligands was derived from mining of scientific literature in PubMed using the term 'HPV AND (T cell epitope OR HLA binding peptide)' as a keyword and from the Immune Epitope Database (IEDB) ([@bau031-B17]). Data annotation and organization -------------------------------- The collected data and information were manually checked. Errors, inconsistencies, ambiguous and conflicting information and duplications were corrected or removed. Sequences containing ambiguous amino acids, such as XGXXNGILW, were removed because they make it impossible to perform computational prediction for HLA binding motifs. If multiple UniProt entries have identical sequences, only one entry is kept. If a UniProt entry sequence is a substring of a longer sequence, the entry is removed and the entry with longer sequence is kept. The semi-structured annotated data were automatically transformed in a unified extensible markup language (XML) format because XML is inherently semi-structured and is suitable for hosting semi-structured data. Three XML files were created for HPVdb containing information on antigens, T cell epitopes and HLA ligands. Information in an HPV antigen record includes antigen name, full name, virus genotype, UniProt ID, UniProt review status, amino acid sequences, as well as T cell epitopes and HLA ligands, if reported. Data classification ------------------- In HPVdb, HPV viruses were classified into two groups based on high and low clinical risk for cancer. Each risk group was further subclassified using the virus classification system suggested by the International Committee on Taxonomy of Virus (ICTV). An interactive diagram presents the classification and facilitates database search by clicking on any of the genotype boxes ([Figure 2](#bau031-F2){ref-type="fig"}). Figure 2.Classification of the viruses in the HPVdb based on cancer risk was done using the virus classification system suggested by the ICTV. Database construction --------------------- The HPVdb was constructed using KB-builder, an in-house developed framework that streamlines the development and deployment of web-accessible immunological knowledge bases. The web interface uses a set of graphical user interface forms with a combination of Perl, PHP, Common Gateway Interface (CGI) and C background programs. Basic analysis tools -------------------- Several basic bioinformatics tools were integrated in the HPVdb, including a keyword search tool to locate HPV antigens, a keyword search tool to locate the T cell epitopes or HLA ligands of interest and Basic Local Alignment Search Tool (BLAST) that enables sequence similarity search ([@bau031-B18]) and multiple sequence alignment (MSA) to compare multiple sequences. Beyond the basic utility of keyword search, the HPVdb also offers options for filtering HPV antigen data based on genotype, proteins, UniProt review status (reviewed or unreviewed) and sequence type (complete or fragment sequence). To search for T cell epitopes or HLA ligands, users may input either an epitope/ligand sequence or an HLA allele of interest in the text box. If nothing is input into the text box, the search result page will show all the T cell epitopes and HLA ligands in the database. To facilitate sequence similarity search, the collected antigen protein sequences were organized into FASTA format and were converted into a searchable format to enable searching using the Basic Local Alignment Search Tool (BLAST) algorithm ([@bau031-B18]). MAFFT, an MSA tool, selected because of its outstanding performance in terms of speed and alignment quality, was downloaded and installed locally ([@bau031-B19]). Specialized analysis tools -------------------------- The HLA binding prediction tool for on-the-fly peptide binding prediction to 15 frequent HLA class I and class II alleles (A\*0101, A\*0201, A\*0301, A\*1101, A\*2402, B\*0702, B\*0801, B\*1501, DRB1\*0101, DRB1\*0301, DRB1\*0401, DRB1\*0701, DRB1\*1101, DRB1\*1301, DRB1\*1501) has been integrated in each HPV antigen record to facilitate efficient antigenicity analysis. NetMHCpan and NetMHCIIpan ([@bau031-B20], [@bau031-B21]) were selected for this purpose based on our previous benchmark studies on the accuracy of online HLA binding prediction servers ([@bau031-B22], [@bau031-B23]). Other data mining tools include the sequence variability analysis tool, the conservation analysis tool for T cell epitopes and HLA ligands and the visualization tool that shows the localization of epitopes in a given individual or aligned HPV protein/genotype combination. Sequence variability analysis can be performed on entries grouped by protein and further narrowed down by virus genotype or subtype and sequence type. The variability analysis at amino acid level is based on calculation of Shannon entropy ([@bau031-B24]) at each position in an MSA. The entropy is calculated using the formula where *H* is the entropy, *x* is the position in the MSA, *i* represents individual amino acids at position *x*, *I* is the number of different amino acids on position *x* and *P~i~* is the frequency of the given amino acid. Conservation of a position, *x*, is defined by the frequency of the consensus amino acid. Semiautomated update and maintenance ------------------------------------ The database will be actively maintained by members of the bioinformatics core, cancer vaccine center, the Dana-Farber Cancer Institute. Bugs and problems will be fixed as they are reported. The database will be updated every 6 months, using automated retrieval systems such as Wget (<http://www.gnu.org/software/wget/>) and data from primary databases such as the Uniprot and the IEDB. Programs have been made to automatically compare the collected data with the existing data in the database and identify new data; the new data will be manually annotated; and the XML data files will be automatically updated to include the annotated new data. On completion of the data update, the last step of the database update is to test all the basic and advanced analysis tools and workflows to ensure they function well with the updated data files. Results ======= Using the NCBI taxonomic identifiers of the 18 HPV genotypes to search the UniProtKB, we collected 5099 antigen entries (as of December 2013), of which 162 entries were reviewed by UniProt experts and 4937 entries were not reviewed. The data cleaning process removed 2318 antigen entries---5 entries had unknown gene names such as 'X' (UniProt ID: Q705E1) and 'Y' (UniProt ID: Q705E0), 83 antigen sequences contained ambiguous amino acid X and 2230 had redundant sequences. The final list has 2781 antigen entries ([Table 1](#bau031-T1){ref-type="table"}). Eight HPV proteins were included in the HPVdb, namely, E1, E2, E4, E5, E6, E7, L1 and L2. A large number of antigen sequences have been reported for proteins E1 (630), E2 (320), E6 (419), E7 (233), L1 (624) and L2 (296). Around one-third of these sequences were derived from species HPV16 (978). We observed that inconsistent letter case has been used in gene names. For example, gene names e7 (UniProt ID: Q9DIH6) and I2 (UniProt ID: Q9DHD4) were listed instead of E7 and L2. All lower case gene names were modified to upper case. In total, 191 verified T cell epitopes and 45 verified HLA ligands were collected from data mining of literature and databases. Table 1.The number of antigen entries in HPVdb grouped by their UniProt review status and type of antigen sequencesSequence TypeReviewedNot reviewedTotalComplete sequence16016841844Fragment2935937Total16226192781 Using the keyword search function of HPVdb, users are able to search antigen records by keywords. Users can further refine the search by selecting a protein name, and/or a virus genotype, and/or a sequence type and/or UniPort review status. [Figure 3](#bau031-F3){ref-type="fig"}A shows the HPVdb antigen search page with parameters selected being Protein E7, virus genotype 18, complete sequence regardless of UniProt review status. [Figure 3](#bau031-F3){ref-type="fig"}B shows the search result table. The accession numbers in the first column are hyperlinked. By clicking on HPV000092, we get a page, as shown in [Figure 3](#bau031-F3){ref-type="fig"}C, displaying information on the antigen. The antigen information table consists of HPVACC (a unique accession number), date, last updated date, antigen name, full name, virus genotype, UniProt ID, UniProt status (reviewed or unreviewed), sequence status (complete or fragment), a list of T cell epitopes and HLA ligands of the antigen with references, antigen amino acid sequence and the embedded HLA binding prediction tool. The T-cell epitope and HLA ligand sequences are hyperlinked to T cell epitope or HLA ligand record tables. Most of the epitope and HLA-ligand records were enriched with additional annotation, e.g. 'defined in healthy donors', 'associated with virus clearance', 'defined in cervical intraepithelial neoplasia (CIN) or cervical cancer patients' and 'defined by mass spectrometry from cervical cancer biopsy specimens'. Most of these peptides were derived from the two oncogenic proteins E6 and E7 of HPV16. Figure 3.Screenshots of HPV antigen search tool and result pages. (**A**) HPV antigen search page. (**B**) The search result page---the accession numbers in the result table are hyperlinked to HPV antigen information pages. (**C**) HPV00092 (UniPort ID: P06788) information table. Users can also search T cell epitopes and HLA ligands by keywords, such as epitope/ligand sequences or HLA allele names. [Figure 4](#bau031-F4){ref-type="fig"}A is a screen shot of a T cell epitope record table---an A\*0201 restricted T cell epitope in HPV type 16 E7 protein sequences. We can learn how conserved the epitope YMLDLQPET is by clicking on the button 'check conservation of T cell epitope T000125'. As shown in [Figure 4](#bau031-F4){ref-type="fig"}B, the epitope exists in all but one HPV type 16 E7 proteins (94.12% conserved). Figure 4.(**A**) A screenshot of a T cell epitope record table in the HPVdb. This table catalogs the relevant information of T cell epitope T000125, i.e. epitope sequence, restricted HLA allele, PubMed id(s) of the reference paper(s) and its characteristics (e.g. information on how the epitope was identified). A multiple sequence alignment of the protein sequences containing the epitope (highlighted) is displayed. (**B**) A screenshot of the conservation analysis result page obtained by clicking on 'check conservation of T cell epitope T000125' button. [Figure 5](#bau031-F5){ref-type="fig"}A shows the sequence variability analysis tool page. The search parameters include virus genotype 16, protein E7 and complete sequence type. [Figure 5](#bau031-F5){ref-type="fig"}B shows the sequence variability analysis result page. On top of the page, there is a plot of entropy (red curve) and percentage of sequences (blue curve) containing the consensus amino acid at position. The consensus sequence is shown below X-axis with conserved positions in blue. A conserved position is one with entropy \<1, gap fraction \<0.1 and consensus amino acid \>90%. The detailed position-by-position amino acid variability information and the consensus sequence are available for download. Figure 5.(**A**) A screenshot of sequence variability analysis tool page. (**B**) Plot of entropy (red curve) and percentage of sequences (blue curve) containing the consensus amino acid. The consensus sequence is shown below X-axis with conserved positions in blue. A conserved position is one with: entropy \<1, gap fraction \<0.1 and consensus amino acid \>90%. BLAST was integrated into the database to enable sequence similarity search. It can be used for protein and genotype identification. The MSA can be performed by selecting a protein name and virus genotype for sequence diversity analysis. The visualization tool provides three peptide display formats for mapping T cell epitopes or HLA ligands. In format 1, T cell epitope and HLA ligands are highlighted on the peptide sequences. In format 2, the peptides are highlighted in the MSA. In format 3, each peptide is shown in a separate line based on the restricting HLA alleles. [Table 2](#bau031-T2){ref-type="table"} summarizes the analysis tools integrated in the HPVdb and their uniform resource locators (URLs) to access them directly. More information of the HPVdb and instructions for using the database are available at <http://cvc.dfci.harvard.edu/hpv/HTML/help.php>. Table 2.The analysis tools integrated in HPVdb and their URLsToolURLReferencesBLAST<http://blast.ncbi.nlm.nih.gov/Blast.cgi>([@bau031-B18])MAFFT MSA<http://www.ebi.ac.uk/Tools/msa/mafft/>([@bau031-B19])NetMHCpan<http://www.cbs.dtu.dk/services/NetMHCpan/>([@bau031-B20])NetMHCIIpan<http://www.cbs.dtu.dk/services/NetMHCIIpan/>([@bau031-B21])Search tool for HPV antigens<http://cvc.dfci.harvard.edu/hpv/HTML/search.php>Search tool for T cell epitope/HLA ligand<http://cvc.dfci.harvard.edu/hpv/HTML/searchT.php>Blast HPVdb<http://cvc.dfci.harvard.edu/hpv/HTML/blast.php>MSA of HPV sequences<http://cvc.dfci.harvard.edu/hpv/HTML/alignment.php>Sequence variability analysis tool<http://cvc.dfci.harvard.edu/hpv/HTML/varability.php>([@bau031-B14])T cell epitope/HLA ligand visualization tool<http://cvc.dfci.harvard.edu/hpv/HTML/viewEpitope.php>Classification of the viruses based on cancer risk<http://cvc.dfci.harvard.edu/hpv/HTML/classification.php>HLA binding prediction toolEmbedded in each antigen entry tableT cell epitope/HLA ligand conservation analysis toolEmbedded in each experimentally validated T cell epitope/HLA entry table; also embedded in each HLA binding prediction result page. A Case Study---The Identification of a Conserved HPV16 E7 T Cell Epitope ======================================================================== Because HPV proteins E6 and E7 are functionally required for cancer initiation and persistence, they offer exceptional targets for immune-based therapies. We have identified, by mass spectrometry, an E7-derived cytotoxic T cell epitope (E7~11--19~) that was presented by cervical cancer cell lines as well as seven of nine HPV16-positive primary tumor cervical cancer biopsy samples ([@bau031-B25]). Our analyses have shown that the number of distinct cytotoxic T lymphocyte epitope targets on a tumor is small, requiring precise focusing of vaccine formulation. In the study, we identified the E7~11--19~, but not the related E7~11--20~ peptide, on all of the established HPV16 tumor cell lines ([@bau031-B25]). The latter peptide, which has received considerable attention in the literature as a possible tumor antigen, was incorrectly selected for vaccine formulation and was clinically ineffective ([@bau031-B28]). The two patients that did not display E7~11--19~ peptide also had a loss of the human cellular thioreductase GILT protein. Thioreductases are proteins that unfold cysteinylated proteins such as HPV E7, which was associated with altered HPV peptidome display in HPV16-driven cervical cancers ([@bau031-B26]). The HPVdb played an important role to facilitate and speed up the study. First, *insilico* predictions of A\*0201 binding peptides (both 9- and 10-mers) were performed (prediction result is not shown). Based on the prediction results, we synthesized 21 peptides. A\*0201 binding assay identified 10 of them as binding peptides. Interferon γ (IFN γ) ELISpot assay was used to test immune recognition of the 10 A\*0201-binding peptides in peripheral blood mononuclear cells isolated from six A\*0201-positive healthy donors. There are only two HPV peptides, E7~11--19~ and E6~29--38~, eliciting spot-forming unit numbers 4- to 5-fold over background in one donor. MS^3^ Poisson detection mass spectrometry identified that the peptide E7~11--19~ (refer to [Figure 4](#bau031-F4){ref-type="fig"}A for the epitope's information) is physically displayed on HPV16-transformed, A\*0201-positive cells. Given the expression of E7~11--19~ on HPV-16 transformed or transfected cell lines, we would like to know whether known strains of HPV16 conserve this epitope. We performed the epitope conservation analysis using the analysis tool embedded in each HPVdb T cell epitope entry page as shown in [Figure 4](#bau031-F4){ref-type="fig"}A. As shown in the analysis result page in [Figure 4](#bau031-F4){ref-type="fig"}B, the epitope is conserved in 16 of 17 (94.12% conserved) HPV16 E7 complete sequences. A single substitution mutation L15V in HPV001854 (UniProt ID: C0KXQ5) resulted in the immune escape. Among the 35 HPV16 cervical cancer or cervicitis patients we analyzed, only a single patient sample contained the HPV001854 sequence variant. Conserved HPV T cell epitopes displayed by HPV transformed tumors such as E7~11--19~ may be the basis of a therapeutic T cell-based cancer vaccine. DISCUSSION ========== HPV infection is prevalent among sexually active population. However, \>95% of infections are temporary and are promptly cleared by the immune system ([@bau031-B29], [@bau031-B30]). The regression of tumor lesions and persistent high-risk infections depend on strong localized antigen-specific T cell responses ([@bau031-B31]). HPV-associated cancers express viral oncoproteins, and they represent ideal 'non-self' cancer antigens for the development of a therapeutic cancer vaccine ([@bau031-B11]). One of the research interests of our group at the Dana-Farber Cancer Institute is the discovery of T cell-based HPV vaccine targets. We have identified, by mass spectrometry, an E7-derived cytotoxic T cell epitope (E7~11--19~) that was presented by cervical cancer cell lines as well as seven of nine HPV16-positive primary tumor cervical cancer biopsy samples. With the data analysis and knowledge discovery capacity provided by the HPVdb, this epitope was found to be highly conserved among HPV16 strains. The information and knowledge derived through the computational analysis enabled by the HPVdb directly added to the potential clinical value of this epitope as a vaccine candidate. It highlights the importance of organized epitope information across various HPV types and strains for researchers in the field. Several data sources provide information on HPV genomic and protein sequences. The HPVdb (<http://ncv.unl.edu/Angelettilab/HPV/Database.html>) was constructed a decade ago and has not been actively maintained ([@bau031-B32]). The Papillomavirus Episteme (PaVE) hosts 241 annotated papillomavirus genomes, 2245 genes and regions, 2004 protein sequences and 47 protein structures (<http://pave.niaid.nih.gov/>) ([@bau031-B33]). The PaVE provides basic bioinformatics tools to facilitate keyword and BLAST search, MSA and protein structure viewing. An external L1-specific typing tool developed by Piet Maes *et al.* ([@bau031-B34]) are integrated to the database to predict whether a new isolate is different from other named viruses and meets the criteria for a new type. The Human Papillomavirus Proteome Database hosts genomic and proteomic information on 150 HPV strains and their 1036 protein sequences and 743 predicted structures ([@bau031-B35]). The HPVdb was developed with a different purpose in mind, to facilitate data mining for HPV vaccinology and immunology. The HPVdb tightly integrates curated data and information on both antigen sequences and immunological epitopes with tailored analysis tools to aid rational vaccine design by discovery of vaccine targets. It is a unique data mining system for knowledge discovery in HPV with applications in T cell immunology and vaccinology. To support T cell epitope discovery and make use of existing information and knowledge, we developed KB-builder, a framework that streamlines the development and deployment of web-accessible immunological knowledge systems. The KB-builder framework aims to speed up the immunological research and vaccine design by providing specialist knowledge bases that host cleaned, well-annotated and structured data suitable for the discovery of new knowledge. We built several prototypes focusing mainly on viral and tumor antigens using this framework for immunological knowledge discovery. Knowledge bases generated using KB-builder enable data mining using defined workflows. User-friendly analysis tools can be used individually or as part of workflows. The main purpose of vaccine knowledge bases is to help identify key experiments and reduce the overall number of required experiments for vaccine research including the discovery and design. Systematic discovery of HPV vaccine targets relies heavily on the availability of accurate, up-to-date and well-organized antigen data. HPV antigen data are available through publications, technical reports and databases. These data vary in granularity and quality and are in various formats. The extraction of knowledge from the data scattered around using data mining techniques remains a challenging task. HPVdb is a specialized bioinformatics database that tightly integrates the content (data) and analysis tools to enable the automation of complex queries and data mining. The HPVdb enables speed up of rational vaccine design by providing accurate and well-annotated data coupled with tailored computational analysis tools. The authors would like to thank Dr Songsak Tongchusak and Dr Karl Munger for helpful advice. The HPVdb was developed under the support from NIH grant UO1 AI090043 and DFCI Cancer Vaccine Center Institutional Fund. Funding ======= Funding for open access charge: NIH grant UO1 AI090043. *Conflict of interest*. None declared. [^1]: Citation details: Zhang,G.L., Riemer,A.B., Keskin,D.B, *et al.* HPVdb: a data mining system for knowledge discovery in human papillomavirus with applications in T cell immunology and vaccinology. *Database* (2014) Vol. 2014: article ID bau031; doi:10.1093/database/bau031

INTRODUCTION ============ Local random-pattern skin flaps are commonly used to reconstruct soft tissue defects. However, flap necrosis remains a difficult challenge for the plastic surgeon. Blood supply deficiency and ischemic-reperfusion injury are generally responsible for flap necrosis \[[@b1-acfs-2019-00381]\]. Various trials have been conducted to reduce distal flap necrosis and achieve optimal results \[[@b2-acfs-2019-00381]-[@b6-acfs-2019-00381]\]. Lidocaine is an effective pharmacologic agent to increase flap survival \[[@b7-acfs-2019-00381],[@b8-acfs-2019-00381]\]. It is a widely used amide-type anesthetic that blocks voltage-gated sodium channels in cell membranes of neurons and has been shown to improve cutaneous blood flow, inhibit platelet aggregation, and suppress neutrophil function \[[@b9-acfs-2019-00381]-[@b11-acfs-2019-00381]\]. Lidocaine is effective in recovery after flap reperfusion \[[@b12-acfs-2019-00381]\] and can improve randompattern skin flap survival \[[@b8-acfs-2019-00381]\]. To minimize unwanted adverse effects associated with systemic application and maintain a positive effect on the flap, lidocaine should be used topically. The fractional ablative laser was introduced as a new technology for drug delivery enhancement \[[@b13-acfs-2019-00381]\]. Many studies have demonstrated that the laser can assist in topical drug delivery into the skin. Various compounds, including topical 5-aminolevulinic acid, methyl 5-aminolevulinate, 5-fluorouracil, ascorbic acid, corticosteroids, diclofenac, imiquimod, methotrexate, allogeneic stem cells, autologous platelet-rich plasma, and vaccines, have been investigated \[[@b14-acfs-2019-00381]-[@b18-acfs-2019-00381]\]. Lidocaine is a topical agent that demonstrated enhanced transdermal absorption in a fractional ablative laser in vivo model. Oni et al. \[[@b18-acfs-2019-00381]\] reported that the amount of local lidocaine uptake can be manipulated by changing the energy setting and ablative depth of the laser. In other words, the possibility of systemic adverse effects of lidocaine can be minimized by controlling the energy of the laser. We hypothesized that fractional ablative laser therapy would enhance the lidocaine effect to improve random-pattern skin flap survival with minimal systemic adverse effects. Therefore, in this study, we investigated the effect of fractional ablative carbon dioxide laser (FxCL) and topical lidocaine spray on random-pattern skin flap survival in rats. METHODS ======= Animals and groups ------------------ The experimental protocol, including the use of animals in the research, was approved by the Institutional Animal Care and Use Committee of Yeungnam University. Forty-eight 2-month-old male Sprague-Dawley rats (weight, 200--250 g) were randomly divided into four groups according to treatment: FxCL+lidocaine (n=12), FxCL (n=12), lidocaine (n=12), and nontreatment (control; n=12). All animals were provided with standard laboratory food and water ad libitum. Surgical procedure ------------------ The surgical procedure was performed under general anesthesia via the intramuscular injection of zolazepam (50 mg/kg, Zoletil; Virbac, Carros, France) and xylazine (5 mg/kg, Rompun; Bayer Korea, Seoul, Korea). To prevent infectious complications, a single dose of cefazolin (60 mg/kg; Yuhan Co., Seoul, Korea) was administered before the procedure and an aseptic technique was rigorously applied during the procedure. After the dorsal skin was shaved with an electric shaver, a 3×9-cm caudally pedicled random-pattern skin flap, called a McFarlane flap, was designed on the skin. The palpable hip joints were used as anatomical landmarks to define the base of the flap. The incision was made, and the entire flap was undermined below the level of the panniculus carnosus. Then, the flap was completely separated from the deep fascia up to its base. After controlling any bleeding, a 3×9-cm silicone sheet was inserted between the muscle layer and skin flap to prevent muscle neovascularization and then immediately sutured back to the donor bed using 4-0 Nylon sutures ([Fig. 1](#f1-acfs-2019-00381){ref-type="fig"}). All procedures were performed by one researcher (YP), and no rats died during the procedure. The flap area was equally divided into three distinct zones: proximal, intermediate, and distal ([Fig. 2](#f2-acfs-2019-00381){ref-type="fig"}). Laser assembly -------------- FxCL (Edge ONE 2.0; Jeisys Medical Inc., Seoul, Korea) with a wavelength of 10,600 nm was used. The dimension of the treatment area of the handpiece was 10×10 mm^2^. The fractional ablative laser total fluence energy setting used was 3.0 J/cm^2^ with 120-μm ablative depth. A probe setup with 100 dots of 1-cm^2^ size was used to deliver laser energy to the skin flaps. Flap treatment in the four groups --------------------------------- Following the surgical procedure, rats in the FxCL+lidocaine and FxCL groups received one pass of FxCL without missed or overlapped areas. Immediately following FxCL treatment, topical lidocaine spray (Xylocaine 10% spray; AstraZeneca, Cambridge, England) was applied in the FxCL+lidocaine and lidocaine groups. The lidocaine concentration of topical lidocaine spray is 0.1 g/mL. From approximately 5 cm above the randompattern skin flap, the researchers perpendicularly sprayed the lidocaine on each proximal, intermediate, and distal zone of the flap. Lidocaine spray was sprayed once in each zone of the flap. The amount of lidocaine spray sprayed at one time was about 0.1 mL. To prevent the lidocaine from penetrating the skin barrier through the incision site, the incision line was covered with sterile gauze. In the FxCL and control groups, 0.9% saline was applied to the skin flaps with the same pump of the topical lidocaine spray. Then, all rats were housed individually in standard experimental cages in an environmentally controlled room and were fed standard rat chow and water ad libitum for 7 days. Assessment of flap survival --------------------------- To quantify the survival area, all flaps were photographed with a digital camera (EOS 600D; Canon, Tokyo, Japan) and the survival area was measured using an image program (ImageJ software; National Institutes of Health). Results were expressed as percentage survival in relation to total surface area of the flap. Neutrophil and capillary vessel density --------------------------------------- A 1×1-cm skin flap around the demarcation line of the necrotic area was harvested including necrotic and survival site samples. All tissue specimens were fixed in 10% neutral buffered formalin for at least 24 hours at room temperature, embedded in paraffin, cut into 4-mm slices, and stained with hematoxylin and eosin (H&E). Another 1×1-cm full-thickness tissue sample in the intermediate zone excluding necrotic tissue was harvested for neutrophil count under a light microscope (×100). The number of erythrocyte-containing capillary vessels was also counted. These counts were performed in five areas under high magnification (×200) in each area to calculate the capillary vessel density per 1 mm^2^. The blood vessels were measured at the level of the papillary dermis inferior to the panniculus carnosus. Immunohistochemical evaluation ------------------------------ Paraffin-embedded 4-μm sectioned tissue samples were immunostained with monoclonal antibody against rat vascular endothelial growth factor (VEGF; 1:100 dilution, ab28364 Abcam; Santa Cruz Biotechnology, Santa Cruz, CA, USA) and CD31 (1:100 dilution; Novus Biologicals, Littleton, CO, USA). Then, the sections were counterstained with Mayer's hematoxylin and mounted. The amount of VEGF- and CD31-positive cells were quantified in five fields per specimen (×400). Statistical analysis -------------------- Data on percentage of flap survival in the photographic evaluation, number of capillary vessels and neutrophils in the histologic evaluation, and VEGF/CD31-positive microvascular density are expressed as mean±standard deviation. One-way analysis of variance with Bonferroni correction was conducted in each group. Statistical analysis was performed using SPSS version 23.0 (IBM Corp., Armonk, NY, USA) with *p*-values \< 0.05 considered statistically significant. RESULTS ======= Assessment of flap survival --------------------------- Percentage of survival area on postoperative day 7 was 69.08%±3.20% in the FxCL+lidocaine group, 57.08%±5.91% in the FxCL group, 58.16%±4.80% in the lidocaine group, and 53.41%±5.43% in the control group. Percentage of survival area in the FxCL+lidocaine group was significantly higher than that in the other three groups (*p*\<0.001). However, differences between the FxCL, lidocaine, and control groups with respect to percentage of survival area were not statistically significant (*p*=0.18) ([Table 1](#t1-acfs-2019-00381){ref-type="table"}, [Fig. 3](#f3-acfs-2019-00381){ref-type="fig"}). Neutrophil and capillary vessel density --------------------------------------- The stratum corneum of the random-pattern skin flap was successfully disrupted by the laser in the FxCL+lidocaine and FxCL groups ([Fig. 4](#f4-acfs-2019-00381){ref-type="fig"}). Full-thickness tissue in the intermediate zone excluding the necrotic tissue presented differently under the light microscope on postoperative day 7. Dense neutrophil infiltration (41.70±8.40) was observed in the control group, whereas it was markedly decreased in the FxCL+lidocaine group (27.20±4.24, *p*\<0.001). Thus, the inflammatory reaction was less severe in the FxCL+lidocaine group than in the control group. There was no statistically significant difference in neutrophil count in the FxCL (37.23±7.15) and lidocaine (35.43±6.41) groups compared with the control group ([Fig. 5](#f5-acfs-2019-00381){ref-type="fig"}). Capillary vessel density (number of vessels in 1 mm^2^) in the dermal papillary layer was 17.74±5.38 in the FxCL+lidocaine group, 18.51±3.57 in the FxCL group, 16.10±6.04 in the lidocaine group, and 18.14±4.88 in the control group, without significant differences among the groups (*p*=0.22). Capillary vessel density in the layer inferior to the panniculus carnosus was 13.91±4.98 in the FxCL+lidocaine group, 15.84±4.09 in the FxCL group, 14.36±6.11 in the lidocaine group, and 13.28±3.94 in the control group. Similarly, there was no significant difference among the groups. Immunohistochemical evaluation ------------------------------ Immunohistochemical evaluation revealed more VEGF-positive cells in the FxCL+lidocaine group than in the other three groups (FxCL+lidocaine, 121.64±19.38 cells/field; FxCL, 92.45±16.11 cells/field; lidocaine, 98.12±17.83 cells/field; control, 86.09±12.51 cells/field; *p*\<0.05) ([Figs. 6](#f6-acfs-2019-00381){ref-type="fig"}, [7](#f7-acfs-2019-00381){ref-type="fig"}). The number of CD31-positive microvascular vessels was also significantly higher in the FxCL+lidocaine group than in the other three groups (FxCL+lidocaine, 31.38±9.36 vessels/field; FxCL, 22.52±3.99 vessels/field; lidocaine, 24.66±8.25 vessels/field; control, 18.45±5.89 vessels/field; *p*\<0.05) ([Figs. 8](#f8-acfs-2019-00381){ref-type="fig"}, [9](#f9-acfs-2019-00381){ref-type="fig"}). DISCUSSION ========== The results of this study showed that flap survival was significantly higher in the FxCL+lidocaine group than that in the other groups. In addition, neutrophil count was significantly lower in the FxCL+lidocaine group than in the other groups, while neoangiogenetic parameters, VEGF and CD31, were significantly higher in the FxCL+lidocaine group than in the other groups. An ideal drug for improving flap survival would have the following characteristics: (1) clinical availability, (2) administration easily, (3) high therapeutic index and safety, (4) reproducibility of effective results, (5) feasibility of postoperative treatment, (6) cost-effectiveness, (7) known mechanism of action, (8) established bioavailability, and (9) protective effect on necrosis \[[@b19-acfs-2019-00381]\]. No material to date meets all of these requirements. However, we believe that topical lidocaine is one of the closest to an ideal pharmacologic agent to enhance flap survival. Lidocaine, a widely used amide-type local anesthetic, has been reported to enhance neoangiogenesis of random-pattern skin flaps \[[@b8-acfs-2019-00381]\]. When the surgeon decides to use lidocaine to improve flap survival, the administration route also has to be decided. Injection has the possibility of intravenous absorption, which may result in unintended higher concentration of the agent. Topical lidocaine, such as a cream or spray, is considered safer than injectable lidocaine for minimizing systemic adverse effects. Furthermore, such drugs can be easily administered post-operatively, even in an outpatient clinic or as bedside treatment, in clinical practice. However, topical lidocaine spray can be interrupted by the natural skin barrier. The therapeutic efficacy of topical drugs is related to their ability to reach target cells through the skin layer. The major rate-limiting step for percutaneous absorption is passage through the stratum corneum. Reperfusion injury is an inflammatory process modulated by complex signaling mechanisms, which ultimately leads to cell death. As inflammation progresses, neutrophils accumulate. Neutrophil count could be used to predict flap injury \[[@b8-acfs-2019-00381]\]. Lidocaine blocks voltage-gated sodium channels in cell membranes of neurons and has been shown to improve cutaneous blood flow, inhibit platelet aggregation, and suppress neutrophil function. The present study demonstrated that neutrophil density was significantly lower in the FxCL+lidocaine group (27.20±4.24, *p*\<0.001) compared with control group. The fractional ablative laser is a peculiar way that makes microscopic vertical holes on skin surface, each surrounded by a thin layer of ablated tissue, constituting microscopic ablation zones \[[@b13-acfs-2019-00381]\]. The laser leaves microscopic intervening areas of unexposed skin that can be remodeled without scarring. The laser facilitates penetration and distribution of topically applied drugs because the ablated laser holes extend into the dermis and act as channels for drug absorption \[[@b14-acfs-2019-00381]\]. Recently, low fluence setting of the fractional ablative laser, such as 2 or 3 J/cm^2^, was reported to enhance topical drug delivery \[[@b20-acfs-2019-00381]\]. Fractional ablative laser therapy with low fluence has some advantages, including lesser pain and tissue damage and faster recovery than conventional fractional ablative laser therapy. In a pilot study, increasing the laser energy led to visible bleeding, edema and exudate in the treated areas. These factors might make it difficult for lidocaine to be absorbed into the skin against the hydrostatic/oncotic pressure pushing fluid out of the skin. In the histologic evaluation in this study, the stratum corneum was successfully disrupted by lower energy of 3 J/cm^2^. In this study, the FxCL group was necessary to determine the effect of the fractional ablative laser alone on flap survival. Based on our results, laser treatment can enhance survival of random-pattern skin flaps independently. Similarly, Prado et al. \[[@b21-acfs-2019-00381]\] reported the effect of low-level laser therapy, which could enhance random-pattern skin flap survival in rats. Low-power lasers have also been reported to have biostimulating effects, including neoangiogenesis, increased blood flow, and fibroblast proliferation. In the FxCL group, the flap received laser treatment only without pharmacologic agents. Although there were no significant differences, all parameters, including percentage of survival area, neutrophil count, vascular density, and VEGF/CD31 expression, represented a more improved result compared with the control group. It is considered that increased vascular perfusion occurred in the FxCL group. Likewise, in the lidocaine group, most parameters demonstrated enhancement of flap survival. It is considered that the impact of lidocaine, including anti-inflammatory, antithrombotic, and ischemic-reperfusion injury prevention effects, could have produced the result in the lidocaine group without FxCL treatment. To evaluate neoangiogenesis, vascular density in high-power fields and VEGF/CD31 expression were examined in this study. Although there was a significant difference in VEGF/CD31 expression in the FxCL+lidocaine group, the number of capillary vessels was not significantly different among the groups. There is a possibility of missing values or overestimation of the number of microvascular structures on H&E staining because a preexisting vessel could be considered as neoangiogenesis. Most importantly, the markedly enhanced effect of FxCL+lidocaine was significantly different from that of FxCL alone and lidocaine alone in all parameters. Thus, the neoangiogenetic and anti-inflammatory effects of lidocaine may be maximized with FxCL treatment. The potential effect of FxCL and topical lidocaine spray on survival of random-pattern skin flaps in rats was investigated. This study showed the positive effect of FxCL on enhancement of random-pattern skin flap survival in rats with lidocaine. Therefore, FxCL with lidocaine spray can be considered a new therapeutic approach to enhance flap viability. **Conflict of interest** No potential conflict of interest relevant to this article was reported. **Ethical approval** The study was approved by the Institutional Animal Care and Use Committee of Yeungnam University (approval No. YUMC2015-011). {#f1-acfs-2019-00381} {#f2-acfs-2019-00381} {#f3-acfs-2019-00381} {#f4-acfs-2019-00381} {#f5-acfs-2019-00381} {#f6-acfs-2019-00381} {#f7-acfs-2019-00381} {#f8-acfs-2019-00381} {#f9-acfs-2019-00381} ###### Percentage of survival area (%) using photograph in four groups Rat no. FxCL+lidocaine group (%) FxCL group (%) Lidocaine group (%) Control group (%) --------------- ---------------------------------------------------------------- ---------------- --------------------- ------------------- 1 74 63 54 53 2 72 62 58 57 3 69 61 57 58 4 67 67 65 48 5 66 52 62 46 6 73 50 58 57 7 64 48 54 52 8 71 56 59 59 9 69 52 50 48 10 74 60 59 62 11 68 54 67 46 12 71 60 55 55 Mean ± SD (%) 69.08 ± 3.20^[a)](#tfn1-acfs-2019-00381){ref-type="table-fn"}^ 57.08 ± 5.91 58.16 ± 4.80 53.41 ± 5.43 FxCL, fractionated carbon dioxide laser. *p*\<0.05. [^1]: This study was funded by the 2016 Yeungnam University Research Grants.

1. Introduction {#sec1-ijerph-17-03948} =============== Chile is recognized for exploiting its mineral resources \[[@B1-ijerph-17-03948]\] on a world basis, a fact that has brought about economic benefits, but at the same time, a negative impact on the environment, one of the most severe being the generation of mining environmental liabilities (MEL) \[[@B2-ijerph-17-03948],[@B3-ijerph-17-03948]\]. These are distributed along the country; some of them are abandoned and without due management and maintenance. According to a survey made in 2017 by the National Geology and Mining Service (SERNAGEOMIN, for its acronym in Spanish) \[[@B4-ijerph-17-03948]\], there are 740 tailings in Chile: 14.1% of them are classified as active, 62.6% as inactive (non-abandoned), and 23.3% as abandoned (inactive). From the three groups, the latter requires the most attention, their management being still uncertain since they are not thoroughly characterized. For this reason, three management alternatives are considered: (1) Recovering metals of interest, mainly from old deposits with high mineral grades, which could be further exploited since current technology can help extract more minerals from them \[[@B5-ijerph-17-03948],[@B6-ijerph-17-03948],[@B7-ijerph-17-03948]\]; (2) utilizing technological solutions for reusing tailings, e.g., as construction aggregates \[[@B8-ijerph-17-03948]\]; (3) tailings disposal in a stable manner \[[@B9-ijerph-17-03948]\]. A high percentage of tailings are old \[[@B10-ijerph-17-03948]\]. Their construction was not subjected to state supervision owing to the lack of regulations, thus representing an even greater risk \[[@B9-ijerph-17-03948]\]. In the last decade, although Chile has gradually increased its mining regulatory framework to benefit the environment, there are still serious weaknesses. The Chilean territory is characterized by different types of climates and soils \[[@B11-ijerph-17-03948]\], thus making background metal concentrations quite different from each other in the different regions. On a national basis, there is a scarce amount of background metal concentrations. This has not allowed the establishment of regulations to assess the risks that the concentration of certain substances in the soil could imply. So, international regulations are used, resulting from the natural soil conditions of foreign countries. Considering that Chile shows a geological soil diversity and a natural mineralogical abundance, the uncertainty concerning the use of these adopted regulations is high. In 2012, the Chilean Ministry of Environment approved the Methodological Guide for Managing Soils with Potential Presence of Contaminants (GMPGSPPC, for its acronym in Spanish) \[[@B12-ijerph-17-03948]\]. In developed countries such as The Netherlands, Canada, and Australia, the establishment of reference values for heavy metals in soils has allowed the improvement of their planning and environmental management, thus becoming a control device for environmental entities. Since Chile does not count on these reference values, although it does have a large amount of soils with a potential presence of contaminants, mainly from mine tailings, the country needs a methodology to estimate the priority to intervene in the sites exposed to them. Risks associated with tailings are critical because the latter are large masses that could suffer physical and/or chemical destabilizations, that is, the collapse or overflow of the solid materials in the dam, along with chemical reactions resulting in acid mining drainage \[[@B13-ijerph-17-03948],[@B14-ijerph-17-03948],[@B15-ijerph-17-03948]\]. Thus, tools to create a record of tailings requiring intervention is urgently needed. Health risks associated with mineral exposures are widely reported in the literature. Particularly, early exposure to tailings has been associated with the prevalence of congenital malformations, diarrhea, asthma, endocrine and neurological disorders, among others, neurodevelopmental delays and deficits, obesity, diabetes, and cancer---some of which have been increasing over the last few decades \[[@B16-ijerph-17-03948],[@B17-ijerph-17-03948]\]. The objective of the study is to estimate whether Chilean tailings involve contamination risks. To do this, an applied methodology combining the use of algebraic equations and international reference values with environmental and demographic criteria related to the environmental risk stated in GMPGSPPC is presented. Finally, an assessment of the potential ecological risk resulting from heavy metal contamination is made by using Hakanson's methodology \[[@B18-ijerph-17-03948]\], which was chosen because it is one of the methods widely used to assess the potential ecological risks of mining soils \[[@B19-ijerph-17-03948],[@B20-ijerph-17-03948],[@B21-ijerph-17-03948],[@B22-ijerph-17-03948],[@B23-ijerph-17-03948],[@B24-ijerph-17-03948],[@B25-ijerph-17-03948]\]. This methodology allowed the creation of a record of Chilean tailings to be prioritized for urgent intervention, control, and management. 2. Materials and Methods {#sec2-ijerph-17-03948} ======================== 2.1. Experimental Data {#sec2dot1-ijerph-17-03948} ---------------------- Experimental data were obtained from the geochemical database of the Chilean Geochemical Characterization Program for tailings from SERNAGEOMIN \[[@B4-ijerph-17-03948]\]. The data corresponded to 631 out of the 740 tailings registered in the country. The characterization involved the measurement of concentrations in g ton^−1^ of 56 elements and species per sample, considering potential risks and economic values associated with elements of value. Other data available in the database were the type of mining originating the tailings, the state of the tailings (active, inactive, abandoned), their location (region, commune, and geographic coordinates), and mass in estimated tons. These data are available only for a small number of deposits. Eight elements and species characterized in the sample were evaluated: As, Cd, Cr, Cu, Fe, Ni, Pb, and Zn. In this study, the geochemical data identification number (IDQ), provided by the SERNAGEOMIN database aforementioned, was used for each measurement to identify tailings. Only 530 from the 631 dams had georeferentiation data, so 530 dams were assessed. 2.2. Tailings Characteristics {#sec2dot2-ijerph-17-03948} ----------------------------- From the 530 tailings, 84 were classified as active, 421 as inactive, and 25 as abandoned. On a regional basis, most tailings were found in the Coquimbo region (62%), the Atacama region (19%), and the Antofagasta region (6%). According to the type of mining originating them, the following tailings were found: Ag-Au (2), Au (66), Au-Cu (15), Au-Cu-Ag (11), Au-Zn (2), Cu (193), Cu-Au (99), Cu-Mo (13), Cu-Au-Ag(1), Cu-Au-Fe (1), Fe (4), Zn(2), Zn-Cu (2), limestone (2), sediment (96), unknown origin (8), and extremely old dams or stockpiles (2). 2.3. Methodology {#sec2dot3-ijerph-17-03948} ---------------- [Figure 1](#ijerph-17-03948-f001){ref-type="fig"} shows the methodology proposed, which consists of three phases. Phase 1: Identifying tailings exceeding the admissibility of the reference concentration values of As, Cd, Cr, Cu, Fe, Ni, Pb, and Zn from the Canadian \[[@B26-ijerph-17-03948]\] and Australian \[[@B27-ijerph-17-03948]\] guidelines and the Dutch norm \[[@B28-ijerph-17-03948]\]. These countries were chosen for the following reasons: (1) The Netherlands is considered as the country with the most experience and development for protecting soil contamination, their norm being the most used in different parts of the world that lack reference values \[[@B29-ijerph-17-03948],[@B30-ijerph-17-03948],[@B31-ijerph-17-03948]\]; (2) Canadian standard values have been used as a reference for soil quality in Chile \[[@B9-ijerph-17-03948],[@B12-ijerph-17-03948]\]; (3) Australia has climatic and geological characteristics quite similar to Chile \[[@B32-ijerph-17-03948]\]. This phase provided a record of tailings whose heavy metal concentration exceeded the values of the three international reference guidelines. Phase 2: The tailings identified in Phase 1 were assessed according to two of the priority criteria established in GMPGSPPC \[[@B12-ijerph-17-03948]\]. This phase allowed us to obtain a record of the tailings representing a higher risk for the health and safety of the population around them. A subrecord of the tailings identified in Phase 1 was obtained, also including criteria considered as a priority for Chile. Phase 3: The potential risk of deposits showing heavy metal concentrations over reference values, which were also considered as critical in Phase 2, according to the evaluation of risk for health and safety, was estimated by using the methodology by Hakanson \[[@B18-ijerph-17-03948]\], who defined mathematical equations and parameters to estimate contaminant-potential and the ecological risk of different heavy metals. The results allowed us to obtain a subrecord from the one obtained in Phase 2, which identifies Chilean tailings requiring urgent intervention. The methodology followed for each phase shown in [Figure 1](#ijerph-17-03948-f001){ref-type="fig"} is described below. ### 2.3.1. Phase 1: Comparison of Heavy Metal Concentrations, Using International Reference Values {#sec2dot3dot1-ijerph-17-03948} #### Reference Frameworks The geochemical data of tailings were compared with the standard values of soils used by the population, established by the Canada Soil Quality Guidelines \[[@B26-ijerph-17-03948]\], the Australian guidelines \[[@B27-ijerph-17-03948]\], and the standards established by the Ministry of Infrastructure and Water Management (MIWM) in the Dutch Soil Regulation Circular \[[@B28-ijerph-17-03948]\]. - Canadian and Australia Guidelines Given the great variability of reference concentration values provided by the Canadian and Australian guidelines, the first parametrization was made by using the equation by Esquenazi et al. \[[@B9-ijerph-17-03948]\]:$$C^{*} = {sign}\left( C \right) \cdot \log\left\lbrack {1 + {{abs}\ }\left( C \right)} \right\rbrack$$ where C is the concentration of the element or substance, and C\* is the corresponding parameterized value. The Canadian and Australian parameterized values for industrial use are shown in [Table 1](#ijerph-17-03948-t001){ref-type="table"}. The experimental values of heavy metal concentrations corresponding to the 530 tailings assessed were also parameterized with Equation (1). Then, they were compared with the reference values in [Table 1](#ijerph-17-03948-t001){ref-type="table"}. - Dutch norm The Dutch norm is based on an algebraic formula that allows the adaptation of its use, depending on soil nature. Its parameters include the standard intervention value (SIV), which depends on two parameters characteristic of the soil: the percentage of organic material weight and the percentage of clay weight. Although this norm does not have a legal value in Chile, it has been the reference most frequently used by SERNAGEOMIN so far, the same as in other countries \[[@B33-ijerph-17-03948],[@B34-ijerph-17-03948],[@B35-ijerph-17-03948],[@B36-ijerph-17-03948]\]. Since the SERNAGEOMIN geochemical database does not include the measurement of clay percentage nor the percentage of organic material, the Dutch norm adapted to mine tailings by Esquenazi et al. \[[@B9-ijerph-17-03948]\] was used. This methodology allows the definition of risk zones for the population and/or the environment by means of three criteria: (1) intervention required, (2) conditional intervention required, and (3) no intervention required \[[@B9-ijerph-17-03948]\]. The soil intervention value (SIV) is defined by Equation (2):$${SIV} = {{SSIV}\ } \cdot \ \frac{A + {B\ } \cdot {\ x}_{C} + {\ B\ } \cdot x_{OM}}{A + 25\ \cdot B + 10\ \cdot C}$$ where A, B, and C are the specific parameters of each metal, and $x_{OM}$ and $x_{C}$ are the percentage values of organic material and clay weight, respectively. [Table 2](#ijerph-17-03948-t002){ref-type="table"} shows the values of A, B, and C for the eight heavy metals to be assessed, assuming 25% clay and 10% organic material. SSIV is the intervention value of a standard soil for residential use, having the following values: As = 27 mg kg^−1^; Cd = 1.2 mg kg^−1^; Hg = 0.83 mg kg^−1^; Pb = 2.10 mg kg^−1^; Co = 35 mg kg^−1^; Cu = 54 mg kg^−1^; Ni = 34 mg kg^−1^ and Zn = 200 mg kg^−1^ \[[@B28-ijerph-17-03948]\]. Adjusted threshold values were estimated for each heavy metal ($C_{F}$) by applying Equation (3) \[[@B9-ijerph-17-03948],[@B37-ijerph-17-03948]\]. $$C_{F} = \frac{SIV}{SSIV} \cdot \left( \frac{A + 25 \cdot B + 10 \cdot C}{B} \right) - \frac{\left( {A + 2C} \right)}{B}$$ To simplify the analysis, $C_{F}$ values were parametrized according to the equation by Esquenazi et al. \[[@B9-ijerph-17-03948]\], defined as $${AC}_{F} = {sign}\left( C_{F} \right) \cdot \log\left( {1 + {abs}\left( C_{F} \right)} \right)$$ If ${AC}_{F} < 0$, no intervention required.If ${AC}_{F} > 2$, intervention required.If $0 < {AC}_{F} < 2$, intervention depends on the values of other parameters.${AC}_{F}$ are the reference values of the Dutch norm. #### Intervention Requirements Due to Contamination The heavy metal concentration values of the 530 tailings under assessment were parameterized by using Equation (1). These were compared with the Canadian and Australian reference values ([Table 1](#ijerph-17-03948-t001){ref-type="table"}) and the values obtained from the Dutch norm equations (Equations (2)--(4)). As a result of the three referential frameworks, the "no intervention required" and "intervention required" criteria were obtained. Additionally, in the case of the adapted Dutch methodology, the "conditional intervention" criterion was obtained. Its conditionality is subjected to the availability of the percentage values of organic material and clay in the tailings. These data are not available from the SERNAGEOMIN geochemical database. The selection of deposits requiring urgent intervention depends on how their concentrations exceed the admissibility indicated by the Canadian and Australian guidelines and the Dutch norm. In this study, the number of metals classified as "intervention required" by the three international reference values was established as an indicator of the initial tailings prioritization. The priority criteria determined in this study were the following: low priority (0--2 metals requiring intervention), medium priority (3--4 metals requiring intervention), and high priority (5 or more metals requiring intervention). ### 2.3.2. Phase 2: Prioritization Criteria, According to the Methodological Guidelines for Soils with the Potential Presence of Contaminants {#sec2dot3dot2-ijerph-17-03948} Once the tailings were classified according to the intervention prioritization described above, a geospatial tool, QGIS, was used and added to the geographic data layers present in Chilean databases to determine their location. To prioritize soils with potential presence of contaminants (SPPC), two out the four criteria indicated by the methodological guidelines approved by the Chilean Ministry of Health to manage them were used \[[@B10-ijerph-17-03948]\]. The objective of these guidelines is to prioritize SPPC, as follows: Distance from tailings to a populated area: less than 2 km---"high priority", 2--3 km---"medium priority", and greater than 3 km---"low priority". Closeness to water bodies: If tailings are close to water bodies, they are considered as "high priority". If not, they are considered as "low priority". [Table 3](#ijerph-17-03948-t003){ref-type="table"} shows the criteria for classifying Chilean tailings priority. The prioritization criterion "extreme" was applied to tailings with two or more red boxes, without disregarding the case of one red box, which also requires "high priority". ### 2.3.3. Phase 3: Potential Risk Estimation {#sec2dot3dot3-ijerph-17-03948} Lastly, the potential environmental risk of tailings identified as the most critical in the previous phases was estimated. To estimate potential tailings, Hakanson's methodology \[[@B18-ijerph-17-03948]\] was used. He defined mathematical equations and parameters to estimate potential contaminants and the ecological risk factors of various heavy metals. Hakanson proposed the "Potential Ecological Risk Index" (PERI) as a quick tool for environmental assessment, resulting in the classification of contamination areas and the identification of toxic substances of interest. PERI provides a simple quantitative method for assessing the ecological risk potential of a contamination situation \[[@B38-ijerph-17-03948]\]. The equations used are the following \[[@B39-ijerph-17-03948]\]:$$C_{f}^{i} = C^{i}/C_{n}^{i}$$ where $C_{f}^{i}$ is the contaminant factor of heavy metals "i", $C^{i}$ is heavy metal concentration, and $C_{n}^{i}$ represents the heavy metal concentration in a nearby area without anthropogenic intervention (background level). $E_{r}^{i}$ is the ecological risk factor defined as $$E_{r}^{i} = T_{r}^{i}{\ast C}_{f}^{i}$$ $T_{r}^{i}$ is the heavy metal toxicity factor. The values of the heavy metal toxicity factors, $T_{r}^{i}$, were Hg (40), Cd (30), As (10), Cu (5), Pb (5), Cr (2), Zn (1), and Ni (6) \[[@B14-ijerph-17-03948],[@B40-ijerph-17-03948]\]. On the basis of contaminant and ecological risk factors, a potential ecological risk index, RI, was determined through the following addition:$${RI} = \sum\limits_{i = 1}^{n}E_{r}^{i}$$ Equation (7) allows us to estimate the potential ecological risk, considering the criteria in [Table 4](#ijerph-17-03948-t004){ref-type="table"}. \[[@B41-ijerph-17-03948]\]. 3. Results and Discussion {#sec3-ijerph-17-03948} ========================= 3.1. Comparison of Intervention Requirements {#sec3dot1-ijerph-17-03948} -------------------------------------------- There is a great discrepancy between the criteria of the three references, although all of them consider the soil for "residential" use, thus showing the need to have soil background values. These are scarcely measured in Chile, being a basic disadvantage for developing soil norms in the country. [Table 5](#ijerph-17-03948-t005){ref-type="table"} shows the intervention requirements for the 530 tailings assessed, according to the three reference frameworks. Owing to the lack of data about the mass percentage of organic material and clay, there is a great number of tailings classified as "uncertain" (conditional intervention) according to the Dutch norm, thus making assessment difficult. Metals most frequently presenting this characteristic were Cr, Cd, and Ni, the percentage of "uncertain" being 73.4%, 77.9%, and 63.0%, respectively. The metals from the dams classified as "uncertain", showing the smallest number of "uncertainties", were Cu, Hg, Hg, and As, with 11.1%, 10.8%, 13.0%, and 13.0%, respectively. [Table 5](#ijerph-17-03948-t005){ref-type="table"} also shows that Cu is the metal requiring the most intervention, according to the three reference guidelines. Based on the reference values of the Dutch, Canadian, and Australian soils, 86.8%, 98.3%, and 65.5% of the dams require intervention, respectively. The Australian reference values are the most demanding for As, Cd, and Hg, all of them being cancerogenic toxic substances, as shown by the great number of tailings requiring intervention due to the presence of these metals. Since Chile and Australia share similar climates, types of soil, and mining, these values must be considered. A comparison was made among the results obtained for the three guidelines. Coincidences resulting from "intervention required" and "no intervention required" for the heavy metals assessed are shown in [Figure 2](#ijerph-17-03948-f002){ref-type="fig"}. In comparing requirements, there was a good agreement between The Netherlands and Australia (87.0%), while the agreement between The Netherlands and Canada amounted to only 26.0% for this metal. Concerning Cu, 67.2% agreement was found between the three guidelines, while the agreement between The Netherlands and Canada reached 88.5%. As to Cr, Cd, and Ni, the three guidelines show significant discrepancies, Cd being the most remarkable because only 1.1% of the 530 tailings agree with this criterion. The greatest agreement was found for Hg; that is, 85.7% of the criteria coincided for the three reference frameworks. A good agreement was also found for Pb and Zn; that is, 64.9% and 63.8%, respectively. [Table 5](#ijerph-17-03948-t005){ref-type="table"} shows a great number of "uncertain" criteria concerning the Dutch norm, thus decreasing the number of comparisons. In the case of Cr, Cd, and Ni, there are 73.4%, 77.9%, and 63.0% tailings that fit into the category of "uncertain" for these metals, respectively, making the application of the proposed methodology difficult. In addition, the number of tailings requiring intervention was analyzed for the number of heavy metals exceeding maximum values, according to the three guidelines. As a result, 304 tailings required the intervention of one heavy metal; 117 tailings, two heavy metals; 77 tailings, three heavy metals; 28 tailings, four heavy metals; 4 tailings, five heavy metals, while none of the tailings required the intervention of six or more heavy metals. The four tailings exceeding the three norms by five heavy metals are considered as "high priority". They were identified by their IDQ geochemical data numbers. One of them is abandoned, while the other three are inactive. All of them contain a number of heavy metals exceeding the reference values of the three guidelines under assessment, that is, five heavy metals. The tailings size ranges from 600 to 1,875,000 t. Three of them are located in Sierra Gorda commune (Antofagasta Region) and one in Río Hurtado commune (Coquimbo Region). [Table 6](#ijerph-17-03948-t006){ref-type="table"} shows their characteristics. In applying the priority criteria mentioned above to the 530 tailings, 32 of them were classified as "high priority"; 195 as "medium priority"; and 303 as "low priority" for heavy metals. Tailings presenting the highest risks are located in Coquimbo, Atacama, and Antofagasta regions, an obvious fact since the country's largest mining activity is concentrated in the Chilean northern zone. [Figure 3](#ijerph-17-03948-f003){ref-type="fig"} shows the tailings' regional distribution. 3.2. Chilean Methodological Guideline Application {#sec3dot2-ijerph-17-03948} ------------------------------------------------- To apply the Chilean guidelines, it is first necessary to establish the location of each tailing. To do this, QGIS was used to geolocate them. This was compared with the layers available in the country, considering regions, communes, rivers, lakes, and populated areas. For geospatial data, a 2-km area of influence was created around the tailings based on data from Census 2012 (Chile, 2012) for the different locations. [Figure 4](#ijerph-17-03948-f004){ref-type="fig"} shows a Coquimbo region map as an example to indicate tailings and populated areas. [Figure 4](#ijerph-17-03948-f004){ref-type="fig"} shows the tailings area of influence of about 2 km around populated areas. An analysis of the 530 tailings revealed that 195 and 290 of them are located at 2 and 3 km from a populated area, respectively. In addition, there are 154 tailings located near rivers, lakes, and estuaries. Their distribution is shown in [Figure 5](#ijerph-17-03948-f005){ref-type="fig"}. The assessment showed that there are 32 tailings classified as "high priority" due to heavy metal concentration, 195 due to closeness to populated areas, and 154 for closeness to water bodies. [Figure 6](#ijerph-17-03948-f006){ref-type="fig"} shows the results of the 530 tailings assessed, according to the number of heavy metals exceeding maximum values and closeness to populated areas by region. [Figure 7](#ijerph-17-03948-f007){ref-type="fig"} shows the results considering heavy metals and closeness to water bodies. [Figure 6](#ijerph-17-03948-f006){ref-type="fig"} shows tailings classified as "extreme"; that is, they present two critical characteristics: (1) the number of heavy metals exceeding maximum reference values is 5 or more, while (2) closeness to populated areas is smaller than 2 km. On the contrary, [Figure 7](#ijerph-17-03948-f007){ref-type="fig"} does not show any "extreme" tailings, indicating that, at the most, the heavy metals exceeding maximum values are 4. 3.3. Tailings Prioritized as "Extreme" {#sec3dot3-ijerph-17-03948} -------------------------------------- By considering heavy metal concentrations and closeness to populated areas and water bodies, tailings requiring special treatment were determined according to the following: (a) They show a high level of heavy metal contamination, (b) they are located within a 2-km radius close to a populated area, and/or (c) they are located within a water body area of influence. Twelve tailings classified as "extreme" were found, all of them located in the Illapel commune (Coquimbo Region). Five heavy metals exceed reference concentration values and are located at less than 2 km from populated areas and water bodies. 3.4. Ecological Risk Analysis {#sec3dot4-ijerph-17-03948} ----------------------------- Hakanson's methodology \[[@B14-ijerph-17-03948]\] was used to determine metals that are environmentally risky. To do this, soil background values are needed (Cu 134.7 g t^−1^; Cr 20.66 g t^−1^; Ni 50.68 g t^−1^; Zn 99.27 g t^−1^; Pb 7.881 g t^−1^; As 2.30 g t^−1^; Cd 1.508 g t^−1^; Hg 0.14 g t^−1^). The analysis using Hakanson's methodology revealed that the five deposits identified as Anta Colla 1, Anta Colla 2, California 0, Esperanza Dos, and NN3 show high potential ecological risk indexes (RIs) greater than 320 and, therefore, are qualified as "showing extremely high ecological risk", according to the criteria in [Table 3](#ijerph-17-03948-t003){ref-type="table"}. [Figure 8](#ijerph-17-03948-f008){ref-type="fig"} shows the location of four of them. As shown in [Figure 8](#ijerph-17-03948-f008){ref-type="fig"}, California 0 and Anta Colla 2 deposits are located close to populated areas. Hence, exposure to them cannot be controlled because their location makes it impossible to keep people away from the sites, the same for NN3 and Anta Colla 1 tailings located in the area. According to Illapel climatological data, rainfall is very low in summer months, while in winter (June and July), the average rainfall is 50 mm. Rainfall is a risk because metals may infiltrate the soil and contaminate nearby or underground water if the soil is permeable, and there is no appropriate insulation. Soils on the lower area of the Choapa River basin are mainly characterized by highly permeable soils and limited capacity for agricultural use, while in the upper area, the soils have characteristics appropriate for culturing fruit trees. The city of Illapel is located at upper Choapa, so its soil is less permeable. [Table 7](#ijerph-17-03948-t007){ref-type="table"} shows the potential ecological risk index values and the ecological risk factor of each heavy metal per tailings, revealing that the heavy metal with the highest ecological risk factor is As, the lowest being Zn, considering the average of the 12 critical deposits. According to Hakanson's ecological risk factor, the level of contamination was low for three of the twelve tailings assessed, that is, Cr, Ni, and Zn. In Anta Colla 2, the level of contamination was medium for Cu, low for Hg, very high for Cd, and extremely high for Pb and As, the latter being recognized for their high toxicity. In Anta Colla 1, the level of contamination was medium for Cd, high for Cu and Hg, and extremely high for Pb and As, the same as above. In California 0, the level of contamination was high for Cu and Cd, very high for Pb, and extremely high for As and Hg. In California 2B, the level of contamination was low for Cu, Pb, and Hg, medium for As, and very high for Cd. In El Arenal, San Jorge 1-2-3, and Tailings Dams 0, 1, and 2, the level of contamination was medium for As and Cu. In NN3 and Pluma de Oro, the level of contamination was low for Cu and medium for As. In Esperanza Dos, the level of contamination was high for Cu and medium for As. In Esperanza Dos and NN3, the level of contamination was extremely high for Hg, while in El Arenal, San Jorge 1-2-3, and Tailings Dams 1 and 2, the level of contamination was medium for Hg. In California 2B, Esperanza Dos, NN3, and San Jorge 1-2-3, the level of contamination was very high for Cd. In El Arenal, Pluma de Oro, and Tailings Dams 0, 1, and 2, the level of contamination was high. Five tailings fit into the category of "extremely high potential ecological risk", that is, Anta Colla 1, Anta Colla 2, California 0, Esperanza Dos, and NN3. The other seven tailings fit into the category of "very high potential ecological risk". These results indicate that the 12 tailings require intervention for their ecological and environmental risks. However, the order of priority is Anta Colla 2, Anta Colla 1, California 0, NN3, and Esperanza Dos. 4. Conclusions {#sec4-ijerph-17-03948} ============== This paper deals with a methodology to prioritize Chilean tailings according to their potential ecological and environmental risks. The application of this methodology shows that five tailings require urgent intervention in Chile; the heavy metals showing the greatest environmental risk being As, Cd, Pb, and Hg, which are recognized for their high toxicity. The comparison of the three reference frameworks showed diverse criteria, thus making the analysis difficult. Only the tailings that do not require intervention, according to the Dutch, Canadian, and Australian reference frameworks, were discarded because they do not involve potential ecological and environmental risks. This allowed us to reduce the costs associated with a more thorough assessment. On the other hand, the application of Chilean criteria, although not compulsory, suggests Chilean soil conditions. Finally, this study reveals the need to count on background soil concentration values and also measurements of clay and organic materials for the tailings. These data will allow a clearer picture of the real risks associated with Chilean tailings. The authors are thankful to "Rapid Implementation Projects---Pilot Scaling", Universidad Católica del Norte, Engineering Project 2030, 16ENl2-71940, for the financial support of this project. Conceptualization, E.J.L. and N.F.G.; methodology, E.J.L., I.L.M., D.A.P., and N.F.G.; validation, N.F.G. and R.J.R.; formal analysis, E.J.L. and R.J.R.; investigation, E.J.L., I.L.M., F.A.Á., R.J.R., D.A.P., and N.F.G.; data curation, F.A.Á. and D.A.P.; writing---original draft preparation, E.J.L.; writing---review and editing, E.J.L.; visualization, F.A.Á.; project administration, E.J.L. All authors have read and agreed to the published version of the manuscript. This research was funded by "Rapid Implementation Projects---Pilot Scaling", Universidad Católica del Norte, Engineering Project 2030, 16ENl2-71940. The authors declare no conflict of interest. {#ijerph-17-03948-f001} {#ijerph-17-03948-f002} {#ijerph-17-03948-f003} {#ijerph-17-03948-f004} {#ijerph-17-03948-f005} {#ijerph-17-03948-f006} {#ijerph-17-03948-f007} ![Tailings showing extremely high risk, according to Hakanson's methodology \[[@B14-ijerph-17-03948]\].](ijerph-17-03948-g008){#ijerph-17-03948-f008} ijerph-17-03948-t001_Table 1 ###### Canadian \[[@B26-ijerph-17-03948]\] and Australian \[[@B27-ijerph-17-03948]\] soil quality reference values. Parameter Canada Australia ----------- ----------- ----------- ------ -------- g ton^−1^ As 12 1.1139 100 2.0043 Cd 10 1.0414 20 1.3222 Cr 64 1.8129 100 2.0043 Cu 63 1.8062 1000 3.0004 Hg 6.6 0.8808 15 1.2041 Ni 45 1.6628 600 2.7789 Pb 140 2.1492 300 2.4786 Zn 250 2.3997 7000 3.8452 ^1^ Soils for residential use. ^2^ Parameterization with Equation (1). ijerph-17-03948-t002_Table 2 ###### Reference values for the SIV calculation of each element, according to the Dutch norm \[[@B28-ijerph-17-03948]\]. Element A B C --------- ----- -------- -------- As 15 0.4 0.4 Cd 0.4 0.007 0.021 Hg 0.2 0.0034 0.0017 Pb 50 1 1 Ni 10 1 0 Zn 50 3 1.5 Cu 15 0.6 0.6 Cr 50 2 0 ijerph-17-03948-t003_Table 3 ###### Criteria for classifying tailings priority requirements. Prioritized Intervention Requirements Closeness to Communities Metal Concentration over Reference Values ^1^ Closeness to Water Bodies --------------------------------------- -------------------------- ----------------------------------------------- --------------------------- -- -- -- -- -- **High** **Medium** **Low** ^1^ Applying the three international referential frameworks, red: requires hight priority, yellow: requires medium priority and green requires low priority. ijerph-17-03948-t004_Table 4 ###### Adjusted grading standard of potential ecological risk of heavy metals in soil \[[@B41-ijerph-17-03948]\]. $\mathbf{E}_{\mathbf{r}}^{\mathbf{i}}$ Contamination Level RI Potential Ecological Risk ---------------------------------------- --------------------- --------------------- --------------------------- $E_{r}^{i} < 30$ Low RI \< 40 Low $30\ \leq E_{r}^{i} < 60$ Middle $40 \leq RI < 80$ Middle $60\ \leq E_{r}^{i} < 120$ High $80 \leq RI < 160$ High $120\ \leq E_{r}^{i} < 240$ Very high $160 \leq RI < 320$ Very high $240\ \leq E_{r}^{i}$ Extremely high $320 \leq RI$ Extremely high ijerph-17-03948-t005_Table 5 ###### Results of the Dutch, Australian, and Canadian soil quality reference guidelines application to 530 Chilean tailings. Reference Requirements As Cu Cr Cd Ni Pb Hg Zn -------------------------- ----------------------- ----- ----- ----- ----- ----- ----- ----- ----- Dutch norm Intervention required 138 460 34 1 0 89 24 55 No intervention required 323 11 107 116 196 372 449 325 Conditional intervention 69 59 389 413 334 69 57 150 Canadian guidelines Intervention required 0 521 238 37 324 255 29 181 No intervention required 530 9 292 493 206 275 501 349 Australian guidelines Intervention required 163 347 30 525 0 171 525 13 No intervention required 367 183 500 5 530 359 5 517 ijerph-17-03948-t006_Table 6 ###### Characteristics of the most critical tailings from a risk viewpoint, according to Dutch, Canadian, and Australian reference frameworks. IDQ Resource Mass (t) State Region Commune ------ ---------- ----------- ----------- ------------- -------------- 1609 Cu 1,875,000 Inactive Antofagasta Sierra Gorda 1665 Cu 600 Abandoned Antofagasta Sierra Gorda 1639 Cu 14,080 Inactive Antofagasta Sierra Gorda 949 Cu-Au 40,005 Inactive Coquimbo Río Hurtado ijerph-17-03948-t007_Table 7 ###### Average heavy metal concentration values and potential ecological risk index for Illapel critical tailings. Identification ^1^ Ecological Risk Factor $\mathbf{\ E_{r}^{i}}$ Potential Ecological Risk Index -------------------- ----------------------------------------------- --------------------------------- ----- ----- ------- -------- ------- ------- ------ Anta Colla 2 53.4 5.6 9.2 0.7 460.4 1702.2 137.2 21.9 2391 Anta Colla 1 64.9 5.5 8.9 0.7 442.6 1607.9 33.3 106.7 2271 California 0 116.5 12.3 8.6 1.1 186.9 469.4 67.7 1280 2142 California 2B 11.5 6.9 7.5 0.5 24.1 40.0 135.8 24.8 251 El Arenal 42.3 5.4 7.9 2.5 60.1 40.0 77.8 48.6 285 Esperanza Dos 81.6 8.7 9.0 0.6 78.7 40.0 123.0 244.3 586 NN 3 26.3 5.2 7.5 0.6 69.1 40.0 132.7 445.7 727 Pluma de oro 7.3 9.7 7.9 0.7 26.6 40.0 80.7 69.5 242 San Jorge 1-2-3 31.8 7.3 6.7 0.9 22.6 40.0 123.3 32.4 265 Tailings dam 0 35.0 11.3 9.6 1.1 50.4 40.0 70.6 27.9 246 Tailings dam 1 45.3 5.6 7.1 1.0 31.5 40.0 103.3 40.0 274 Tailings dam 2 35.2 6.0 7.6 1.0 40.4 40.0 100.1 32.4 263 ^1^ Identification according to National Geology and Mining Service (SERNAGEOMIN) \[[@B4-ijerph-17-03948]\].

Introduction {#Sec1} ============ This project was undertaken on the behalf of the Italian Association of Poland Syndrome (AISP), a non-profit organization supporting clinical, social and scientific activities for the benefit of persons affected by Poland syndrome (PS), with the aim of providing best-practices recommendations for the management of affected individuals. This work is based on peer-reviewed scientific literature intermingled with the personal experience of the authors who are familiar with PS and come from different areas of expertise. The consensus was reached through a slightly modified Delphi process \[[@CR1]\], which allows agreement based on guide practice when published evidence is lacking. People living with PS usually face a diagnostic "odyssey" undergoing multiple radiological, clinical and genetic tests, which are not always suitable. Published materials and personal experiences were here discussed: the provided recommendations should be considered a consensus based on the opinion of clinical experts with years of experience on PS, as further levels of evidence are not always available in the literature. People with PS typically miss one pectoralis major muscle \[[@CR2]\]. In most affected individuals, underdevelopment selectively affects the sternocostal head of the affected pectoralis major muscle with consequent asymmetry of the chest. In some cases, additional omolateral muscles of the chest, shoulder and upper limb, may be missing or underdeveloped. Bones of the rib cage can be also affected with shortened/missing/supranumerary ribs and fused/malformed/supranumerary vertebrae, potentially leading to more severe thoracic deformations and respiratory distress. Mammary gland and nipple hypo/aplasia also occur along with sparse or ectopic omolateral axillary hair \[[@CR3]\]. A subset of PS patients also have hand malformations affecting the same side and commonly including brachydactyly with or without syndactyly of the central fingers \[[@CR4], [@CR5]\]. Additional upper limb anomalies of PS include mild shortening of forearm bones and mild underdevelopment of the arm muscles; but these features are not easily detected unless accurate measurements are carried out. In the absence of bony involvement, musculoskeletal manifestations of PS usually have trivial or minor health implications and, therefore, can go unnoticed for years. By contrast, severely affected individuals with multiple muscle and bony abnormalities of the chest, upper limb or both are recognized at birth. In a minority of the cases, PS combines with extra-musculoskeletal congenital defects such as lung or kidney malformations and dextrocardia \[[@CR6]\]. Rarely, chest and/or upper limb abnormalities of the PS type occur bilaterally \[[@CR7]\]. Although bilaterality might be an underestimated feature of PS, the existence of a PS variant symmetrically affecting both sides of the thorax is still a matter of debate \[[@CR8]\]. Main text {#Sec2} ========= Etiology and pathogenesis of PS {#Sec3} ------------------------------- To date, the etiology and pathogenesis of PS are still unknown. Most PS cases are sporadic. This evidence together with the striking asymmetry characterizing the typical patients are arguments against a mendelian etiology for PS. Nevertheless, a genetic etiology cannot be completely excluded due to the existence of familial cases, as well as the possibility of genetic mosaicism potentially explaining also the sporadic ones \[[@CR9], [@CR10]\]. Indeed, sporadic cases could be also explnained by the intervention of genetic and environmental factors according to a multifactorial poligenic model. Concerning the pathogenesis of PS, there are many available hypotheses in literature. The most commonly agreed one is that dysembryological process leading to PS is caused by a vascular defect of the subclavian artery (which nurtures the developing scapulo-humeral girdle region) around the 45th day of embryonic life. This causes an early insufficiency of inflow to the distal limb and to the breastplate, eventually leading to the main musculoskeletal features of PS. According to this model, the vascular insult might be secondary to a variety of environmental and constitutional factors which remain undefined \[[@CR11]\]. Alternatively, and considering PS as a polygenic trait, it may be due to the involvement of genes regulating embryonic development, in particular of pectoral girdle, whose variants could be transmitted to the patient from healthy parents and be at the origin of PS. Methods {#Sec4} ======= Consensus decision making to reach agreement among participants {#Sec5} --------------------------------------------------------------- A working group of 35 experts has been set up, including clinical and molecular geneticists, pediatricians, plastic, thoracic and hand surgeons, psychologists, radiologists and physical therapists, to formulate best practice recommendations for PS based on the consensus reached through the Delphi process, which has been recognized as a valid approach to ensure relative anonymity and economy. Three experts (MC, IB, LM and MT) have identified five main areas to be explored: 1) Diagnosis, 2) Surgical treatments and specific medical approaches on the major symptoms, 3) Plastic surgery and rehabilitation, 4) Psychological issues and social assistance, 5) Clinical follow-up and general management. Each participant was asked to provide provisional recommendations (*N* = 103) based on their personal experience and bibliographic knowledge: the PubMed/MEDLINE electronic databases were searched for suitable articles from 1990 to 2019. All PS synonyms were searched in MeSH terms, title and abstract. Subsequently, the panel members independently evaluated the initial recommendations proposed regarding the topics listed above in an online survey (100% response rate): the participants were asked in two subsequent rounds of surveys whether they agreed, disagreed (they were also asked to provide a reason) or were unable to comment on any of them. The findings resulted in three possible grade levels: i) certainly useful/strong literature, ii) possibly useful/modest literature, or iii) Good Clinical Practice (GCP) but no relevant literature available. The recommendations were accepted if ≥50% of the experts agreed. For the last step, the recommendations approved by the previous surveys were discussed in a face-to-face consensus meeting held on October 2019 and hosted in the Policlinico Hospital in Modena (number of experts participating: 21). Disagreements were resolved through discussion or asking for the opinion of a third expert. Although consensus was reached, comments were considered to better clarify the recommendations. An expert without the right to vote (MC) facilitated the entire process. Results {#Sec6} ======= Diagnosis {#Sec7} --------- Unilateral agenesis or hypoplasia of the (sternocostal head of the) pectoralis major muscle is currently considered pathognomonic of PS (Table [1](#Tab1){ref-type="table"}) that is still exclusively a clinical diagnosis (for the additional features, refer to the Principal diagnostic criteria section). Table 1Recommendations for diagnosis of PS (major complication)GradeConsensus agreementR1.1Major complications are related to the severity of thoracic and upper limb defects; there are different complications, such as functional and structural limitations of compromised thoracic region and of upper limb.Definitely useful/strong literature86,7%R1.2Aesthetic problems can determine psychological serious issues of patients and their parents. Psychological issues may be of different severity according to the gender and the education level.Definitely useful/strong literature100%R1.3After the diagnosis of PS, US is indicated to exclude intra-abdominal, renal and heart structural anomalies.Possibly useful/modest literature100%Fig. 1Transverse over the sternum. Probe placed transversely on the sternum (s) with the insertion of sterno-costal component of pectoralis major muscle (\*) on both side of the manubrium and body of sternum, providing immediate information regarding any asymmetry of this component. In order to check the lower part of sterno-costal component is necessary moving with the probe distally to the sixth costal cartilage (c) and the first fibers of rectus abdominis (arrow) ### Clinical presentation {#Sec8} The clinical diagnosis of PS may be suspected in neonatal/pediatric age in front of patients who present i) asymmetric hypoplasia of the pectoral muscles and any abnormalities of the thoracic cage, or ii) anomalies affecting the pectoral muscles unilaterally. Patients are rarely reported with bilateral pectoralis major muscle agenesis/hypoplasia. In adulthood, patients with unexplained muscle hypoplasia may be diagnosed with PS, especially if associated with asymmetric of the thoracic cage, anomalies of the mammary region (hypoplasia or agenesis of the breast and of the unilateral nipple) and abnormalities of the hand. ### Clinical diagnosis {#Sec9} Since a gene(s) responsible for PS has not yet been identified, the diagnosis of PS is clinical and based on the recognition of the characteristic recurrent pattern of features along with an appropriate differential diagnosis. Agenesis or hypoplasia of the pectoralis major muscle is currently the cardinal feature mandatory for the diagnosis (Table [2](#Tab2){ref-type="table"}). The pectoral muscle anomaly is generally easily observed by asking the patient to push the palms of the hands against each other with the arms positioned in front of the body \[[@CR12]\]. Additional commonly reported features include: Absence or hypoplasia of other chest muscles: small pectoral muscle, anterior tightened muscle, gran dorsal muscle, deltoid muscleAnomalies of the thoracic cage: agenesis or hypoplasia of one or more costal segments, pectus carinatum, excavatum, clavicular hypoplasia, pulmonary herniationAbnormalities of the mammary region: agenesis or hypoplasia of the breast, areola and nippleAbnormalities of the upper limb and shoulderAbsence or asymmetric reduction of axillary hairsOther associated skeletal anomalies: Sprengel anomaly (congenital elevation of the scapula secondary to absence of the upper portion of the serratus anterior muscle), radioulnar synostosis, emivertebre, vertebral fusions.Table 2Recommendations for diagnosis of PS (principal diagnostic criteria)GradeConsensus agreementR2.4The mandatory feature of PS is the agenesis or hypoplasia of the pectoralis major muscle (the sterno-costal head is always affected). In most cases, PS is unilateral. Presumed bilateral PS needs a more extensive differential diagnosis. Additional diagnostic criteria are hypo/aplasia of the omolateral mammary gland and nipples, and malformations of the omolateral upper limb (limited to or more severely affecting the central rays).Definitely useful/strong literature93,8%R2.5The diagnosis is made through the physical examination of the patient; an ultrasound of the pectoralis muscles is important but not strictly necessary for the diagnosisDefinitely useful/strong literature92,9%R2.6The sterno-costal head of the pectoralis major muscle is involved in most frequently; the other heads of the pectoralis major muscle and the pectoralis minor muscle are involved in different percentages of patientsDefinitely useful/strong literature93,3%R2.7The latissimus dorsi muscle may be involved too in a minority of casesDefinitely useful/strong literature92,3%R2.8Many variable phenotypical characteristics can be associated but we cannot diagnose PS in the absence of the basic diagnostic criterionDefinitely useful/strong literature92,9%R2.9The concurrence of rare internal organ malformations, such as kidney agenesis or destrocardia, may ease prenatal detection, but also in these cases, the underlying PS is recognized postnatallyDefinitely useful/strong literature71,4%Fig. 2Sagittal parasternal with the upper edge of the probe on the clavicle. Probe on the sagittal axis of the clavicle, immediately lateral of sterno-clavicular joint: at the upper edge of probe there is the clavicle (cl) where clavicular component of pectoral major (p1) takes origin; a cleavage plane (arrow head) with the sterno-costal component (p2) is visible. Starting from the midportion of the clavicle and continuing laterally, the subclavius muscle (sb) is identified underneath the bone, parallel to the long axis of clavicule, crossed below by axillary artery (\*) and vein (collapsed in the image) and brachial plexus cord They are also described as associated malformations: genitourinary malformationscardiac malformationshepatic and/or biliary tract malformations. ### Radiological examination {#Sec10} Ultrasound (US) represents the first line diagnostic imaging method to support or confirm the clinical diagnosis of PS, due to its easy accessibility, cost-effective and radiation free modality. An accurate description of soft tissue and chest wall morphological abnormalities, not always easy to identify during clinical assessment, is of outstanding importance since to different types of malformation correspond a different therapeutic approach. Evaluation by US is therefore recommended (Table [3](#Tab3){ref-type="table"}) to assess pectoralis muscle (major and minor), breast characteristics of the two halves of the thorax, and could be used as screening exam for cardiac and renal abnormalities \[[@CR13]\]. Associated rib abnormalities are detectable by US but, when suspected, imaging work-up should be completed with other diagnostic modalities described later. In case of specific clinical concern, by means of US performed by radiologists with expertise in musculoskeletal pathology, is possible to examine complex dysmorphism of distal upper extremities: i.e. mapping of hand muscle could address surgery and rehabilitation, moreover in pediatric patient less suitable for MRI scanning. Table 3Recommendations for radiological examinationGradeConsensus agreementR3.10Evaluation by ultrasound is recommended to assess pectoralis major, subcutaneous tissue and breast characteristics of the two halves of the thorax. US examination, because of its more availability, not radiation exposure for the patients and cost-effectiveness, should be the first line (and often the only one) imaging tool in order to confirm the clinical suspect of PS (agenesia or hypoplasia of the pectoralis major muscle) and to assess the severity of anomalies.Definitely useful/strong literature92,3%R3.11Evaluation by CT (without contrast medium) is recommended only for abnormalities of the rib cage that require surgery in non-adult patientsPossibly useful/modest literature100%R3.12Pre-natal suspect of PS is sometimes possible as a collateral finding in routine US morphological examination if there are severe bone manifestations involving hands or rib cage, but it has to be confirmed with clinical examination after birth. On the basis of the current state of knowledge no further radiological analysis are indicated in uterus if parents are affected by the syndrome.GCP (no literature available)100%R3.13Radiological workup includes: ultrasound of the pectoralis region, chest X-ray, cardiac evaluation with echocardiography, abdominal US, other examinations on the basis of the specific phenotypeDefinitely useful/strong literature100%R3.14Ultrasound is able to categorize pectoralis major abnormalities in three classes (i.e. i, total agenesis of the muscle; ii, agenesis of the sternocostal part with normal costoclavicular component; iii, partial agenesis of the sternocostal part with normal costoclavicular component) as well as to recognize regional anomalies affecting the pectoralis minor and regional vesselsDefinitely useful/strong literature90,9%R3.15MR imaging has limited indications, particularly when ultrasound is non-conclusivePossibly useful/modest literature77,8%R3.16Chest radiography should be obtained routinely for gross evaluation of the rib cage and the heartDefinitely useful/strong literature88,9%R3.17Imaging is not always necessary for the diagnosis of PS, however it may be helpful for the surgical planningDefinitely useful/strong literature77,8%R3.18CT or MRI often more clearly depict the absence of the pectoralis major muscle and allows better appreciations of other nearby associated musculoskeletal anomalies but should not be indicated in a routine radiologic evaluation in a primary diagnostic phaseDefinitely useful/strong literature88,9%R3.19US examination should be performed with an high frequency probe, with a musculoskeletal preset, and should be bilateral and comparative, detecting all the 3 heads of the pectoralis major muscle (clavicular, sternocostal and abdominal) with sagittal and trasverse scans, the pectoralis minor muscle and the mammary glandDefinitely useful/strong literature100%R3.20X-rays of the thorax or of the ribs are not specific for PS and not often necessary in diagnostic phase but can help showing associated malformation of the rib cagePossibly useful/modest literature90,0%R3.21Only in few selected patients should be considered a complete radiologic study of the skeletal and muscles of hands, forearms, upper arms and/or scapulas with X-rays or CT (other than MRI) in order to better detect complex anomalies clinically evident and define a pre-surgical assessment.GCP (no literature available)62,5%R3.22Chest X-Ray if there is clinical suspicion of rib agenesis. CT scan if severe deformity of the rib cage is observed; cardiac and renal US evaluations could be performed to exclude cardiac or renal anomalies.Possibly useful/modest literature100%R3.23In complex deformities of the chest wall, CT scan may provide a more detailed depiction of bone anomalies and vascular relationships. Multi-imaging evaluation may be needed in case of hand deformities to support clinical assessment.Definitely useful/strong literature81,8%Fig. 3Transverse over the coracoid immediately inferior to the clavicle and medial to gleno-humeral joint. Probe placed transversely over the coracoid (\*\*), found immediately inferior to the clavicle and medial to gleno-humeral joint; on the medial side of the coracoid is possible to appreciate the pectoralis minor muscle (pm). Superficial to the coracoid is located the anterior head of deltoid (da), to differentiate from clavicular component of pectoralis major (p) identifying the cephalic vein in between them (\*): pectoralis major muscle intramuscular tendon (arrow) US examination should be performed with an high frequency probe, with a musculoskeletal preset, and should be bilateral and comparative, detecting all the 3 heads of the pectoralis major muscle (clavicular, sternocostal and abdominal) with sagittal and transverse scans, the pectoralis minor muscle and the mammary gland. Thereby US allows to classify different pectoralis major abnormalities into three morphologic groups: i, total agenesis of the muscle; ii, agenesis of the sterno-costal part with normal clavicular component; iii, partial agenesis of the sterno-costal part with normal costoclavicular component. Moreover US is able to recognize regional anomalies affecting the pectoralis minor and regional vessels \[[@CR14]\]. For evaluation of pectoralis major and minor muscle, this "4 view" standard assessment is proposed (Table [4](#Tab4){ref-type="table"}, Figs. [1,](#Fig1){ref-type="fig"}[2](#Fig2){ref-type="fig"}, [3](#Fig3){ref-type="fig"}, and [4](#Fig4){ref-type="fig"}). In Prenatal setting, US screening is strongly limited by the difficulties in appreciating such a thoracic asymmetry. Only severe bone manifestations involving hands or rib cage might be identified on occasion of US morphological examination, but fetal movements often reduce the detection rate especially in low-risk pregnancies. Any prenatal findings is aspecific and need to be confirmed with clinical examination after birth \[[@CR15]\]. On the basis of the current state of knowledge no further radiological analysis are indicated in uterus if parents are affected by the syndrome. A complete radiological workup as well as US of the pectoralis region, echocardiography and abdominal ultrasound, inckudes also X ray of the thorax if rib cage abnormalities are clinically suspected \[[@CR16]\]. In selected case of severe chest malformation, thoracic CT and /or MRI are appropriate for a thorough pre-surgical assessment. In particular CT scan with contrast may provide a more detailed depiction of bone anomalies and vascular relationships. Only in few selected patients should be considered a complete radiologic study of the skeletal and muscles of hands, forearms, upper arms and/or scapula with X-rays or CT (other than MRI) in order to better detect complex anomalies clinically evident and define a pre-surgical planning \[[@CR17], [@CR18]\]. Wide consensus is reached among experts about the "second line" role of CT and MRI in PS diagnostic-therapeutic process. Both CT and MRI clearly depict the absence of the pectoralis major muscle and allows better appreciations of other nearby associated musculoskeletal and inner organs anomalies but, considering radiation exposure of CT, necessity for anesthesia in younger pediatric patient who undergoes MRI other than lower cost-effectiveness, should not be indicated as first approach. In setting of pre-surgical planning, MRI can provide the most of necessary information without exposing patient, often in young age, to the risk of radiations. Regarding thoracic MRI, in order to obtain a highly defined picture of a structure in constant movement (breathing related), specific sequences and adjustments are required. Table 4"4 view" standard assessment for evaluation of pectoralis major and minor muscleProbe PositionDescriptionTransverse over the sternumMoving in cranio-caudal direction to demonstrate the sternal component of the pectoralis major muscle over the sternocostal junctions and comparing to the opposite side in order to detect any asymmetry (Fig. [1](#Fig1){ref-type="fig"})Sagittal parasternal with the upper edge of the probe on the clavicle.Moving laterally to demonstrate the clavicular component of pectoralis major. Switching color Doppler on to demonstrate the position of cephalic vein as a landmark to distinguish the clavicular component of the pectoralis major from the anterior component of deltoid (Fig. [2](#Fig2){ref-type="fig"})Transverse over the coracoid immediately inferior to the clavicle and medial to gleno-humeral jointOn the medial side of the coracoid is possible to appreciate the tiny tendon by which pectoralis minor muscle takes origin. Moving caudally the muscle appears between the pectoralis major superficially and rib with interposed intercostal muscle on the depth (Fig. [3](#Fig3){ref-type="fig"})Transverse on the arm in external rotationMoving in cranio-caudal direction from the humeral head along the tendon of the long head of the biceps up to the myotendinous junction to demonstrate the overlying pectoralis major tendon and its insertion into the humerus (Fig. [4](#Fig4){ref-type="fig"})Fig. 4Transverse on the arm in external rotation. Probe placed transversely on the arm in external rotation: pectoralis major tendon (arrows) inserting on the lateral labrum of bicital groove; the tendon is located superficial to long head of the biceps (lb), short head of the biceps (sb) and coracobrachialis muscle (cb). Moving the probe medially keeping the same orientation, the pectoralis minor appears in between pectoralis major and thoracic wall, and two neurovascular bundles for pectoralis muscle are visualized (lateral pedicle white arrow head, medial pedicle black arrow head) ### Differential diagnoses {#Sec11} Literature is scanty on the differential diagnosis of PS. Until the preparation of this manuscript, 33 papers appeared in PubMed under the research strings \["Poland syndrome" AND "differential diagnosis"\] and \["Poland sequence" AND "differential diagnosis"\], or \["Poland's syndrome" AND "differential diagnosis"\] and \["Poland's sequence" AND "differential diagnosis"\]; but, among them, differential diagnosis of PS was discussed only in four of them \[[@CR19]--[@CR22]\]. Of these papers, two were dedicated to the radiological differential diagnosis of unilateral hyperlucent lung at standard radiographs \[[@CR20], [@CR21]\], one is a systematic review on controversies in PS \[[@CR22]\], and the remaining is a case report pointing out a possible overlap between PS and localized (thoracic) lipoatrophy \[[@CR19]\]. Accordingly, a satisfactory differential diagnosis of PS is carried out by the intersection of practitioner's experience on human malformation patterns and the "complexity" of the presenting phenotype. Scenarios of differential diagnosis of PS include: Unilateral PS without additional skeletal/limb anomalies;Unilateral PS with additional skeletal/limb anomalies;Bilateral PS;PS "plus" (i.e. PS with additional anomalies not regularly considered part of the PS phenotypic spectrum);Unilateral hyperlucent lung at thoracic X-rays. The diagnosis of PS is usually clear-cut in the "typical" patient presenting with isolated unilateral hypo/aplasia of the major pectoris muscle with or without hypo/aplasia of the overlaying mammary gland/areola/nipple (Table [5](#Tab5){ref-type="table"}). The most common differential diagnoses include thoracic asymmetry due to clinically significant thoracic scoliosis and/or anomalies of the bony structures (sternum, ribs, vertebrae), isolated congenital mammary gland/areola/nipple asymmetry (particularly, in women) or unilateral hypo/aplasia, and consequences of thoracic traumas/surgery. Additional disorders that may present cause localized hypo/atrophy of the thoracic soft tissues include localized lipoatrophy \[[@CR19]\], morphea or localized scleroderma, Becker nevus syndrome and Parry-Romberg syndrome (extending to the thoracic soft tissues). Finally, thoracic asymmetry might be also the result of unilateral diaphragmatic acquired or congenital abnormalities. Physical examination accurately assessing the soft tissues of the thoracic cage in combination with scrutiny of the past medical history and, in selected cases, radiological investigations support the practitioner in the differential diagnosis. In patients presenting with (usually) unilateral or asymmetric upper limb involvement, the identification of the concurrent omolateral pectoralis anomaly, which is considered necessary and pathognomonic of PS, separates patients with PS from those with partially overlapping upper limb malformations. Considering the pattern of upper limb involvement in PS, the most common differential diagnosis is ectrodactyly, which defines a developmental deficiency of the distal central rays of the limb resulting in hypo/aplasia of the central digits extending, in some patients, to the corresponding metapods. The typical presentation of hand ectrodactyly is a two- or three-finger hand with a cleft extending up to the carpal bones. Less frequently, the typical brachysyndactyly observed in PS might be confused with the more common transverse terminal limb deficiencies. In these cases, differential diagnosis is usually resolved at physical examination of the thoracic cage. Patients with bilateral PS are rare in clinical practice, although some authors speculated that, in PS, bilateral features should not surprise and could be underestimated in the current literature \[[@CR8]\]. In PS, bilateral presentation rises additional, though exceptional differential diagnoses mainly in the field of skeletal dysplasias leading to thoracic cage hypoplasia. Accordingly, one case with "bilateral" PS has been published, which has been subsequently reassigned to the diagnosis of "thoracic dysplasia" \[[@CR23], [@CR24]\]. In this scenario, physical examination by an expert in human malformation patterns and total-body skeletal radiograph allows differential diagnosis. Exceptionally, PS presents in combination with additional anomalies, which are nosologically separated from PS. Sprengel deformity (i.e. hypoplasia, and upper and medial dislocation of the scapula) has been recently recognized as the most common accompanying feature of PS \[[@CR25]\]. PS can also associate with other (usually) sporadic malformation patterns (with unknown or heterogeneous etiology) affecting the cephalic pole and/or the upper segments of the musculoskeletal system. In particular, PS combines recurrently with Moebius syndrome in the literature and this phenotype has been defined Poland-Moebius syndrome \[[@CR26]\]. In single reports, PS concurs with Adams-Oliver syndrome \[[@CR27]\], Klippel-Feil syndrome \[[@CR28]\], facio-auriculo-vertebral dysplasia \[[@CR29]\], and frontonasal dysplasia \[[@CR30]\]. Carey-Fineman-Ziter syndrome is an ultrarare congenital myopathy with marked facial weakness and additional features, also including PS \[[@CR31]\]. Carey-Fineman-Ziter syndrome is easily differentiated from the "simpler" PS by the unique pattern of presenting features. It has been recently associated with recessive variants in *MYMK* \[[@CR32]\]. Unilateral agenesis of the major pectoris muscle (with or without hypoaplasia of the overlaying mammary gland/areola/nipple) generates a difference in radiolucency of the lungs at standard X-rays. Such a radiological presentation is in common with a wide variety of acquired disorders and congenital conditions, whose extent is much beyond the scope of these recommendations. It is relevant, however, to note that, in addition to PS, many different pleural, lung parenchymal, pulmonary vasculature, central airways and mediastinal anomalies, as well as technical issues (e.g. patient rotation and lateral decentering) may cause unilateral hyperlucency of the lungs at standard X-rays \[[@CR20], [@CR21]\]. Table 5Recommendations for differential diagnosesGradeConsensus agreementR5.24In unilateral PS, differential diagnosis includes: (i) congenital or acquired thoracic soft-tissue anomalies (including isolated unilateral mammary gland/areola/nipple hypo/aplasia, localized lipoatrophy, morphea, Parry-Romberg syndrome, Becker nevus syndrome, surgery, traumas); (ii) asymmetry of the thoracic bony structures, due to thoracic scoliosis and/or bony anomalies; (iii) unilateral acquired or congenital defects of the diaphragm.Possibly useful/modest literature100%R5.25In bilateral PS, differential diagnosis should also consider skeletal dysplasias affecting the rib cages ("thoracic dysplasia").Possibly useful/modest literature77,8%R5.26In PS with upper limb anomalies, differential diagnosis should consider ectrodactyly, and, in a minor extent, transverse upper limb defects.Possibly useful/modest literature100%R5.27In complex phenotypes, it should be considered that PS commonly present with Sprengel deformity and less frequently with Moebius sequence (Poland-Moebius syndrome). Single patients combining PS with Adams-Oliver syndrome, Klippel-Feil sequence, facio-auriculo-vertebral dysplasia and frontonasal dysplasia have been also described. Finally, PS can be also part of the Carey-Fineman-Ziter, a congenital myopathy due to recessive variants in *MYMK*.Definitely useful/strong literature91,7% ### Communicating the diagnosis {#Sec12} The diagnosis of PS is arrived at in most cases within the first year of life \[[@CR33]\]. It is often communicated to parents who are unaware of the presence of the disorder \[[@CR34]\]. As for any newborn with congenital anomalies the diagnostic pathway and process should be communicated to the parents by explaining that the waiting time is necessary to receive the results of all requested investigations for an accurate evaluation (Table [6](#Tab6){ref-type="table"}). The diagnosis should be addressed to both parents, since they are the "care coordinator", a role that should be filled formally by a provider of the healthcare system \[[@CR35]\]. The information should be delivered gradually and adapted to the family's socio-cultural level. Once the medical team is confident in the diagnosis of PS, it is important to underline that PS is not progressive and that, in the absence of severe rib cage malformations, its survival rate is comparable with the general population. A normal psychomotor development is expected and the malformation does not affect the growth. Correct education allows the development of manual skills even in the carriers of severe malformations. The communication of the diagnosis should occur in the presence of a psychologist or if this is not possible, an external psychological help should be suggested to support families and patients in coping with the disease. The continuity of care must be guaranteed and, if the specialist surgeon is not available in the clinical team, the contacts of the most appropriate specialist (s) for surgery must be given to the patients/families during the diagnosis and the follow-up. They should be also informed about parent/patient organizations \[[@CR36]\] such as AISP. A minority of cases is diagnosed in adolescence or adulthood. Dissatisfaction with body shape can be a source of distress that may significantly affect the dimension of one's health. Puberty, in particular, is characterized by many sometimes dramatic physical, physiologic, biochemical and personality changes that occur when teens are particularly sensitive to the opinions of others \[[@CR37]\]. Table 6Recommendations for communicating the diagnosisGradeConsensus agreementR6.28In the communication of the diagnosis it is important to underline that PS is not progressive congenital anomaly and does not associate with abnormal psychomotor development.Definitely useful/strong literature92,9%R6.29In cases without severe thoracic malformation, the communication of the diagnosis should include a statement on the survival rate which is comparable with the general population.Definitely useful/strong literature100%R6.30As PS is an exclusion diagnosis, the parents and family should be informed on the need of waiting the results of all requested investigations before fixing the diagnosis.Definitely useful/strong literature84,6%R6.31The communication of the diagnosis should be made in the presence of a psychologist. If the psychological support is not available at the time of the diagnosis an external psychological should be advised.Definitely useful/strong literature90,9%R6.32Contacts of the most appropriate specialist(s) for surgery must be given to the patients/families during the diagnosis and the follow-up. They should be also informed about available local, national and international support groups and/or patients associationsDefinitely useful/strong literature92,9% ### Genetic counseling {#Sec13} Most cases are sporadic, affected children of unaffected parents with no family history of congenital anomalies. Rare familial cases are reported in literature with a rate ranging from 4.2% (PS families with recurrence of pectoral muscle defects) to 8.4% (Poland-like syndrome families with PS index case and ≥ 1 relative(s) showing normal pectoral muscles but with upper limb and/or thoracic anomalies common in PS) \[[@CR9]\]. In the presence of sporadic cases the risk of recurrence, i.e. the risk that the same couple has a second affected child, is low. For familial cases, the recurrence risk is higher than that of the general population and must be calculated in accordance to the possible inheritance pattern (Table [7](#Tab7){ref-type="table"}). It is therefore of fundamental importance to clinically evaluate the parents of affected children to exclude they could present mild, underdiagnosed forms that could therefore turn an apparently sporadic case into a familial case thus increasing the risk of recurrence. Genetic counseling is essential in all cases of PS and should be focused on informing patients that: the pathogenic mechanisms underlying PS are still unknown;most PS patients present in isolated form;most cases of PS are sporadic but about less than 10% show familial recurrence, with higher prevalence in males, which suggests a genetic, hereditary basis;to date, different models of transmission have been hypothesized, including autosomal recessive (recurrence risk of 25%) and autosomal dominant with incomplete penetrance (up to a 50% risk of recurrence);sporadic cases can be explained by the appearance of de novo mutations;non-recurrent genomic rearrangements have been occasionally associated with PS \[[@CR38], [@CR39]\].Table 7Recommendations for genetic counselingGradeConsensus agreementR7.33At the moment, PS is considered a sporadic disorder. Familial recurrence is really exceptional. Therefore, general counseling for sporadic cases in families with a previously affected child or for affected adults with negative family history should be reassuring concerning the chance of recurrence in a future pregnancy.Definitely useful/strong literature100%R7.34Genetic counseling in families with multiple affected individuals should prompt the revision of the clinical diagnosis. If the diagnosis is confirmed in multiple affected individuals, genetic counseling should consider specific Mendelian inheritance patterns or, perhaps, multifactorial inheritance.Definitely useful/strong literature87,5%R7.35Because the molecular basis of PS is mostly unknown a confirmatory molecular test applicable in prenatal diagnosis is not available and genetic counseling remains unsupported by laboratory tests.Definitely useful/strong literature100%R7.36Cytogenomic, molecular and exomic investigations should be limited to the more complex cases; no routine laboratory investigation is available for confirming the diagnosis of PS.Possibly useful/modest literature100%R7.37As PS is a congenital disorder, genetic counselling is recommended in all casesGCP (no literature available)100% Surgical treatments and specific medical therapies on the major symptoms {#Sec14} ------------------------------------------------------------------------ There is not any definitive solution to treat PS as well as any other genetically determined malformation syndrome. However, a surgical treatment of the major abnormalities and medical complications are appropriate and do not differ from that of other similar isolated malformations in the general population. ### Thoracic surgery {#Sec15} In the literature, there are no strict guidelines or indications regarding thoracic surgery in PS (Table [8](#Tab8){ref-type="table"}). In the experience of the group of pediatric and thoracic surgeons who are the authors of the present recommendations, thoracic surgery is not always necessary in PS \[[@CR40]\]. The indication to thoracic surgery are usually related to severe rib cage anomalies, either regarding anterior chest wall (pectus excavatum, carinatum or combination of both) or ribs (rib agenesis). In case of anterior chest wall defect, the indication to the surgery can be cosmetical or functional. If the heart or the lungs are compressed by a pectus excavatum, a retrosternal metallic bar (Nuss procedure) is helpful to lift the sternum up and releiving the visceral compression. In case of asymmetric anterior chest wall without heart/lung compression, cosmesis can be improved with cartilage resection or osteotomy or both. In young patients, with a compliant chest wall, conservative treatments as FMF brace or vacuum bell can be used. In case of rib agenesis, surgery aims to give more protection to the thoracic organs, although no reports of traumatic complications have been reported in PS patients to the best of our knowledge. Thoracoplasty for rib agenesis is performed also to abolish paradoxycal respiration and stabilize the lungs and rib cage during forced inspiration/exhalation cycle. In order to obtain the stabilization of rib cage, different surgical solutions can be chosen, mainly using non absorbable prosthetic mesh or metallic prosthesis. Goretex mesh can be used successfully especially if the rib defect is not too wide, usually in case of absence of one or two ribs. If the defect involves three or more ribs, it is better to reinforce the rib cage by using a metallic prosthesis. In recent years, 3D printed metallic prostheses have been successfully implanted in PS patients. Goretex mesh is usually used above the metallic prosthesis to avoid skin erosion. It is usually better to correct a severe chest wall anomaly (if present) before breast/pectoral reconstruction, to improve the final result. In some cases, chest wall anomaly can be corrected at the same time of plastic surgery first step. The combined approach has the advantage of optimizing skin incision, reducing the surgical procedures, the number of hospitalizations and the dyscomfort for the patient. Regarding the best age for surgery, it is usually not necessary to operate the patients before 12-13 years of age. Ideally, thoracic reconstruction around 14-15 years of age when the thoracic growth is already advanced is our preferred time. Table 8Recommendations for thoracic surgeryGradeConsensus agreementR8.38Respiratory symptoms are not common in PS patients. Lack of protection of lungs and other thoracic organs due to the rib cage defect (rib agenesis) does not indicate per se thoracoplasty during childhood.Definitely useful/strong literature57,1%R8.39It is better to avoid non resorbable materials before 12 years of age.Possibly useful/modest literature80,0%R8.40Conservative methods (vacuum bell, FMF or corset for pectus carinatum) are promising tools to treat pectus excavatum and carinatum associated with PS in young patients.Definitely useful/strong literature100%R8.41The management of pectus excavatum and pectus carinatum should be evaluated for each case and can be carried out through conservative strategies (vacuum bell, braces) or interventional ones (Nuss procedure, surgical treatment of pectus carinatum), although surgery in absence of respiratory symptoms should be postponed at least until the beginning of adolescence, towards the completion of the growth of the thoracic wallDefinitely useful/strong literature57,1%R8.42PS can be classified in minimal (only pectoral defect), partial (thoracic or upper arm variant) and complete formDefinitely useful/strong literature88,9%R8.43TBN classification is useful to classify the thoracic defect in PSDefinitely useful/strong literature100%R8.44Early evaluations of patients optimizes the treatment and is better for psychological reasons.Possibly useful/modest literature100%R8.45In selected cases, 3D Printing and new technologies can be helpful to build prosteses custom made for thoracic reconstruction in PSDefinitely useful/strong literature100%R8.46Combined surgical treatment (thoracic and plastic surgery) can reduce the number of surgical procedures.Definitely useful/strong literature100% ### Hand surgery {#Sec16} Hand anomalies in PS can show many forms of clinical presentation varying from a hypoplasic but functional hand to severe grade of symbrachydactyly with one or more absent fingers (Table [9](#Tab9){ref-type="table"}). Moreover, there are a group of patients which are not affected by hand malformation, presenting the only absence of pectoral muscles. Clinical examination should not be limited to the hand but the entire upper limb must be checked because of the chance of malformations affecting the wrist, the elbow and the shoulder. Classifying the anomalies of the hand represents an important step which has a direct correlation with the choice of treatment; many classifications have been proposed but the majority were focused exclusively on the hand missing the likelihood of an involvement of other district of the upper limb. In 2012, a new system has been published and it can be considered more useful because it takes into consideration the entire upper limb \[[@CR5]\]. Eight types of clinical aspects have been considered: I:Absence of hand/upper limb anomaliesII:Hypoplastic hand without morphologic and functional anomaliesIII:Symbrachydactyly with 5 functional fingers and possible morphologic anomalies of the phalanges and partial range of motionIV:Symbrachydactyly with some functional fingersV:Symbrachydactyly with absent or nonfunctioning fingersVI:Classic hand anomalies of PS with proximal radioulnar synostosisVII:Classic hand anomalies of PS with congenital high scapulaVIII:Other associated anomaliesTable 9Recommendations for hand surgeryGradeConsensus agreementR9.47It is mandatory to check all the upper limb in order to identify any malformations (shoulder, elbow) associated with hand anomaliesDefinitely useful/strong literature100%R9.48Reconstructive planning should be adapted to the type of deformity of the handDefinitely useful/strong literature100%R9.49The correction of syndactyly should begin between 12 and 24 months of life; if the first web space is involved, surgery should be performed between 6 and 12 months.Definitely useful/strong literature100%R9.50If phalanx are absent, two options should be proposed to the parents: microvascular digital transfer from the foot or non-microvascular free phalangeal transfer from the footDefinitely useful/strong literature100%R9.51The patient must be followed until the end of skeletal growth because recurrence of syndactyly, secondary to scar hypetrophy, may be possibleDefinitely useful/strong literature100%R9.52When a recurrence occurs, it should be corrected during adolescence in order to reach a definitive resultDefinitely useful/strong literature100%R9.53We recommend to use the following classification (useful for treatment therapeutic of hand function) of Hand and Upper Limb anomalies in PS:I Absence of hand/upper limb anomaliesII Hypoplastic hand without morphologic and functional anomaliesIII Symbrachydactyly with 5 functional fingers and possible morphologic anomalies of the phalangesand partial range of motion (ROM)IV Symbrachydactyly with some functional fingersV Symbrachydactyly with absent or nonfunctioning fingersVI Classic hand anomalies of PS with proximal radioulnar synostosisVII Classic hand anomalies of PS with congenital high scapulaVIII Other associated anomaliesDefinitely useful/strong literature100%R9.54Types I and II (R2.11) do not need any surgical treatments, which, however, is necessary for type III and, in particular, for types IV and V, to improve hand functionDefinitely useful/strong literature100%R9.55The reconstruction of the second and third webspaces can be delayed until 18 months of age without adverse effect on hand function or fine motor developmentDefinitely useful/strong literature100%R9.56Early surgery is recommended for border digits as syndactyly between digits of disparate length may result in flexion contracture or angular deformity.Definitely useful/strong literature100%R9.57Minor syndactyly, such as observed in PS, can be treated by the usual methods of local enlargement plasty of the first commissure: trident plasty (YV double Z), Z plasty at four tattersDefinitely useful/strong literature100%R9.58The use HA scaffold for skin regeneration in syndactyly release surgery in young children represent a valid alternative to the use of skin graftsDefinitely useful/strong literature100%R9.59The first wound care is recommended after 3 weeks post-surgeryDefinitely useful/strong literature66,7% Reconstructive planning should be discussed with the parents, adapting it to the type of deformity. Children with normal hand (type I) or with hypoplasic but functional hand (type II) do not need any surgical treatment which is necessary instead for type III and in particular for types IV and V in order the improve hand function. The issues the surgeon must deal with are the presence of syndactyly and the absence of phalanges (typically the middle phalanx) or entire fingers. Timing of surgery depends on the type of malformation. The correction of syndactyly should begin between 12 and 24 months of life, however, in case of syndactyly between finger of disparate length, early surgery is recommend between 6 and 12 months in order to reduce the risk of flexion contracture or angular deformities \[[@CR41]\]. The reconstruction of the second and third web spaces can be delayed until 18 months of age without adverse effect on hand function development. Surgical techniques do not differ from those adopted in other forms of syndactyly by means of local flaps (VY -- Double Z -- Z plasty at four tatters) to reconstruction the web space as well as the digital surfaces. Surgeons may need to use a skin graft for the uncovered areas; the volar surface of the wrist or the groin are the most common donor site. However, the use of hyaluronic acid scaffold for skin regeneration in syndactyly release surgery may be take into consideration as an alternative to the use of skin graft, due to the good results reported in young patients \[[@CR42]\]. After surgery an occlusive bandage should be applied and the first wound care is recommended after 2 or 3 weeks. However, some patient may need an early dressing in case of infection, exudate or insufficient care of the dressing by the parents. In case of symbrachydactyly with phalanx absence, two options may be proposed to the parents, according to the grade of malformation: i) Non microvascular free phalangeal transfer from the foot, or 22) Microvascular digital transfer from the foot. The first option should be proposed in case of symbrachydactyly with the presence of the proximal phalanx in order to increase the length of the single ray. Conversely, microvascular transfer is the only option in case of a peromelic hand or in the rare situation of a congenital metacarpal amputation. However, parents are often frightened from the idea of using the foot as donor site as well as from the risk of failure of a microsurgical transfer whereas they usually accept better the proposal of a free phalangeal transfer also considering the lesser functional and aesthetic result. Finally, physicians must follow the patient until the end of skeletal of skeletal growth, because recurrence of syndactyly, secondary to scar hypertrophy, may be possible. When a recurrence occurs, it should be corrected, when possible, during adolescence in order to reach a definitive result. ### Plastic surgery {#Sec17} Reconstructive surgery in PS has a mainly aesthetic meaning and it aims to improve quality of life (Table [10](#Tab10){ref-type="table"}). The asymmetry involving the breasts, in fact, is often moderate or severe and it is made even more complex by the underlying thoracic malformations. Social aspects in case of bodily malformation cannot be ignored, both for male and female patients. A single defect, in fact, can affect the overall physical appearance up to create a variable degree of uneasiness in social relations \[[@CR43]\]. Patients affected by PS seem to experience high level of body uneasiness during adolescence \[[@CR37]\]. The reconstructive path, for this reason, can start during puberty and it depends strictly on the characteristics of the mammary / thoracic malformation, on the anatomical characteristics of the patient and on the psychological condition linked to the malformation. First of all it is important to classify the malformation: TNB classification is the most recent specific classification of thoracic, breast and nipple malformations which guides the surgeon in choosing the surgical path \[[@CR40], [@CR44]\]. A customized approach is therefore necessary even in relation to patient's age, expectations and needs \[[@CR43]\]. Patients suffering of PS can contact the surgeon at any age depending on the time of diagnosis. Specialist interviews in the neonatal period and/or before puberty are only intended to reassure caregivers. Reconstructive plan can benefit from the modern breast reconstructive surgery techniques and should be the least invasive and debilitating for the patient given the main aesthetic purpose of the reconstruction and the musculoskeletal malformative characteristics that often affect the chest. Muscle transpositions should be used in selected cases after an accurate evaluation of the pros and cons of the procedure based on the anatomical characteristics (back, shoulder, posture, ...) and life habits (sport activity, work activity, hobbies, ...) of the patient. Muscle transpositions shouldn't be used in non-adult patients, or even in patients who have not fully completed their psycho-physical development and who have not clearly outlined their social and working life habits \[[@CR33]\]. Reconstructive surgery can help especially younger patients (teenagers) to experience the aesthetic stigmata of the disease less traumatically, accompanying them on a path of growth and positive development of their body image. Soft tissue reconstructive surgery cannot be definitive; breast, muscles and adipose tissue involved in the malformative feature are tissues in continuous evolution based on age, physiological changes (pregnancy) and life habits and, for this reason, it may be necessary to rework them over time to maintain the desired aesthetic result. Table 10Recommendations for plastic surgeryGradeConsensus agreementR10.60Reconstructive plan should be the least invasive and debilitating for the patient given the main aesthetic purpose of the reconstructionDefinitely useful/strong literature90,0%R10.61Autologous fat graft should be the first surgical procedure but it is strictly dependent to the grade of deformities, BMI index and chest wall involvement.Definitely useful/strong literature62,5%R10.62Breast implants are the simplest solution to obtain missing breast volumesDefinitely useful/strong literature100%R10.63The contralateral breast should be reworked as little as possible, especially in young nulliparous patientsPossibly useful/modest literature80%R10.64Skin expansion should be planned if side-affected nipple dislocation exceeds 2 cm (N2 in TNB classification) or if the side-affected breast is absent (N3,B2 in TNB classification).Definitely useful/strong literature75%R10.65Muscle transpositions should be used in strictly selected cases after an accurate evaluation of the pros and cons of the procedure based on the anatomical characteristics (back, shoulder, posture, ...) and life habits (sport activity, work activity, hobbies, ...) of the patient.Possibly useful/modest literature100%R10.66Muscle transpositions shouldn't be used in non-adult patients, or even in patients who have not fully completed their psycho-physical development and who have not clearly outlined their social and working life habits.Definitely useful/strong literature100% The rehabilitation plan must be scheduled after each intervention on the basis of illness-specific criteria and specific surgical procedures (Table [11](#Tab11){ref-type="table"}). Table 11Recommendations for physical therapiesGradeConsensus agreementR11.67It's necessary to monitor vertebral column and thoracic symmetry during the growthDefinitely useful/strong literature100%R11.68To evaluate scapulo-thoraco-humeral dynamics is recommendedPossibly useful/modest literature100%R11.69To evaluate symmetry of development of the upper limb muscular masses is recommendedPossibly useful/modest literature100%R11.70Pysiatrist visit to assess the feasibility of a reconstructive intervention of transposition of the gran dorsal muscle is recommendedGCP (no literature available)66,7%R11.72To evaluate use of upper limb gestures is recommendedGCP (no literature available)100%R11.73The following evaluations to correct aesthetic/functional balance are highly recommended and must be supportive in surgery decision:a) to monitor the step of psycho-motor development in the upper limb use,b) to evaluate the active and passive range of motion if are different,c) to evaluate the single prehensile movements age-related and to observe the preferred patterns of usage,d) to measure the pinch and grip strengthPossibly useful/modest literature100% Psychological issues and social assistance {#Sec18} ------------------------------------------ Currently, there is no scientific literature on psychological support for patients, and their families, suffering from PS (Table [12](#Tab12){ref-type="table"}). In contrast, clinical practice highlights the importance of psychological support for the patient and his family. Furthermore, the psychologist appears to be an important resource also within the multidisciplinary specialist team both for drafting the protocols for taking charge \[[@CR45]\] and for supporting colleagues during the various stages of the diagnostic, care and assistance process \[[@CR46]\]. Since the syndrome is increasingly diagnosed at birth, psychological support must first be directed to parents to elaborate painful experiences that could affect the attachment and care of the child and thus reinforce parenting skills. Subsequently, as the child reaches more and more autonomy, the support is necessary both for the parents and for the subject himself to avoid developing feelings of inadequacy and consequent socialization problems. About the choice to undergo surgery, especially in adolescence, it is useful that the decision is reached with the help of the psychologist, after assessing the subject's psychological conditions, in full awareness and self-determination so that the result is satisfactory \[[@CR47]\]. It is not recommended to take the intervention exclusively on the advice of medical specialists. When the diagnosis is late, in adolescent or adult age, the psychodiagnostic evaluation is strongly recommended to check how the anomalies caused by the syndrome have affected the subject's life \[[@CR48]\], self-image and relationships with others \[[@CR49]\]. The request for correction of anomalies must be carefully considered and examined in depth to exclude unrealistic and unrealizable expectations \[[@CR50]\]. Table 12Recommendations for psychological supportGradeConsensus agreementR12.74To perform a psychological evaluation before facing the reconstructive intervention related to thoracic and/or mammary deformityGCP (no literature available)100%R12.75Psychological support is fundamental since the diagnosis of the PS to avoid the neurosis onsetGCP (no literature available)100%R12.76To elaborate the diagnosis and to reinforce parental capacitiesGCP (no literature available)100%R12.77We recommend a preliminary assessment of psychological condition in adolescent patients, to guide self- consciousness and knowledge of individual needs in order to reach self -determination about surgical operationGCP (no literature available)100%R12.78We recommend previous assessment of psychological condition in adult patients, to help on accepting the condition and to face up the associate consequences, including surgical treatment or therapyGCP (no literature available)100%R12.79The psychologist is an important resource for the entire multidisciplinary team during the diagnostic, care and assistance process, helping to build the best path for the specific situation (for example, for the communication of the diagnosis)Definitely useful/strong literature100% Clinical follow-up and general management {#Sec19} ----------------------------------------- Following the diagnostic classification, people affected with PS are followed by periodic follow-up visits to manage the main complications of the disease, such as functional and aesthetic issues (abnormalities of limbs, paroxysmal movements of the chest wall, reduced lung function, muscle weakness, scoliosis). Pediatric and plastic surgery examination for thoracic abnormalities (ribs, condrosternal cartilages, pectoral muscles, mammary gland) involves periodic clinical examinations and surgical correction in the most complex cases (Table [13](#Tab13){ref-type="table"}). Table 13Recommendations for clinical follow-upGradeConsensus agreementR13.80Thoracic surgery in pediatric age has to be planned and performed by pediatric surgeons.Definitely useful/strong literature100%R13.81The role of the pediatric thoracic surgeon includes: evaluation of thoracic symmetry, assessment of ribs anomalies, evaluation of the sternum, long-term follow-up, possible surgical treatment.Definitely useful/strong literature100%R13.82The following evaluations have to be planned for each newly diagnosed patient \< 18 yrs. of age: Pediatric thoracic surgeon clinical evaluation, Chest x-ray, Cardiac evaluation with cardiac US, Genetic counseling, Orthopedic evaluation, Plastic surgeon evaluation for pre-adolescents and teenagers, Abdominal ultrasound.Definitely useful/strong literature87,5%R13.83We recommend for child patients a preliminary assessment of psychological condition, to guide the adequate development of body image and to prevent inferiority feelings.GCP (no literature available)100% The orthopedic evaluation and the hand surgery visit are indicated for the management of postural and upper limb anomalies, with a focus in the hand function. Periodical follow-up and physiotherapy in the most complex cases are also recommended (Table [14](#Tab14){ref-type="table"}). Since in (rare) PS cases are also described for the presence of genitourinary (renal agenesis, renal ectopia, vesico-ureteral reflux and cryptorchidism) and cardiac (dextrocardia and inter-atrial defect) malformations it is advised to perform relevant specialist assessments when appropriate. Regarding risks and potential complications such as breast implant-associated anaplastic large cell lymphoma (BIA-ALCL) in case of breast reconstruction with textured breast implant device, periodic clinical examination is recommended, as one would do to screen for implant complications such as capsular contracture. Table 14Recommendations for clinical follow-upGradeConsensus agreementR14.84Annual follow-up in case of surgery, especially in case of reconstruction with breast or pectoral implant (medical examination, ultrasound) is recommendedDefinitely useful/strong literature100%R14.85Provide adequate explanations about the need to perform more or less invasive surgical adjustments to maintain the symmetry between the two hemilates in relation to the physiological changes of the bodyGCP (no literature available)83,3%R14.86The patient with PS need to be assisted by a multidisciplinary team (coordinated by a Case Manager) tailored on the basis of the real needs of the patient/family. In general the team should involve the following specialists: Pediatric/thoracic surgeon, Plastic surgeon, Orthopedic surgeon, Hand surgeon, Radiologist, Geneticist, Psychologist, Cardiologist, Ophthalmologist, and other professionals as neededDefinitely useful/strong literature100%R14.87A strong relationship with Patients Advocacy Organizations, both national and international ones, is crucial for the best care of patients with PSDefinitely useful/strong literature100%R14.88If there is not functional limitations there is no need of surgery but patients could decide to undergo surgery for aesthetic reasons.Definitely useful/strong literature70% Final remarks {#Sec20} ------------- The development of best standards of care and commonly approved procedures is urgently needed for all rare diseases ([www.rarebestpractices.eu](http://www.rarebestpractices.eu)). In fact, to advance knowledge of PS the primary goal is defining evidence-based guidelines for those affected and their caregivers (Table [15](#Tab15){ref-type="table"}). The present working group has formulated a total of 91 recommendations for the diagnosis and management of PS, based on a systematic review of literature and a consensus procedure. In total, 69 recommendations within 14 general principles, ranging from diagnosis to therapeutic approaches have been accepted with \> 90% agreement among experts. When summarizing the degree of evidences available for the recommendations proposed in this review, some are "Good Clinical Practice" based on clinical experience of the authors. Key topics include assessment of the rib cage and other associated skeletal anomalies or abnormalities. Careful monitoring of the status and well-being of patients by an experienced multidisciplinary team and validated scores for the progression and follow-up are required in order to perform a structured assessment of the outcomes, as are long-term follow-up studies to clarify the risks of complications of therapeutic approaches. Given the rarity of PS, international collaboration is essential to provide specific support even in geographic areas lacking expert specialists. Table 15General recommendationsGradeConsensus agreementR15.89Our knowledge of epidemiology of PS should be improvedDefinitely useful/strong literature100%R15.90The precise cause of PS is not known yet: further studies are urgent to find the reasons why PS occurs but research should focus into the etiopathogenesis of PSDefinitely useful/strong literature100%R15.91Standardization of protocols on a national and international basis is neededDefinitely useful/strong literature100% Conclusions {#Sec21} =========== To conclude, this initiative is based on the personal opinion of experts based on the best available evidence and provides recommendations for the options of diagnosis and treatment of PS patients, in order to improve the outcome for those affected. It will now be important to extend the discussion and the acceptability of our recommendations to a wider community of clinicians and families: a large international meeting on PS is expected to be scheduled in 2021 under the direct management of the AISP. In fact, one of the main purposes of the AISP is to improve the care and follow-up of people living with PS. For this reason the AISP has recently developed, in close collaboration with some clinical centers for the treatment of PS in Italy, the first registry for PS available in the world. The registry of PS is a tool for the systematic collection of personal and phenotypic data of affected people. Through the digitalization of high-quality clinical data, the registry aims to improve knowledge of PS and maximize natural history and epidemiologic studies. The point of view of patients and caregivers {#Sec22} -------------------------------------------- Purpose of AISP has always been to enhance the network of specialists and medical facilities who collaborates each other in order to allow the best possible care for patients on the Italian territory. Thanks to the drafting of this document, AISP did one important step closer to reach this aim and hope to provide clinicians and patients diagnostic and therapeutic approaches in PS as uniform as possible. AISP will surely advertise this report on its website and its social channels. This document will be shared with other European associations of PS patients and rare disease associations in Italy and in Europe and with the ERN of rare diseases. With the help of the clinicians involved in the draft, the document will be distributed to scientific society, medical structures, general practitioners and pediatricians, asking them to collaborate in its sharing. Moreover an informative brochure will be produced in order to enable patients to fully understand the document. The update of these recommendations will be promoted by AISP as part of its constant scientific activity and will be an integral part of the scientific program of AISP National Conferences. PS : Poland Syndrome AISP : Associazione Italiana Sindrome di Poland GCP : Good Clinical Practice FMF : Freire Martinez Ferro TBN : Thoracic/Breast/Nipple-areola complex ERN : European References Network MRI : Magnetic Resonance Imaging CT : Computed tomography **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors are grateful to families for their cooperation and are also indebted to Ms. Clare Groarke for revising the manuscript. This work has been promoted within the European Reference Network on Rare Bone Disorders (ERN-BOND). The first version of the manuscript was drafted by IB, LM, MB, NC, MT, LS, and MC. All participants had contributed to proposing, scoring and discussing the recommendations in detail. All authors read and approved the final version. Any research funds was being made available. Data sharing not applicable as no datasets were generated during the current study. Not applicable. Individual Informed Consent was obtained where appropriated. The authors declare that they have no competing interests.

Introduction {#sec1_1} ============ Current research demonstrates that pterygium is a proliferative disease with hyperplastic growth of the corneoconjunctival fibrovascular tissue onto the cornea \[[@B1]\]. Surgical therapy can be used to successfully manage pterygia; however, recurrence remains a problem \[[@B2]\]. To reduce recurrence, surgical management may include autoconjunctival grafting \[[@B3]\], lamellar keratoplasty \[[@B4]\], amniotic membrane transplantation \[[@B5]\], and mitomycin C (MMC) \[[@B6]\]. Autologous conjunctival grafting has been shown to be the best technique, as it has low recurrence and adverse events rates \[[@B7]\]. Technically easier and therefore widely applied with primary pterygium excision is MMC. MMC is an antibiotic isolated from *Streptomyces caespitosis*, which interferes with RNA, DNA and protein synthesis. The topical use to prevent pterygium recurrence was first described in 1963 in Japan \[[@B8]\]. Though significantly reducing recurrence rates, MMC can cause scleral necrosis, secondary glaucoma, iritis, cataract, corneal edema, corneal perforation, and endophthalmitis \[[@B9]\]. The overall incidence of scleral necrosis after pterygium surgery has been reported as 0.2-4.5% with higher rates being linked to the use of MMC \[[@B10]\]. Scleral thinning can occur as early as one week postoperatively or even years after the use of topical MMC and is independent of intra- or postoperative delivery \[[@B11]\]. The management of surgically induced scleral necrosis associated with MMC post-pterygium is complex and shows variable success. The purpose of this report is to present a case of a serious complication after intraoperative MMC administration in order to illustrate a successful management strategy: the use of surgical repair by lamellar keratoplasty and amniotic membrane along with the control of ocular surface inflammation. Case Report {#sec1_2} =========== A 57-year-old male was referred to the Department of Ophthalmology at the University Hospital of Zurich, Zurich, Switzerland, for the treatment of progressive painful peripheral corneal melting in the left eye two weeks after pterygium surgery with intraoperative MMC. Primary excision had been performed at another institution, using a bare sclera technique combined with application of 0.2 mg/ml MMC to the scleral bed for 2 min. Primary closure of the conjunctiva had been achieved by a continuous suture with Vicryl® 8-0 (Ethicon Inc., Blue Ash, Ohio, USA). The postoperative management consisted of Tobradex® eye drops tid (3 mg tobramycin and 1 mg dexamethasone/1 ml; Alcon Laboratories Inc., Fort Worth, Tex., USA). After one week, the patient was changed to Floxal SDU® eye drops (Ofloxacin; Bausch and Lomb, Rochester, N.Y., USA) and 5 g vitamin A® ointment tid (vitamin A; Bausch and Lomb). At first presentation, his best corrected visual acuity (OD +1.75 sph/øcyl; OS +1.25/øcyl) was 20/20 in the right eye and 20/60 in the left eye. The patient reported he had strictly adhered to the prescribed postoperative regimen and denied any other medical conditions, including any rheumatic disease. On slit-lamp examination, the left eye showed a nasal crescent-shaped area of corneal ulceration measuring 2.5 × 2.0 mm reaching Descemet\'s membrane with adjacent severe conjunctival and scleral injection (**fig. [1](#F1){ref-type="fig"}**). Anterior chamber assessment showed concomitant mild cellular and fibrin reaction. His corneal smears, looking for evidence of a bacterial, viral and/or fungal infection, were negative. Complete blood count, electrolytes and inflammatory markers were also within normal ranges. In particular, immunological testing did not reveal any pathological elevation of ANA or ANCA. The patient was admitted for pulse corticosteroid therapy with 0.1% Dexafree® UD eye drops every 2 h (dexamethasone; Théa Pharma, Clermont-Ferrand, France) and 250 mg Solu-Medrol® tid intravenously (methylprednisolone; Pfizer AG, New York, N.Y., USA). Floxal SDU eye drops were given every 2 h as a prophylaxis against infection. To inhibit collagenolysis, the patient was treated with 100 mg Vibramycin® bid orally (doxycycline; Pfizer, New York, N.Y., USA) and 10% acetylcysteine eye drops qid. In addition, 500 mg Valtrex® tid orally (Valacyclovir; GlaxoSmithKline, London, UK) was prescribed until the diagnostic viral results were reported negative. Corneal melting stabilized under this treatment and ocular surface inflammation was reduced. Eight days after admission, when inflammation was under control, debridement of the corneal ulcer, recession of the adjacent conjunctiva, lamellar keratoplasty and amnion membrane transplantation covering the entire epithelial defect were successfully performed. The first postoperative day showed a well-positioned lamellar corneal transplant. Local corticosteroid and antibiotic treatment was tapered and systemic corticosteroid therapy was ceased ten days postoperatively. By the third postoperative week, increased superonasal fibrovascular activity was observed at the limbus. This was interpreted as impending recurrence of the pterygium and the patient received 0.5 ml Kenacort® 40 mg/ml subconjunctivally (triamcinolone; Dermapharm AG, Grünwald, Germany). Local antibiotic treatment with Floxal SDU was ceased four weeks postoperatively. 0.1% Dexafree UD eye drops were continued tid until ten weeks postoperatively and then replaced by FML Liquifilm® eye drops bid (fluorometholone; Allergan AG, Irvine, Calif., USA). On follow-up, four and a half months after surgery, spectacle-corrected visual acuity was 20/20. The Javal-Schiötz Ophthalmometer® (Haag-Streit AG, Bern, Switzerland) showed an irregular astigmatism of −1.5 dpt at 100°, which was confirmed with the Oculus-Pentacam® (Oculus GmbH, Wetzlar, Germany). Slit-lamp examination revealed a lamellar transplant, a few stromal parapupillary opacities nasally and a deep and quiet anterior chamber. The conjunctiva bordering the nasal cornea was slightly inflamed with an area of intensive vascularization and thickening (**fig. [2](#F2){ref-type="fig"}**). To minimize ocular surface inflammation, all sutures were removed and a subconjunctival injection of 0.6 ml Kenacort was given. Additionally, FML Liquifilm drops were replaced with 0.1% Dexafree UD qid. After a few weeks of slight regression, regrowth of the fibrovascular pannus was observed at the edge of the lamellar graft and 5 mg/ml bevacizumab eye drops (hospital pharmacy, University Hospital of Zurich, Zurich, Switzerland) were then given topically for 4 weeks. Six months postoperatively another subconjunctival Kenacort depot was placed. Superficial inflammation was then stabilized. A new formation of anterior synechiae extending through the three o\'clock position at the former site of corneal melting was detected. At 14 months postoperatively a slight increase in vascularization was noted at the edge of the amnion patch. A fourth subconjunctival injection of Kenacort was given and the patient was continued on FML Liquifilm drops once daily for another 4 months. No recurrence was noted at 24 months. Discussion {#sec1_3} ========== This case illustrates a severe complication after bare sclera excision of a pterygium with intraoperative application of MMC to the scleral bed. Successful management was achieved through control of the inflammatory process and the combination of a lamellar keratoplasty and amnion membrane patch to prevent the impending corneal perforation. The amnion membrane transplant (with its anti-inflammatory properties) together with locally and systemically applied corticosteroid drugs were used to control the inflammation. Adjunctive therapy with subconjunctival steroids and anti-vascular endothelial growth factor (VEGF) drugs was needed to treat a minor recurrence in the postoperative period. The aims of the primary surgery were not only to fully excise the pterygium, but also to prevent its recurrence. Mitomycin is utilized as ancillary treatment due to its antiproliferative effects. These irreversible effects of MMC might be the only cause of our patient\'s postoperative complication. On the other hand, apoptosis and necrosis might also have been triggered by the inflammatory properties of the pterygium and its excision. This severe complication threatened the tectonic integrity of the globe. Our primary goal, therefore, had been to surgically restore the integrity of the cornea, which was successfully achieved with lamellar keratoplasty. MMC-induced complications are associated with ocular surface inflammation, which can lead to a further deterioration of the clinical situation. Our goal, therefore, was not only to restore the full corneal thickness, but also to control the ocular surface inflammation with corticosteroids and tetracyclines. An additional anti-inflammatory effect was achieved by the amnion membrane transplantation. Furthermore, the timely recognition and treatment of an impending recurrence, with subconjunctival triamcinolone injection and anti-VEGF drugs, were crucial in stopping the recurrence. Several studies have recently emphasized the importance of VEGF in the development and recurrence of pterygium \[[@B12]\]. Despite the above-mentioned treatment, the formation of anterior synechiae extending through the three o\'clock position at the former site of corneal melting could not be prevented. It remains unclear whether the synechiae were triggered by corneal microperforations most probably existing at referral or as a consequence of surgical intervention. Since the growth of the synechiae was progressive, a possible etiology is chronic inflammation, which may have been due to the irreversible damage to the tissue on the cell level caused by MMC. Recently described surgical options to reduce the recurrence of pterygium are recommended alternatives to bare sclera surgery \[[@B9]\]. The conjunctival autograft described by Kenyon et al. \[[@B3]\] shows low recurrence rates with minimal complications and is nowadays the preferred choice of many surgeons. In addition to the conjunctival autograft, limbal stem cell transplantation has been introduced to reconstruct the excised pterygial area as anatomically and physiologically correct as possible to theoretically reduce the recurrence rate \[[@B13]\]. However, some studies have shown that the recurrence rates after limbal stem cell transplantations are comparable with conjunctival autograft surgery alone \[[@B14]\]. Another surgical option in the primary treatment of pterygium is amniotic membrane transplantation. Despite the antiscarring, anti-angiogenic, and anti-inflammatory properties of amnion membrane, recent studies have found no advantage in terms of recurrence rates compared to conventional treatments, and some authors have even reported unacceptably high recurrence rates \[[@B5]\]. Recently, fibrin glue has been used to avoid suture-related complications \[[@B15]\]. The major drawbacks regarding fibrin glue are its costs, bovine contents, and the potential risk of transmitting infections \[[@B9]\]. Our case demonstrates a successful management of a severe MMC-induced complication. The corneal integrity was effectively repaired with lamellar keratoplasty and the inflammation was controlled by amniotic membrane grafting as well as the application of corticosteroids and tetracyclines. Considering the potential adverse events with the use of MMC, we believe that despite the technical challenge and the longer surgical time, the conjunctival autograft technique should be used to treat primary pterygia. Disclosure Statement {#sec1_4} ==================== The authors have no conflicts of interest or sources of funding to declare. {#F1} {#F2}

Introduction ============ Monitoring of apoptosis, the programmed cell death, in various pathology conditions such as degenerative diseases, tumor treatment, and myocardial infarction would provide necessary information for medical team. One of the early biochemical changes of apoptotic cells is exposure of PS on the external surface of the plasma membrane ([@B1]-[@B6]). A noninvasive method of detecting this process such as nuclear imaging using radiopharmaceuticals could assist the diagnosis and therapy of apoptosis related pathologies, which is useful in preclinical and clinical studies in human. Radioligands that have affinity and bind to PS are good candidates for noninvasive imaging of apoptosis ([@B7]-[@B9]). Annexin V with high affinity for PS has been radiolabeled with ^99m^Tc and ^18^F in apoptosis imaging studies. Annexin V (35 kDa) is a natural intracellular human phospholipid binding protein which needs extracellular Ca^2+^ for binding to PS (approximately 2.5 mM Ca^2+^). Its rapid clearance from body, short blood half life, and inability to deference between apoptosis and necrosis makes the optimal imaging of apoptotic cells difficult ([@B10]-[@B11]). Regarding high potential of peptides in diagnosis and therapy, radiopeptides with high affinity to PS independent of Ca^2+^ are appropriate candidates for apoptosis imaging. In the last decade various groups have tried to introduce peptides with high affinity for PS through phage display technology ([@B12]-[@B16]). The results of research by Burtea *et al* in 2009 showed that peptide LIKKPF has high affinity for PS (K~d~= 2×10^-9^M). The peptide was conjugated with DTPA, labeled with Gadolinium, and tested as MRI contrast agent with apoptotic Jurkat cells and animal model of liver apoptosis ([@B16]). Because of low sensitivity of imaging agent, the result was moderately successful. Kapty *et al* radiolabeled LIKKPF with ^18^F prosthetic group N-succinimidyl-4-\[^18^F\] fluorobenzoate (\[^18^F\] SFB) and determined the affinity of radiolabeled peptide ([@B17]-[@B18]). The radiochemical purity (RCP) was low. In order to improve RCP, a cystein residue was added to LIKKPF and the thiol reactive N-\[6-(4-\[^18^F\] fluorobenzylidene) aminooxyhexyl\] maleimide (\[^18^F\] FBAM) was used as a prosthetic group. Although the modification of LIKKPF to CLIKKPF and radiolabeling with ^18^FBAM increased RCP but cyctein residue resulted in lower affinity. For routine clinical SPECT imaging, ^99m^Tc is the radionuclide of choice. This is because it has good radiation physical characteristics (IT, 140 keV (90), 6h) and is available through an inexpensive ^99^Mo/^99m^Tc generator. Furthermore, it is well developed and has varied chemistry ([@B19]). The application of radiopeptides as apoptosis imaging agents in assessing the therapeutic response of tumors to radio/chemo therapy, extent of damage after myocardial infarction, and neurodegenerative diseases has encouraged our research group for design and synthesis of new peptides. To reach the goal, the LIKKPF with nanomolar affinity to PS was chosen as a template. In this study, the biological activity of synthesized HYNIC-LIKKPF radiolabeled with ^99m^Tc was evaluated. The affinity of radiolabeled LIKKPF with ^99m^Tc was assessed *in-vitro* using camptothecin treated Jurkat cells and *in-vivo* using mouse model of liver apoptosis by intraperitoneal injection of lipopolysacharid (LPS). Moreover, the stability, labelling efficiency, RCP, and LogP of radiolabeled peptide were determined. Experimental ============ For this study, the amino acids and resin were obtained from Bachem (Bubendorf, Switzerland). Coupling reagents, HOBT and DIC were purchased from Sigma-Aldrich (St. Luis, MO, USA). Succinimidyl-N-Boc-HYNIC were purchased from ABX advanced Biochemical compounds GmbH (Radeberg, Germany). FITC-conjugated Annexin V was from Biolegend (San Diego, USA) and was used according to the manufacturer's instructions. Propidium iodide (PI), Camtothecin, and lipopolysacharide were purchased from Sigma-Alderich. All of the chemicals, solvents and reagents were of analytical quality and used without further purification. Silica gel 60 F~254~ pre-coated aluminium sheets from Merck were used for TLC. Using normal serum, ^99m^TcO~4~^-^ was eluted from a ^99^Mo/^99m^Tc generator (Pars-isotope, Tehran, Iran). The distribution of radioactivity on TLC was determined using a TLC Scanner Mini-Scan, MS.1000. This was equipped with flow count B-FC-1000 and gamma detector MS3200, (Bioscan, Washington, USA). Mass-Spectra was recorded on LC-MS Triple Quad 6410 Agilent Technologies using series 1200 HPLC system (Tokyo, Japan) column: C-18, 250 × 4.6 mm, 5 m, mobile phase: A:H~2~O + 0.1% TFA, B: Acetonitrile, flow rate: 1 mL/min, 20 L, total run time: 40 min. A NaI well counter (Triathler multilabel tester, Hidex, Finland) and a dose calibrator (Atomlab 100, Biodex, NY) were used to measure low and high levels of radioactivity, respectively. Flow cytometric analysis were performed using a flow cytometer equipped with its accompanying software (FACSCalibur and CellQuestPro, respectively, Becton Dickinson). *Synthesis of peptide LIKKPF functionalized with HYNIC (HYNIC-LIKKPF)* Synthesis of peptide LIKKPF was done by using standard Fmoc strategy as described previously ([@B20]). Briefly, the peptide sequence Leu-Ile-Lye-Lye-Pro-Phe was assembled on Wang Resin with two equiv of N-α-Fmoc-protected amino acid (Phe) (Fmoc-Phe-OH) and two equiv HOBt and DIC as a coupling reagent in six steps. Removal of protecting group was carried out with 10% piperidine in DMF. Coupling of HYNIC to the last amino acid was done with 2 equiv of HATU, 3 equiv of DIPEA, two equiv of Succinimidyl-N-Boc-HYNIC in dry DMF. After shaking for 45 min at room temperature (RT), the solution was removed and the resin was washed with DMF and CH~2~Cl~2~, respectively. The completeness of the coupling reaction was checked by a Kaiser Test. The cleavage of peptide from resin was checked using cocktail TFA/TIS/H2O (95: 2.5: 2.5) for 45 min ([@B21]). The solvents evaporated and peptide was precipitated with diethyl ether. The identity of peptide was confirmed by LC-MS. *Stability of HYNIC-LIKKPF in serum* 50 µL of peptide (1 µg/µL in ethanol) was added to 450 µL of fresh human serum and incubated for 5, 10, 30, 60 and 120 min at 37˚C. After incubation time, 500 µL of acetonitrile was added to precipitate serum proteins. The mixture was centrifuged at 10000 g for 10 min. The supernatant was analyzed with LC-MS. *Labeling of HYNIC-LIKKPF with* ^99m^ *Tc* A detailed protocol of radiolabeling reaction of HYNIC-LIKKPF with ^99m^Tc is available from the authors ([@B20]). The optimal condition of labelling: ^99m^Tc-HYNIC-LIKKPF: RCP \> 90%, (15-1000) μg HYNIC-LIKKPF, 5 mg EDDA, 10 mg tricine, 7μg SnCl~2~, pH 5-6, 100˚C for 30 min. *Log P determination* Partition coefficient of radiolabeled peptide was determined in the mixture of PBS (pH 7.4) and n-octanol. 10 µCi of labeled peptide was added in 1 mL of PBS/n-octanol (1: 1) mixture and vials were vortexed for 4 min and centrifuged at 5000 g for 10 min to separate two phases (aqueous and organic). 100 µL of each phase was measured in a gamma counter. The Log P values were obtained in ratio of activity in organic phase to aqueous phase. *Stability of radiolabeled peptide in serum* The stability of ^99m^Tc-HYNIC-LIKKPF was studied as previously reported ([@B20]). Briefly, 50 µCi of labeled peptide was added to 450 µL of fresh human serum and incubated at 37˚C. At different time points plasma proteins were precipitated out by reacting with 0.5 mL acetonitrile. The mixture was centrifuged at 10000 g for 10 min. The activity bound to the plasma protein was measured by counting the activity associated with the precipitate. The supernatant was analysed by Radio-TLC \[acetonitrile: water (95: 5)\]. *In-vitro studies* *Induction of apoptosis* Human leukemia cells (Jurkat J6 cell, Pastur Institute, Tehran, Iran) were cultured in RPMI 1640 medium supplemented with 10% FBS, penicillin, and streptomycin. Apoptosis was induced on cells with camptothecin dissolved in DMSO to final concentration of 2 and 6 µM in growth medium and incubated in 5% CO~2~ incubator at 37˚C for 24 and 4 h respectively ([@B22]). Induction of apoptotsis and percent of apoptotic cells were examined and measured by flow cytometry using Annexin V-FITC and PI. Camptothecin treated and untreated cells were harvested after 4 and 24 h, centrifuged for 5 min at 700 g, washed with cold PBS, and suspended in 1 mL cold binding buffer (HEPES 10 mM, NaCl 140 mM, BSA 1 mg/mL, CaCl~2~ 2.5 mM, pH7.4). After addition of 5 L (0.1 g/L) Annexin V-FITC and 15 min incubation at RT, 10 L PI solution (1 g/L) was added and fluorescence intensity was measured using 2 color flow cytometry. Untreated cells and cells incubated with DMSO were considered as control. *Binding studies of radiolabeled peptide* A binding assay was performed in triplicate in the presence of increasing amounts of radiolabeled peptide using apoptotic Jurkat cells. 500 L of cell suspension in binding buffer was added to 1 mL Eppendorf tubes and incubated with increasing concentrations of radiolabeled peptide (2 × 10^5^ cells/mL, ^99m^Tc-HYNIC-LIKKPF: S.A = 40 Ci/mmoL, 20 - 100 nM), for 30 min at room temperature. At the end of incubation times, the mixture was centrifuged (700 g, 5 min), the cell pellet was washed with cold binding buffer (3X) and the radioactivity of pellet was measured. Untreated cells used as negative control. For each radioligand concentration, nonspecific binding was determined by incubation of cells with excess amount of unlabeled peptide (100X of maximum concentration of radiolabeled peptide). The binding study was done in the presence and absence of calcium in binding buffer. *In-vivo studies* Balb/C adult mice (6-8 week old) were used and obtained from the breeding facility of the Department of Pharmacology and Toxicology, School of Pharmacy, Shahid Beheshti University of Medical Sciences. All animal studies were conducted in accordance with the guidelines established by the Shahid Beheshti University of Medical Sciences. Biodistribution studies were done in normal and apoptotic Balb/C mice (pre-treated and non-treated with cold peptide) with groups of 3 mice per each time point. Liver apoptosis was induced by intraperitoneal injection (IP) of lipopolysacharid (LPS, Escherichia Coli, Serotype 055: B5, Sigma) dissolved in normal saline at a dose of 0.5 mg/Kg, 12 h before experiments ([@B23]). Pre-treated mice were received 200 µg cold peptide 30 min before injection of radiolabeled peptide. *Biodistribution studies of* ^99m^ *Tc-HYNIC-LIKKPF* One Hundred µCi of radiolabeled peptide (1 Ci/1 µL in saline) was injected via the tail vein of normal and apoptotic mice model, pre-treated and non-treated. The animals were sacrificed at 30 min, 2 h and 4 h post injection (n = 3 for each time point). Interested organs and tissues were separated, weighted, and counted. The results were reported as percentage of injected dose per gram of organ (%ID/g). *SPECT/CT imaging of normal and apoptotic mice* 100 µCi of ^99m^Tc-HYNIC-LIKKPF was injected via the tail vein of normal and apoptotic mice model. The mice were anesthetized intraperitoneally with ketamine HCl. Anesthesia was maintained by booster injections of ketamine. The SPECT/CT imaging was performed on the SIEMENS Symbia T clinical SPECT/CT scanner. The mice were placed in supine position and CT scans performed for anatomical reference and attenuation correction (spatial resolution 1.25 mm, 80 kV, 150 mAs) with a total CT scanning time of 10 sec. Dynamic SPECT images were acquired for 15 min (60 sec per frame) starting 30 sec after injection. The acquision matrix size was 256 × 256 and a zoom factor of 2.67 was used. The total counts of the liver, kidneys and bladder with related ratios were measured. SPECT/CT scans of healthy and liver apoptosis Balb/c mice were done 60 min after injection. The acquision protocol was flash 3D with 256 × 256 matrix size, 2.29 zoom factor and ninety 10-second views with both detectors. SPECT data were acquired with a noncircular orbit. Reconstruction was performed using the flash 3D algorithm with 4 iterations and 20 subsets. Transmission data were reconstructed into a matrix of equal size by means of filtered back-projection, yielding a co-registered image set. Results ======= We have already reported the synthesis and radiolabeling of HYNIC-LIKKPF with ^99m^Tc ([@B20]). The stability of HYNIC-LIKKPF was determined in serum. The peptide functionalized with HYNIC was incubated in fresh human serum (10X) at 37˚C. After percipataion of serum proteins, the supenatant was analyzed with LC-MS. The results showed peptides were nearly 100% intact after 12 h. The optimum conditions of radiolabeling of HYNIC-LIKKPF with ^99m^Tc as previously reported is RCP \> 90%, (15-1000) μg peptide, 5 mg EDDA, 10 mg tricine, 7 μg SnCl~2~, 5-10 mCi ^99m^Tc, pH 5-6, at 100˚C for 30 min. Log P of radiolabeled peptides was determined in PBS buffer and n-octanol. ^99m^Tc-HYNIC-LIKKPF showed a Log P~o/w~= -0.87. Stability of radiolabeled peptide was checked in human fresh serum at 5, 10, 30, 60 and 120 min of incubation. The stability was analyzed by measuring the activity of precipitated serum proteins and TLC on supernatant. Results show that ^99m^Tc radiolabeled peptide was 95% intact and only 5% of ^99m^Tc released or transferred to serum proteins after 2 h. Apoptosis was induced in Jurkat cells by Camptothecin, a quinoline based DNA topoisomerase I inhibitor using two protocols (2 M, 24 h and 6 µM, 4 h) ([@B22]). Results showed that apoptosis is achieved with both protocols. Apoptosis induction was confirmed by Flow cytometry using Annexin V-FITC and PI. There was less than 1% of cells dead, 78% apoptotic, 7% necrotic, and 14% unstain ([Figure 1](#F1){ref-type="fig"}). {#F1} The affinity (K~d~) of radiopeptide to PS was determined by saturation binding studies. Jurkat cells were incubated with increasing amount of radiolabeled peptides at RT. After incubation time, cell bound radiopeptide was separated from free by centrifuge. The specific binding (SB) was determined (Total bound-Non-specific bound, TB-NSB = SB). A plot of SB against concentration of radiopeptide was used for calculation of affinity (k~d~) and B~max~. The curve was fitted according to a sigmoidal dose response profile using Graph Pad Prism. Results are shown in [Table 1](#T1){ref-type="table"} and [Figure 2](#F2){ref-type="fig"}. ###### Biological characteristics of radiolabeled peptide -------------------------------------------------------------------------------------------------------------------------------------------------------------- **Concentration** **LogP** **K** ~d~ **B** ~max~\ **nmol/cell** ---------------------- ------------------- -------------------------------------- --------------------------------------- ------------------------------------ ^99m^Tc-HYNIC-LIKKPF (20-100) nM -0.87 (90 ± 21.06) nM (1.141 × 10^-13^± 1.514 × 10^-14^) LIKKPF 2.51[\*](#TFN1){ref-type="table-fn"} 2 nM [\*](#TFN1){ref-type="table-fn"} -------------------------------------------------------------------------------------------------------------------------------------------------------------- calculated by Burtea *et al* (16). {#F2} Biodistribution studies of radiolabeled LIKKPF was done in normal and mice model of liver apoptosis, pre-treated and non-treated. Mice were sacrificed at different time points, organs were separated, weighted, and activity measured. Results were reported as percentage of injected dose per gram of organs (%ID/g). The distribution of radioactivity after IV injection of ^99m^Tc-HYNIC-LIKKPF in normal and liver apoptosis mice is summarized as a function of time in [Figure 3](#F3){ref-type="fig"}. The most prominent organs of uptake in normal mice include the kidneys, bladder, intestine, and colon. The remaining activity in the carcass and organs 30 min after injection was less than 15% of injected dose. It is apparent that the radioactivity rapidly cleared from blood and excreted in urine, suggesting renal clearance. Uptake in liver of apoptotic non-treated mice was 0.85% at 30 min (2.5 times normal values), 0.81% at 2 h (6.25 times normal values), which decreased to 0.05% at 4 h post injection. Kidney uptake was less than normal mice at all-time points. The renal activity decreased 22%, 23%, and 70% at 30 min, 2 and 4 h post injection, respectively. Colon uptake was 1.5 times above normal values 4 h post injection, suggesting hepatic clearance as well as kidney clearance. Hepatic uptake was inversely proportional to renal uptake in apoptotic non-treated mice. The pattern of bio distribution of apoptotic mice pre-treated was similar to normal mice. {#F3} Generally speaking, accumulation of radio peptides in liver of apoptotic non-treated mice model was higher compare to normal and apoptotic pre-treated mice. One Hundred µCi of ^99m^Tc-HYNIC-LIKKPF was injected via the tail vein of normal and apoptotic mice model. As shown in [Figure 4](#F4){ref-type="fig"}, the summation of the dynamic frames in the first 15 min clearly shows the significant activity concentration of tracer in kidneys, bladder and liver as it was seen in bio distribution studies as well. The liver to total frame counts were 7% and 9% for normal and liver apoptosis mouse model, respectively. For investigating the uptake behavior, the liver total counts were also measured in the SPECT/CT images of the mice acquired 60 min after injection. Based on mean count per number of pixels and data normalization of images, liver uptake has increased 63% compare to normal mouse. {#F4} Discussion ========== One of the early signs of apoptosis, within a few hours of apoptotic stimulus, is PS exposure on the extracellular face of the plasma membrane. PS is an abundant target. The most studied imaging agent for apoptosis is Annexin V and its fragments so far. Because of limitations of Annexin V (suboptimal pharmacokinetics, inability to distinguish apoptosis from necrosis, calcium dependent) other agents have been introduced such as small peptides and molecules ([@B1]-[@B11]). There is a need to develop a radioligand with better biodistribution profile for apoptosis imaging. Our group is currently working on design of new peptides as imaging agents for PS. The LIKKPF, introduced by Burtea *et al* with nanomolar affinity for PS, was selected as template. In this study, synthesis, conjugation, radiolabeling, and biological properties of radiolabeled LIKKPF was evaluated *in-vitro* and *in-vivo*. Peptide LIKKPF was successfully synthesized via Fmoc strategy and functionalized with HYNIC at N-terminal. The structures were confirmed with LC-MS. The functionalized peptide was stable in human serum at 37˚C for at least 2 h. HYNIC conjugated LIKKPF was radiolabeled with ^99m^Tc (RCP \> 90%). The radiolabeled LIKKPF was stable in human serum for at least 2 h at 37˚C. LogP of radiolabeled LIKKPF was determined. The LogP value indicates high hydrophilicity of radiolabeled LIKKPF. The calculated LogP of LIKKPF by Burtea *et al* using ACDLabs 12.0 software was 2.51. Since each functional group of molecule contributes to the overall LogP, conjugation of HYNIC to LIKKPF and radiolabeling with ^99m^Tc could be one of the reasons for significant reduction of LogP compare to original phage display LIKKPF. The reduced LogP of radiolabeled peptide explains its fast clearance from blood and excretion through kidneys. The radiolabeled LIKKPF was tested *in-vitro* and *in-vivo*. The affinity of radiolabeled peptide was determined in saturation binding studies using camptothecine treated Jurkat cells. Results revealed that binding of radiolabeled LIKKPF to PS is not calcium dependent and radiolabeled peptide bind specifically to apoptotic cells. The level of binding to apoptotic cells was 2.5 times higher than control cells. The affinity of radiolabeled LIKKPF was lower than the original phage displayed peptide calculated by Burtea *et al* (ELISA using phage, K~d~= 2.5 nM). Kapty *et al* have also reported low affinity ([@B18]). There are two main reasons for low affinity of radiolabeled LIKKPF to PS. First, since more than 1 copy of peptide contribute in binding to target, the original phage displayed peptide has higher affinity with a better and more suitable conformation than single peptide ([@B13]). The second reason is low hydrophobicity of radiolabeled LIKKPF (negative LogP), which is responsible for peptide lipid interaction. Interaction of peptide with membrane lipid and polar PS head depends on its sequence. The positively charged residues (predominantly basic) provide electrostatic interactions with polar PS head group, while hydrophobic side chain residues hold peptide attached to membrane through hydrophobic interactions, which allows electrostatic interactions with the neighboring basic residues (3, 16). Electrostatic interactions of LIKKPF with PS is achieved by lysine residues that are ionized at physiological pH 7.4, which suggest their interaction with the polar head of PS as was mentioned by Burtea *et al* ([@B16]). Hydrophobic residues, leucine, isoleucine, proline and phenylalanine thought to be responsible for hydrophobic interaction. The affinity of fluorescent tagged LIKKPF and CLIKKPF determined by Kapty *et al* was 0.69 µM and 1.9 µM, respectively. They argued that addition of cysteine residue at N-terminus and fluorescent tag are responsible for low affinity of peptides ([@B18]). Although the affinity of radiolabeled LIKKPF was less than original phage peptide, the level of binding to apoptotic cells was 2.5 times higher than control cells. In this study, a mouse modal of liver apoptosis was used for *in-vivo* experiments by injection of LPS intraperitoneally 12 h before experiment. Bio distribution studies showed higher liver uptake of radio peptide in apoptotic non-treated mouse model compare to normal and pre-treated mouse. The *in-vivo* imaging results were consistent with bio distribution studies. *In-v*ivo imaging of normal and mouse model of liver apoptosis also showed a high concentration of radio peptide in liver of mouse model observed by SPECT/CT 60 min after injection. The percentage of whole body activity per organ determined by region of interest (ROI) image analysis correlated with the percentage of ID/g organ determined by bio distribution studies. The measurements show the high uptake of the liver in apoptotic mice model with 63% increase compare to normal mouse. There is negligible liver uptake in the healthy mouse ([Figure 4](#F4){ref-type="fig"}). In conclusion, electrostatic and hydrophobic interactions of peptide with membrane are necessary for PS binding ([@B3], [@B13], [@B16]). The peptide should be attached to membrane for electrostatic interactions with the neighboring residues. The more hydrophobic peptides should be considered in designing new peptides for PS binding. *Abbreviations* HYNIC: 6-Hydrazinonicotinamide, TFA: trifluoroacetic acid, Tricine: N-\[tris (hydroxymethyl) methyl\] glycine, HOBT: n-hydroxy benzotriazole, DIC: diisopropylcarbodiimide, DIPEA: diisopropylethylamine, DMF: dimethylformamide, TIS: triisobutylsilane, MEK: methyl ethyl ketone, EDDA: ethylenediamine diacetate, Fmoc: 9-fluoroenylmethoxycarbonyl, RCP: radiochemical purity. This study was supported by funds from Iranian National Science Foundation (INSF).

(J Am Heart Assoc. 2019;8:e011605 DOI: 10.1161/JAHA.118.011605.)30977410 Clinical PerspectiveWhat Is New?Quantitative flow ratio (QFR) showed higher correlation and diagnostic performance for the prediction of fractional flow reserve than instantaneous wave‐free ratio; however, the absolute efficacy of QFR to predict instantaneous wave‐free ratio was also excellent. In addition, diagnostic performance of fractional flow reserve, instantaneous wave‐free ratio, and QFR was comparable when coronary flow reserve was used as a reference method.What Are the Clinical Implications?As QFR does not require additional interrogation with a pressure wire or administration of hyperemic agents, and shows significantly shorter measurement time than invasive physiologic assessment, this might represent a more simple, safe, and cost‐effective method to guide revascularization.With excellent diagnostic agreement of QFR with both fractional flow reserve and instantaneous wave‐free ratio, QFR‐guided strategy might be a promising tool for improving the adoption rate of physiology‐based revascularization. Introduction {#jah34031-sec-0008} ============ The presence of inducible myocardial ischemia is the prerequisite indication for percutaneous coronary intervention. In this regard, a pressure‐derived physiologic index, fractional flow reserve (FFR), has been the standard invasive method to evaluate the functional significance of epicardial coronary artery stenosis.[1](#jah34031-bib-0001){ref-type="ref"}, [2](#jah34031-bib-0002){ref-type="ref"} Recently, instantaneous wave‐free ratio (iFR), which does not require hyperemia, was developed as an alternative for FFR,[3](#jah34031-bib-0003){ref-type="ref"} and 2 randomized controlled trials demonstrated comparable clinical outcomes between iFR‐ and FFR‐guided strategies with less use of revascularization after iFR‐guided treatment.[4](#jah34031-bib-0004){ref-type="ref"}, [5](#jah34031-bib-0005){ref-type="ref"} On this ground, recent guidelines recommend the measurement of FFR or iFR in defining the functional significance of intermediate epicardial coronary stenoses as a class IA recommendation.[1](#jah34031-bib-0001){ref-type="ref"} However, the adoption rates of FFR‐ or iFR‐guided percutaneous coronary intervention are still low in real‐world practice.[6](#jah34031-bib-0006){ref-type="ref"} As alternative methods to evaluate functional significance of epicardial coronary stenosis, functional coronary imaging has recently emerged, allowing wire‐free functional assessment of stenosis severity based on a computational fluid dynamics model or mathematical assumptions of coronary flow. Quantitative flow ratio (QFR) is a 3‐dimensional quantitative coronary angiography (QCA)--based computation of FFR, and previous studies have demonstrated excellent correlations and diagnostic agreements with FFR.[7](#jah34031-bib-0007){ref-type="ref"}, [8](#jah34031-bib-0008){ref-type="ref"}, [9](#jah34031-bib-0009){ref-type="ref"}, [10](#jah34031-bib-0010){ref-type="ref"}, [11](#jah34031-bib-0011){ref-type="ref"}, [12](#jah34031-bib-0012){ref-type="ref"} However, as QFR is calculated using simulated hyperemic status of coronary circulation,[7](#jah34031-bib-0007){ref-type="ref"}, [8](#jah34031-bib-0008){ref-type="ref"} this might show discrepancies with iFR‐based treatment decision‐making, as this latter index is measured during resting status.[13](#jah34031-bib-0013){ref-type="ref"}, [14](#jah34031-bib-0014){ref-type="ref"} Nevertheless, data for diagnostic performance of QFR compared with iFR as a reference standard are still limited.[15](#jah34031-bib-0015){ref-type="ref"} In addition, as FFR and iFR are pressure‐derived surrogates of coronary flow, and QFR uses simulated coronary flow, the comparison of diagnostic performance of these indices using other independent reference tests, such as coronary flow reserve (CFR), would be more reasonable.[13](#jah34031-bib-0013){ref-type="ref"}, [16](#jah34031-bib-0016){ref-type="ref"} For these reasons, we sought to evaluate the diagnostic performance and agreement of QFR using FFR or iFR as reference standards, and also evaluated these 3 physiologic indices using CFR as an independent reference standard. Methods {#jah34031-sec-0009} ======= Anonymized patient‐level data will be made available by the corresponding author upon reasonable request. Study Population {#jah34031-sec-0010} ---------------- The study population was derived from previously published studies.[17](#jah34031-bib-0017){ref-type="ref"} Briefly, between April 2016 and June 2018, a total of 118 consecutive patients with acute myocardial infarction (AMI) who underwent clinically indicated physiologic assessment for nonculprit stenosis with visual stenosis of 40% to 80% were included from the prospective Institutional Registry of Samsung Medical Center.[17](#jah34031-bib-0017){ref-type="ref"} AMI was defined as the third universal definition of MI.[18](#jah34031-bib-0018){ref-type="ref"} Data from 203 patients with stable ischemic heart disease (SIHD) were selected from the prospective multicenter registry of comprehensive physiologic assessment, which enrolled consecutive patients who underwent clinically indicated invasive coronary angiography and physiologic assessment from 5 university hospitals in Korea (Samsung Medical Center, Seoul National University Hospital, Inje University Ilsan Paik Hospital, Keimyung University Dongsan Medical Center, and Ulsan University Hospital) ([clinicaltrials.gov](http://clinicaltrials.gov): NCT02186093).[19](#jah34031-bib-0019){ref-type="ref"} Among these populations, 82 patients with AMI (105 nonculprit vessels) and 182 patients with SIHD (253 vessels) were included in the current analysis after excluding vessels without available QFR because of anatomical or angiographic limitations in QFR analysis (Figure [1](#jah34031-fig-0001){ref-type="fig"}). The excluded patients (57 patients, 17.8%) had limited coronary angiographic image quality for QFR analysis (calibration failure, ostial disease, insufficient projections, tortuous vessels, overlapping of vessels, or poor contrast filling). The study protocol was approved by the institutional review board of each participating center and was conducted according to the principals of the Declaration of Helsinki. All patients provided written informed consent before enrollment. {#jah34031-fig-0001} Invasive Angiography and Measurement of Physiologic Indices {#jah34031-sec-0011} ----------------------------------------------------------- Coronary angiography was performed with standard techniques. After administration of intracoronary nitrate (100 or 200 μg), angiographic views were obtained. QCA was performed with optimal projections using validated software (CAAS II, Pie Medical System). From QCA, minimal lumen diameter, reference vessel size, and lesion length were measured and percentage diameter stenosis was calculated. After diagnostic angiography, coronary physiologic indices were obtained as previously described.[19](#jah34031-bib-0019){ref-type="ref"} After engagement of a 5‐7F guide catheter without side holes in the coronary artery, the pressure‐temperature sensor guide wire (Abbott Vascular) was calibrated and equalized to aortic pressure. Then, it was placed at the distal segment of a target vessel. Before each physiologic measurement, intracoronary nitrate (100 or 200 μg) was administered. Intravenous infusion of adenosine (140 μg/kg per min through a peripheral vein) or intracoronary bolus injection of nicorandil (2 mg) was used to induce hyperemia.[20](#jah34031-bib-0020){ref-type="ref"}, [21](#jah34031-bib-0021){ref-type="ref"} Resting distal to aortic coronary pressure (Pd/Pa) was calculated as the ratio of mean aortic pressure (Pa) to mean distal coronary arterial pressure (Pd). iFR was calculated as the mean Pd divided by the mean Pa during the diastolic wave‐free period. The resting tracing data were extracted and the iFR was calculated using automated algorithms acting over the wave‐free period during a minimum of 5 beats, as previously described.[14](#jah34031-bib-0014){ref-type="ref"} FFR was acquired during maximal hyperemia and was defined as the lowest value of mean hyperemic Pd/Pa. CFR was calculated as resting mean transit time divided by hyperemic resting mean transit time. To derive resting mean transit time, a thermodilution curve was obtained by using 3 injections (4 mL each) of room‐temperature saline in both resting and hyperemic states. After every measurement, the pressure wire was pulled back to the guide catheter and the presence of pressure drift was checked. All coronary physiologic measurements were performed after diagnostic angiography in patients with SIHD or after percutaneous coronary intervention for the culprit vessel in the nonculprit vessel of patients with AMI. Coronary physiologic data were collected and validated at a core laboratory in a blinded fashion. The cutoff values of FFR ≤0.80, iFR ≤0.89, resting Pd/Pa ≤0.92, and CFR ≤2.0 were used in the current study.[13](#jah34031-bib-0013){ref-type="ref"}, [14](#jah34031-bib-0014){ref-type="ref"} Computation of QFR {#jah34031-sec-0012} ------------------ Three‐dimensional QCA and analysis of QFR were performed by an independent core laboratory with dedicated software (QAngio‐XA 3D, version 1.2, Medis) in a blinded fashion for clinical data or invasive FFR, iFR, resting Pd/Pa, or CFR values, as previously described.[8](#jah34031-bib-0008){ref-type="ref"}, [9](#jah34031-bib-0009){ref-type="ref"} Briefly, end‐diastolic frames of 2 optimal angiography projections, which were separated with angles of at least 25°, were selected and used for 3‐dimensional model reconstruction. The 3‐dimensional contour model of the segment of interest and its reference vessel were constructed in an automated manner and manual correction of contour was performed, if necessary. After acquisition of fixed QFR, estimated contrast coronary flow was calculated using thrombolysis in myocardial infarction (TIMI) frame‐count adjustment, which indicated the frames where contrast entered and exited the segmented part of the vessel.[8](#jah34031-bib-0008){ref-type="ref"} With application of TIMI frame‐count adjustment in the calculation method, the software automatically calculated the contrast QFR value. The cutoff value of contrast QFR ≤0.80 was used in the current study.[8](#jah34031-bib-0008){ref-type="ref"}, [9](#jah34031-bib-0009){ref-type="ref"} Statistical Analysis {#jah34031-sec-0013} -------------------- Categorical variables are presented as numbers and relative frequencies. Continuous variables are presented as mean and SD or median with interquartile range according to their distributions, which were checked by the Kolmogorov‐Smirnov test. ANOVA was used for comparison of continuous variables among the groups. Repeated measure correlation coefficients (*r*) using linear mixed model were calculated to assess the correlations among FFR, iFR, CFR, and QFR for adjustment of multivessel measurements within a patient. The differences of correlation coefficients were tested by the Fisher r‐to‐z transformation. The agreement between QFR and FFR or between QFR and iFR was tested by Bland‐Altman plots. Diagnostic performances of QFR were presented with sensitivity, specificity, positive predictive value (PPV), negative predictive value, and diagnostic accuracy. Diagnostic performances were compared using McNemar test or weighted generalized score statistic. Discriminant function was evaluated using area under the curve (AUC) and 95% CIs in receiver operating curve analysis, and AUC was compared with the DeLong method. To evaluate interindividual variability in QFR assessment, 2 independent researchers analyzed 30 randomly selected cases. The paired measurements were compared using paired sample *t* test. All probability values were 2‐sided, and *P*\<0.05 was considered statistically significant. The statistical package R, version 3.4.3 (R Foundation for Statistical Computing) was used for statistical analysis. Results {#jah34031-sec-0014} ======= Baseline Patient and Lesion Characteristics {#jah34031-sec-0015} ------------------------------------------- Baseline patient and lesion characteristics are presented in Table [1](#jah34031-tbl-0001){ref-type="table"}. The mean age was 60.6±13.3 years and 203 patients (76.9%) were men. Mean percentage diameter stenosis, FFR, iFR, CFR, and QFR were 53.1±19.0%, 0.80±0.13, 0.88±0.12, 3.14±1.30, and 0.81±0.14, respectively. The proportions of FFR ≤0.80, iFR ≤0.89, and CFR ≤2.0 were 39.9%, 40.5%, and 23.4%, respectively. The distributions of FFR, iFR, and QFR according to clinical presentation are shown in [Figure S1](#jah34031-sup-0001){ref-type="supplementary-material"}. Regarding interindividual variability in QFR assessment, the QFR values from 2 independent researchers were nearly the same without significant differences (0.792±0.107 versus 0.794±0.109, *P*=0.919). ###### Baseline Patient and Lesion Characteristics Patient Characteristics (N=264) -------------------------------------------------- ------------ Demographics Age, y 60.6±13.3 Men 203 (76.9) Body mass index, kg/m^2^ 24.1±3.2 Hypertension 133 (50.4) Diabetes mellitus 87 (33.0) Hypercholesterolemia 156 (59.1) Current smoker 47 (17.8) Family history of coronary artery disease 20 (17.4) Prior myocardial infarction 16 (6.1) Left ventricular ejection fraction, % 60.9±10.1 Clinical presentations SIHD 182 (68.9) non--ST‐segment--elevation myocardial infarction 55 (20.8) ST‐segment--elevation myocardial infarction 27 (10.2) Lesion Characteristics (N=358) ----------------------------------- ------------ Lesion location LAD 223 (62.3) LCX 68 (19.0) RCA 67 (18.7) Quantitative coronary angiography Reference vessel diameter, mm 3.14±0.61 Minimum lumen diameter, mm 1.50±0.72 Diameter stenosis, % 53.1±19.0 Lesion length, mm 15.6±10.0 Invasive physiologic indices Fractional flow reserve 0.80±0.13 Instantaneous wave‐free ratio 0.88±0.12 Resting Pd/Pa 0.92±0.09 Quantitative flow ratio 0.81±0.14 Coronary flow reserve 3.14±1.30 Values are expressed as mean±SD or number (percentage). LAD indicates left anterior descending artery; LCX, left circumflex artery; RCA, right coronary artery; Pd/Pa, distal to aortic coronary pressure; SIHD, stable ischemic heart disease. John Wiley & Sons, Ltd Diagnostic Performance of QFR to Predict FFR or iFR {#jah34031-sec-0016} --------------------------------------------------- QFR was significantly correlated with FFR and iFR; nevertheless, QFR showed a significantly higher correlation coefficient with FFR than iFR (*r*=0.863 versus 0.740, *P*\<0.001) ([Figure S2](#jah34031-sup-0001){ref-type="supplementary-material"} and Table [2](#jah34031-tbl-0002){ref-type="table"}). The higher correlation of QFR with FFR than with iFR was similar in both AMI nonculprit and SIHD vessels (*r*=0.857 versus 0.741 in patients with SIHD, *P*\<0.001; 0.884 versus 0.739 in patients with AMI, *P*=0.002) ([Figure S2](#jah34031-sup-0001){ref-type="supplementary-material"} and Table [2](#jah34031-tbl-0002){ref-type="table"}). The agreement between QFR and FFR was also higher than that between QFR and iFR in the total population (bias±SD: 0.002±0.140 versus −0.070±0.188, *P*\<0.001), and the higher agreement of QFR with FFR than with iFR was also similarly observed in both AMI nonculprit and SIHD vessels ([Figure S3](#jah34031-sup-0001){ref-type="supplementary-material"}). ###### Comparison of Diagnostic Performance of QFR to Predict FFR or iFR FFR as Reference iFR as Reference *P* Value ------------------------- ------------------------------- ------------------------------- ----------- Total population Sample size 264 Patients with 358 vessels 264 Patients with 358 vessels Correlation coefficient 0.863 (0.800--0.907) 0.740 (0.631--0.820) \<0.001 Sensitivity, % 92.3 (87.9--96.7) 80.0 (73.5--86.5) \<0.001 Specificity, % 89.8 (85.7--93.8) 82.2 (77.0--87.3) \<0.001 PPV, % 85.7 (80.2--91.2) 75.3 (68.5--82.1) 0.002 NPV, % 94.6 (91.5--97.7) 85.8 (81.0--90.6) 0.004 Diagnostic accuracy, % 90.8 (90.7--90.8) 81.3 (81.2--81.4) \<0.001 SIHD Sample size 182 Patients with 253 vessels 182 Patients with 253 vessels Correlation coefficient 0.857 (0.779--0.909) 0.741 (0.612--0.831) \<0.001 Sensitivity, % 90.1 (84.0--96.2) 78.4 (69.8--87.0) 0.012 Specificity, % 89.5 (84.8--94.2) 81.8 (75.9--87.7) \<0.001 PPV, % 82.8 (75.4--90.3) 69.7 (60.6--78.7) 0.006 NPV, % 94.2 (90.5--97.9) 87.7 (82.5--92.9) 0.066 Diagnostic accuracy, % 89.7 (89.7--89.8) 80.6 (80.5--80.8) \<0.001 AMI (nonculprit) Sample size 182 Patients with 105 vessels 182 Patients with 105 vessels Correlation coefficient 0.884 (0.737--0.951) 0.739 (0.461--0.885) 0.002 Sensitivity, % 96.2 (90.9--101.4) 82.5 (72.6--92.3) 0.008 Specificity, % 90.6 (82.7--98.4) 83.3 (72.8--93.9) 0.067 PPV, % 90.9 (83.3--98.5) 85.5 (76.1--94.8) 0.371 NPV, % 96.0 (90.6--101.4) 80.0 (68.9--91.1) 0.027 Diagnostic accuracy, % 93.3 (93.2--93.4) 82.9 (82.6--83.1) 0.009 Values are expressed as estimates with 95% CIs. Correlation coefficient was calculated based on per‐vessel analysis using mixed linear model for adjustment of multivessel measurements within a patient. The differences of correlation coefficients were tested by the Fisher r‐to‐z transformation. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic accuracy were calculated based on per‐vessel analysis and compared using McNemar test or weighted generalized score statistic. AMI indicates acute myocardial infarction; FFR, fractional flow reserve; iFR, instantaneous wave‐free ratio; QFR, quantitative flow ratio; SIHD, stable ischemic heart disease. John Wiley & Sons, Ltd Diagnostic performances of QFR to predict FFR ≤0.80 or iFR ≤0.89 are shown in Table [2](#jah34031-tbl-0002){ref-type="table"}. With FFR as a reference standard, the sensitivity, specificity, positive PPV, negative predictive value, and diagnostic accuracy of QFR were 92.3%, 89.8%, 85.7%, 94.6%, and 90.8%, respectively. With iFR as a reference, the sensitivity, specificity, PPV, negative predictive value, and diagnostic accuracy of QFR were 80.0%, 82.2%, 75.3%, 85.8%, and 81.3%. All indices of diagnostic performance of QFR were higher when FFR was used as a reference than iFR, regardless of clinical presentation (all *P*\<0.05) (Table [2](#jah34031-tbl-0002){ref-type="table"}). The distributions of target vessels were not significantly different according to the concordance or discordance of QFR, compared with FFR or iFR (*P*=0.692 for QFR with FFR; 0.659 for QFR with iFR) ([Table S1](#jah34031-sup-0001){ref-type="supplementary-material"}). The discriminant functions of QFR to predict FFR ≤0.80 or iFR ≤0.89 are shown in Figure [2](#jah34031-fig-0002){ref-type="fig"} and Table [3](#jah34031-tbl-0003){ref-type="table"}. The AUC values of QFR for FFR ≤0.80 or iFR ≤0.89 were 0.953 (95% CI, 0.932--0.974) and 0.880 (95% CI, 0.844--0.917), respectively (*P* for comparison\<0.001). In both patients with SIHD and those with AMI, the AUC values of QFR for FFR ≤0.80 was significantly higher than those for iFR ≤0.89. However, there was no significant difference in AUC of QFR between the nonculprit vessel of patients with AMI and those with SIHD, regardless of the reference methods used (Table [3](#jah34031-tbl-0003){ref-type="table"}). {#jah34031-fig-0002} ###### Comparison of Discriminant Function of QFR to Ischemic Reference Standard FFR as Reference 264 Patients With 358 Vessels iFR as Reference 264 Patients With 358 Vessels *P* for Comparison[a](#jah34031-note-0006){ref-type="fn"} ----------------------------------------------------------- ------------------------------------------------ ------------------------------------------------ ----------------------------------------------------------- --------- --------- Total population 0.953 (0.932--0.974) \<0.001 0.880 (0.844--0.917) \<0.001 \<0.001 SIHD 0.946 (0.919--0.974) \<0.001 0.876 (0.828--0.923) \<0.001 0.011 AMI 0.967 (0.936--0.998) \<0.001 0.882 (0.817--0.947) \<0.001 0.021 *P* for comparison[b](#jah34031-note-0007){ref-type="fn"} 0.320 NA 0.876 NA NA Area under the curve (AUC) values were calculated based on per‐vessel analysis and compared with the DeLong method. QFR indicates quantitative flow ratio. *P* value for comparison according to the reference tests (fractional flow reserve \[FFR\] or instantaneous wave‐free ratio \[iFR\]). *P* value for comparison according to the clinical presentations (nonculprit of acute myocardial infarction \[AMI\] or stable ischemic heart disease \[SIHD\] vessels). John Wiley & Sons, Ltd The discriminant function of QFR to predict resting Pd/Pa ≤0.92 was also evaluated and the AUC values of QFR for iFR ≤0.89 or resting Pd/Pa ≤0.92 were not significantly different (Figure [2](#jah34031-fig-0002){ref-type="fig"}). In comparison of discriminant functions among QFR, iFR, and resting Pd/Pa to predict FFR≤0.80 as a reference, QFR showed significantly higher AUC than iFR (0.953 versus 0.918, *P*=0.023) or resting Pd/Pa (0.953 versus 0.909, *P*=0.023); however, there was no significant difference between iFR and resting Pd/Pa (*P*=0.682) ([Figure S4](#jah34031-sup-0001){ref-type="supplementary-material"}). Comparison of FFR, iFR, and QFR Using CFR as an Independent Reference Standard {#jah34031-sec-0017} ------------------------------------------------------------------------------ The discriminant functions of FFR, iFR, and QFR were modest with CFR as a reference standard, although iFR showed a significantly higher discriminant function than FFR or QFR (Figure [3](#jah34031-fig-0003){ref-type="fig"} and Table [4](#jah34031-tbl-0004){ref-type="table"}). However, when CFR ≤2.0 was used as a reference standard, the overall diagnostic accuracies of FFR, iFR, and QFR were not significantly different (65.4%, 70.6%, and 64.9%, respectively; all *P* values in pairwise comparisons \>0.05, overall comparison *P*=0.061) as with sensitivity, specificity, PPV, and negative predictive value (Table [4](#jah34031-tbl-0004){ref-type="table"}). {#jah34031-fig-0003} ###### Comparison of Diagnostic Performance and Discriminant Function of FFR, iFR, and QFR With CFR as a Reference Standard FFR iFR QFR ------------------------ ------------------------------------------------------------- --------------------------------------------------------------------------------------------------------- ------------------------------------------------------------- AUC 0.682 (0.600--0.764)[a](#jah34031-note-0009){ref-type="fn"} 0.765 (0.691--0.838)[b](#jah34031-note-0010){ref-type="fn"} ^,^ [c](#jah34031-note-0011){ref-type="fn"} 0.677 (0.596--0.758)[a](#jah34031-note-0009){ref-type="fn"} Sensitivity, % 61.1 (48.1--74.1) 68.5 (56.1--80.9) 64.8 (52.1--77.6) Specificity, % 66.7 (59.7--73.6) 71.2 (64.5--77.9) 65.0 (57.9--72.0) PPV, % 35.9 (26.1--45.7) 42.0 (31.7--52.4) 36.1 (26.5--45.6) NPV, % 84.9 (78.9--90.8) 88.1 (82.8--93.4) 85.8 (79.9--91.7) Diagnostic accuracy, % 65.4 (65.2--65.6) 70.6 (70.4--70.7) 64.9 (64.7--65.1) Values are expressed as estimates (95% CIs). Area under the curve (AUC) values were calculated based on per‐vessel analysis and compared with the DeLong method. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic accuracy were calculated based on per‐vessel analysis and compared using McNemar test or weighted generalized score statistic. CFR indicates coronary flow reserve. *P*\<0.05 compared with instantaneous wave‐free ratio (iFR). *P*\<0.05 compared with fractional flow reserve (FFR). *P*\<0.05 compared with quantitative flow ratio (QFR). John Wiley & Sons, Ltd Among the total population, QFR showed discordance mainly with iFR rather than FFR. QFR disagreed with iFR in 18.7% (67/358 vessels), including 38 vessels with QFR ≤0.80 and iFR \>0.89 (10.6%), and 29 vessels with QFR \>0.80 and iFR ≤0.89 (8.1%) (Figure [4](#jah34031-fig-0004){ref-type="fig"} and Table [5](#jah34031-tbl-0005){ref-type="table"}). CFR was significantly higher in vessels with iFR \>0.89 than with iFR \<0.89, regardless of QFR values (QFR \>0.80 group: CFR 3.48±1.19 for iFR \>0.89 versus CFR 2.59±1.13 for iFR ≤0.89, *P*=0.016; QFR ≤0.80 group: CFR 3.66±1.46 for iFR \>0.89 versus CFR 2.52±1.18 for iFR ≤0.89, *P*\<0.001). However, CFR was comparable between the vessels with QFR \>0.80 and QFR ≤0.80, in both the iFR \>0.89 group (*P*=0.896) and the iFR ≤0.89 group (*P*=0.997) (Figure [4](#jah34031-fig-0004){ref-type="fig"} and Table [5](#jah34031-tbl-0005){ref-type="table"}). {#jah34031-fig-0004} ###### Physiologic and Angiographic Differences in Vessels Among 4 QFR and iFR Agreement Groups QFR \>0.80 and iFR \>0.89 (n=175) QFR ≤0.80 and iFR \>0.89 (n=38) QFR \>0.80 and iFR ≤0.89 (n=29) QFR ≤0.80 and iFR ≤0.89 (n=116) *P* Value ----------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------ ----------- Quantitative coronary angiography Reference vessel diameter, mm 3.20±0.62[a](#jah34031-note-0013){ref-type="fn"} 3.26±0.59[a](#jah34031-note-0013){ref-type="fn"} 2.77±0.56[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 3.12±0.59[a](#jah34031-note-0013){ref-type="fn"} 0.003 Minimal lumen diameter, mm 1.82±0.70[c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 1.30±0.54[b](#jah34031-note-0014){ref-type="fn"}, [a](#jah34031-note-0013){ref-type="fn"} 1.71±0.63[c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 1.03±0.52[b](#jah34031-note-0014){ref-type="fn"}, [a](#jah34031-note-0013){ref-type="fn"} \<0.001 Diameter stenosis, % 44.6±17.2[c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 61.2±12.7[b](#jah34031-note-0014){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} 40.1±17.5[c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 66.6±13.8[b](#jah34031-note-0014){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} \<0.001 Lesion length, mm 12.3±7.2[c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 17.9±8.8[b](#jah34031-note-0014){ref-type="fn"} 12.1±7.2[d](#jah34031-note-0016){ref-type="fn"} 20.5±12.2[b](#jah34031-note-0014){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} \<0.001 Invasive physiologic indices FFR 0.89±0.05[c](#jah34031-note-0015){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.81±0.06[b](#jah34031-note-0014){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.84±0.04[b](#jah34031-note-0014){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.67±0.11[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} \<0.001 QFR 0.91±0.06[c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.74±0.05[b](#jah34031-note-0014){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.87±0.05[c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.66±0.11[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} \<0.001 iFR 0.95±0.03[a](#jah34031-note-0013){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.93±0.03[a](#jah34031-note-0013){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.86±0.03[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 0.75±0.13[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} ^,^ [a](#jah34031-note-0013){ref-type="fn"} \<0.001 CFR 3.48±1.19[a](#jah34031-note-0013){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 3.66±1.46[a](#jah34031-note-0013){ref-type="fn"} ^,^ [d](#jah34031-note-0016){ref-type="fn"} 2.59±1.13[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} 2.52±1.18[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} \<0.001 Resting mean transit time 0.94±0.48[a](#jah34031-note-0013){ref-type="fn"} 0.97±0.40[a](#jah34031-note-0013){ref-type="fn"} 0.60±0.25[b](#jah34031-note-0014){ref-type="fn"} ^,^ [c](#jah34031-note-0015){ref-type="fn"} 0.82±0.36 \<0.001 Hyperemic mean transit time 0.29±0.18[d](#jah34031-note-0016){ref-type="fn"} 0.28±0.12 0.25±0.11 0.36±0.16[b](#jah34031-note-0014){ref-type="fn"} 0.007 IMR, U 21.5±13.3[d](#jah34031-note-0016){ref-type="fn"} 17.1±8.5 17.6±8.9 14.0±6.6[b](#jah34031-note-0014){ref-type="fn"} \<0.001 IMR \>25U 23.7% 13.8% 10.0% 9.0% 0.055 CFR indicates coronary flow reserve; FFR, fractional flow reserve; IMR, index of microvascular resistance. *P*\<0.05 compared with quantitative flow ratio (QFR) \>0.80 and instantaneous wave‐free ratio (iFR) ≤0.89. *P*\<0.05 compared with QFR \>0.80 and iFR \>0.89. *P*\<0.05 compared with QFR ≤0.80 and iFR \>0.89. *P*\<0.05 compared with QFR ≤0.80 and iFR ≤0.89. John Wiley & Sons, Ltd Discussion {#jah34031-sec-0018} ========== The current study evaluated the diagnostic performance of QFR to define functionally significant epicardial coronary stenoses using FFR or iFR as reference standard methods. In addition, these physiologic indices were also evaluated using CFR as an independent reference standard. The main findings were as follows. First, QFR showed excellent correlation and diagnostic performance for both invasive pressure--derived physiologic indices (FFR and iFR), regardless of clinical presentation. Second, the correlation, diagnostic performance, and discriminant function of QFR were better for FFR than for iFR. Third, when CFR was used as a reference standard, iFR showed the highest discriminant function in comparison with FFR or QFR, and the discordance between QFR and iFR was mainly driven by differences in the coronary flow (CFR). However, the diagnostic accuracies of FFR, iFR, and QFR were not statistically different using CFR as a reference standard. Invasive Physiologic Indices in Contemporary Practice {#jah34031-sec-0019} ----------------------------------------------------- As coronary revascularization is only beneficial when reversible myocardial ischemia is present, there have been numerous efforts to detect the presence of myocardial ischemia. Although many noninvasive tests to assess myocardial ischemia are available, a previous study reported a low diagnostic yield and limited PPV of these tests.[22](#jah34031-bib-0022){ref-type="ref"} Currently, FFR, a pressure‐derived physiologic index, has been validated in several large‐scale randomized controlled trials and is considered a standard invasive method to define the functional significance of an epicardial coronary stenosis.[2](#jah34031-bib-0002){ref-type="ref"} Recently, another invasive physiologic method, iFR, which can be measured in the resting state without induction of hyperemia, was introduced.[4](#jah34031-bib-0004){ref-type="ref"}, [5](#jah34031-bib-0005){ref-type="ref"} Previous studies reported about 10% to 30% disagreement between FFR and iFR in the classification of functional significance of coronary lesions[2](#jah34031-bib-0002){ref-type="ref"}; however, recent large‐scale randomized controlled trials reported comparable clinical outcomes between FFR‐ and iFR‐guided revascularization strategies with less revascularization after the iFR‐guided strategy.[4](#jah34031-bib-0004){ref-type="ref"}, [5](#jah34031-bib-0005){ref-type="ref"} Consequently, both FFR‐ and iFR‐guided revascularization strategies are included in the recent guidelines with a class 1A recommendation.[1](#jah34031-bib-0001){ref-type="ref"} Nevertheless, the adoption rate of FFR or iFR in real‐world practice is still limited.[6](#jah34031-bib-0006){ref-type="ref"}, [23](#jah34031-bib-0023){ref-type="ref"} Emerging Role of QFR as an Alternative to Invasive Physiologic Assessment {#jah34031-sec-0020} ------------------------------------------------------------------------- The low adoption rates of FFR‐ or iFR‐guided revascularization strategies might be associated with additional coronary instrumentation and cost of drugs and devices, prolonged procedural time, possible patient discomfort with adenosine administration for FFR measurement, limited confidence with the results, experience, or personal beliefs.[24](#jah34031-bib-0024){ref-type="ref"} To overcome these limitations, there have been several efforts to develop new techniques for assessing functional significance of coronary lesions from noninvasive tests[25](#jah34031-bib-0025){ref-type="ref"} or without additional pressure wire interrogation. QFR is an angiography‐derived method that provides functional assessment of coronary stenoses from computation of 3‐dimensional QCA and TIMI frame count without additional pressure wire or induction of hyperemia. Previous studies reported excellent correlation and diagnostic agreement of QFR compared with FFR.[7](#jah34031-bib-0007){ref-type="ref"}, [8](#jah34031-bib-0008){ref-type="ref"}, [9](#jah34031-bib-0009){ref-type="ref"}, [10](#jah34031-bib-0010){ref-type="ref"}, [11](#jah34031-bib-0011){ref-type="ref"}, [12](#jah34031-bib-0012){ref-type="ref"} Furthermore, QFR analysis showed significantly lower measurement time than FFR.[11](#jah34031-bib-0011){ref-type="ref"} Nevertheless, QFR has not been thoroughly evaluated using other reference standards beyond FFR. As the current guideline also recommends the use of iFR‐guided decision as class IA recommendation,[1](#jah34031-bib-0001){ref-type="ref"} it is important to evaluate the diagnostic ability of QFR in defining functionally significant stenosis using iFR as a reference standard. In a previous study by Emori et al,[15](#jah34031-bib-0015){ref-type="ref"} QFR showed good correlation with iFR as well as with FFR. Although diagnostic accuracy of QFR was numerically higher with FFR than with iFR (94% versus 74%), that difference was not statistically significant, probably because of insufficient sample size. The current study demonstrates the excellent correlation and diagnostic agreement of QFR not only with FFR but also with iFR. However, QFR was more strongly correlated with FFR than iFR, and diagnostic performance and discriminant function of QFR were better with FFR than with iFR. These results were consistently found in both SIHD and AMI nonculprit vessels. Considering that QFR is derived from the mathematical assumptions for hyperemic status of coronary circulation, the better correlation and diagnostic agreement of QFR with FFR than with iFR seems to be natural. As QFR was originally designed to predict FFR, the current results show the features of QFR rather than which one is correct or which one is better. Furthermore, diagnostic accuracy (81.3%) and discriminant function (AUC=0.725) of QFR for iFR in the current study were also good. Therefore, the findings of this study support the clinical value of QFR in determining the functional significance of coronary stenoses from the iFR point of view. It should be noted that QFR showed wider variance in lesions with a lower range of FFR and iFR ([Figure S3](#jah34031-sup-0001){ref-type="supplementary-material"}). This might be related to the intrinsic limitation of any simulated indices including computed tomography--derived FFR, as these simulated indices cannot inherently reflect the collateral flow or combined microvascular dysfunction.[25](#jah34031-bib-0025){ref-type="ref"}, [26](#jah34031-bib-0026){ref-type="ref"}, [27](#jah34031-bib-0027){ref-type="ref"} Discrepancies Between QFR and Invasive Pressure--Derived Indices on the Basis of CFR {#jah34031-sec-0021} ------------------------------------------------------------------------------------ The current study also evaluated FFR, iFR, and QFR using CFR as an independent reference standard. As with previous studies, iFR showed better discriminant function for CFR compared with FFR or QFR. In addition, the discordance between QFR and iFR was mainly caused by differences in patient‐specific CFR. In the algorithm of contrast‐QFR model, the fixed‐QFR value, which assumes an empiric coronary flow, is adjusted to the patient\'s specific coronary flow using TIMI frame count.[7](#jah34031-bib-0007){ref-type="ref"} Although contrast QFR provided more accurate prediction of invasive FFR than fixed QFR,[7](#jah34031-bib-0007){ref-type="ref"}, [8](#jah34031-bib-0008){ref-type="ref"} the correlation between TIMI frame‐count--based volumetric flow rate reserve in the algorithm of contrast QFR[7](#jah34031-bib-0007){ref-type="ref"} and actual Doppler‐derived CFR was modest (*r*=0.62).[28](#jah34031-bib-0028){ref-type="ref"} Current results imply that contrast QFR would have limited ability to reflect the actual patient‐specific CFR, and the discordance between QFR and iFR might originate from the fundamental difference between the 2 indices. Furthermore, in the case of significant microvascular dysfunction, any types of computational or simulated methods including QFR might have limited diagnostic accuracy.[29](#jah34031-bib-0029){ref-type="ref"} Nevertheless, the diagnostic accuracy of QFR was not significantly different from those of FFR or iFR with CFR as a reference standard. These results imply that FFR, iFR, and QFR might share similar limitations as a surrogate marker of coronary flow. In addition, it also implies that the fundamental difference of FFR, iFR, and QFR with CFR would not have substantial influence to a binary decision using those surrogate indices of coronary flow in routine clinical practice. However, considering the difference in revascularization rates between FFR‐ and iFR‐guided treatment in the DEFINE‐FLAIR (Functional Lesion Assessment of Intermediate Stenosis to Guide Revascularisation) or iFR‐SWEDEHEART (Instantaneous Wave‐free Ratio versus Fractional Flow Reserve in Patients with Stable Angina Pectoris or Acute Coronary Syndrome) trials, where iFR guidance resulted in 5% fewer revascularizations than FFR guidance, it should be noted that QFR‐guided decision‐making might result in slightly higher revascularization rates than iFR‐guided decision‐making. Clinical Implications {#jah34031-sec-0022} --------------------- The current study evaluated the diagnostic performance of QFR using other reference standards beyond FFR. Although QFR showed higher correlation and diagnostic performance for prediction of FFR than iFR, the absolute efficacy of QFR to predict iFR was also good. In addition, the diagnostic accuracies of FFR, iFR, and QFR were not significantly different when CFR was used for a reference standard. As QFR does not require additional interrogation with a pressure wire or administration of hyperemic agents, and shows significantly shorter measurement time than invasive physiologic assessment,[11](#jah34031-bib-0011){ref-type="ref"} this may represent a more simple, safe, and cost‐effective method to guide revascularization. With excellent diagnostic agreement of QFR with both FFR and iFR, which are the current standard methods to define functional significance of epicardial coronary stenosis, a QFR‐guided strategy might be a promising tool to improve the adoption rate of physiology‐based revascularization. Ongoing randomized controlled trials evaluating superiority in clinical outcomes following a QFR‐guided strategy compared with an angiography only--guided strategy (FAVOR III China, NCT03656848) would clarify the value of QFR in daily clinical practice. Limitations {#jah34031-sec-0023} ----------- There are several limitations in this study. First, clinical outcome data were not evaluated; therefore, we could not evaluate the prognostic implications of QFR‐guided treatment. Second, as QFR is an angiography‐based method, its accuracy depends on the quality of images and optimal projection. Indeed, about 17% of patients and 26% of vessels were excluded from the QFR analysis based on improper quality of angiographic images. Although these rates were high in the current study, they might be lower with the use of recommended angiographic projections and simultaneous on‐site real‐time QFR analysis. Third, because the contrast QFR model computes TIMI frame count to adjust the simulation of coronary flow, the quality of contrast injection would have a potential influence on the accuracy of the contrast QFR model. However, there has been no standardization method of contrast injection technique and there was no direct evidence that evaluated the influence of contrast injection technique and the accuracy of contrast QFR. Fourth, various agents (intravenous adenosine or intracoronary nicorandil) were used for hyperemia induction in this study. However, it is reported that all of these agents have similar hyperemic efficacy without systemic bias in FFR measurement.[20](#jah34031-bib-0020){ref-type="ref"}, [21](#jah34031-bib-0021){ref-type="ref"} Fifth, CFR can be affected by microvascular function and does not essentially represent epicardial coronary stenosis alone. However, considering the relatively small proportion of vessels with high index of microvascular resistance as well as low CFR (5.7% of total vessels), the influence from the significant microvascular dysfunction might be minimal in the current analysis. Last, contrast FFR, which is another nonhyperemic pressure ratio, was not available in the current study. Conclusions {#jah34031-sec-0024} =========== QFR has a high correlation and agreement with respect to both invasive pressure--derived indices, FFR and iFR, although better when FFR is used as the comparator. The differences observed between QFR and iFR may be explained by CFR. As a pressure‐derived index not depending on wire or adenosine, QFR might be a promising tool for improving the adoption rate of physiology‐based revascularization. Disclosures {#jah34031-sec-0025} =========== Dr Joo Myung Lee received a research grant from St. Jude Medical (Abbott Vascular) and Philips Volcano. Dr Joo‐Yong Hahn received a research grant from St. Jude Medical (Abbott Vascular). Dr Bon‐Kwon Koo received an institutional research grant from St. Jude Medical (Abbott Vascular) and Philips Volcano. The remaining authors have no disclosures to report. Supporting information ====================== ###### **Figure S1.** Distributions of fractional flow reserve (FFR), instantaneous wave‐free ratio (iFR), and quantitative flow ratio (QFR). Distributions of FFR, iFR, and QFR are presented in (**A**) total population, (**B**) vessels of stable ischemic heart disease, or (**C**) nonculprit vessels of acute myocardial infarction. **Figure S2.** Correlations between fractional flow reserve (FFR), instantaneous wave‐free ratio (iFR), and quantitative flow ratio (QFR) according to clinical presentation. Correlation of QFR with (**A**) FFR or with (**B**) iFR is presented. r~rm~ indicates repeated measure correlation coefficient. **Figure S3.** Agreement between quantitative flow ratio (QFR) and fractional flow reserve (FFR) or between QFR and instantaneous wave‐free ratio (iFR). Bland‐Altman plots are presented (**A**) between QFR and FFR or (**B**) between QFR and iFR. **Figure S4.** Discriminant functions of quantitative flow ratio (QFR), instantaneous wave‐free ratio (iFR), and resting distal to aortic coronary pressure (Pd/Pa) to predict fractional flow reserve (FFR) as a reference standard. Discriminant functions of QFR, iFR, and resting Pd/Pa to predict FFR ≤0.80 are presented. AUC indicates area under the curve. ###### Click here for additional data file. [^1]: Dr Hwang and Dr K. H. Choi contributed equally to this work.

Introduction {#Sec1} ============ Polarization is one of the most fundamental yet fascinating properties of transverse waves, as light waves, gravitational waves, and sound waves in solids. For polarization studies, the polarization of light waves is a representative being accessible easily to experimental demonstrations, and allows characterizing as well as manipulating polarization in all its features. While homogeneously distributed states of polarization have been investigated comprehensively, recent research focuses on spatially inhomogeneous distributions of polarization, i.e. the coexistence of several states of polarization in complex transverse light fields. These so-called Poincaré beams^[@CR1],\ [@CR2]^ include combinations of linear states of polarization known as vector beams^[@CR3]^ as well as elliptical and circular polarization states in spatial structures. Moreover, Poincaré beams are strongly connected to singular optics^[@CR4]^ due to the appearance of different types of vectorial singularities within light fields revealing structured polarization^[@CR5]^. Different techniques have been proposed to generate inhomogeneous distributions of states of polarization as e.g. the superposition of spatial modes^[@CR6],\ [@CR7]^, the application of polarization holograms^[@CR8]^, q-plates^[@CR9],\ [@CR10]^, or plasmonic metasurfaces^[@CR11]^ and dynamic techniques using spatial light modulators^[@CR12]--[@CR16]^ and digital micro-mirror devices^[@CR17]^. Recently the combination of polarization modulation with amplitude or phase modulation has been investigated^[@CR18]--[@CR21]^ more intensively. However, a dynamic technique to modulate extended polarization structures is essential for investigations of advanced polarization singularity formations. In this paper, we present the experimental realization of dynamically modulated Poincaré beams. Primarily, the polarization of Laguerre-Gaussian (LG) beams is designed according to the structure of its amplitude distribution. Furthermore, different combinations of spin and orbital angular momentum are modulated in a helical LG beam. Finally, polarization singular structures are enabled which is demonstrated by a spiraling Poincaré beam containing various kinds of polarization singularities. All results are experimentally implemented by a dynamic modulation technique which is discussed within the method section. It is realized by a single spatial light modulator which is in combination with a pair of wave plates capable of generating all kinds of fully-polarized polarization states. As we use a pixelated display, various combinations of coexisting states of polarization become possible in a spatially highly resolved pattern. To verify high modulation quality of extended polarization patterns, a spatially resolved measurement of Stokes parameters (see method sections) is employed^[@CR22]^. The advantage of our approach is the automatic alignment of horizontal and vertical polarization components due to the dynamic modulation technique. Opposed to previous studies^[@CR23]--[@CR25]^, an explicit superposition and alignment of orthogonally polarized modes is not necessary. This avoids propagation instabilities of (singular) polarization fields due to misalignment and gives the opportunity to investigate the longitudinal evolution of such structures. Results {#Sec2} ======= To access simultaneous holographic modulation of amplitude, phase as well as polarization, we employ a two step method experimentally implemented by a split-screen configuration^[@CR26]^ of a parallel aligned nematic liquid crystal spatial light modulator (SLM) as depicted in Fig. [1(a)](#Fig1){ref-type="fig"}. In the first step, a combined amplitude and phase encoding technique^[@CR27],\ [@CR28]^ allows the realization of e.g. higher order laser modes as Hermite-, Laguerre- or Ince-Gaussian beams. However, the method is not restricted to these modes but enables to modulate arbitrary light fields. In the second step, a dynamic polarization modulation system (DPMS) is implemented using the second half of the SLM (see Methods below) to realize spatially polarization structured Poincaré beams. Furthermore, our system enables to demonstrate orbital and spin angular momentum coupling as well as to create and study polarization singularities as V-points, C-points, L-lines and C-lines as discussed in the following.Figure 1Polarization structured LG beams. (**a**) Experimental setup: polarization measurement system (II) and DPMS combined with phase and amplitude modulation (I). C: camera, (H/Q)WP: (half/quarter) wave plate, M: mirror, P: polarizer. (**b**) Schematic of the subspace (*θ* ~1~ = −45°, *θ* ~2~ = 90°). (**c**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{LG}}}_{\mathrm{5,3}}^{e}$$\end{document}$ with azimuthally and (**d**) $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{LG}}}_{\mathrm{7,9}}^{o}$$\end{document}$ with radially modulated polarization structure. Polarization states are depicted by red or blue ellipses/circles indicating right and left elliptical/circular polarization, respectively. Linear polarization states are shown by green lines. Stokes parameters S~1,2,3~ ∈ \[−1, 1\] (**e**--**g**,**k**--**m**) as well as the ellipticity *ε* ∈ \[−1, 1\] (**h**,**n**) and the orientation *ψ* ∈ \[−*π*/2, *π*/2\] (**i**,**o**) of polarization ellipses are given. Corresponding side-by-side holograms are shown in (**j**,**p**). LG beams exhibiting discrete spatial polarization structures {#Sec3} ------------------------------------------------------------ Complex light fields is a common term in modern optics for a variety of light fields tailored spatially in one or various parameters. In this context, generation of extra-cavity higher order light modes is an established field enabled by employment of liquid crystal spatial light modulators. We modulate such higher order light modes which are not only structured in amplitude and phase, but also in polarization. Therefore, we generate higher order Laguerre-Gaussian (LG~*n*,*l*~) modes and simultaneously modulate polarization states spatially^[@CR29]^. In Fig. [1](#Fig1){ref-type="fig"} two examples are given: An even $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{LG}}}_{5,3}^{e}$$\end{document}$ beam with azimuthal variation of six different polarization states shown in (c, e-j), and an odd $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{\rm{L}}{\rm{G}}}_{7,9}^{o}$$\end{document}$ beam with radial variation of four different polarization states shown in (d, k-p). To verify modulation quality, the state of polarization is determined by spatial Stokes parameter measurements (s. method section) and visualized by plotting corresponding polarization ellipses (see Fig. [1(c,d)](#Fig1){ref-type="fig"}). To increase visibility, the polarization ellipses are not drawn for every but for every fifth pixel and plotted on top of the measured intensity distribution (depicted in gray scale). Right-handed, left-handed and linear polarization states are illustrated in red, blue, and green, respectively, as indicated in Fig. [1(b)](#Fig1){ref-type="fig"}. Experimental results show clearly separated areas of different polarization states as intended by the applied holograms (Fig. [1(j,p)](#Fig1){ref-type="fig"}). Remarkably, measured polarization states are not only resolved in their orientation *ψ* (s. Fig. [1(i,o)](#Fig1){ref-type="fig"}) but also in their ellipticity *ε* = tan(*χ*) (s. Fig. [1(h,n)](#Fig1){ref-type="fig"}) defined by the ratio of semi-major and semi-minor axes of the polarization ellipse. In both examples, all states of polarization are chosen to be part of a ring-like subspace of the Poincaré sphere (compare details of DPMS in method section) illustrated in Fig. [1(b)](#Fig1){ref-type="fig"}, which means S~1~ = 0 while S~2,3~ ∈ \[−1, 1\]. This is also verified by the experimental results given in Fig. [1(e--g,k--m)](#Fig1){ref-type="fig"}. Corresponding side-by-side holograms for the modulation of higher order Laguerre-Gaussian beams with structured polarization are shown in Fig. [1(j,p)](#Fig1){ref-type="fig"}. The shown LG-Poincaré beams have not been constructed by aligned superposition of two separately generated orthogonal modes but by direct modulation enabled by a liquid crystal display. Therefore, the spatial distribution of phase, amplitude, and polarization values becomes almost independent and gives a new degree of freedom to tailor complex light fields, i.e. it becomes possible to determine spatial distributions of polarization states within higher order light modes. In addition an auto-alignment is given due to direct polarization modulation by the DPMS. This opportunity can even be used to manipulate secondary parameters connected to certain spatial phase or polarization distributions as the optical angular momentum of light. In the following section we demonstrate such a case and show the new dimension of singular light patterns. Spatial modulation of spin and orbital angular momentum {#Sec4} ------------------------------------------------------- Independent modulation of phase and polarization allows for spatial structuring of angular momentum carrying beams given by combination of spin (SAM) and orbital (OAM) angular momentum. We explore this opportunity by investigation of a higher order helical LG beam, which is visualized in Fig. [2](#Fig2){ref-type="fig"}. Due to its phase structure, the helical LG~5,3~ beam carries orbital angular momentum (OAM, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{L}}_{z}$$\end{document}$) proportional to 3*ħ* per photon^[@CR30]^. In addition, we introduce a spatially varying spin angular momentum (SAM, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{S}}_{z}$$\end{document}$) by spatial polarization modulation of linear ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{S}}_{z}=0$$\end{document}$), right- ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{S}}_{z}=-\hslash $$\end{document}$) and left- ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{S}}_{z}=+\hslash $$\end{document}$) circular polarization states. The handedness of the circular polarization can be seen in Fig. [2(a,b)](#Fig2){ref-type="fig"} where measured polarization ellipses and the corresponding ellipticity (*ε*) are shown. Therefore the total amount of angular momentum $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{J}={\overrightarrow{L}}_{z}+{\overrightarrow{S}}_{z}$$\end{document}$ is spatially modulated as indicated in Fig. [2(h)](#Fig2){ref-type="fig"} showing a cross-section of the beam along the white dashed line in Fig. [2(a)](#Fig2){ref-type="fig"}. High quality of polarization modulation can also be seen from the orientation *ψ* of polarization states (Fig. [2(c)](#Fig2){ref-type="fig"}) as well as the three Stokes parameters (Fig. [2(e--g)](#Fig2){ref-type="fig"}).Figure 2Spatial polarization modulation of a helical LG~5,3~ beam: (**a**) Polarization ellipses based on measured Stokes parameters S~1,2,3~ ∈ \[−1, 1\] shown in (**e**--**g**). Ellipticity *ε* ∈ \[−1, 1\] and orientation *ψ* ∈ \[−*π*/2, *π*/2\] of polarization states are given in (**b**) and (**c**). (**d**) Sketch of employed subspace of DPMS. (**h**) Distribution of spin (*S* ~*z*~), orbital (*L* ~*z*~) and total (*J*) angular momentum per photon along the white dashed line in (**a**). The given example demonstrates the capability to tailor optical angular momentum structures by spatial phase and polarization modulation. Beside a general interest of singular optics to use this opportunity to investigate fundamental properties of such light fields, especially domains sensitive to optical angular momentum will profit from our investigations, as quantum communication and optical tweezers and its applications in optofluidics and biophotonics^[@CR31]^. Holographic modulation of polarization singular structures {#Sec5} ---------------------------------------------------------- Holographic polarization modulation not only allows for spatial variation of different polarization states, but also gives the opportunity to create certain polarization structures which reveal a continuous polarization flow field including polarization singularities^[@CR4],\ [@CR32],\ [@CR33]^. In Fig. [3](#Fig3){ref-type="fig"} a polarization spiral embedding a vector singularity (V-point, i.e. point of undefined polarization) in its center is realized^[@CR34]^. It is surrounded by intertwining lines of linear and left- or right-handed circular polarizations, so called L- and C-line singularities. In case of L- or C-line singularities the handedness or orientation of polarization ellipses is undefined, respectively. In between two adjacent line singularities elliptical polarization states form a smooth transition required for a continuous polarization flow field. To distinguish between different polarization singularities, they are sketched by green (L-line), red (right-handed C-line) and blue (left-handed C-line) lines and a black dot (V-point) in Fig. [3(a)](#Fig3){ref-type="fig"}. To realize the polarization spiral experimentally, the DPMS is set to create polarization states on the meridian of the Poincaré sphere of the S~2,3~-plane (illustrated in Fig. [3(b)](#Fig3){ref-type="fig"}) by choosing orientations *θ* ~1~ = −45° and *θ* ~2~ = 90° for the first and second quarter wave plate, respectively (comp. method section). Experimental results of the spiraling polarization field resulting from the phase hologram Δ*ϕ* ~SLM~ in Fig. [3(k)](#Fig3){ref-type="fig"} are given in Fig. [3(j,l--r)](#Fig3){ref-type="fig"} and can be compared to theoretical simulations shown in Fig. [3(a--i)](#Fig3){ref-type="fig"}. The V-point is characterized by a conserved topological quantum number^[@CR32]^, namely the Poincaré field index $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\sigma }_{P}=\frac{1}{2\pi }{\oint }_{C}d{\varphi }_{\mathrm{2,3}}=-2$$\end{document}$ estimated (counterclockwise) around the V-point from the phase *ϕ* ~2,3~ = arctan(S~3~/S~2~) of the Poincaré field shown in Fig. [3(i) and (r)](#Fig3){ref-type="fig"}. In addition the phase *ϕ* ~1,2~ = arctan(S~2~/S~1~) is shown in Fig. [3(e) and (n)](#Fig3){ref-type="fig"}. Continuity of polarization parameters can be well seen from ellipticity of the polarization distribution (Fig. [3(c,l)](#Fig3){ref-type="fig"}) as well as the Stokes parameters (Fig. [3(f--h,o--q)](#Fig3){ref-type="fig"}). High experimental reproducibility of simulated singular structures generated by the DPMS is given in all three coordinates of the Poincaré sphere (i.e. Stokes parameter S~1~, S~2~, S~3~) as well as ellipticity and orientation of polarization ellipses (*ε*, *ψ*).Figure 3Polarization spiral containing various polarization singularities^[@CR34]^: The polarization modulation hologram Δ*ϕ* ~SLM~ in (**k**) results in a polarization field shown in ((**a**) simulation) and ((**j**) measurement). Appearing polarization singularities are sketched in (**a**) by green (L-line), red (right-handed C-line) and blue (left-handed C-line) lines and a black dot (V-point). The V-point is characterized *σ* ~*P*~ = −2 estimated from the phase *ϕ* ~2,3~ of the Poincaré field in (**i**) and (**r**). (**a**--**i**,**k**) show simulations while (**j**,**l**--**r**) give experimental results of Stokes parameters S~1,2,3~, ellipticity and orientation of the polarization ellipses (*ε*, *ψ*), as well as phases of complex Stokes fields *ϕ* ~1,2~, *ϕ* ~2,3~. The direct modulation of spatial structures of polarization states as demonstrated in this example, allows to study complex Poincaré beams and polarization singular light fields. Auto-alignment of polarization components enables high modulation quality and enhanced stability compared to interferometric methods. This facilitates good comparability of experimental results and simulations. In particular, complex polarization singularity structures^[@CR35]^ become realizable. For example, higher order singularities in form of polarization flowers and webs as well as hybrid polarization structures can be generated^[@CR33],\ [@CR36]^. In addition, beside transverse appearance and characteristics, especially dynamics of complex light fields, i.e. their longitudinal propagation, reveal a broad range of interesting findings as the evolution of phase and polarization singularities. As already phase as well as polarization singularities reveal interesting evolutions if considered independently, we expect distinctive dynamics if evolving polarization singularities are combined with naturally unfolding phase singularities. Dynamics provoked by customized light fields, generated by our dynamic holographic method, will be investigated in future. Methods {#Sec6} ======= Any polarization can be defined by two orthogonal states, e.g. horizontal and vertical linear polarization, and is characterized by its amplitude ratio *A* ^÷^ = *A* ~*v*~/*A* ~*h*~ and phase shift Δ*ϕ* = *ϕ* ~*v*~ − *ϕ* ~*h*~ between these two components. A polarization sensitive device which only affects one of both components can be used to alter *A* ^÷^ and/or Δ*ϕ* and therefore change the polarization of light. For this purpose, nowadays liquid crystal spatial light modulators (SLM) are the device of choice as they allow for a computer controlled and spatially resolved (pixelwise) variation of the complex field^[@CR27]^. While the polarization component parallel to the liquid crystal director (*P* ~\|\|~) is shifted in amplitude and phase, its orthogonal component (*P* ~⊥~) is not altered for a phase-only and parallel aligned nematic-liquid-crystal SLM as used here^[@CR29],\ [@CR33],\ [@CR34]^. The relative amplitude and phase also depend on the input polarization ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{P}}_{{\rm{in}}}$$\end{document}$). Accordingly, a dynamic polarization modulation system (DPMS) should consist of a SLM enclosed by two rotatable wave plates (WP) as explained in the following. Dynamic polarization modulation system -- DPMS {#Sec7} ---------------------------------------------- A rotatable wave plate is able to transform linear polarized light into elliptical or circular polarization states in dependence of the orientation angle *θ* of the fast axis of the wave plate with respect to the initial polarization. In our experimental realization depicted in Fig. [4](#Fig4){ref-type="fig"} the first rotatable wave plate WP~1~ sets an initial amplitude ratio $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{W{P}_{1}}^{\div}={A}_{\perp }/{A}_{\parallel }$$\end{document}$ of *P* ~⊥~ and *P* ~\|\|~ and an initial phase shift $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Delta }}{\varphi }_{W{P}_{1}}$$\end{document}$ of the light impinging on the SLM. Thus, the respective orientation angle *θ* ~1~ defines the size of a subspace of states of polarization accessible by the SLM. Subsequently, the second rotatable wave plate WP~2~ allows to choose a set of polarization states determined by its orientation angle *θ* ~2~ to finally yield the subspace of the whole DPMS being accessible at once, as explained later on. Summing up, the parameter of the DPMS are a spatially resolved amplitude factor *A* ~SLM~(*x*, *y*) to influence the amplitude ratio *A* ^÷^, a spatially resolved phase shift Δ*ϕ* ~SLM~(*x*, *y*) (both induced by the SLM), retardations of the two wave plates Γ~1,2~(*n*) = *λ*/*n* with ($\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$n\in {\mathbb{N}}$$\end{document}$) and their orientations *θ* ~1,2~ of the fast axis. Figure [4(a)](#Fig4){ref-type="fig"} shows the setup including the DPMS (III) and a polarization measurement system (II) which is described later on. The orientation *θ* is defined as the angle between the input polarization and the fast axis of the WP (counterclockwise) in the direction of propagation. Exemplary, an orientation angle of *θ* = −45° is depicted in Fig. [4(b)](#Fig4){ref-type="fig"}.Figure 4Schematic of (**a**) an experimental setup including (III) the DPMS and (II) a polarization measurement system; (**b**) the orientation angle *θ* of a wave plate (WP) defined as the inclination of its fast axis to the linear horizontal input polarization, here shown for *θ* = −45°. C: camera, (H/Q)WP: (half/quarter) wave plate, M: mirror, P: polarizer. ### Mathematical description of DPMS by Jones formalism {#Sec8} For a comprehensive analysis of accessible subspaces and influence of parameters, the resulting states of polarization are represented on a Poincaré sphere which is a unit sphere spanned by the Stokes parameters (S~1~, S~2~, S~3~) with all fully-polarized states located on its surface. Linear polarization states can be found on the equator, while right circular and left circular polarization states mark the north and south pole, respectively. All possible output polarizations $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{P}}_{{\rm{out}}}$$\end{document}$ (=subspace) can be calculated with the Jones formalism to be$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\overrightarrow{P}}_{{\rm{out}}}={M}_{{\rm{WP}}}({n}_{2},{\theta }_{2}){M}_{{\rm{SLM}}}({\rm{\Delta }}{\varphi }_{{\rm{SLM}}},{A}_{{\rm{SLM}}}){M}_{{\rm{WP}}}({n}_{1},{\theta }_{1}){\overrightarrow{P}}_{{\rm{in}}},$$\end{document}$$where the matrices of the reflective SLM and the two rotatable wave plates are given by$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\begin{array}{rcl}{M}_{{\rm{WP}}}(n,\theta ) & = & (\begin{array}{cc}\cos (\theta ) & -\,\sin (\theta )\\ \sin (\theta ) & \cos (\theta )\end{array})\times (\begin{array}{cc}\exp (-i\frac{\pi }{n}) & 0\\ 0 & \exp (i\frac{\pi }{n})\end{array})\\ & & \times (\begin{array}{cc}\cos (\theta ) & \sin (\theta )\\ -\,\sin (\theta ) & \cos (\theta )\end{array}),\\ {M}_{{\rm{SLM}}}({\rm{\Delta }}{\varphi }_{{\rm{SLM}}},{A}_{{\rm{SLM}}}) & = & (\begin{array}{cc}-{A}_{{\rm{SLM}}}(x,y)\exp (-i{\rm{\Delta }}{\varphi }_{{\rm{SLM}}}(x,y)) & 0\\ 0 & 1\end{array})\mathrm{.}\end{array}$$\end{document}$$ ### Spatial phase retardation of polarization components {#Sec9} We use this formalism to evaluate various combinations of parameters and map each corresponding subspace of polarization states in dependence of the phase shift Δ*ϕ* ~SLM~ of the SLM on the Poincaré sphere, as shown in Fig. [5](#Fig5){ref-type="fig"}. If the amplitude ratio *A* ^÷^ is not effected by the SLM (*A* ~SLM~(*x*, *y*) = 1) the subspace of the DPMS can be described as the intersection of a plane and the surface of the Poincaré sphere which gives a circle in a three dimensional space, as illustrated in Fig. [5(a)](#Fig5){ref-type="fig"}. This circle is uniquely describable by the vector $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{r}$$\end{document}$(*r*, *ϑ*, *φ*) = (*r*(*θ* ~1~), *ϑ*(*θ* ~2~), *φ*(*θ* ~2~)) from the origin of the sphere to the center *C* of the circle in spherical coordinates and its radius *R*(*θ* ~1~). In many applications a 1:1 ratio of *P* ~\|\|~ and *P* ~⊥~ (i.e. $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{{\rm{WP}}}_{1}}^{\div}=1$$\end{document}$) is preferred which is facilitated by a retardation of *n* ~1~ ∈ {2, 4} of WP~1~. Furthermore a quarter wave plate (*n* ~1~ = 4) used as WP~1~ shows lower angle sensitivity which simplifies its handling in an experimental setup. In Fig. [5(b,d)](#Fig5){ref-type="fig"} several subspaces in dependence of *θ* ~1~ are shown for a combination of two half and two quarter wave plates (used for WP~1,2~), respectively. Although in both cases WP~1~ is rotated about 45°, the circles are distributed over the whole Poincaré sphere for the half wave plates, while for the quarter wave plates only one half is covered, resulting in lower aberrations per angle uncertainty. In general, it is possible to access every polarization state by at least one combination of wave plates. An overview of all parameters used in Fig. [5](#Fig5){ref-type="fig"} can be found in Table [1](#Tab1){ref-type="table"}. Figure 5Dynamic polarization modulation: (**a**) Subspace of a DPMS. (**b**--**f**) Subspaces in dependence of WP~1~, influencing the radius *R* of the circle. Position of *C* depends on WP~2~ as investigated for different parameters in (**e**--**f**). Experimental results are shown in (**g**--**i**). Parameters for all sub-figures can be found in Table [1](#Tab1){ref-type="table"}.Table 1Retardation Γ(*n*) = *λ*/*n* and orientation angle *θ* ~1,2~ parameters of the two wave plates WP~1,2~ of simulations and experiments shown in Fig. 5.Fig. 5n~1~*θ* ~1~ (deg) from:step:ton~2~*θ* ~2~(deg) from:step:to(b)22.5:5:45290(c)4−4522.5:2.5:45(d)4−90:−5:−135490(e)445445:5:90(f)4−12, 3, 40:2.5:180(g)4−90:−5:−135490(h)445445:5:90(i)4−1.520:10:90(i)4−1.5415:15:165To shift the subspace on the surface of the Poincaré sphere the second wave plate (WP~2~) is used. Its retardation determines the trajectory of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{r}$$\end{document}$ which follows a characteristic curve depending on *n* ~2~ and *θ* ~2~. For a half wave plate used as WP~2~ this trajectory is identical with the equator. Comparable to the longitudes of a sphere, Fig. [5(c)](#Fig5){ref-type="fig"} represents several subspaces of radius *R* = 1 (*θ* ~1~ = 45°) for a variation of *θ* ~2~ in case of WP~2~ being a half wave plate. If WP~2~ is replaced by a quarter wave plate, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{r}$$\end{document}$ follows a lemniscate trajectory (i.e. figure of eight) projected on the surface of the sphere and centered around the input polarization. Thus replacing the half wave plate used as WP~2~ enables to shift $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{r}$$\end{document}$ of the subspace to elliptical and even circular polarization states and the subspace is no longer restricted to linear polarization states. For *R* = 1 a rotation of WP~2~ (=quarter wave plate) leads to a variable inclination of the circle's plane with the equatorial plane, as presented in (e). As polarization states can be part of several circles, the flexibility of the DPMS is very high to choose between different combinations of parameters to spatially control the state of polarization of a light field. Especially a variable wave plate used as WP~2~ allows shifting $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{r}$$\end{document}$ to nearly arbitrary positions on the Poincaré sphere, as illustrated in (f), where *θ* ~1~ = −1° while *θ* ~2~ is varied for *n* ~2~ = {2, 3, 4}. Advantages of our setup compared to previous methods are the two independently rotatable wave plates which allow to tilt a subspace (circle) around the Poincareé sphere (comp. Fig. [5](#Fig5){ref-type="fig"}) and gain higher flexibility. Further, polarization control by a single SLM and auto-alignment of polarization components are benefits of our method. Sub- Fig. [5(g--i)](#Fig5){ref-type="fig"} show the experimental results of Stokes parameter measurements of polarization states modulated with the DPMS. In the first example (g) WP~1~ is rotated in −5° steps by 45° while WP~2~ is fixed in analogy to the simulation in (c). It can be seen, that our measurements match theoretical predictions very well. In (h) WP~2~ is varied while WP~1~ is fixed. Here some deviations can bee seen which can be explained by an aberration of the retardation of the wave plate (within the retardance accuracy of *λ*/200 of the multi-order wave plate). This is verified by the measurement in (i) where $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\overrightarrow{r}$$\end{document}$ does not exactly match the S~3~ axis for *θ* ~2~ = ±45° as expected for WP~2~ being a quarter wave plate, as well. As the retardation of a wave plate typically cannot be influenced, we recommend to use high quality wave plates, only. ### Setting a spatial amplitude ratio of polarization components {#Sec10} In general, any state of polarization is describable as combination of a linear horizontal and a linear vertical polarization state and is explicitly defined by their phase shift Δ*ϕ* = *ϕ* ~*v*~ − *ϕ* ~*h*~ and amplitude ratio *A* ^÷^ = *A* ~*v*~/*A* ~*h*~. Both parameters can be influenced homogeneously for the whole beam profile by setting the orientation of the WPs. In addition, the SLM allows for a spatially resolved modification of phase shift of polarization components by introducing Δ*ϕ* ~SLM~(*x*, *y*) as discussed in the previous section. By this method already a plethora of polarization structures can be realized as for example all possible combinations of linear polarizations, i.e. vector beams. However, it is desired to influence the amplitude ratio spatially as well, giving access to almost any polarization structure. Therefore, we upgrade our method to not only allow polarization states represented on single rings on the Poincaré sphere but to enable polarization structures formed by polarization states given by continuous segments of the surface of the Poincaré sphere. For this purpose, spatial modulation of the amplitude ratio *A* ^÷^ of perpendicular polarization components is implemented. We developed a new method, which is related to combined amplitude and phase modulation^[@CR27]^, but now applied to polarization modulation. More specific, spatially resolved variation of phase Δ*ϕ* and amplitude *A* ^÷^ ratio of polarization components is realized by a phase-only SLM. Therefore, the before mentioned phase Δ*ϕ* ~SLM~(*x*, *y*) is accompanied by an additional amplitude factor $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{\rm{SLM}}}^{^{\prime} }(x,y)$$\end{document}$. The idea is to employ a blazed grating *φ* ~*B*~ to diffract portions of one polarization component to a higher diffraction order (i.e. off-axis propagation) to affect the amplitude ratio *A* ^÷^ of modulated and non-modulated polarization components *P* ~⊥~ and *P* ~\|\|~ within the zeroth order. The zeroth order transmission function of the phase-only function $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$f(x,y)={e}^{i{A}_{{\rm{SLM}}}^{^{\prime} }{\phi }_{B}}$$\end{document}$ is given by$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${T}_{{0}^{th}}(x,y)={e}^{i{A}_{{\rm{SLM}}}^{^{\prime} }(x,y)\pi }\cdot {\rm{sinc}}(-{A}_{{\rm{SLM}}}^{^{\prime} }(x,y\mathrm{)).}$$\end{document}$$ If we choose $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{\rm{SLM}}}={\rm{sinc}}(-{A}_{{\rm{SLM}}}^{^{\prime} })$$\end{document}$ with *A* ~SLM~, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{\rm{SLM}}}^{^{\prime} }\in \mathrm{[0,}\,\mathrm{1]}$$\end{document}$ Equation ([2](#Equ3){ref-type=""}) employs the desired amplitude factor *A* ~SLM~. To eliminate the additional phase term, one has to multiply by its complex conjugate. Employed within the DPMS this technique directly effects the amplitude ratio$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${A}_{{\rm{DPMS}}}^{\div}(x,y)={A}_{{{\rm{WP}}}_{1}}^{\div}\cdot \frac{1}{{A}_{{\rm{SLM}}}(x,y)}\cdot {A}_{{{\rm{WP}}}_{2}}^{\div}$$\end{document}$$of the polarization components and can be combined with the modulation of the phase shift$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\rm{\Delta }}{\varphi }_{{\rm{DPMS}}}(x,y)={\rm{\Delta }}{\varphi }_{{{\rm{WP}}}_{1}}+{\rm{\Delta }}{\varphi }_{{\rm{SLM}}}(x,y)+{\rm{\Delta }}{\varphi }_{{{\rm{WP}}}_{2}}$$\end{document}$$between polarization components as discussed in the previous section. In Fig. [6](#Fig6){ref-type="fig"} a comparison of subspaces of the DPMS is shown (a) with phase retardation only, and (b) with simultaneous modulation of phase retardation and amplitude ratio. It can be clearly seen, that the subspace which also effects the amplitude ratio covers a continuous segment of the surface of the Poincaré sphere and is not limited to a single ring. The orientation and size of the surface depends on Γ~1,2~(*n*), *θ* ~1,2~, Δ*ϕ* ~SLM~(*x*, *y*) and *A* ~SLM~(*x*, *y*).Figure 6Spatial amplitude retardation of polarization components: Subspace of DPMS with (**a**) phase retardation only, and (**b**) phase retardation and additional influence to the amplitude ratio of polarization components (*θ* ~1,2~ = −45°). Spatially resolved Stokes parameter measurements {#Sec11} ------------------------------------------------ In the experiment a reflective SLM is illuminated under a small angle by a linearly polarized laser (*λ* = 532 nm). A half wave plate (HWP in Fig. [4(a)](#Fig4){ref-type="fig"}) allows to choose a linear horizontal input polarization which is parallel to the director of the liquid crystal display and a polarization measurement system depicted in Fig. [4(a)](#Fig4){ref-type="fig"} (II). To evaluate the polarization modulation of the DPMS, intensity (*I*) measurements behind a rotating quarter wave plate (rotation angle *θ* ~3~) and a fixed linear polarizer are related to the corresponding Stokes vector of the polarization state^[@CR22]^ by Fourier analysis:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${S}_{0}=\frac{2}{N}\sum _{j=1}^{N}{I}_{j}-2{I}_{j}\,\cos (4{\theta }_{{3}_{j}})$$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${S}_{1}=\frac{8}{N}\sum _{j=1}^{N}{I}_{j}\,\cos (4{\theta }_{{3}_{j}})$$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${S}_{2}=\frac{8}{N}\sum _{j=1}^{N}{I}_{j}\,\sin (4{\theta }_{{3}_{j}})$$\end{document}$$ $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${S}_{3}=\frac{4}{N}\sum _{j=1}^{N}{I}_{j}\,\sin (2{\theta }_{{3}_{j}}),$$\end{document}$$where *N* \> 8 is the number of measurements (minimum given by sampling theorem). To yield normalized Stokes parameters *S* ~1,2,3~ have to be divided by *S* ~0~. Within the experimental setup, we employed a camera to enable spatially resolved Stokes parameter measurements. While for homogenous polarization distributions it is possible to take average values of several camera pixels, for the complex spatially structured Poincaré beams a pixel wise evaluation of Stokes parameters is performed. Opposite to other methods frequently used for vector beam analysis, as e.g. intensity observations behind a rotating polarizer, here beside orientation also ellipticity information of polarization states is recorded. Conclusion {#Sec12} ========== Simultaneous modulation of spatial phase, amplitude, and polarization values by a holographic method enables to tailor complex Poincaré beams, advanced optical angular momentum structures and various kinds of polarization singularities. To show the potential of the method, we demonstrate the generation of higher order LG modes combined with spatial modulation of polarization states. However, our holographic modulation system (DPMS) is not restricted to LG beams, but allows generating any light mode together with linear, elliptic and circular states of polarization in various combinations. Furthermore, spatial combination of spin and orbital angular momentum, as shown for a higher order helical LG beam, can be employed in quantum communication and classical entanglement^[@CR37]^. Moreover, such polarization tailored light beams are of interest for high NA focusing and its applications, where three-dimensional polarization structures including z-polarization components can be tailored^[@CR38]^. Of special interest are experimental results and simulations of complex Poincaré beams embedding different types of polarization singularities. Realization of polarization singularities (V-points, L-lines and C-lines) and their analysis are of general importance as singular structures are not only present in optics but (among others) also occur in astronomy or Bloch walls of magnetic systems. Although already several polarization singularities have been modulated with the DPMS, our system will lead to experimental realizations of even more complex polarization structures^[@CR32],\ [@CR33],\ [@CR35],\ [@CR36]^ combining several types of singularities to whole polarization networks. Additionally, we expect distinctive dynamics if (unstable) polarization singularities are combined with (unfolding) phase singularities. Finally, capabilities of the DPMS have been evaluated within the method section. It has been characterized with respect to the phase of the SLM, the retardation of the wave plates and their orientation to demonstrate the enormous flexibility of the method. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors thank the German Research Foundation (DFG) for financial support of this project in the frame of the German-Chinese transregional research project TRR61. C.A. and E.O. conceived the experiment and conducted simulations. C.S. and E.O. carried out experiments. All authors discussed the results and reviewed the manuscript. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

Introduction {#sec0005} ============ New automated blood cell analyzers can provide information about individual cell characteristics, such as hemoglobin content of reticulocytes, hemoglobin content of mature erythrocytes, and percentages of microcytic erythrocytes and hypochromic cells. These parameters have been used in the diagnosis of iron deficiency anemia (IDA), β-thalassemia (β-Thal),[@bib0120; @bib0125; @bib0130] and anemia of chronic disease (ACD).[@bib0135; @bib0140] The differentiation between these three conditions is very important as the clinical approach is different in each particular diagnosis. As reticulocytes have a normal life span of one or two days, information concerning the hemoglobin content of young red cells is a good indication of iron availability and an early marker of iron-deficiency erythropoiesis.[@bib0145] Reticulocyte hemoglobin equivalent (Ret-He) reflects real-time information on the synthesis by young erythrocytes in bone marrow. Other available parameters are the percentage of red cells with Hb content equivalent to or less than 17 pg (%HypoHe), and the percentage of red cells with a volume less than 60 fL (% MicroR),[@bib0120] which corresponds to a sub-population of mature red cells exhibiting evidence of insufficient iron content. A mathematical formula using %MicroR and %HypoHe (MHe) proposed by Urrechaga et al.[@bib0150] tested discriminant indices in healthy individuals, β-Thal and IDA patients. A sensitivity of 97.4% and specificity of 97.1% was reported in differentiating β-Thal from mild IDA. Anemia associated with chronic inflammation, infection or malignancy is the most common anemia in hospitalized patients. Although stainable iron is present in bone marrow, elevated levels of inflammatory cytokines interfere in erythropoiesis leading to a hyporegenerative anemia and defective iron incorporation into erythrocyte progenitors. Reduced concentrations of circulating iron and normal or increased iron stores characterize the state of functional iron deficiency.[@bib0155] Anemia of inflammation can be associated with absolute iron deficiency (ACD combi) generally in patients with inflammatory disease and chronic blood loss. Differentiation between ACD and ACD combi is clinically important, but in the clinical practice, differentiation is difficult using conventional biomarkers such as ferritin concentration and transferrin saturation.[@bib0160] The soluble transferrin receptor/log ferritin ratio may be useful in distinguishing ACD from ACD combi.[@bib0165] The aim of this study was to analyze the effectiveness of new laboratory parameters related to mature red blood cells and reticulocytes to differentiate three conditions related to iron deficiency: IDA, ACD and ACD combi. Moreover, the performance of these parameters was tested to distinguish IDA from β-Thal, two common causes of microcytic anemia. Methods {#sec0010} ======= This project was approved by the Ethics Committee of the Faculty of Medical Sciences, Universidade Estadual de Campinas (UNICAMP), São Paulo, Brazil. All samples were selected from routine collections and informed consent was waived. Peripheral blood samples from 117 adult patients with anemia (Hb \< 12.0 g/dL for women and Hb \< 14.0 g/dL for men) were selected from routine workload. Blood analysis had been requested by general practitioners, in general to investigate anemia. Patients were classified according to iron status analysis (commercial kits from Roche Diagnostics Germany): IDA when serum iron (SI) levels were \<45 mg/dL for men and \<30 mg/dL for women; percentage transferrin saturation (%TS) \<15%; serum ferritin (SF) \<30 μg/L for men and \<13 μg/L for women. Patients were classified as ACD when SI levels were normal or reduced (40--160 mg/dL and 30--160 mg/dL for men and women, respectively), %TS normal or decreased (30--50%), normal or high SF (30--400 μg/L and 13--150 μg/for men and women, respectively) and C-reactive protein (C-RP) \> 5 mg/dL (Tina-Quant C-Reactive Protein, Roche Diagnostics, Germany). Soluble transferrin receptor (sTfR) levels (Roche Diagnostics, Germany) were measured in all samples, and the sTfR/log ferritin ratio was used to identify iron deficiency in patients with ACD. Patients with ACD having a sTfR/log ferritin ratio \> 2.06 or sTfR \> 3.71 μg/mL (cut-off values indicative of iron deficiency in our laboratory) were classified as ACD combi. Twenty-six patients had the diagnosis of β-Thal according to the hemoglobin A~2~ level determined by high performance liquid chromatography (HPLC- Variant II -- Hemoglobin Testing System, Bio-Rad Laboratories, Inc., CA, USA). Patients with β-Thal associated with other kinds of anemia, patients with reticulocytosis or pancytopenia, individuals who had received transfusions within the previous three months, and patients on iron replacement therapy were excluded from the study. A control group (CG) was compounded by healthy individuals with no clinical signs or symptoms of disease, including acute inflammatory/infectious conditions, normal hematologic findings, and C-RP \< 5 mg/L. Healthy individuals were students or laboratory staff, all of whom donated blood samples on a voluntary basis. Determination of erythrocyte and reticulocyte parameters was performed using the Sysmex XE-5000 automated hematological analyzer (Sysmex, Kobe, Japan), which provides the following parameters: Ret-He, %MicroR, hemoglobin content of erythrocytes obtained from the optical counting of red blood cells (RBC-He), and %HypoHe. The MHe index was calculated as: %MIcroR-%HypoHe.[@bib0150] Mann--Whitney test was applied to compare groups. Receiver operator characteristic analysis (ROC) was used to evaluate the accuracy of the parameters to differentiate between types of anemia. The level of significance was set for a *p*-value \<0.05. Data were analyzed using the SPSS statistics program for Windows (version 13.0. SPSS Inc. Chicago, IL, USA). Results {#sec0015} ======= According to adopted criteria, individuals were classified as IDA (42 patients), β-Thal (25 individuals), ACD combi (22 patients), ACD (28 patients) and CG (54 individuals). [Table 1](#tbl0005){ref-type="table"} describes the demographic characteristics and laboratorial data of the patients and the CG, and [Table 2](#tbl0010){ref-type="table"} shows the iron status measurements used to classify the patients in different groups. As expected the β-Thal group had the highest %MicroR ([Table 3](#tbl0015){ref-type="table"}). The %HypoHe was higher in the β-Thal group compared to other groups except for IDA patients. However, as the microcytic cells were more abundant in the β-Thal group, when these two parameters were associated in the MHe index, this difference became more evident and statistically significant. The Mann--Whitney test identified no difference between the IDA and ACD combi groups in respect to any parameter. When the ACD and ACD combi groups were compared, the RBC-He and Ret-He were significantly lower in ACD combi patients (*p*-value = 0.016 and *p*-value = 0.003, respectively). Furthermore, the %MicroR and %HypoHe were significantly higher in ACD combi patients, as was the MHe index (*p*-value = 0.001, *p*-value = 0.003 and *p*-value = 0.014, respectively). Although the ACD group had a sTFR/log ferritin ratio below the cut-off indicative of iron deficiency, the Ret-He and RBC-He values were significantly lower than the CG whereas the %HypoHe, %MicroR and MHe index were higher (*p*-value \< 0.001 for all comparisons). The best test for differentiating IDA from β-Thal was the MHe index (area under curve \[AUC\]: 0.977; 95% confidence interval \[95/% CI\] 0.950--1.005). A cut-off of 13.8 was very sensitive (96.2%) and specific (92.7%) to identify IDA patients. The parameter %MicroR was effective (AUC 0.886: CI: 95% 0.810--0.963) to detect iron deficiency with sensitivity and specificity for values \<25.0% of 84.6% and 78.0%, respectively. The ROC curve for %HypoHe was the most effective to differentiate between ACD and ACD combi although with a moderate AUC (0.785: CI: 95% 0.661--0.909), sensitivity 72.7%, and specificity 71.4% (cut-off 1.8%). The %HypoHe was the best parameter to distinguish between IDA and ACD (AUC: 0.835; 95% CI: 0.737--0.933). The sensitivity and specificity of a %HypoHe \< 2.45% to identify ACD were 75.4% and 70.4%, respectively. The results of the ROC analysis were lower for RBC-He (AUC: 0.809; 95% CI: 0.696--0.922), Ret-He (AUC: 0.780; 95% CI: 0.661--0.899), and %MicroR (AUC: 0.785; 95% CI: 0.662--0.908). The capacity of the tests to discriminate IDA from ACD combi was not satisfactory and the AUC was less than 0.700 for all these parameters. Discussion {#sec0020} ========== The diagnostic accuracy of reticulocyte parameters has been tested by many authors, especially to diagnose iron deficiency in patients submitted to dialysis.[@bib0145; @bib0165; @bib0170] Measurement of the reticulocyte content is helpful in detecting early stages of iron deficiency prior to the development of anemia.[@bib0145; @bib0175; @bib0180] A reduction in reticulocyte hemoglobin has been observed in other conditions besides iron deficiency, such as in hemoglobinopathies.[@bib0185; @bib0190] The effectiveness of reticulocyte parameters to diagnose IDA and ACD was reported in other research.[@bib0125; @bib0155; @bib0195] In a recent study with geriatric patients, the authors concluded that Ret-He is no better than classic indices, such as mean cell hemoglobin and mean cell hemoglobin concentration, in differentiating between IDA and ACD.[@bib0140] Our results showed that, although the Ret-He value was lower in IDA than ACD, the accuracy of the test to distinguish both types of anemia was moderate, and less effective compared to %HypoHe. The potential of Ret-He was demonstrated in a study with patients with chronic rheumatic disease and anemia. The predictive value of RetHe was tested in response to oral iron therapy and, according to the authors, the findings support the role of Ret-He as a marker for iron responsiveness.[@bib0200] Additional extended red blood cell parameters besides Ret-He have been tested, such as %MicroR, and %HypoHe, and indices generated by combining them.[@bib0205] The results regarding the differentiation between IDA and β-Thal are promising, although the optimum cut-off varies according to the population studied and the criteria adopted to classify anemia.[@bib0205; @bib0210] The MHe index was first proposed and tested by Urrechaga et al. [@bib0195] The performance of this index was better (sensitivity 98.0% and specificity 95.9) than other published indices. The authors suggest that samples with MHe values \>11.5 can be chosen for further analysis to confirm the diagnosis of thalassemia; these data are coincident with our results. The MHe index was the best parameter to discriminate IDA from β-Thal, although the cut-off value used was different from the value described by Urrechaga et al.[@bib0195] probably because patients were not stratified according to the severity of anemia. As far as we know, no other reports exist about the efficiency of expanded erythrocyte parameters in identifying patients with anemia of inflammation associated with absolute iron deficiency. The clinical usefulness of the determination of %Hypo has long been recognized in differentiating between iron-deficient and iron-sufficient patients with chronic kidney disease receiving erythropoietin stimulating agents.[@bib0215] The current study applied the sTfR/log ferritin ratio to identify iron deficiency in patients with ACD. Measurement of the sTfR is considered a good indicator of functional iron status, as it does not suffer any influence of systemic inflammation unlike the SI, transferrin, and ferritin levels.[@bib0220; @bib0225] Thus, the transferrin receptor synthesis is stimulated when there is a reduction of functional iron as shown by the results of this study. Iron deficiency is characterized by an increase in sTfR levels and low ferritin values, while in anemia of inflammation, transferrin receptor levels are slightly affected whereas serum ferritin increases greatly.[@bib0230] In the clinical practice this differentiation is important because iron supplementation is beneficial to ACD combi patients, but may be deleterious for ACD patients. According to the results of this study, even though the sensitivity and specificity were moderate, the best parameter for distinguishing ACD combi from ACD is %HypoHe, followed by MicroR. It is interesting to note that there was no difference between the IDA and ACD combi groups in any test, different to what was observed when ACD was compared with IDA. In fact, the absolute iron deficiency associated to ACD increases the number of microcytic and hypochromic red cells, causing cell features similar to IDA. ACD patients showed evidence of reduced iron availability for erythropoiesis, but the disturbance of the iron metabolism in functional iron deficiency was less remarkable than in the association of ACD with IDA. In practical terms the incorporation of new cell indices can speed up the diagnosis of IDA, β-Thal and ACD, and consequently target more quickly and more precisely the subsequent confirmatory exams in order to introduce appropriate treatment. On the other hand, the difficulty in identifying absolute iron deficiency in patients with inflammatory conditions remains. Therefore, the challenge persists, and other studies are needed to identify a parameter with clinical decision value. Conflicts of interest {#sec0025} ===================== The authors declare no conflicts of interest. ###### Demographic characteristics and hematological data of patients and the control group. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Parameter IDA\ β-Thal\ ACD combi\ ACD\ CG\ (*n* = 42) (*n* = 25) (*n* = 22) (*n* = 28) (*n* = 54) ------------------------ ------------------------------------------------------- ------------------------------------------- ------------------------------------------- -------------- -------------- Age (years) 42.0[a,b](#tblfn0005 tblfn0010){ref-type="table-fn"}\ 54.0[a](#tblfn0005){ref-type="table-fn"}\ 65.6[b](#tblfn0010){ref-type="table-fn"}\ 51.0\ 41.0\ (16.0--82.0) (15.0--76.0) (20.0--81.0) (17.0--82.0) (16.0--75.0) Gender male/female (%) 26.2/73.8 66.7/33.3 42.9/57.1 63.6/36.4 35.8/62.3 Hb (g/dL) 10.1\ 12.0\ 9.9\ 9.5\ 13.7\ (6.2--13.6) (6.9--14.7) (7.3--12.2) (6.6--12.8) (12.0--16.1) MCV (fl) 75.2\ 65.7\ 74.4\ 81.1\ 87.9\ (61.2--79.6) (58.4--70.8) (58.5--88.9) (61.4--95.2) (80.5--94.8) MCH (pg) 22.5\ 20.4\ 24.2\ 26.8\ 29.1\ (17.0--26.1) (18.4--24.4) (18.4--28.9) (19.6--32.0) (25.4--32.2) RDW (%) 16.4\ 16.3\ 16.6\ 15.5\ 13.2\ (13.4--21.0) (14.0--19.0) (13.7--28.2) (13.3--25.6) (12.0--15.5) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Values as medians, minimum and maximum. IDA: iron deficiency anemia: β-Thal: heterozygous beta-thalassemia; ACD combi: anemia of chronic disease associated with absolute iron deficiency; ACD: anemia of chronic disease; CG: control group; Hb: hemoglobin; MCV: mean cell volume; MCH: mean cell hemoglobin; RDW: red cell distribution width. *p*-value = 0.030. *p*-value = 0.023. ###### Biochemical data. ---------------------------------------------------------------------------------------------------- Parameters IDA\ β-Thal\ ACD combi\ ACD\ CG\ (*n* = 42) (*n* = 25) (*n* = 22) (*n* = 28) (*n* = 54) ------------------- ------------- ---------------- ---------------- ---------------- --------------- SI\ 24.5\ 94.0\ 23.5\ 28.0\ 100.0\ (mg/dL) (7.0--73.0) (49.0--224.0) (12.0--99.0) (9.0--98.0) (49.0--185.0) TS (%) 6.1\ 31.9\ 9.2\ 11.6\ 29.5\ (1.5--15.5) (22.4--68.2) (4.5--25.7) (3.0--42.2) (17.6--50.1) SF (μg/L) 8.2\ 224.5\ 254.5\ 163.7\ 59.6\ (3.2--23.4) (66.8--1619.0) (27.8--2000.0) (42.1--2000.0) (19.7--407.9) sTfR (μg/mL) 9.4\ 7.4\ 7.0\ 2.7\ 2.4\ (1.4--28.8) (2.6--25.5) (3.5--22.4) (0.5--6.2) (1.6--3.8) sTfR/log ferritin 10.4\ 2.6\ 3.4\ 1.1\ 1.3\ (1.9--40.9) (1.3--10.9) (1.3--9.6) (0.2--3.8) (0.8--2.2) ---------------------------------------------------------------------------------------------------- Values as medians, minimum and maximum. IDA: iron deficiency anemia: β-Thal: heterozygous beta-thalassemia; ACD combi: anemia of chronic disease associated with absolute iron deficiency; ACD: anemia of chronic disease; CG: control group; SI: serum iron, TS: transferrin saturation; SF: serum ferritin; sTfR: soluble transferrin receptor. ###### Reticulocyte and red cell indices for patients and the control group. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Parameters IDA\ β-Thal\ ACD combi\ ACD\ CG\ (*n* = 42) (*n* = 25) (*n* = 22) (*n* = 28) (*n* = 54) ------------- ------------------------------------------- -------------- ------------------------------------------- ------------------------------------------- -------------- Ret-He (pg) 25.2\ 23.0\ 26.0[b](#tblfn0020){ref-type="table-fn"}\ 29.7[c](#tblfn0025){ref-type="table-fn"}\ 35.1\ (16.9--32.6) (21.1--30.2) (19.2.--37.0) (21.0--39.5) (31.0--39.2) RBC-He (pg) 23.4[a](#tblfn0015){ref-type="table-fn"}\ 21.9\ 26.4[b](#tblfn0020){ref-type="table-fn"}\ 28.7[c](#tblfn0025){ref-type="table-fn"}\ 31.8\ (16.7--35.3) (18.8--26.3) (19.5--30.5) (20.8--34.7) (28.0--34.2) HypoHe (%) 6.2\ 9.4\ 4.4[b](#tblfn0020){ref-type="table-fn"}\ 1.4[c](#tblfn0025){ref-type="table-fn"}\ 0.2\ (0.4--44.5) (1.4--20.4) (1.1--31.5) (0.3--15.3) (0.1--0.8) MicroR (%) 13.5[a](#tblfn0015){ref-type="table-fn"}\ 29.2\ 14.2[b](#tblfn0020){ref-type="table-fn"}\ 5.55[c](#tblfn0025){ref-type="table-fn"}\ 1.4\ (0.4--48.5) (19.3--54.2) (0.6--59.0) (1.2--48.9) (0.5--4.0) MHe 5.8[a](#tblfn0015){ref-type="table-fn"}\ 23.0\ 7.2[b](#tblfn0020){ref-type="table-fn"}\ 4.05[c](#tblfn0025){ref-type="table-fn"}\ 1.15\ (−9.8--22.2) (8.1--34.6) (0--27.5) (0.5--36.3) (0.3--3.7) ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Values as medians, minimum and maximum. Mann--Whitney test was applied to compare groups. IDA: iron deficiency anemia: β-Thal: heterozygous beta-thalassemia; ACD combi: anemia of chronic disease associated with absolute iron deficiency; ACD: anemia of chronic disease; CG: control group; Ret-He: reticulocyte hemoglobin content; RBC-He: red blood cell hemoglobin content; HypoHe: percentage of hypochromic erythrocytes; MicroR: percentage of hypochromic erythrocytes; MHe: MicroR-HypoHe index. *p*-value \< 0.01 (IDA × β-Thal). *p*-value ≤ 0.01 (ACD × ACD combi). *p*-value \< 0.001 (ACD × CG).

See related research by Martin *et al*., <http://breast-cancer-research.com/content/16/2/R38> When treatment is based solely on clinicopathological risk factors, many women with estrogen receptor-positive/human epidermal growth factor receptor 2-negative (ER^+^/HER2^−^) tumors are overtreated -- subjected to morbidity from cytotoxic chemotherapy for negligible benefit. Identifying patients safely treated by endocrine therapy alone has driven the development of prognostic gene expression assays. In the previous issue, Martin and colleagues describe the third clinical validation of EndoPredict (EP; Sividon Diagnostics GmbH, Cologne, Germany) \[[@B1]\], a second-generation multigene test trained to predict distant recurrence in ER^+^/HER2^−^ tumors, and by extension the need for adjuvant chemotherapy. EP was previously validated in prospective--retrospective analyses of endocrine-treated postmenopausal ER^+^/HER2^−^ breast cancer patients in two clinical trials (ABCSG-6 and ABCSG-8) \[[@B2]\]. Genomic-based assays developed from complex, high-dimensional data are susceptible to overfitting. Clinical validation must be performed in entirely independent datasets using predefined, locked-down classifier algorithms and analysis plans. Martin and colleagues' study exemplifies the rigor required by Simon and colleagues for a formal prospective--retrospective study to contribute to generating level IB evidence \[[@B3]\]. The authors present EP validation results in ER^+^/HER2^−^ patients from the GEICAM/9906 clinical trial of node-positive women treated with contemporary chemotherapy. The prognostic ability of EP remains robust in this higher risk group, identifying a 10-year metastatic rate of 7% in the predefined EP low-risk group (versus 30% in the high-risk group). Including tumor size and nodal status as the EPclin classifier identifies a small (13%) but impressively low-risk cohort of women who experienced no distant recurrences at 10 years. As the whole of this cohort received chemotherapy, the clinical utility of this finding (to avoid chemotherapy) is difficult to infer, although the 100% metastasis-free survival in both the fluorouracil--epirubicin--cyclophosphamide and the fluorouracil--epirubicin--cyclophosphamide--paclitaxel randomized arms does imply no benefit from adding paclitaxel. The authors speculate that the EPclin low-risk group (recurrence-free at 10 years) in this 5-year endocrine therapy-treated population may identify women not needing extended endocrine therapy, consistent with the ABCSG-8 trial \[[@B4]\]. This promising idea must be interpreted cautiously given that only 16 of the 74 EPclin low-risk patients had 10-year follow-up data. Both the aTTom and ATLAS trials have shown that survival benefits of extended hormonal therapy become more apparent after year 10 \[[@B5],[@B6]\]. Several multigene prognostic assays have now been developed for use in ER^+^/HER2^−^ breast cancers. First-generation assays including MammaPrint (MammaPrint: Agendia, Amsterdam, The Netherlands) and Onco*type* DX (Oncotype: Genomic Health, Redwood City, CA, USA) suffered from early methodological issues, most seriously a failure to maintain rigorous separation between training and validation sets, and inclusion of nonluminal and/or HER2^+^ tumors in their training sets, thereby allowing these high-risk tumors to skew outcome-related gene selection away from the relevant patient group \[[@B7]-[@B9]\]. MammaPrint was specifically trained around early relapse (within 5 years) in node-negative women, most having received no adjuvant systemic therapy, and has not been shown to predict late recurrence outside the original training-validation cohort. Onco*type* DX heavily weighed the tamoxifen-only arm of the NSABP-B20 trial in its training set, where most recurrences occurred within 5 years, and has diminished prognostic ability beyond year 5 \[[@B10]\]. More recently, building upon biological and technical advances and more rigorous approaches to validation, second-generation multigene tests have been developed, including the Breast Cancer Index (BCI: bioTheranostics, San Diego, CA, USA), PAM50 (PAM50: NanoString Technologies Inc., Seattle, WA, USA) and EP. The Breast Cancer Index combines a molecular grade index (quantifying tumor grade-associated genes) and a two-gene ratio, *HOXB13*:*IL17BR*, related to estrogen signaling \[[@B11]\]. PAM50, unlike signatures trained around outcome, was developed as a biological classifier of the major intrinsic molecular subtypes of breast cancer \[[@B12]\]. These three assays predict both early and late recurrences \[[@B4],[@B10],[@B13]\]. IHC4 and Mammostrat (Mammostrat: Clarient, Inc., Aliso Viejo, CA, USA) immunohistochemical panels are also prognostic in early breast cancer \[[@B14],[@B15]\]. IHC4 uses standard pathology markers (ER, progesterone receptor, HER2 and Ki67) to provide prognostic information comparable with Onco*type* DX \[[@B14]\]. Immunohistochemical staining and scoring does suffer from limited analytical reproducibility, probably contributing to Martin and colleagues' identification of low Ki67 scores (\<14%; a published cutoff point for good-prognosis luminal A tumors) in a surprisingly high fraction (almost three-quarters) of this node-positive cohort \[[@B16]\]. Each of these gene expression and immunohistochemical panels identifies a good prognosis group that may not need chemotherapy. Emerging evidence suggests that some panels identify women at such low risk of late recurrence that they may safely avoid extended endocrine therapy. For high-risk women, however, the question is not one of chemotherapy versus no chemotherapy, but rather a question of which chemotherapy agent(s) will be most effective for which patients -- a true predictive indication. In Martin and colleagues' report, the EP score did not predict benefit from adding weekly paclitaxel to fluorouracil--epirubicin--cyclophosphamide chemotherapy. Outcome-trained signatures from nonchemotherapy populations are unlikely to predict between chemotherapy regimens; Table [1](#T1){ref-type="table"} summarizes some relevant features of the referenced molecular signatures, including predictive studies. ###### Overview of selected multigene signatures for breast cancer **Assay; platform, clinical material** **Training parameter** **Approval or endorsement** **Analytical validity: published assay validation** **Clinical validity: prognosis validation** **Predicting treatment benefit using randomized clinical trials** **Randomized prospective trials** ----------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------ ----------------------------------------------------- --------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------- --------------------------------------------------- ---------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------- Breast Cancer Index; RT-PCR, FFPE (central) Outcome (ER^+^, pN0, endocrine-treated women) MGI component -- biology (tumor grade related genes) H:I component -- outcome (recurrence in tamoxifen-treated women) No No ATAC \[[@B13]\], Stockholm \[[@B17]\], multiple nonrandomized trial cohorts No^a^ No No No No No EndoPredict; RT-PCR, FFPE (distributed) Outcome (distant recurrence in endocrine-treated ER^+^/HER2^−^ pN0/pN^+^ women) CE Mark Yes \[[@B18],[@B19]\] ABCSG6 \[[@B2]\], ABCSG8 \[[@B2]\], GEICAM/9906 \[[@B1]\] No No No No GEICAM/9906 \[[@B1]\] (failed to predict benefit) No IHC4; IHC, FFPE (distributed) Outcome (distant recurrence in ER^+^ endocrine-treated women) No No ATAC \[[@B14]\], TEAM \[[@B20]\] No No No No No No MammaPrint; microarray, fresh and FFPE (central) Outcome (5-year metastasis rate in pN0 women) FDA (fresh): risk for distant metastasis, \<61 years, stage I and II, tumor ≤5 cm and node-negative No Multiple nonrandomized trial cohorts including RASTER No No No No No MINDACT prognosis validation (to report 2015) Mammostrat; IHC, FFPE (central) Outcome (unselected cohort of breast cancer patients) No No NSABP-B14, NSABP-B20 \[[@B15]\] multiple nonrandomized trial cohorts No No NSABP-B20 (±CMF) \[[@B15]\] (all women benefit -- high risk benefit the most) No No No Onco*type* DX; RT-PCR, FFPE (central) Outcome (recurrence in mainly tamoxifen-treated ER^+^, pN0 women) NCCN, ASCO, St. Gallen (role for identifying women that may benefit from chemotherapy) Yes \[[@B21]\] NSABP-B14 \[[@B9]\], NSABP-B28 \[[@B22]\], SWOG8814 \[[@B23]\], multiple nonrandomized trial cohorts NSABP-B14 \[[@B24]\] (largest benefit in quantitative ER high/recurrence risk low patients) No NSABP-B20 (±CMF) \[[@B25]\], SWOG8814 (±CAF) \[[@B23]\] (benefit from chemotherapy with high recurrence score) No NSABP-B28 \[[@B22]\] (failed to predict a benefit) TAILORx (node-negative, to report 2015) RxPONDER (one to three positive nodes, recruiting) PAM50 (research based assay); RT-PCR and microarray, FFPE and fresh (distributed) Biology (identification of major molecular subtypes) N/A research assay No NCIC-MA5 \[[@B26]\], NCIC-MA12 \[[@B27]\], GEICAM/9906 \[[@B28]\], multiple nonrandomized trial cohorts NCIC-MA12 \[[@B27]\] (luminal subtype predicts benefit) NOAH \[[@B29]\] (HER2-enriched benefits the most) No NCIC-MA5 \[[@B26]\] (CMF vs. CEF; epirubicin benefit in HER2-enriched subtype only) GEICAM/9906, CALGB/9342 and CALGB/9840 \[[@B30]\] (low proliferation score predicts weekly paclitaxel benefit) No Prosigna; nCounter, FFPE (distributed) Biology (subtype); outcome (ROR score) CE Mark, Health Canada, FDA: prediction of 10-year DRFS in ER^+^, node 0 to 3, postmenopausal women treated with endocrine therapy Yes \[[@B31]\] ATAC \[[@B32]\], ABCSG08 \[[@B33]\] No No No No No RxPONDER (one to three nodes, recruiting; embedded additional analysis) CAF, cyclophosphamide, doxorubicin, fluorouracil; CEF, cyclophosphamide, epirubicin, fluorouracil; CMF, cyclophosphamide, methotrexate and fluorouracil; DRFS, distant relapse-free survival; ER, estrogen receptor; FDA, US Food and Drug Administration; FFPE, formalin-fixed paraffin-embedded; HER2, human epidermal growth factor receptor 2; H:I, *HOXB13*:*IL17BR*; IHC, immunohistochemistry; MGI, molecular grade index; N/A, not applicable; pN0, pathological lymph node-negative; pN^+^, pathological lymph node-positive; ROR, risk of recurrence; RT-PCR, reverse transcription polymerase chain reaction. Breast Cancer Index: bioTheranostics, San Diego, CA, USA; EndoPredict: Sividon Diagnostics GmbH, Cologne, Germany; IHC4: MammaPrint: Agendia, Amsterdam, The Netherlands; Mammostrat: Clarient, Inc., Aliso Viejo, CA, USA; Onco*type*: Genomic Health, Redwood City, CA, USA; PAM50: NanoString Technologies Inc., Seatlle, WA, USA; Prosigna: NanoString Technologies Inc., Seattle, WA, USA. ^a^Nested cohort study using material from NCIC CTG MA.17 -- HOXB13/IL17BR predictive of benefit from extended letrozole. What does the future hold for gene expression signatures? Cheaper and faster next-generation sequencing has been touted as the pinnacle of personalized medicine, destined to render multigene expression assays obsolete. However, the genetic complexity of tumors (copy number variations, chromosome-scale structural changes, thousands of mutations, epigenetic changes and intratumoral genetic heterogeneity) is proving even more complex than anticipated. Much as the increased detail from electron microscopy never did replace light microscopy for cancer diagnosis, the broader signatures detected by representative gene expression profile assays, reflecting clinically significant patterns common across many patients, are likely to remain relevant for important treatment decisions. Abbreviations ------------- EP: EndoPredict; ER: Estrogen receptor; HER2: Human epidermal growth factor receptor 2. Competing interests ------------------- TON reports a proprietary interest in the PAM50 assay, which has been licensed to Nanostring Technologies. ZK has no competing interests.

Introduction {#s1} ============ Rheumatoid arthritis (RA) is a systemic autoimmune disease with a heterogeneous clinical spectrum. At present, our ability to predict this heterogeneity is poor, likely because of limitations in understanding its molecular complexity. Recent efforts to improve outcome in patients with RA have focused on the early stages of the disease [@pone.0085248-Schoels1], when aggressive treatment can slow progression and change long-term course. Currently, another crucial therapeutic goal is to delay/prevent progression of undifferentiated arthritis (UA) and one of the challenges faced by rheumatologists is the classification of these patients. Thus, many studies have attempted to identify prognostic markers in early RA or UA that correlate with disease progression in order to establish which patients are at risk for poor outcome or, conversely, which are destined to have a more benign disease so that overtreatment can be avoided [@pone.0085248-Smolen1]. However, criteria for this clustering of patients are scarce. To date, validated biomarkers of severity such as rheumatoid factor (RF) or anti-citrullinated peptide antibodies (ACPA) do not enable us to identify those patients who require more intensive treatment. Although challenging, identification of such predictive tools would help us tailor treatment. Vasoactive intestinal peptide (VIP) is produced by neural, endocrine, and immune cells. It has anti-inflammatory and immunoregulatory effects [@pone.0085248-Gomariz1], which are mediated by three G-protein--coupled receptors (VPAC1, VPAC2, and PAC1) [@pone.0085248-Couvineau1], [@pone.0085248-Harmar1]. First studies evaluating the endogenous role of VIP were performed in transgenic mice harboring a chimeric VIP gene which showed reduced brain VIP content and deficiencies in learning abilities supporting an important function for VIP in vivo [@pone.0085248-Gozes1]. Recently, it has been described that VIP-deficient mice display certain physiological abnormalities [@pone.0085248-Szema1], [@pone.0085248-Colwell1] and exhibit reduced mortality and impaired proinflammatory responses to lipopolysaccharide-induced endotoxemia [@pone.0085248-Abad1], suggesting that defects in the innate arm of immunity may occurs in the chronic absence of VIP. Numerous reports have focused on the effects of VIP treatment. Thus, administration of VIP has demonstrated therapeutic effects in several murine models of inflammatory/autoimmune diseases [@pone.0085248-Delgado1]--[@pone.0085248-Abad3]. In a murine model of collagen-induced arthritis, administration of VIP reduced joint inflammation and destruction, thus decreasing the inflammatory response and inducing a shift in the Th1/Th2 balance [@pone.0085248-Delgado1], [@pone.0085248-Juarranz1]. Research in human models has confirmed the modulatory effects of VIP in *ex vivo* assays with fibroblast-like synoviocytes and peripheral blood lymphocytes from patients with RA through down-regulation of proinflammatory mediators [@pone.0085248-Juarranz2]--[@pone.0085248-GutirrezCaas1]. Decreased expression of VIP was recently reported in the synovial fluid of patients with osteoarthritis (OA) and poor radiological progression, indicating a protective role for VIP [@pone.0085248-Jiang1]. Moreover, patients with juvenile idiopathic arthritis have lower serum levels of VIP than healthy controls [@pone.0085248-ElSayed1]. Consequently, the role of endogenous VIP in the pathophysiology of RA is becoming clearer. Our hypothesis is that VIP serum levels are correlated with disease severity. Therefore, the objective of this work was to assess VIP levels during the follow-up of patients with early arthritis (EA) and to explore its potential value as a biomarker in RA. Methods {#s2} ======= Ethics Statement {#s2a} ---------------- The register protocol was reviewed and approved by the Ethics Committee for Clinical Research at the Instituto de Investigación Sanitaria La Princesa. All patients were informed about the study and signed an informed consent form before inclusion in the EA register. Patients and Controls {#s2b} --------------------- The study sample comprised 91 patients enrolled on our EA register. The inclusion criteria included more than 1 swollen joint for at least 4 weeks and symptoms for less than a year. Only data from patients fulfilling the 1987 American College of Rheumatology criteria for RA [@pone.0085248-Arnett1] within the 5-year follow-up (n = 69) or with chronic undifferentiated arthritis (n = 22) were analyzed. Patients with diagnosis of spondyloarthritis, connective tissue diseases or crystal induced arthritis, were excluded from the analysis. The register protocol included 5 visits (baseline, 6, 12, 24, and 60 months), and at each one we recorded the following data in an electronic database: clinical and demographic information; disease duration at the beginning of follow-up; 28-joint Disease Activity Score (DAS28) [@pone.0085248-Prevoo1]; global disease activity on a 100-mm visual analogue scale assessed by both the patient and the physician; Health Assessment Questionnaire score (HAQ; Spanish version) [@pone.0085248-EsteveVives1]; and the results of laboratory tests including erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), rheumatoid factor (RF, assessed by nephelometry; positive \>20 IU/ml), and anti-citrullinated peptide antibody (ACPA, measured by enzyme immunoassay \[EIA\]: Euro-Diagnostica Immunoscan RA; positive \>50 IU/ml). Healthy donors (n = 100) were recruited from the Centro de Transfusiones de la Comunidad de Madrid. Following the Spanish Personal Data Protection Law, their demographic information was confidential. Measurement of Serum VIP {#s2c} ------------------------ VIP levels were determined using a competitive EIA with a commercially available kit according to the manufacturer\'s instructions (Phoenix Pharmaceutical, Karlsruhe, Germany). We previously optimized sample preparation procedure to obtain optimal results. First, it was tested the extraction of peptides from serum using a SEP-COLUMN containing 200 mg of C18, according to the manufacturer's recommendations. Second, serum samples were concentrated by using a lyophilizer to assay different concentrations. Then, VIP levels were determined. The best results were obtained in reconstituted dried extract (2∶1) without the use of columns. The assay was performed as follow, briefly, the samples were freeze-dried and dissolved in EIA buffer (2∶1), added to an immunoplate pre-coated with a secondary antibody, and incubated with biotinylated VIP and a primary antibody whose Fab fragment competitively binds to the biotinylated peptide and targeted peptide in samples. After washes, the wells were incubated with streptavidin-horseradish peroxidase, which catalyzed the oxidation of the substrate solution. The reaction was stopped using the stop solution and absorbance measured at 450 nm. A standard curve of known concentration was established. The concentration in the samples was determined by extrapolation to this curve and by applying the corresponding dilution factor. Samples from each patient were assayed twice. The minimum detectable concentration was 0.12 ng/ml of sample, with an intra-assay and interassay variation of ≤5% and 15%, respectively. Since serum samples were stored at --80°C for different periods of time until assay (range, 1 to 112 months), we first analyzed the effect of sample frozen-time on the ability of our assay to detect VIP. As shown in [Figure S1](#pone.0085248.s001){ref-type="supplementary-material"}, no significant correlation was observed between VIP serum level and frozen-time. Statistical Analysis and New Variables {#s2d} -------------------------------------- Normally distributed quantitative variables were expressed as the mean ± standard deviation, while non-normally distributed variables were expressed as the median and interquartile range (IQR). Qualitative variables were described using an estimation of the proportions. Variables with a normal distribution were analyzed using the *t* test, while the Mann-Whitney or Kruskal-Wallis tests were used for variables with a non-normal distribution. A χ^2^ test or the Fisher exact test was used to compare categorical variables. Considering that the raw data contained considerably left-shifted values ([Figure S2A](#pone.0085248.s002){ref-type="supplementary-material"}), we decided to censor values higher than 1000 pg/ml in this figure ([Figure S2B](#pone.0085248.s002){ref-type="supplementary-material"}) to avoid missing information if patients with very high VIP values were excluded. In addition, since the censored variable does not show a Gaussian distribution, data were normalized through logarithmic transformation ([Figure S2C](#pone.0085248.s002){ref-type="supplementary-material"}). Additional variables were defined to further describe the role of VIP in the progress of arthritis. VIP levels were considered to be low when the concentration was below the 25^th^ percentile of its concentration in the healthy population (370 pg/ml). The intensity of disease-modifying anti-rheumatic drugs (DMARDs) treatment (IDT) was assessed as the number of days of treatment with each DMARD during follow-up adjusted for weighted coefficients, as described elsewhere [@pone.0085248-Gonzlezlvaro1]. To identify factors that influenced VIP levels during follow-up (logarithmic transformation of the censored VIP levels as the dependent variable), we used data from 340 visits in the 88 patients with all information available in at least 2 visits to fit a population-averaged model by means of generalized linear models nested by patient and visit using the *xtgee* command of Stata 12 for Windows (StataCorp LP, College Station, Texas, USA). The population-averaged generalized estimating equations were first modeled by adding all variables with a *p* value \<0.15 in the bivariate analysis. The final models were constructed using a quasi-likelihood estimation based on the independence model information criterion [@pone.0085248-Pan1] and Wald tests, after removing all variables with *p*\>0.15. To determine the usefulness of baseline VIP levels for predicting disease progress, we performed an ordered logistic model using the *ologit* command of Stata 12. The dependent variable was the disease activity level at the 2-year follow-up, using the cut-off points for DAS28 proposed by Prevoo et al. (\<2.6, remission; 2.6 to 3.2, low disease activity; 3.2 to 5.1, moderate disease activity; \>5.1 high disease activity) [@pone.0085248-Prevoo1]. In this model, the dependent variable considered remission as 0 and low, moderate, and high disease activity as 1, 2, and 3. Thus, the ordered logistic analysis can also estimate cut-off points that aid the interpretation of the coefficients for each independent variable according to the levels of the dependent variable. The analysis was modeled as described above for *xtgee.* The significance level was set at p\<0.05. Results {#s3} ======= Characteristics of Patients with EA. VIP Levels in Patients and Healthy Donors {#s3a} ------------------------------------------------------------------------------ A total of 91 patients with RA (n = 69) or UA (n = 22) were included in the study. The main difference between these two subgroups was a higher frequency of female patients in the RA group, in which more severe disease at baseline and a higher prevalence of RF and ACPA positivity were also found ([Table 1](#pone-0085248-t001){ref-type="table"}). No significant differences were detected between the groups in variables expressing a subjective component such as pain, global disease assessment by patient, tender joint count, and HAQ. 10.1371/journal.pone.0085248.t001 ###### Baseline characteristics of patients with early arthritis. {#pone-0085248-t001-1} Rheumatoid artritis(n = 69) Undifferentiated arthritis(n = 22) Total(n = 91) P value --------------------------- ------------------------------------------------------ ------------------------------------------------------ --------------------------------------------------- --------- Age (years) 54 \[45--66\] 55 \[44--69\] 54 \[45--66\] n.s. Female gender (%) 78 54 73 0.004 Disease duration (months) 5 [@pone.0085248-Gomariz1]--[@pone.0085248-Colwell1] 7 [@pone.0085248-Gomariz1]--[@pone.0085248-Delgado2] 5.4 \[3.2--8.4\] n.s. GDAPa 46 \[29--60\] 46 \[28--58\] 46 \[28--60\] n.s. GDAPh 50 \[30 - 72\] 30 \[14--50\] 42 \[25--66\] 0.002 VAS Pain 52 \[25--70\] 48 \[13--60\] 50 \[20--70\] n.s. HAQ 1.125 \[0.625--1.750\] 0.875 \[0.375--1.375\] 1 \[0.625--1.625\] n.s. TJC 6 [@pone.0085248-Smolen1]--[@pone.0085248-Abad2] 3 \[0--10\] 6 [@pone.0085248-Schoels1]--[@pone.0085248-Abad2] n.s. SJC 7 [@pone.0085248-Gomariz1]--[@pone.0085248-Delgado1] 3 \[0--6\] 5 [@pone.0085248-Smolen1]--[@pone.0085248-Abad1] 0.004 ESR 32 \[19--54\] 17 \[8--34\] 30 \[15--46\] 0.006 CRP (mg/dl) 0.92 \[0.2--1.96\] 0.48 \[0.1--0.82\] 0.72 \[0.16--1.7\] 0.047 DAS28 (0--10) 5.0 \[4.0--6.1\] 3.6 \[3.1--4.5\] 4.8 \[3.5--6.0\] 0.002 RF (%) 54 18 45 0.004 ACPA+ (%) 60 14 49 \<0.001 VIP (pg/ml) 342 \[262--440\] 375 \[243--490\] 352 \[261--444\] n.s. Data are shown as the median or percentage. GDAPa: global disease assessment by patient; GDAPh: global disease assessment by physician; VAS: visual analog scale; HAQ: health assessment questionnaire; TJC: tender joint count; SJC: swollen joint count; ESR: erythrocyte sedimentation rate; CRP: C-reactive protein; DAS28∶28-joint count Disease Activity Score; RF: rheumatoid factor; ACPA: anti-citrullinated peptide antibodies; VIP: vasoactive intestinal peptide. n.s.: not significant. Considering serum levels of VIP at baseline, no significant differences were detected between the groups ([Table 1](#pone-0085248-t001){ref-type="table"}) or between patients and healthy controls ([Figure 1](#pone-0085248-g001){ref-type="fig"}). Nevertheless, the distribution of the VIP serum concentration was considerably heterogeneous, especially among RA patients, and ranged from 100 to \>1000 pg/ml. In addition, the lowest levels of VIP tended to be more frequent in RA and UA patients than in healthy donors ([Figure 1](#pone-0085248-g001){ref-type="fig"}). {#pone-0085248-g001} Low VIP Levels are Associated with Higher Disease Activity in Patients with EA {#s3b} ------------------------------------------------------------------------------ Despite a significant decrease in disease activity during follow-up ([Figure 2A](#pone-0085248-g002){ref-type="fig"}), we could not detect a parallel down-regulation of serum VIP levels ([Figure 2B](#pone-0085248-g002){ref-type="fig"}). Therefore, we performed a multivariate analysis to achieve a more accurate appraisal of the potential relationship between VIP levels and disease activity in patients with EA. Our data showed that elderly patients expressed slightly but significantly higher serum levels of VIP during follow-up ([Table 2](#pone-0085248-t002){ref-type="table"} and [Figure S3](#pone.0085248.s003){ref-type="supplementary-material"}). In addition, VIP levels tended to be significantly higher in visits at which patients were treated with TNF blockers. Interestingly, after adjustment for these confounders, we observed a significant and negative correlation between disease activity, estimated through DAS28, and VIP serum levels ([Table 2](#pone-0085248-t002){ref-type="table"}, model 1). {ref-type="table"}, model 2.](pone.0085248.g002){#pone-0085248-g002} 10.1371/journal.pone.0085248.t002 ###### Variables associated with VIP serum levels during follow-up of patients with early arthritis. {#pone-0085248-t002-2} Model 1 Model 2 ------------------- -------------- --------- -------------- --------- Gender Male Reference Reference Female 0.108±0.067 0.106 0.105 0.086 Diagnosis RA n.i. n.s. n.i. n.s. UA ACPA n.i. n.s. n.i. n.s. Age (by 10 yr) 0.05±0.019 0.013 0.05±0.018 0.008 GDA Pat n.i. n.s. n.i. n.s. GDA Phy n.i. n.s. n.i. n.s. HAQ n.i. n.s. n.i. n.s. SJC n.i. n.s. n.i. n.s. MS n.i. n.s. n.i. n.s. DAS28 −0.027±0.014 0.045 -- -- Disease activity Remission -- -- Reference Low −0.094±0.043 0.029 Moderate −0.187±0.049 \<0.001 High −0.113±0.057 0.046 Leflunomide(mg/d) n.i. n.s. n.i. n.s. TNF blockers 0.183±0.075 0.01 0.165±0.065 0.011 The longitudinal analysis was performed with data (logarithmic transformation of censored VIP levels; see [figure S2](#pone.0085248.s002){ref-type="supplementary-material"}) from 340 visits corresponding to the 88 patients with all information available in, at least, two visits. Model 1 was fitted using the continuous value of DAS28 as measure of disease activity while model 2 includes the categorical variable based in cut-off values of this index. The average number of visits by patient was 3.9. The table shows all the variables reaching p\<0.15 at the bivariate analysis (see Methods section for further information on multivariable analysis modeling). Coeff: coefficient; CI: confidence interval; RA: rheumatoid arthritis; UA: undifferentiated arthritis; ACPA: anti-citrullinated peptide antibodies; yr, year; GDA: global disease assessment; Pat: patient; Phy: physician; HAQ: health assessment questionnaire; SJC: swollen joint count; MS: morning stiffness; DAS28, 28-joint Disease Activity Score; TNF: tumor necrosis factor; n.s.: not significant; n.i.: not included. Next, we analyzed the concentration of VIP at the different visits after clustering the results in four groups: remission, low, moderate, and high disease activity. As shown in [Figure 2C](#pone-0085248-g002){ref-type="fig"}, we observed lower levels of VIP in those visits with moderate or high activity compared to visits at which patients were in remission and those differences reached statistically significance after adjusting by confounders ([Table 2](#pone-0085248-t002){ref-type="table"}, model 2). However, this approach did not reveal whether inflammation induced low VIP levels or those patients with low VIP showed more intense disease activity. Low Baseline VIP Serum Levels as a Biomarker of Disease Severity {#s3c} ---------------------------------------------------------------- Next, we studied whether having low VIP serum levels at baseline could predict long-term disease activity. As [Figure 3A](#pone-0085248-g003){ref-type="fig"} shows, those patients who continued to have high or moderate disease activity levels after a two-year follow-up had lower baseline VIP serum levels. This behaviour was detected despite the observed trend toward higher prescription of DMARDs in patients with low VIP serum levels ([Figure 3B](#pone-0085248-g003){ref-type="fig"}). {#pone-0085248-g003} In view of the various confounding factors that could influence disease activity level after two years of follow-up, we performed a multivariate ordered logistic regression to determine which variables contribute to an increased level of disease activity. Female patients showed an odds ratio (OR) of 4.3 for poorer disease outcome than male patients ([Table 3](#pone-0085248-t003){ref-type="table"}). In addition, patients requiring combined therapy displayed an OR of 6.6 for remaining at a higher disease activity level than those who did not require DMARDs. After adjustment for these confounders, having low baseline VIP levels was associated with a higher level of disease activity after two years of follow-up. However, marked interference with ACPA positivity was observed ([Table 3](#pone-0085248-t003){ref-type="table"}). Thus, patients with low VIP serum levels but negative ACPA had an OR of 6.1 for being at a higher disease activity level than those with negative ACPA levels and normal VIP serum levels ([Table 3](#pone-0085248-t003){ref-type="table"} and [Figure 3C](#pone-0085248-g003){ref-type="fig"}). By contrast, patients with positive ACPA levels showed a nonsignificant trend toward lower disease activity regardless of VIP levels ([Table 3](#pone-0085248-t003){ref-type="table"} and [Figure 3C](#pone-0085248-g003){ref-type="fig"}). In addition, the model revealed that the effect of gender, treatment, and baseline VIP level was more intense when the level of disease activity increased (β coefficients for cutpoints in [Table 3](#pone-0085248-t003){ref-type="table"}). 10.1371/journal.pone.0085248.t003 ###### Effect of low baseline VIP levels in serum [on disease activity after a two-year follow-up](http://ondiseaseactivityafteratwo-yearfollow-up). {#pone-0085248-t003-3} Odds ratio (95% CI) P -------------------------------- ----------------------- ------- Age (years) \<45 Ref. 45--65 -- n.s. \>65 -- n.s. Gender Man Ref. Woman 4.35 (1.46--12.99) 0.008 Diagnosis RA Ref. UA -- n.s. Positive RF -- n.s. Interaction VIP, ACPA Normal VIP, ACPA-negative Ref. Low VIP, ACPA-negative 6.11 (1.28--29.22) 0.023 Normal VIP, ACPA-positive 0.68 (0.24--1.9) 0.458 Low VIP, ACPA-positive 0.24 (0.03--1.85) 0.170 DMARD treatment None Ref. Monotherapy 2.2 (0.30--12.96) 0.437 Combined therapy 6.62 (0.84--51.93) 0.071 Baseline DAS28 -- n.s. Cutpoints β coeff. (95% CI) Remission/Low disease activity 1.59 (--0.40 to 3.59) -- Low/Moderate disease activity 2.46 (0.44--4.49) -- Moderate/High disease activity 5.97 (3.57--8.38) -- DAS28∶28-joint Disease Activity Score; HAQ: Health Assessment Questionnaire; Coeff.: coefficient; Ref.: reference; RA: rheumatoid arthritis; UA: undifferentiated arthritis; ACPA: anti-citrullinated peptide antibody; n.s.: not significant. In terms of treatment, ACPA positivity was associated with higher prescription of DMARDs, as occurred when low VIP levels were associated with ACPA positivity ([Figure 3D](#pone-0085248-g003){ref-type="fig"}). More heterogeneous regimens were found in patients with low VIP levels (treating physician blinded) and negativity for ACPA, but they undoubtedly received more intensive treatment than the population that met neither of these conditions ([Figure 3B and 3D](#pone-0085248-g003){ref-type="fig"}). Discussion {#s4} ========== Rheumatoid arthritis is a complex heterogeneous disease resulting from the interaction between genetic and environmental triggers and from the intervention of key molecules that modulate its severity. To date, there are no published data demonstrating an association between abnormal regulation of VIP and an increased risk for developing RA. However, growing evidence supports the ability of VIP to regulate the intensity of the inflammatory process and the immune response that contribute to the pathogenesis of rheumatic diseases [@pone.0085248-GonzalezRey1]. In this work, we showed that the lowest values of serum VIP levels were clustered in the EA group, despite no significant differences compared with healthy donors. Furthermore, our results reveal an inverse correlation between disease activity and VIP concentration in serum. Thus, patients with lower VIP levels showed higher DAS28 scores, and, conversely, higher VIP levels were detected in the group of patients in remission and in those with significantly weaker disease activity. Several experimental findings support this observation. Treatment with VIP reduced the incidence and severity of arthritis in murine models of RA by decreasing the production of proinflammatory cytokines and chemokines, and inducing a shift in the Th phenotype from a Th1-type toward a Th2-type response and generating efficient regulatory T cells [@pone.0085248-Delgado1], [@pone.0085248-GonzalezRey2]. In human fibroblast-like synoviocytes from RA patients, VIP downregulated the expression and production of proinflammatory cytokines, chemokines and COX2 as well as the production of IFNβ, CXCL8, and the matrix metalloproteinase-3 induced by TLR ligands [@pone.0085248-Juarranz2], [@pone.0085248-Carrin1], [@pone.0085248-PrezGarca1]. Given the above evidence on the role of VIP, it seems reasonable that patients in whom production of this immunoregulatory peptide did not increase have more severe autoimmune and inflammatory responses and poor outcomes. In close correlation with our data, expression of VIP in synovial fluid and articular cartilage from patients with OA was negatively associated with progressive joint damage as a potential indicator of disease severity, thus suggesting that VIP could play a protective role in progression of OA [@pone.0085248-Jiang1]. Moreover, low serum levels of VIP have been described in patients with juvenile idiopathic arthritis who show clinical evidence of cardiac autonomic neuropathy associated with a parasympathetic dysfunction. In these patients, a significant positive association was found between cardiac autonomic neuropathy and major disease manifestations, including activity; therefore, the authors suggested that serum VIP should be assessed in these patients [@pone.0085248-ElSayed1]. Although several parameters (eg, RF, ACPA, ESR, and CRP) have been proposed as predictors of long-term outcome of EA, they enable us to classify only 65% of patients [@pone.0085248-Smolen1], [@pone.0085248-Gonzlezlvaro1], [@pone.0085248-deVriesBouwstra1]--[@pone.0085248-Szodoray1]. Furthermore, as RF and ACPA largely overlap, identifying new prognostic markers would be essential for providing complementary information in order to improve the performance of a biomarker-guided strategy. In this setting, the association between VIP levels and disease activity reported here suggests that baseline serum VIP could be a feasible biomarker that enables EA patients to be stratified for therapeutic decision making. In fact, our data support that measurement of serum VIP levels at baseline enhances the predictive value of ACPA in determining long-term outcome in patients with EA. Thus, ACPA-negative patients with low VIP serum levels had a higher degree of disease activity than ACPA-negative patients with normal serum VIP levels and an even higher degree of disease activity than ACPA-positive patients. This finding likely reflects selection bias, since the presence of ACPA, a recognized marker of poor prognosis to which the attending rheumatologist was not blinded, was associated with more intense prescription of DMARDs. As a consequence, ACPA-positive patients with normal VIP showed similar disease activity to ACPA-negative patients with normal VIP. Interestingly, although the difference was not significant, ACPA-positive patients with low VIP serum levels had a slightly higher degree of disease activity and a greater requirement for DMARDs than ACPA-positive patients with normal VIP serum levels. On the other hand, ACPA-negative patients with low VIP levels received more heterogeneous treatment, probably because of the lack of a suitable marker, since physicians were entirely unaware of the value of this parameter. An additional advantage of VIP as a prognostic marker is its scarce variation during follow-up, which indicates that VIP levels depend on the individual and do not change owing to external factors during follow-up. Thus, VIP concentrations could predict the outcome of arthritis independently of the treatment prescribed and the phase of the disease. Our study is subject to a series of limitations. First, our data show that elderly patients have slightly but significantly higher serum levels of VIP. However, as demographic information in the healthy volunteers was confidential, we were unable to draw a comparison with the patients. In this regard, the correlation between VIP levels and gender and age has received little attention in the literature. The only study that does examine this correlation found no differences in VIP concentrations between men and women with secondary hyperparathyroidism undergoing dialysis [@pone.0085248-Dez1]. Nevertheless, variation attributable to age in our study is lower than individual variability among EA patients ([Figure S3](#pone.0085248.s003){ref-type="supplementary-material"}). Second, the absence of a pre-established treatment strategy in our unit means that variations in the prescription of DMARDs by the rheumatologists involved in the study could interfere with the results. Third, considering the dispersion of VIP values owing to the variability of ELISA, we had to apply mathematical modifications to optimize the statistical analysis, thus hampering implementation in daily clinical practice. Our study has the following strengths. First, exhaustive and rigorous collection of data on treatment with DMARDs goes some way to resolving the second limitation. VIP levels were measured retrospectively; therefore, the rheumatologists were blinded to serum VIP levels but not to ACPA reactivity when they selected therapy. Consequently, they were able to prescribe more intensive treatment to ACPA-positive patients. Second, we included a high number of samples and clinical data (about 4 visits per each of the 91 patients). In conclusion, our study demonstrates that EA patients with low baseline VIP levels have worse disease outcome. Validation of the predictive value of serum VIP levels would enable earlier and more intensive treatment of patients with the most aggressive forms of RA. Supporting Information {#s5} ====================== ###### **Frozen storage does not affect the measurement of serum VIP levels.** VIP concentration at serum from samples stored from one month to 10 years is shown. The red line represents the linear prediction obtained with the command *lfit* of Stata 12. (TIF) ###### Click here for additional data file. ###### **Normalization of the VIP serum levels variable in order to obtain a distribution closer to Gaussian.** (TIF) ###### Click here for additional data file. ###### **Correlation between age and VIP serum levels.** Data are shown as dot plot and the exponential linear prediction with 95% confidence interval at ages 20, 30, 40, 50, 60, 70 and 80. These data were obtained using the command *marginsplot* of Stata 12 after performing the multivariable analysis displayed in [Table 2](#pone-0085248-t002){ref-type="table"} (dependent variable logarithmic transformation of censored VIP). (TIF) ###### Click here for additional data file. The authors are grateful to the patients of the EA Register and to the Centro de Transfusiones de la Comunidad de Madrid for recruiting healthy donors. [^1]: **Competing Interests:**A patent application (P201230827) for the use of VIP as a marker of prognostic in autoimmune diseases has been deposited by the Fundación de Investigación Biomédica del Hospital Universitario de la Princesa at the Oficina Española de Patentes y Marcas. The full name of the patent application is: VIP use as a prognostic marker of autoimmune diseases. This does not alter the authors' adherence to all the PLOS ONE policies. [^2]: Conceived and designed the experiments: CM AMO RPG IG-A. Performed the experiments: CM YJ AL IVS IG-A AMO JL. Analyzed the data: IG-A CM RPG AMO. Wrote the paper: CM IG-A RPG RG-V.

Background & Summary {#Sec1} ==================== Rapid advancements in computational methods for chemical synthesis planning and automated reaction mechanism generation, especially in the area of machine learning, are causing a significant shift in how such problems are tackled. Deep learning approaches are replacing conventional quantitative structure-activity relationships often based on support vector machines, decision trees, or linear methods like partial least squares^[@CR1],[@CR2]^. These new systems are becoming widely available for computer-aided retrosynthesis^[@CR3]^, reaction outcome prediction^[@CR3]^, high-throughput virtual screening^[@CR4]^, and more general molecular property prediction^[@CR5],[@CR6]^. Computational approaches are also increasingly common in reaction mechanism generation due to the large number of species and reactions that are generally required for accurate descriptions of phenomena like pyrolysis, combustion, and atmospheric oxidation^[@CR7]--[@CR9]^. Frequently, this involves characterizing chemical pathways with quantum chemistry^[@CR8]^, but deep learning methods have also recently been applied to estimate thermochemistry during mechanism generation^[@CR10],[@CR11]^. While computers already outperform humans at qualitatively predicting reaction products^[@CR12],[@CR13]^ and successful yield predictions have been demonstrated for limited datasets^[@CR14],[@CR15]^, quantitative reaction information is still elusive in large databases like Reaxys^[@CR16]^, Pistachio^[@CR17]^, and the United States Patent and Trademark Office database^[@CR18]^. Reaction yield, time, and some quantitative conditions like temperature are sometimes available, but there is usually no information on reaction kinetics. If such data were available, calculation of derived properties---such as minimum reaction times and branching ratios---would be possible. Our goal is to provide a quantitative dataset of reactions that enables the calculation of such data and can lead to more efficient drug design and help in deciding which reactions are important in mechanism generation. Computationally generating a dataset of reactions is significantly more complex than only calculating stable equilibrium structures because transition states (TSs) of chemical reactions cannot be enumerated in the same manner as stable molecules. Even if the reactant and product structures are known, the exact TS geometry has to be found via a human-guided search or with expensive automated TS finding methods. Here, we use automated potential energy surface exploration to generate the dataset of reactions, which has been shown to be successful in cases when many reaction pathways have to be evaluated^[@CR19]--[@CR21]^. More specifically, we rely on the growing string method^[@CR22]^ to automatically optimize reaction paths and TSs. We report quantum chemical data on more than 16,000 reactions in the form of reactants, products, and TSs at the B97-D3/def2-mSVP level of theory and 12,000 reactions at the *ω*B97X-D3/def2-TZVP level of theory. The data include the raw output from geometry optimizations and frequency calculations in addition to atom-mapped SMILES, activation energies, and enthalpies of reaction. All reactions are gas-phase calculations involving up to seven carbon, oxygen, or nitrogen atoms per molecule. The reactants are sampled from GDB-7, a subset of GDB-17^[@CR23]^, meaning that all reactions have a unimolecular reactant but potentially multi-molecular products. Figure [1](#Fig1){ref-type="fig"} illustrates the dataset generation process and the resulting space of reactions in terms of their activation energies and enthalpies of reaction.Fig. 1Reaction data generation and visualization of reaction space. During data generation, many reactants are optimized, hundreds of reaction paths for each reactant are searched with an automated transition state finding method, and the resulting products are optimized. The reaction space spans a wide range of activation energies and is visualized with a bivariate kernel density estimate (using a Gaussian kernel) of the probability density of the activation energy and enthalpy of reaction. The visualization encompasses both forward and reverse reactions. Methods {#Sec2} ======= Overview {#Sec3} -------- The dataset generation procedure started by selecting molecules from GDB-7^[@CR23]^, generating conformers, and optimizing the lowest-energy conformer. An exhaustive set of driving coordinates subject to valence and connectivity constraints were generated for each reaction. Reaction paths were calculated with the growing string method^[@CR22]^, which searched along each of the driving coordinates. Products and TSs discovered in this way were reoptimized, duplicate reactions were removed, and checks were performed to verify the reactions. The generated reactions were then refined at a higher level of theory. Because of the large number of density functional theory (DFT) calculations required, the massively parallel nature of the calculations was exploited by running thousands of calculations in parallel on a supercomputer. Reactant optimization {#Sec4} --------------------- Because of the unfavorable scaling of quantum chemical calculations, we only considered molecules with at most seven heavy atoms (C, N, O). *All* molecules with six or fewer heavy atoms were selected from GDB-7 (\~770) and a random selection of \~430 molecules were selected from the set with seven heavy atoms. Starting from the SMILES strings, we embedded several hundred conformers for each molecule using the RDKit^[@CR24]^ with the ETKDG distance geometry method^[@CR25]^ and relaxed their geometries using the MMFF94 force field implemented in RDKit. The lowest energy structure was selected for each molecule and optimized at both the B97-D3/def2-mSVP with Becke-Johnson damping level of theory^[@CR26]^ and the *ω*B97X-D3/def2-TZVP^[@CR27]^ level of theory with Q-Chem 5.1^[@CR28]^. We ascertained that none of the molecules contained imaginary frequencies. All calculations, including the subsequent string method calculations, were done in the singlet state and used a spin-unrestricted ansatz because the bond distortions occurring in the corresponding TSs might be better treated with an unrestricted formulation. The def2-mSVP basis set in the Karlsruhe *def2* basis set family^[@CR29]^ is a modified version of def2-SV(P), which corrects for an overestimation of bond lengths involving hydrogen^[@CR30]^. All DFT calculations used the *SG-2* standard quadrature grid, which is of sufficient quality for B97-based functionals^[@CR31]^. Potential energy surface exploration {#Sec5} ------------------------------------ The most demanding and most time-intensive step of the reaction generation process is the optimization of reaction paths to the minimum energy paths (MEPs) containing the correct TS structures. We accomplished this in an automated fashion by using the single-ended growing string method (GSM)^[@CR22]^ at the B97-D3/def2-mSVP level of theory. GSM performs the reaction path optimization using a set of delocalized internal coordinates, which means that the resulting MEPs may be slightly different than those obtained via a reaction path following procedure in mass-weighted internal coordinates^[@CR32]^. Single-ended methods only require a reactant structure to find reactions whereas double-ended methods additionally require knowledge of the product^[@CR33],[@CR34]^. *A priori* specification of the product can be problematic when there is no simple elementary step connecting reactant and product. Single-ended GSM solves this issue by only requiring a set of driving coordinates to initiate the reaction path search. In our case, the driving coordinates are specified as bond transformations in terms of primitive internal coordinates. The direction given by the primitive internal coordinate vector is projected onto the nonredundant delocalized internal coordinates^[@CR35]^, which is the space in which the reaction path optimization occurs. This results in a single tangent vector that represents all of the driving coordinates simultaneously. Importantly, this allows all other coordinates to change without constraint during the optimization, thus allowing necessary angle, torsion, and even additional bond changes to occur. Once a path has been grown, the entire path is optimized towards the MEP while monitoring the number of TSs along the path and truncating it if more than one TS is detected---ensuring that the reaction is elementary. As a result of this, not all bond changes given in the driving coordinates are guaranteed to occur. Towards the end of the path optimization, an exact TS search takes place guided by curvature information from the string. In order to obtain many reactions, we generated an exhaustive list of driving coordinate sets for each reactant subject to a few constraints. Because elementary reactions usually involve few bond changes, we specified that at most two bonds could be broken, at most two bonds could be formed, and a total of at most three bonds could be changed. A "bond" in this sense ignored bond orders and only considered whether two atoms were connected to each other. Note that these constraints were only selected to ensure a computationally tractable number of driving coordinates. As described in the previous paragraph, these limits did not apply during the actual path optimization, they were only used to specify the initial search direction. We also ignored driving coordinates involving only a single bond change as these would likely correspond to barrierless associations or dissociations. Driving coordinates involving equivalent hydrogens were not included. Equivalent hydrogens only differ in their atom indices, *e.g*., a hydrogen atom that is part of a methyl group is considered to be equivalent to another hydrogen in the same methyl group. Lastly, the driving coordinates were further limited based on the valences of the expected product structures. Hydrogen atoms must have one bond, carbons can be connected to a minimum of two and a maximum of four atoms, oxygen to a minimum of one and a maximum of two, and nitrogen to a minimum of one and a maximum of three. This process usually resulted in several hundred sets of driving coordinates per reactant. Following each GSM calculation, the endpoint of the paths were subjected to additional geometry optimizations to ensure that the product structures were at a minimum. For each reactant, there were many duplicate reactions. Instead of discarding all of them, up to four duplicates of the same reaction were retained for additional TS optimization in case some of the optimizations fail. While GSM already produces a mostly optimized TS structure, the additional optimization step ensured that the TSs were optimized to high accuracy. Reaction verification and extraction {#Sec6} ------------------------------------ After the additional TS optimizations, duplicate reactions were filtered out again. If duplicates were present, the lowest-barrier reaction was retained. Differences in barrier height may arise due to different TS conformers. Although GSM provided an optimized MEP for each reaction, it is possible that some reactions containing incorrect transition states remained. These were filtered out according to a normal mode analysis described in the Technical Validation section. To convert from three-dimensional geometries to SMILES^[@CR36]^, connections and bond orders could be perceived with Open Babel^[@CR37]^. However, there were cases where the derived bond orders were chemically unreasonable, for example, when the resulting SMILES contained adjacent radical atoms which most likely correspond to double bonds. To eliminate unreasonable structures, we converted the Open Babel molecule to InChI^[@CR38]^, which only treats bond orders implicitly and resolves the issue. A downside to using InChI is that tautomers are assigned the same string, but this can be circumvented by converting to a nonstandard InChI containing a fixed-hydrogen layer. Additionally, atom ordering was lost in the InChI conversion. We reconstructed the atom map by converting to an RDKit molecule and determining the graph isomorphism between the original molecule and the RDKit molecule without considering bond orders. In the future, an alternative procedure for perceiving SMILES could be implemented based on natural bond orbital analysis^[@CR39]^. The activation energies were extracted by adding the zero-point energies from a harmonic vibrational analysis to reactant, product, and TS energies and computing the difference between resulting TS and reactant energies. Similarly, enthalpies of formation were determined based on the difference of product and reactant energies. Refinement {#Sec7} ---------- B97-D3/def2-mSVP strikes a reasonable balance between cost and accuracy for potential energy surface exploration, but does not provide particularly accurate energies. Therefore, we refined the discovered pathways using *ω*B97X-D3/def2-TZVP. As mentioned earlier, reactants were already optimized with *ω*B97X-D3/def2-TZVP. Reactions were extracted as described in the preceding subsection, but some duplicates were retained to increase the probability of successful reoptimization. Only the duplicate with the smallest activation energy was retained in the end. Products and TSs were then reoptimized with *ω*B97X-D3/def2-TZVP and the final high-level reactions were extracted as before. Data Records {#Sec8} ============ Q-Chem output files, extracted SMILES, activation energies, and enthalpies of formation are available for 16,365 B97-D3/def2-mSVP reactions and for 11,961 *ω*B97X-D3/def2-TZVP reactions^[@CR40]^. The raw log files are stored in two compressed archive files, b97d3.tar.gz and wb97xd3.tar.gz for B97-D3/def2-mSVP and *ω*B97X-D3/def2-TZVP data, respectively. Each archive contains a separate folder for each reaction labelled rxn\#\#\#\#\#\#, where \#\#\#\#\#\# denotes the reaction number padded with zeros. Within each folder are the three log files for a reaction, r\#\#\#\#\#\#.log for the reactant, p\#\#\#\#\#\#.log for the product, and ts\#\#\#\#\#\#.log for the transition state. Each log file contains the output of a geometry optimization and harmonic vibrational analysis. Atom-mapped SMILES, activation energies, and enthalpies of formation for each reaction are listed in the comma-separated values files b97d3.csv and wb97xd3.csv for the B97-D3/def2-mSVP and *ω*B97X-D3/def2-TZVP levels of theory, respectively. The reactions are listed in the same order as the corresponding folders in the archive files. The columns in the comma-separated values files are explained in Table [1](#Tab1){ref-type="table"}.Table 1A description of the columns in the comma-separated values files.Column labelDescriptionidxReaction indexrsmiReactant SMILESpsmiProduct SMILESeaActivation energy (kcal mol^−1^)dhEnthalpy of reaction (kcal mol^−1^) During the potential energy surface exploration, many duplicate reactions were encountered which were filtered out. Additionally, reactions that did not pass the tests described in the Technical Validation section were removed from the final list. Nonetheless, all of these calculations also produced optimized transition states, although the reactants and products were not verified for many of them, and duplicate transition states exist. These data may still prove to be useful if only transition state structures are required or if additional calculations are done to obtain the corresponding reactants and products. Therefore, the log files for all successfully optimized transition states at both levels of theory are stored in ts_with_dup_b97d3.tar.gz and ts_with_dup_wb97xd3.tar.gz. There are 69,366 B97-D3/def2-mSVP transition states and 24,987 $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\omega $$\end{document}$B97X-D3/def2-TZVP transition states. Technical Validation {#Sec9} ==================== Although the growing string method produces an optimized minimum energy path that should contain the correct TS in most cases, insufficient path discretization and reoptimization of TS geometries can lead to convergence failures or result in incorrect transition states. We performed several checks to filter out incorrect reactions. We ensured that all TSs have exactly one imaginary frequency. Reactions were also removed if the energy during the TS optimization changed by more than 3 kcal mol^−1^ relative to the highest energy on the growing string path. The most important check that we performed was to verify that the atomic displacements for the imaginary frequency matched the bond changes that occurred going from the proposed reactant to product. For each proposed reaction, we determined which bonds were changing in the reaction and ensured that the imaginary frequency normal mode displacements along those bonds were larger than the displacements along all the other bonds. This indicated that movement along the reaction coordinate mostly involved atoms undergoing significant change in the reaction. After all these changes, there is still the possibility that some of the transition states are incorrect. As a final check, we removed all of the reactions where the imaginary frequency of the transition state was less than 100 cm^−1^ in magnitude, as these typically correspond to conformational changes. To avoid excessive computational cost, DFT methods had to be used to generate the reaction dataset. The functional chosen for the string method calculations, B97-D3, does not provide accurate activation energies, but was selected due to its low computational cost. However, *ω*B97X-D3 has been shown to yield excellent quantitative barrier heights with a 2.28 kcal mol^−1^ root-mean-square deviation from reference data that is estimated to be more than ten times as accurate as the best density functionals^[@CR41]^, which makes this data very useful. Therefore, the following analysis was only completed for the *ω*B97X-D3 data. In order to show that the dataset provides a reasonably diverse set of reactions spanning many different chemistries even though constraints were set on the number of atoms and driving coordinate generation parameters, it is necessary to characterize the types of reactions. Figure [1](#Fig1){ref-type="fig"} already shows that the range of activation energies and enthalpies of formation is very large. Even high-energy reactions involving barriers of up to 200 kcal mol^−1^ are included in the dataset. If the data are used to learn reaction prediction models, including such high-energy paths is important in order to not bias models towards the low-energy regions. Figure [2](#Fig2){ref-type="fig"} shows that even though the driving coordinates were limited to three bond changes, significantly more complex reactions involving more bond changes occur in the dataset. Nonetheless, most elementary reactions predominantly occur with only two or three bond changes. Furthermore, the median activation energy increases with an increasing number of bond changes, which is expected.Fig. 2Activation energy distributions. The distribution of activation energies split by the number of bond changes in the *ω*B97X-D3 reactions. Bond changes only consider changes in connectivity between atoms, irrespective of bond order. The distributions are scaled to have equal area. Instead of simply counting the number of bond changes, the reactions can be classified based on the types of bonds that are changed. Figure [3](#Fig3){ref-type="fig"} shows that all combinations of bond changes between H, C, N, and O atoms occur in the dataset with many examples present for all reaction types. H--H changing reactions are the rarest because they only correspond to hydrogen molecule formation.Fig. 3Bond change types. The number of times each type of bond change occurs in the *ω*B97X-D3 reactions. For example, C--N denotes both forming a bond between C and N atoms and breaking a bond between the atoms. This also includes a change in the bond order between the two atoms. Lastly, we characterized the reaction diversity by automatically extracting a set of general templates. We only focused on the reactive center by using RDKit's *GetReactingAtoms* method to isolate atoms changing in the reaction. The molecular fragments in the reactants and products identified as the reactive center were then concatenated together to form the reaction template. In addition to the connectivity of the reacting atoms, the only features considered were atom identity, charge, aromaticity, and bond type. Figure [4](#Fig4){ref-type="fig"} shows the results of this automated extraction. Many templates only have a single reaction example and only the eight most popular templates have more than 100 reaction examples, highlighting the diversity present in the dataset.Fig. 4Automatically extracted reaction templates. The reactions were grouped with very general reaction templates that only consider connectivity of atoms in the reactive center, atom identity, charge, aromaticity, and bond type. The top 20 templates are denoted with SMARTS strings^[@CR45]^. Usage Notes {#Sec10} =========== With the exception of the growing string method code, which is available from the developers of the method^[@CR42]^, and the Q-Chem quantum chemistry package, all code necessary to reproduce the generated data is available on GitHub^[@CR43]^. The repository contains several scripts, which should be run in the following order:parse_qm9.py: Converts the QM9 data directory^[@CR44]^, which contains the GDB-9 SMILES along with quantum mechanically derived properties, to a pickled file containing a list of MolData objects, which store the information in QM9 as Python objects.make_opt_jobs.py: Performs conformer searches and makes Q-Chem input files for optimization of reactant geometries based on the QM9 SMILES. The geometry optimizations themselves have to be performed with Q-Chem outside of the code, preferably in a massively parallel fashion on a supercomputer.create_gsm_jobs.py: Reads the geometry optimization outputs of the reactant optimizations, generates driving coordinates, and writes the files required for the GSM calculations. The GSM code has to be compiled separately^[@CR42]^. The GSM calculations also have to be run separately and should produce output files with a gsm\#.out format, where \# corresponds to each reaction path.create_prod_optfreq_jobs.py: Reads the string endpoints from the successfully completed GSM calculations and writes the Q-Chem input files for the product optimizations.create_ts_optfreq_jobs.py: Extracts the TS geometries from the GSM output files, removes duplicate reactions using the output from the product optimizations, and writes the Q-Chem input files for additional TS optimizations.extract_reactions.py: Extracts the unique reactions using the reactant, product, and TS optimization outputs in the form of a comma-separated values file containing SMILES, activation energies, and enthalpies of reaction. Can also write the file path information of all relevant log files to the CSV output, which can be used to copy the log files for every reaction.refine_reactants.py: Writes Q-Chem input files for reoptimization of the reactants at the higher level of theory.refine_products_and_ts.py: Uses the same method as implemented in extract_reactions.py to extract reactions and write Q-Chem input files for the reoptimization of products and TSs at the higher level of theory. After running the Q-Chem jobs, extract_reactions.py can be run again to extract the high-level reactions. If desired, the levels of theory and the reaction generation settings can be changed in the config folder. **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We gratefully acknowledge financial support from the the DARPA Make-It program under contract ARO W911NF-16-2-0023. This research used resources of the National Energy Research Scientific Computing Center (NERSC), a U.S. Department of Energy Office of Science User Facility operated under Contract No. DE-AC02-05CH11231. C.A.G. performed all potential energy surface exploration, quantum chemistry calculations, and SMILES and energy extraction. L.P. performed the reaction template analysis. All authors worked on the manuscript. The code used to generate the data is freely available on GitHub under the MIT license^[@CR43]^. Further details on how to use it to generate the data are given in the Usage Notes. The authors declare no competing interests.

Background ========== Duchenne muscular dystrophy (DMD) is a severe, X-linked genetic muscle-wasting disorder characterised by progressive muscle weakness that culminates in respiratory failure and premature death. The disease affects approximately 1:3,500 live male births worldwide, and affected boys are usually wheelchair bound by their early teens and experience a severely reduced quality of life. DMD is caused by mutations in the dystrophin (*dmd*) gene resulting in very low levels or a complete absence of the dystrophin protein, a key structural element of muscle fibres that renders them highly susceptible to damage. As a consequence, dystrophic muscles are characterised by inflammation and ongoing cycles of degeneration and regeneration. This environment limits muscle regenerative capacity and there is concomitant replacement of formerly functional muscle fibres with adipose and fibrotic material (reviewed in \[[@B1]\]). The *mdx* mouse is the most commonly used animal model of DMD. It arose from a natural mutation in a colony of C57BL/10 mice in which a premature stop codon was introduced into the *dmd* gene, resulting in a complete loss of the dystrophin protein \[[@B2],[@B3]\]. While this model is a good genocopy of the human disease, the pathology of the *mdx* mouse does not mimic that observed in DMD. *Mdx* mice exhibit a mild pathology with a slightly reduced lifespan (18--24 months) and functional decline of hindlimb muscles becoming evident at approximately 18 months of age \[[@B4]\], which is most likely a result of modifier genes in the C57BL/10 strain that can alter the effect of dystrophin loss on skeletal muscle pathology. Unlike the hindlimb muscles, the diaphragm of *mdx* mice undergoes progressive deterioration of muscle structure and function and is therefore a preferred muscle when examining the efficacy of potential therapeutic agents of clinical relevance \[[@B5]-[@B7]\]. As DMD is caused by mutations in a single gene, one of the most promising therapies is through gene replacement. However, while gene replacement or correction studies are likely to provide an eventual cure for DMD, several barriers need to be overcome including the presence of fibrosis within dystrophic skeletal muscles (reviewed in \[[@B8]\]). Fibrosis not only creates a physical barrier, but also replaces the muscle fibres that can be targeted, limiting the efficacy of cell- and gene-based therapies. Attenuating fibrotic infiltration may be needed to optimise gene, cell and pharmacological therapies. Various agents with antifibrotic properties have been trialled to reduce fibrosis deposition in skeletal muscle. Suramin, a TGF-β inhibitor, and interleukin-15 (IL-15) have been shown to reduce muscle fibrosis but can have side effects when administered systemically \[[@B9]-[@B11]\]. Another compound with antifibrotic properties is tranilast, an orally bioavailable antiallergic agent that has been approved for use in the human population in Japan and South Korea since 1982 for the treatment of bronchial asthma, atopic dermatitis and allergic rhinitis \[[@B12]-[@B17]\]. Since that time, the effectiveness of tranilast as a therapeutic agent for a range of fibrotic disorders and its mechanism of action have been studied extensively both *in vitro* and *in vivo* (reviewed in \[[@B18]\]). In 1992, Suzawa and colleagues \[[@B19]\] demonstrated that tranilast suppressed release of profibrotic cytokines from monocyte-macrophages *in vitro*\[[@B19]\], highlighting tranilast's antifibrotic properties. Tranilast has subsequently been demonstrated to reduce tuberointerstitial and heart fibrosis in diabetic rat models and to block TGF-β-induced fibrosis *in vitro* and *in vivo*\[[@B20]-[@B24]\]. In addition, tranilast administration was found to be efficacious in reducing muscle fibrosis in the Bio14.6 hamster model of limb-girdle muscular dystrophy and reducing serum creatine kinase levels in *mdx* dystrophic mice, effects they suggest may be a result of tranilast-mediated inhibition of the Ca^2+^-permeable growth factor-regulated channel (TRPV2 or GRC) \[[@B18],[@B25]\]. In this study, we report that short-term administration of tranilast in *mdx* mice decreases fibrosis in skeletal muscle and improves the resistance to muscle fatigue. Together these findings demonstrate that tranilast has therapeutic potential to combat fibrosis in muscle diseases such as DMD. Results ======= Tranilast does not alter skeletal muscle mass or strength --------------------------------------------------------- At the end of the 9-week treatment period, the tibialis anterior (TA), soleus (SOL), extensor digitorum longus (EDL), plantaris (PLANT), gastrocnemius (GAST), quadriceps (QUAD) and heart muscles from both non-treated and treated *mdx* mice were significantly larger than those from non-treated and treated control mice (Figure [1](#F1){ref-type="fig"}A). These differences cannot be attributed to differences in food intake as daily intake was not different between strains or treatment groups (data not shown) and averaged 3.3 g/mouse/day. Administration of tranilast did not significantly alter the mass of any of the tested skeletal muscles in control or *mdx* mice (Figure [1](#F1){ref-type="fig"}A). Consequently, 9-week treatment with tranilast did not affect whole-body strength (Figure [1](#F1){ref-type="fig"}B) or mobility (Figure [1](#F1){ref-type="fig"}C), as assessed by grip strength and rotarod performance. {#F1} Fibrotic deposition is decreased in muscles of tranilast-treated dystrophic mice -------------------------------------------------------------------------------- The TA and diaphragm muscles of *mdx* mice contained three- and nine-fold more fibrosis, respectively, compared with control mice (Figure [2](#F2){ref-type="fig"}). Tranilast administration to young *mdx* mice for 9 weeks resulted in a significant three-fold (31%) decrease in fibrosis in the diaphragm compared with untreated *mdx* mice (Figure [2](#F2){ref-type="fig"}A-C, *P* \< 0.05). A similar trend (30% decrease, *P* = 0.10) was observed in the TA muscles of *mdx* mice. The level of fibrosis in the TA muscles and diaphragm of control animals was naturally very low and unchanged with tranilast administration (Figure [2](#F2){ref-type="fig"}C). Analysis of gene expression revealed that Collagen 1a1, 3a1, 5a1 and TGFβ genes were all more highly expressed in the diaphragm of *mdx* compared to C57BL/10 mice, but tranilast treatment had no effect on the expression levels of these genes (Table [1](#T1){ref-type="table"}). Tranilast treatment resulted in a change in fibre type distribution in the TA muscles of *mdx* mice with an increased proportion of type IIa fibres with a concomitant decrease in type IIb/x fibres compared with muscles from untreated *mdx* mice (Table [2](#T2){ref-type="table"}). No significant differences were observed between tranilast-treated and control *mdx* mice in fibre cross-sectional area or oxidative enzyme capacity in either the TA or diaphragm muscles (Table [2](#T2){ref-type="table"}). {#F2} ###### Tranilast administration does not cause a long-term change in fibrotic gene expression in the diaphragm   **CON + VEH** **CON + TRA** ***mdx-*VEH** ***mdx-TRA*** ------------ --------------- --------------- --------------- --------------- **Col1a1** 1.00±0.09 0.83±0.13 3.42±0.42\* 3.37±0.33\* **Col2a1** 1.00±0.28 1.23±0.47 0.79±0.20 0.92±0.34 **Col3a1** 1.00±0.10 0.89±0.13 3.72±0.26\* 3.27±0.34\* **Col4a1** 1.00±0.06 0.87±0.11 1.31±0.14 1.35±0.13 **Col5a1** 1.00±0.09 0.91±0.08 3.46±0.24\* 3.40±0.36\* **COl6a1** 1.00±0.06 0.87±0.11 1.31±0.14 1.35±0.13 **Fn1** 1.00±0.14 1.34±0.07 1.00±0.14 1.01±0.18 **TGFβ1** 1.00±0.11 0.86±0.07 1.48±0.04\* 1.45±0.12\* At the conclusion of the treatment period RNA was extracted and real-time PCR analysis was used to examine the expression level of genes known to be associated with fibrosis. \**P* \< 0.05 *mdx* vs. control. ###### **Tranilast administration causes a shift in fibre type proportions in the TA muscles of*mdx*mice**   **CON + VEH** **CON + TRA** ***mdx*** **+ VEH** ***mdx*** **+ TRA** -------------------------------- ---------------- ----------------- --------------------- --------------------- **Tibialis anterior**  **Fibre CSA (μm**^**2**^**)** 1,811.4 ± 54.9 1,859.7 ± 74.3 1,890.2 ± 54.0 1,932.2 ± 113.6  **SDH intensity (o.d.)** 1,645.4 ± 57.4 1,622.3 ± 36.6 1,497.2 ± 74.1^†^ 1,464.7 ± 48.9^†^  **% Type IIa fibres** 11.0 ± 3.1 9.1 ± 1.6 4.9 ± 0.6 10.8 ± 0.5\* **Diaphragm**  **Fibre CSA (μm**^**2**^**)** 1,010.6 ± 69.3 949.8 ± 69.8 764.9 ± 39.9^†^ 874.6 ± 65.3^†^  **SDH intensity (o.d.)** 2,004.0 ± 63.9 1,725.6 ± 212.3 1,897.3 ± 46.2^†^ 1,793.1 ± 124.1^†^  **% Type IIa fibres** 63.8 ± 1.6 67.8 ± 3.9 71.5 ± 2.6^†^ 78.6 ± 3.3^†^ At the conclusion of the treatment period the fibre cross-sectional area (CSA), fibre type and oxidative enzyme activity were determined by histological evaluations of cross-sections stained with H&E and reacted for succinate dehydrogenase (SDH) activity. ^†^*P* \< 0.05, group main effect; \**P* \< 0.05, tranilast-treated vs. vehicle-treated *mdx* mice. Tranilast administration improves resistance to muscle fatigue in dystrophic mice --------------------------------------------------------------------------------- Dystrophic *mdx* mice exhibited a \~40% reduction in diaphragm- and TA-specific force compared with control (Figure [3](#F3){ref-type="fig"}A/B). Nine-week treatment with tranilast did not improve whole-body strength or mobility and did not improve maximum force-producing capacity in the TA or diaphragm muscles of control or *mdx* mice (i.e., for TA muscle: C57BL/10 control 1,825 ± 35 mN vs. C57Bl/10 tranilast 1,743 ± 55 mN; *mdx* control 1,456 ± 88 mN vs. *mdx* tranilast 1,177 ± 139 mN). However, force production during a 4-min fatiguing stimulation protocol was improved in both the diaphragm and TA muscles of tranilast-treated *mdx* mice (Figure [3](#F3){ref-type="fig"}B, C). {#F3} Tranilast impairs glucose tolerance in control and dystrophic mice ------------------------------------------------------------------ To check whether tranilast administration altered glucose handling in control and dystrophic mice we also performed a glucose tolerance test. Dystrophic *mdx* mice exhibited impaired glucose tolerance as evidenced by a 100% higher glucose response following a single intraperitoneal injection of glucose (1 g/kg) (expressed as area under the curve, Figure [4](#F4){ref-type="fig"}, *p* \< 0.05). While basal blood glucose levels were not affected by tranilast administration, \~20% increased peak blood glucose levels were observed in treated control and *mdx* mice compared with untreated mice during the GTT (Figure [4](#F4){ref-type="fig"}). Furthermore, the blood glucose response was \~70% higher in tranilast-treated control and *mdx* mice compared with untreated mice (Figure [4](#F4){ref-type="fig"}). {#F4} Discussion ========== The identification of pharmacological agents that can prevent, reduce and/or resolve fibrotic deposition has great potential for enhancing therapies for DMD and other muscle-wasting disorders. While gene and cell therapies will eventually provide the cure for the single-gene muscle-wasting disorders, the efficacy of these approaches is likely to be hampered by the presence of significant fibrosis within affected skeletal muscles. Here we have demonstrated that one agent, tranilast, successfully reduces fibrotic deposition in skeletal muscles of *mdx* dystrophic mice. Tranilast has been administered to sarcoglycan-deficient Bio14.6 hamsters, a rodent model of limb-girdle muscular dystrophy (LGMD). Treatment of 30-day-old hamsters for 120 days significantly decreased fibrosis in skeletal muscle and reduced serum creatine kinase levels and the number of centrally nucleated muscle fibres, indicating reduced muscle fibre breakdown and regeneration \[[@B25]\]. That study also observed a reduction in serum creatine kinase levels after a 30-day treatment in 30-day-old *mdx* mice \[[@B25]\]. We have subsequently demonstrated that oral administration of tranilast (\~300 mg/kg) to young mice for 9 weeks significantly reduced fibrotic accumulation by \~30% in the diaphragm muscles of *mdx* mice (Figure [2](#F2){ref-type="fig"}A, B, C). We observed a similar trend towards a decrease in fibrosis accumulation in the TA muscles of treated *mdx* mice but this was not statistically significant (*P* = 0.10). This is most likely due to the low levels of fibrosis in the TA muscles compared with those in the diaphragm of *mdx* mice. The observed decrease in the diaphragm, which is the most severely affected of the muscles in the *mdx* mouse, indicates that tranilast was able to reduce fibrotic accumulation. We did not observe a change in the maximum force production of the TA muscle or diaphragm muscle strips, which is not surprising since a 30% reduction in fibrosis in these muscles of *mdx* mice would only increase the available functional tissue in the TA and diaphragm by 1.5% and 5%, respectively. We did not observe an improvement in whole-body strength or mobility of the treated mice or in maximum force of diaphragm muscle strips. Other compounds with specific muscle effects and antifibrotic properties such as IL-15 and sildenafil have been shown to improve the muscle force-producing capacity \[[@B26],[@B27]\]. However, we did observe an improvement in the fatigability of both the diaphragm and TA muscles in treated *mdx* mice, indicating a potential improvement in muscle function. This improved fatigue resistance was not due to any changes in the muscle oxidative capacity, as there were no differences in muscle fibre oxidative capacity (SDH) between control and tranilast-treated mice (Table [1](#T1){ref-type="table"}). Although tranilast administration successfully decreased fibrotic tissue infiltration in dystrophic skeletal muscle, one concern was that it also resulted in impaired glucose tolerance in both dystrophic and control mice. Tranilast has been shown to inhibit insulin secretion in rats \[[@B28]\], so long-term treatment may not be possible using this specific compound. However, the effective human dose of tranilast for fibrotic pathology has been shown to be 5 mg/kg, and subsequent safety information regarding the use of tranilast in humans at this dose has not indicated toxicity issues associated with effects on insulin secretion \[[@B18]\]. It remains to be determined whether this dose, which is significantly lower than the dose used in the present study, would be sufficient to alter fibrosis in human skeletal muscle. Therefore it is possible that newer generation drugs may be required for therapeutic application. To this end, more targeted drugs based on the structure of tranilast are being developed that can attenuate interstitial fibrosis in the hearts of diabetic rats without causing hyperglycaemia \[[@B29]\]. Conclusion ========== Interventions to minimise fibrosis are important not just for skeletal muscle diseases but also to enhance functional recovery after serious muscle injuries. Although tranilast decreased fibrosis in dystrophic skeletal muscles, functional benefits were limited to modest improvements in fatigue resistance with impaired glucose tolerance also being a limiting factor. These issues need to be overcome in order to improve the therapeutic relevance and efficacy. Together these findings demonstrate that administration of potent antifibrotic compounds such as tranilast and newer drugs could help preserve skeletal muscle structure to ultimately increase the efficacy of pharmacological, cell and gene replacement/correction therapies for muscular dystrophy and related disorders. Methods ======= Animals ------- Three-week-old male C57BL/10 (*n* = 32) and C57BL/10ScSn *mdx* (*mdx*) (*n* = 32) mice were obtained from the Animal Resources Centre (ARC), WA, Australia. All experimental protocols were approved by the Animal Ethics Committee of The University of Melbourne and conducted in accordance with the Australian code of practice for the care and use of animals for scientific purposes as stipulated by the National Health and Medical Research Council (Australia). Mice were allocated into one of four groups: (1) control group treated with vehicle (CON + VEH, *n* = 16); (2) control group treated with tranilast (CON + TRA, *n* = 16); (3) *mdx* group treated with vehicle (*mdx* + VEH, *n* = 16); (4) *mdx* group treated with tranilast (*mdx* + TRA, *n* = 16). Tranilast administration ------------------------ Three-week-old control (C57BL/10) and *mdx* mice received tranilast (kindly provided by Dr. Spencer Williams, Department of Chemistry, The University of Melbourne) for a period of 9 weeks. Mice received standard laboratory chow (Speciality Feeds, Glen Forrest, WA, Australia) with or without the addition of 2.7 g tranilast/kg. Food was made available *ad libitum* based on the assumption that mice would consume approximately 4 g of feed per day resulting in a dose of 400 mg/kg/day of tranilast. This dose has been shown previously to reduce collagen infiltration in heart \[[@B20]\] and kidney \[[@B30]\] of diabetic rats. Food intake was monitored throughout the treatment period \[[@B20]\] and as mice consumed an average 3.3 g/day, daily tranilast intake averaged 293 mg/day. Measurement of whole body strength ---------------------------------- Whole body strength, whole body mobility and coordination were assessed in control C57BL/10, treated C57BL/10, control *mdx* and treated *mdx* mice at 2 (mobility) or 3 (body strength) days prior to endpoint by means of a grip strength meter (Columbus Instruments, Columbus, OH) and rotarod performance test (Rotamex-5, Columbus Instruments) as described previously \[[@B31]\]. Glucose tolerance test ---------------------- Glucose tolerance tests were performed on control C57BL/10, treated C57BL/10, control *mdx* and treated *mdx* mice 5 days prior to endpoint. Following an overnight fast, a basal blood sample was taken from the tail vein (23 G needle) and analysed for glucose concentration using a glucometer. Mice then received an intraperitoneal injection of glucose solution (1 g/kg body mass). At 15, 30, 60, 90 and 120 min after the injection of the glucose solution, a blood sample was collected from the tail vein (23 G needle) and analysed for glucose concentration. Assessment of contractile properties of skeletal muscle and tissue collection ----------------------------------------------------------------------------- At the conclusion of the treatment period, mice were anesthetised with sodium pentobarbitone (Nembutal; 60 mg/kg; Sigma-Aldrich) *via i.p*. injection. The methods for assessment of the contractile properties of the mouse tibialis anterior (TA) muscle *in situ* have been described in detail previously \[[@B32]\]. At the conclusion of the contractile measurements *in situ*, the TA muscle was carefully excised, blotted on filter paper and weighed on an analytical balance, followed by freezing in thawing isopentane for later histological examination. Soleus (SOL), extensor digitorum longus (EDL), plantaris (PLAN), gastrocnemius (GAST) and quadriceps (QUAD) muscles were excised, blotted on filter paper, trimmed of their tendons and adhering tissue, weighed and frozen in liquid nitrogen. The entire diaphragm and rib cage were then surgically excised and costal diaphragm muscle strips composed of longitudinally arranged full-length muscle fibres were isolated and prepared for functional assessment *in vitro*, as described in detail elsewhere \[[@B32]\]. Upon completion of the functional analyses, the diaphragm muscle strip was trimmed of tendon and any adhering non-muscle tissue, blotted once on filter paper and weighed on an analytical balance. The muscle strips were then frozen in thawing isopentane for later histological examination. Mice were killed as a consequence of diaphragm and heart excision while deeply anesthetised. Skeletal muscle histology and fibrosis -------------------------------------- Serial sections (5 μm) were cut transversely through the diaphragm and the TA muscle using a refrigerated (-20°C) cryostat (CTI Cryostat; IEC, Needham Heights, MA). Sections were stained with haematoxylin and eosin (H&E) to reveal the general muscle architecture. Muscle fibre cross-sectional area, oxidative enzyme capacity and fibre type were determined as described previously \[[@B33]\]. Briefly, TA and diaphragm sections were reacted histochemically for succinate dehydrogenase (SDH) activity and immunoreacted with antibodies to laminin (Sigma-Aldrich) and myosin IIa, N2.261 (The University of Iowa, Department of Biology, Iowa City, IA; USA) in order to determine the oxidative capacity, CSA of individual myofibers and proportion of type IIA fibres respectively. Muscle collagen content was assessed from Van Gieson-stained cross-sections \[[@B32]\]. Digital images of stained sections were obtained using an upright microscope with a camera (Axio Imager D1, Carl Zeiss, Wrek, Göttingen, Germany), controlled by AxioVision AC software (AxioVision AC Rel. 4.8, Carl Zeiss Imaging Solutions, Wrek, Göttingen, Germany). Images were quantified using AxioVision 4.8 software. Analysis of gene expression --------------------------- At the conclusion of the treatment period, diaphragm muscles were excised and total RNA was extracted using a commercially available kit, according to the manufacturer's instructions (PureLink RNA Mini Kit, Invitrogen). The RNA concentration was determined by a spectrophotometer at 260 nm and subsequently transcribed into cDNA using the Superscript VILO cDNA synthesis kit (Invitrogen). Real-time RT-PCR was performed as described previously \[[@B34]\] using the following forward and reverse primer sequences: Col1a1, 5′-CACCCTCAAGAGCCTGAGTC-3′and 5′-GTTCGGGCTGATGTACCAGT-3′; Col2a1, 5′-GCCAAGACCTGAAACTCTGC-3′ and 5′-GCCATAGCTGAAGTGGAAGC-3′; Col3a1, 5′-ACCAAAAGGTGATGCTGGAC-3′ and 5′-GACCTCGTGCTCCAGTTAGC-3′; Col4a1, 5′-AAAGGGAGAAAGAGGCTTGC-3′ and 5′-CCTTTGTACCGTTGCATCCT-3′; Col5a1, 5′-GGTCCCTGACACACCTCAGT-3′ and 5′-TGCTCCTCAGGAACCTCTGT-3′; Col6a1, 5′-CCCCATTGGACCTAAAGGAT-3′ and 5′-TCTCCCACTTCACCCTCATC-3′; Fn1, 5′-ACCACCCAGAACTACGATGC-3′ and 5′-GGAACGTGTCGTTCACATTG-3′; TGFβ1, 5′- TGAGTGGCTGTCTTTTGACG-3′ and 5′- TCTCTGTGGAGCTGAAGCAA-3′. Gene expression was quantified using a cycle threshold (C~T~) method, whereby a ΔC~T~ was calculated by subtracting the 18S C~T~ from the gene C~T~. The relative gene expression was then calculated using the expression 2^-ΔCT^. Statistical analyses -------------------- Data were analysed with the GraphPad Prism software. Statistical significance was determined using a two-way analysis of variance (ANOVA; for comparison of treatment and group), with significance set at *P* \< 0.05. A Tukey post hoc multiple comparison test was used where appropriate to determine significance between groups. For fatigue data comparing multiple time points, a two-way repeated measures ANOVA was used. Values are presented as mean ± SEM. Abbreviations ============= CSA: Cross-sectional area; DMD: Duchenne muscular dystrophy; EDL: Extensor digitorum longus; GAST: Gastrocnemius; LGMD: Limb girdle muscular dystrophy; mdx: Muscular dystrophy X-linked; PLAN: Plantaris; QUAD: Quadriceps; SOL: Soleus; TA: Tibialis anterior; TGFβ: Transforming growth factor-β; TRA: Tranilast; VEH: Vehicle. Competing interests =================== None of the authors have any competing interests. Authors' contribution ===================== KS, SMG, DIS, RK and GSL conceived and designed the experiments. MT, KS, AC and TN performed the experiments. KS, MT, SMG, DIS, RK and GSL have analysed and interpreted the data. KS, MT, DIS, RK and GSL have drafted and revised the manuscript. All authors have given final approval of the latest version of the manuscript. Acknowledgements ================ Supported by a Research Collaboration Grant from The University of Melbourne. K.S. is supported by an Early Career Fellowship from the National Health and Medical Research Council.

{#sp1 .nil7}

Background {#S0005} ========== Mortuary workers like other health workers are exposed to the hazards of blood borne pathogens like Human Immunodeficiency Virus, Hepatitis B and C among others at work. \[[@CIT0001],[@CIT0002],[@CIT0003]\]. Indeed health workers are exposed to up to twenty- seven different blood borne pathogens at work \[[@CIT0004]\]. Previous studies suggest that occupational exposure to blood and body fluids is responsible for 2.5% of Human Immunodeficiency Virus and 40% of Hepatitis B and Hepatitis C among health care workers worldwide \[[@CIT0005],[@CIT0006]\]. In the control of hazards facing health workers, baseline and periodic assessment of exposure to these hazards is an important strategy which is useful as a decision making tool in risk assessment and management of occupational hazards. There are many studies that focus on exposure assessment to blood borne pathogens among health workers especially nurses and doctors with most studies indicating a higher frequency of exposure to blood and body fluids among nurses \[[@CIT0007]--[@CIT0009]\]. Little attention has been paid, however to exposure assessment among mortuary workers even though they are also at risk of contracting diseases from exposure to blood borne pathogens like other categories of health workers. The objective of this study was to carry out an assessment of exposure to blood among mortuary workers in south west Nigeria. The information obtained will be useful in designing programmes to protect the health of mortuary workers as well as to assess the effect any programme put in place based on gaps identified. Methods {#S0006} ======= A descriptive cross sectional study was conducted among all mortuary workers in six teaching hospitals in South West Nigeria namely the teaching hospitals at Lagos (2), Ibadan (1), Ife (1), Sagamu (1) and Osogbo (1) cities in total there were 80 mortuary workers. Mortuary workers are often overlooked in studies on health and safety among health workers even though they are also at risk. The majority of mortuary workers in Nigeria work in teaching hospitals, where conditions of service are better, and, as such, are more likely to use best practices. In the city of Lagos, there are some workers in the very few private mortuaries. Mortuary workers in this study include; porters, attendants and others working in mortuary but excludes doctors (pathologists). The study material was a 15- item questionnaire which was self -administered by the mortuary workers. The study material was in three sections. Section A focused on the socio-demographic characteristics of the workers; section B focused on exposure to blood from various sources while section C focused on Hepatitis B vaccination status. The questionnaire was designed by the investigators based on objectives and results from other related studies. Face and content validation was carried out on study material prior to data collection. Ethical issues were addressed in the study by the fact that permission was sought from the management of the hospitals as well as the mortuary workers themselves before data was collected. Participation was voluntary and confidentiality of the information obtained was ensured by restricting access of data tools to only the investigators. Ethical approval was not sought from an institutional body because no invasive procedure was carried out during the study. Data collection took place from March to May 2008. Data was analysed with Epi-info 2002 and presented in tables and figures. In testing association between socio-demographic variables and exposure, odds ratio with 95% confidence intervals were derived. Results {#S0007} ======= A total of 76 completed questionnaires were retrieved and analysed giving a response rate of 95%. [Table 1](#T0001){ref-type="table"} shows the socio- demographic characteristics of respondents. Most of the respondents were aged between 30--39 years with the mean age being 38.2 years. Almost all (92%) of the respondents were males and (data on faith and tribe was collected as part of socio-demography but was not presented because of limited value and space constraints) married. Majority (69.4%) of the workers were in the junior category and slightly more than one third (36%) had worked for less than 5years. Similarly slightly more than one third (36%) had worked for more than 10 years. ###### Socio-demographic characteristics of respondents Variable Frequency (%) ------------------------------ --------------- **Age group(years) N=76** 20--29 15(19.7) 30--39 29(38.2) 40--49 21(27.6) 50--59 11(14.5) **Sex (N=76)** Male 69(92) Female 6(8) **Educational status(N=73)** None 3(4.1) Primary 24(32.9) Secondary 24(32.9) Post-secondary 22(30.1) **Category of worker(N=72)** Senior 22(30.6) Junior 50(69.4) **Work duration(N=75)** Less than 5years 27(36) 5--10years 21(28) Greater than 10 years 27(36) Senior workers are on grade level 8 and above while junior workers are below grade level 8. Education and year of experience play role in category of worker [Table 2](#T0002){ref-type="table"} shows the exposure to blood among the respondents in the previous one year prior to data collection. Fifty three workers (85.5%) had been exposed through blood splash. Similarly 50 (73.5%) had experienced needle-stick injury. Exposure to blood through cuts on skin at work occurred in 48(72.7%) of the workers. Exactly half (50 %) of workers had all three exposures. [Table 3](#T0003){ref-type="table"} shows factors associated with exposure to various forms of exposure to blood. ###### Exposure to blood in the past one year through various sources among respondents Variable Frequency (%) --------------------------- --------------- Blood splash(N=62) 53(85.5%) Needle stick injury(N=68) 18(26.5%) Cuts on skin(N=66) 48(72.7%) All sources (N=68) 34(50%) ###### Factors associated with various sources of exposure to blood among respondents in the past one year Factors associated with exposure Needle-stick injury Blood splash Cuts on skin ----------------------------------------------------- --------------------- ------------------- -------------- ------------------- ---------- ------------------- -- **Sex of worker** Male (N=61) 46(75.4) 3.07(0.36--24.95) 49(80.3) 4.08(0.47--33.59) 44(72.1) 2.59(0.31--20.96) Female(N=6) 3(50) 1(reference) 3(50) 1(reference) 3(50) 1(reference) **Category of worker** Senior(N=23) 14(66.7) 0.69(0.19--2.59) 16(76.2) 0.62(0.14--2.91) 13(61.9) 0.56(0.16--2.01) Junior(N= 43) 32(74.4) 1(reference) 36(83.7) 1(reference) 32(74.4) 1(reference) **Work duration** Less than 5years(N=23) 15(65.2) 1(reference) 22(95.7) 1(reference) 16(69.6) 1(reference) Five years and above(N=45) 35(77.7) 1.87(0.52--6.46) 31(68.9) 0.10(0.00--0.78) 32(71.1) 1.08(0.30--3.63) **Respondent has ever attended training on safety** Yes(N=43) 29(67.4) 0.91(0.25--3.03) 30(69.8) 0.49(0.10--1.91) 28(65.1) 1.20(0.37--3.84) No (N=23) 16(69.6) 1(reference) 19(82.6) 1(reference) 14(60.1) 1(reference) With regards to needle-stick injury, males were 3.07 times as likely to experience it compared to females (95% confidence interval of 0.36--24.95). Senior workers were 0.69 times as likely to experience needle-stick injury compared to junior workers (95% confidence interval of 0.19--2.59). Workers who had worked five years and above were 1.87 times as likely to experience needle-stick injury compared to those who had worked under five years(95% confidence interval 0.52 --6.46). In addition, workers who had ever attended a training programme on safety were 0.91 times as likely to experience needle-stick injury compared to those who had not (95% confidence interval 0.25--3.03). Regarding occurrence of blood splash, males were 4.08 times as likely to experience it compared to females (95% Confidence interval of 0.47--33.59). Senior workers were 0.62 times as likely to experience blood splash compared to junior workers (95% confidence interval of 0.14--2.91). Workers who had worked five years and above were 0.10 times as likely to experience blood splash compared to those who had worked under five years (95% confidence interval of 0.00--0.78,). In addition, workers who had ever attended a training programme on safety were 0.49 times as likely to experience blood splash compared to those who had not (95% confidence interval 0.10--1.91). With regards to cuts on skin at work, males were 2.59 times as likely to experience it compared to females (95% confidence interval of 0.31--20.96). Senior workers were 0.56 times as likely to experience cuts on skin compared to junior workers (95% confidence interval of 0.16--2.01). Workers who had worked for at least five years were 1.08 times as likely to experience cuts on skin compared to those who had worked under five years(95% confidence interval 0.30 --3.63). In addition, workers who had ever attended a training programme on safety were 1.20 times as likely to experience cuts on skin compared to those who had not (95% confidence interval 0.37--3.84). [Table 4](#T0004){ref-type="table"} shows the Hepatitis B vaccination status of workers who had been exposed to needle-stick injury. Among the workers with a history of needle-stick injury, 26(57.8%) had received a dose of hepatitis B vaccine in the past. However, only 5 of the workers with a history of needle-stick injury (10.4%) had completed three doses of hepatitis B required for protection. Similarly among workers with history of blood splash, 29(59.2) had received a dose of hepatitis B vaccine in the past with 8(27.6%) of them completing three doses. Among workers with history of cuts on skin, 22(52.4%) of them had received a dose of hepatitis B vaccine in the past. However, only 3(13.6%) had completed three doses of Hepatitis B vaccine. In all of this, no titre measurement to confirm protection was done. ###### Hepatitis B vaccination status of respondents exposed to needle stick injury, cuts and blood splash Variable Needle-stick injury Frequency (%) Cuts Frequency (%) Blood splash Frequency (%) -------------------------------------------------------- ----------------------------------- -------------------- ---------------------------- **Received a dose of hepatitis B vaccine in the past** Yes 26(57.8) 22(52.4) 29(59.2) No 19(42.2) 20(47.6) 20(40.8) **Number of doses received (N=26)** One 6(23.1) 6(27.3) 6(20.7) Two 15(57.7) 13(59.1) 15(51.7) Three 5(19.2) 3(13.6) 8(27.6) Discussion {#S0008} ========== This, to the best of our knowledge, is the first study targeting health and safety of mortuary workers in south west Nigeria with regards to exposure to blood. This is an often overlooked category of workers even though they face risks to their health from exposures at work. Findings from the study will be useful to managers of these institutions, programme managers and all those interested in occupational health and safety. More than two-thirds of the health workers had been exposed to blood either through cuts, blood splash or needle-stick injury. This finding is much higher than the finding of a study in India with a exposure rate of 32.7% \[[@CIT0010]\], It is also higher than the 43% reported in a study in Iran \[[@CIT0011]\] but is similar to the 79% reported in another study in Southern Iran \[[@CIT0012]\]. In this study, blood splash was the most common form of exposure to blood even though needle-stick injury was also common. This is in agreement with other studies which rank needle-stick injury as a very important source of exposure \[[@CIT0013],[@CIT0014]\]. Exposure through the skin was also common occurring in almost three -quarters of respondents. This necessitates the provision of personal protective devices such as gloves, aprons and boots. Any intervention based on findings of this study will need to focus on availability and utilization of these personal protective devices as well as training on their proper use. It is worth noting that half of the workers had experienced exposure from all sources indicating the high levels of exposure among workers and implications on their health. The factors associated with exposure to blood were also explored in this study. Regarding exposures to needle-stick injury, male sex and having worked for more than 5 years was associated with greater exposure to needle-stick injury compared to being female and having worked for less than five years. However, the confidence interval of the odds ratio suggests chance has a role in the findings. Indeed the wide confidence interval suggests that sample size may be inadequate to explore the association. On the other hand, senior workers and those who had attended a training programme on safety were less likely to have needle-stick injury compared to those who were junior and those who had neverattended any training programme on safety. However, chance may explain findings as shown from confidence intervals. Regarding exposure to blood splash, males were more likely to experience it compared to females, even though chance may explain findings. On the other hand, workers who were in senior category and who had attended a training programme were less likely to experience blood splash compared to junior workers and those yet to attend a training programme even though chance may explain findings. Workers who had spent 5 years and more were less likely to experience blood splash and this was statistically significant as shown from the confidence intervals. This finding of association between work duration and occurrence of blood splash is in keeping with the findings of a study which showed that more experienced workers were less likely to have exposure to blood \[[@CIT0015]\]. Even though males were more likely to experience blood splash, chance may explain the findings. This is in contrast with findings from a study where males were more likely than females to have injury as source of exposure to blood \[[@CIT0016]\]. The small number of females in study may be responsible for the difference seen. Regarding exposure to cuts on skin, males were more likely than females to experience it though chance may provide an explanation. In addition, workers who had worked for more than five year were more likely to report it compared to those who had worked less than five years even though chance may explain the findings. On the other, senior workers were less likely to experience cuts compared to junior workers even though chance may explain the findings.The sample size may be inadequate to explore associations as seen from wide confidence intervals. Among workers who had needle-stick injury, cuts, and blood splash only 19.2%, 13.6% and 27.6% respectively had completed the three doses of Hepatitis B required for protection. This is much lower than the findings of a study in China where over two-thirds of workers were immunized against Hepatitis B \[[@CIT0017]\] but is comparable to findings from other West African countries \[[@CIT0018]\]. This has grave implications for the health and safety of the workers. However, confirmation of protection by antibody estimation was not done. It will be interesting to explore reasons for low percentage of workers who completed all three doses of Hepatitis B vaccine. It may be that the workers were not motivated to complete all doses or that vaccines were not always available. Further research is required in this regard. This study is the first study to the best of our knowledge on assessment of exposure to blood among mortuary workers. The findings have documented high exposure through blood splash and cuts on skin. The study however, has its limitations. The sample size was inadequate to explore associations and antibody estimation was also not done. Nevertheless, the findings will be useful in designing interventions to safe guard the health of the workers as well as for medical audit activities. Conclusion {#S0009} ========== In conclusion, this study demonstrates that in this population of mortuary workers exposure to blood was very common. However, majority of workers with needle-stick injury, cuts and blood splash had not received three doses of Hepatitis B vaccine required for their protection. There is a need to vaccinate all mortuary workers with three doses of Hepatitis B vaccine. In addition, education of workers on risk and training programmes on safety at work will be useful in safe guarding health of mortuary workers. We wish to thank the following for their assistance in the retrieval of completed questionnaires in the participating hospitals Dr S.S Soyemi, Lagos state University teaching hospital Lagos state; Dr B.M Duduyemi University College Hospital Ibadan Oyo state; Dr (Mrs) G.D Omoniyi- Esan Obafemi Awolowo University teaching hospital Ile-ife Osun state; Dr Ojemakinde Ladoke Akintola teaching hospital Osogbo Osun state; Dr E.K Abudu Olabisi Onabanjo Teaching Hospital Sagamu Ogun state. We also wish to thank all mortuary workers who participated in the study. Competing interests {#S0010} =================== The authors declare that they have no competing interests Authors' contributions {#S0011} ====================== OBE and ACC, designed and collected data, OBE and OAT analysed data, OKA wrote discussion. All authors read and approved the final version of the manuscript.

GENETIC EPIDEMIOLOGY ==================== Bipolar disorder (BP) is a disabling psychiatric condition with a considerable public health impact, affecting 1--2% of the general population ([@B89]). Family, twin, and adoption studies show that genetic factors play a leading role in is etiology. Family studies suggest that compared to the general population the risk of BP is 5--10 times greater in siblings of a proband with the disorder ([@B21]) and twin studies provide estimates of heritability for BP that range from 80 to 90%, ranking it among the most heritable of psychiatric disorders ([@B9]). However, conclusive findings from molecular studies to identify relevant genetic factors have proven elusive, and the genetic architecture remains unresolved. Although results from sequencing analyses may reveal rare variants with large effects on the level of the individual, single variants are not likely to influence BP risk at the population level ([@B22]). Rather, multiple genetic phenomena likely contribute to BP risk, including the additive and interactive effects of many variants, each contributing a small effect, as well as heterogeneity across populations. This has lead to an interest in identifying the molecular pathways that are disrupted by these risk variants, any number of which may underlie an individual's susceptibility to BP ([@B7]). GENOME-WIDE ASSOCIATION STUDIES =============================== Encouraged by the evidence from genetic epidemiology studies, considerable effort has been expended over the past two decades to identify susceptibility genes for BP. The completion of the Human Genome Project and advances in high-throughput genotyping technology spurred a new generation of genome-wide association studies (GWAS) making it possible to test for genetic associations at sites of common polymorphic variation in an unbiased manner across the entire genome. The first GWAS of BP with individual level genotype data came from the Wellcome Trust Case Control Consortium (WTCCC; [@B90]). The WTCCC simultaneously conducted a GWAS of seven different disorders, including BP. Two thousand cases for each disorder and a common set of 3,000 controls were genotyped at \<500,000 single-nucleotide polymorphisms (SNPs) across the genome. The analysis of data on BP revealed no findings that were considered genome-wide significant. Soon afterward, another GWAS of BP was reported from the Systematic Treatment Enhancement Program for Bipolar Disorder (STEP-BD; [@B80]). In this study, a total of 1,461 cases with BP from the STEP-BD trial and the University College of London (UCL) and 2,008 controls were genotyped. As with the WTCCC, none of the findings were genome-wide significant. Given the heritability of BP, the lack of genome-wide significant findings and generally inconsistent results from these early GWAS highlighted the fact that the common variants are likely to represent genetic effects of modest size. It was not until large samples from several different studies were combined in a mega-analysis that compelling findings for BP began to emerge. In 2008, [@B30] reported a GWAS of 4,387 BP cases and 6,209 controls from the WTCCC, STEP-UCL, and ED-DUB-STEP2 datasets with over 1.8 million genotyped and imputed SNPs. This was the largest GWAS to date and it reported the first genome-wide significant association in BP \[rs10994336, odds ratio (OR) = 1.45, *p* = 9.1 × 10^-9^\] at *ANK3* (ankyrin G). [@B77] subsequently replicated the findings with *ANK3* in a candidate gene study using a sample that had previously been part of a pooled GWAS ([@B11]). They found significant associations with two SNPs in *ANK3*, including the SNP reported by [@B30]; rs9804190, OR = 1.32, *p* = 3 × 10^-6^; rs10994336, OR = 1.54, *p* = 1.7 × 10^-5^). The two SNPs were \~340 kb apart with little linkage disequilibrium between them, suggesting the possibility of independent loci contributing to allelic heterogeneity. Three GWAS ([@B78]; [@B81]; [@B45]) were subsequently reported, including one ([@B81]) of 1,001 BP cases and 1,033 controls of European-American ancestry and 345 BP cases and 670 controls of African-American ancestry from the National Institute of Mental Health (NIMH) Genetic Association Information Network (GAIN), a second ([@B78]) on 2,076 BP cases and 1,676 controls of European-American ancestry that partially overlapped with the first, and a third ([@B45]) on 1,000 BP cases and 1,000 controls of Han Chinese ancestry. None of the findings from these three GWAS met criteria for genome-wide significance, but provided further confirmatory evidence for an association with *ANK3*. [@B78] carried out a fixed effects meta-analysis of rs10994336 with the previously published findings and provided the strongest evidence yet of association with *ANK3* (OR = 1.47, *p* = 1.1 × 10^-10^). To improve the power to detect susceptibility variants in BP and other complex psychiatric disorders, a landmark initiative referred to as the Psychiatric GWAS Consortium (PGC) was undertaken to assemble existing samples of BP, schizophrenia, depression, attention-deficit hyperactivity disorder (ADHD), and autism and carry out the largest GWAS ever conducted with these disorders ([@B66]). The mega-analysis led by the PGC-BP ([@B65]) brought together an international sample of 7,481 BP cases and 9,250 controls with imputed genotype data on \~2.5 million SNPs. The top finding was again in *ANK3* with an OR = 1.35 (*p* = 7.1 × 10^-9^) at rs10994397, which is approximately 99 kb from rs10994336, the SNP that was highlighted in the original Ferreira et al.'s (2008) study. When combined with an independent replication sample of 4,496 BP cases and 42,422 controls, the most significant finding to emerge was at rs4765913 in *CACNA1C* with an OR = 1.14 (*p* = 1.52 × 10^-8^). The convergence of findings around *ANK3* and *CACNA1C* has suggested a biologically plausible etiologic mechanism in BP. *ANK3* is an adaptor protein that regulates the distribution of voltage-gated ion channels, and *CACNA1C* is the alpha 1C subunit of the L-type voltage-gated calcium channel. This lent further support to the burgeoning hypothesis that BP may be an ion channelopathy ([@B27]; [@B26]; [@B5]; [@B6]; [@B16]). Pathway-based analyses of GWAS data, which collectively tests for associations of SNPs across sets of genes within defined molecular pathways instead of testing each SNP independently ([@B7]), has provided further evidence for this hypothesis. A recent pathway-based analysis ([@B7]) was carried out with data from the NIMH GAIN and WTCCC GWAS using gene sets based on gene ontology (GO) categories. Results from this analysis implicated genes involved in the structure and regulation of ion channels. Of the 16 gene sets that were statistically significant (*p* \< 0.01) in the NIMH GAIN discovery sample, nine were ion channel/binding/transporter sets. Eight of these nine were replicated in the WTCCC sample. Notably, the most significant group in the NIMH GAIN sample, which was also replicated in the WTCCC sample, was a gene set involved in voltage-gated ion channel activity. Since only four genes overlap in the gene sets between samples, it appears that the shared molecular function (mediation of voltage-gated ion channel activity) rather than a few highly significant results common across studies drove the results. Another pathway-based analysis ([@B46]) used a convergent functional genomics (CFG) approach to identify gene sets of interest. They first extracted nominally significant signals (*p* \<0.05) from three GWAS studies (WTCCC, German, and NIMH GAIN datasets; [@B29]; [@B90]; [@B11]), and prioritized potential candidates based on independent, converging lines of evidence from various bioinformatics resources. Results implicated gene sets involved in calcium and potassium regulation, as well as cellular adhesion, thus providing further evidence implicating dysregulation of ion channel machinery in BP. In the largest GWAS of psychiatric illness to date, a recent analysis from the PGC Cross-Disorder Group ([@B23]) combined various approaches to examine a potentially shared genetic contribution to five psychiatric disorders: major depressive disorder, bipolar disorder, schizophrenia, autism spectrum disorders, and ADHD (total *N* = 33,332 cases and 27,888 controls). Methods included multinomial logistic regression with model selection, polygenic risk score analysis, pathways analysis, and an enrichment analysis of expression quantitative trait loci (eQTL) data. In the primary analysis, two of the four genome-wide significant results were in calcium channel genes (*CACNA1C* and *CACNB2*), and model selection supported a contribution of these variants to multiple disorders. Additionally, results from the pathways analysis suggested that genes involved in calcium channel activity have pleiotropic effects among multiple disorders. These findings suggest that variants which alter calcium channel functioning contribute not only to BP, but also to a shared, fundamental mechanism that contributes to a broader definition of psychopathology. GENETIC FINDINGS WITH POTASSIUM ION CHANNELS ============================================ Because of the findings with *CACNA1C*, most of the attention has understandably focused on the role of calcium ion channels in BP. However, the possibility that potassium ion channels may also contribute to the etiology of BP was suggested by a re-analysis of GWAS data from the NIMH GAIN sample (reported in a companion paper in this issue). Given that *ANK3* encodes a protein known to molecularly interact with other membrane bound proteins in neurons, we sought to test whether interactions between *ANK3* and genes encoding these other proteins may contribute to BP susceptibility. Using the STRING 9.0 bioinformatics database ([@B85]), which collates existing evidence of protein--protein interactions, we identified putative interactors with ANK3. We then tested for interactions between SNPs in *ANK3* and in these interacting proteins in association with BP. The most significant findings were between *ANK3* and *KCNQ2* (*p* = 3.18 × 10^-8^), which remained significant after accounting for multiple testing. These interactions were driven by two SNPs in *KCNQ2*, rs2282150 (an intronic SNP) and rs2297385 (a synonymous coding SNP), interacting with 16 different SNPs in *ANK3* that all fall within the boundaries of ankyrin repeats, which are functional domains known to mediate protein--protein interactions. We were able to replicate these associations using data from the WTCCC sample. We also found suggestive evidence of interactions between *ANK3* and *KCNQ3* (*p* = 1.86 × 10^-5^). *KCNQ2* and *KCNQ3* encode proteins that form hetero-tetramer complexes making up the voltage-gated potassium ion channels in neurons. Interestingly, previous genetic studies have implicated potassium related genes in BP. For example, of the over 40 linkage regions that had been reported in genome-wide linkage studies of BP ([@B56]), one of the most prominent implicated chromosome 8q24, a region that includes *KCNQ3*. The 8q24 region initially reached a logarithm (base 10) of odds (LOD) score of 2.39 in a genome-wide scan by the Johns Hopkins Mood Disorders Research Group ([@B31]). This finding was strengthened by including additional pedigrees ([@B52]) and markers \[[@B8]; non-parametric linkage (NPL) = 3.13 and 3.25, respectively\]. It was also replicated in an independent sample by a German group ([@B18]; LOD = 3.6). A meta-analysis using primary genotype data from 11 genome-wide linkage scans again indicated 8q as one of only two significant regions in the genome (LOD = 3.4; [@B53]). [@B93] fine mapped a 3.4-Mb region under the linkage peak with 249 informative SNPs. This family-based association analysis with 155 nuclear families identified a nominally significant SNP (rs1901090, *p* \< 0.05) and haplotype (rs2945733--rs1901090, *p* = 0.003) near the gene. Finally, this region was further probed with 2,756 observed and 15,552 imputed genotypes ([@B94]), the most significant of which (rs2673582, *p* = 4.80 × 10^-5^) lies 27 kb upstream of *KCNQ3*. *KCNQ2* has also been implicated in BP by candidate gene association studies. One study ([@B12]) found two significant associations with SNPs in KCNQ2 (*p* \< 0.05). The results from this study are particularly interesting given the functional consequences of these variants within *KCNQ2*, which suppressed the potassium channel activity. The authors hypothesized that this might result in hyperexcitable neurons, which are characteristic of manic and hypomanic states. Other candidate gene studies of BP have implicated *KCNN3*, a voltage-independent calcium-activated potassium channel, which is thought to mediate neuronal excitability via the slow component of synaptic afterhyperpolarization that follows an action potential event. The highly polymorphic CAG-repeat array in exon 1 had been implicated in BP ([@B17]; [@B92]; [@B38]; [@B51]; [@B71]). However, a meta-analysis of pooled evidence from these individual studies was null ([@B35]). The authors of the meta-analysis noted that the polymorphism in question, or one in LD with it, might yet be associated with a subphenotype of the disorder, such as anticipation or symptom severity. Indeed, a more recent analysis of a schizophrenic sample found an association between the long CAG repeats (which reduce the channel's functioning) and better cognitive performance on tasks that measured the ability to discriminate, select, and execute ([@B37]). A series of gene expression studies have provided further evidence for a role of potassium ion channels in BP. Using post mortem striatum samples of human brains from the Stanley Foundation Brain Collection \[including the nucleus accumbens (Str-NAc) and the lateral cerebellar hemisphere\], one group ([@B82]) analyzed expression of 72 different ion channel subunits. They found altered gene expression patterns in BP, with a significant up-regulation of *KCNQ2* (fold change: 1.51, *p* = 0.039677), *KCNQ3* (fold change: 7.00, *p* = 0.005305), *KCNA4* (fold change: 2.03, *p* = 0.002), as well as significant up-regulation of two α- and β-subunits of voltage-gated type I sodium channel genes (*SCN1A* and *SCN1B*) in the lateral cerebellar hemisphere. In the Str-NAc, they found a significant down-regulation of *KCNS3* (fold change: -1.67, *p* = 0.00082), *KCNA1* (fold change: -1.59, *p* = 0.00396), and up-regulation of *KCNN3* (fold change: 1.57, *p* = 0.04719). Due to the known ability of mood-stabilizing drugs to directly interfere with ion channel activity, they also looked for differential expression among the 72 genes between subjects on versus off mood stabilizers. Interestingly, the only significant genes were *KCNQ3* (*p* = 0.012) and *SCN1B* (*p* = 0.004) in the cerebellum. In a gene expression study using an animal model, whole-brain mRNA from mice treated with antipsychotics (clozapine, haloperidol, or olanzapine) was examined ([@B25]). The results of the study suggested that treatment with antipsychotics altered regulation of two genes encoding subunits of the K~v~1 potassium voltage-gated channel, shaker related subfamily: *KCNA1* and *KCNAB1*, as well as the K~v~ channel-interacting protein *KCHIP*. The authors of the study concluded that the modulation of neuronal voltage-gated ion channels, particularly potassium and sodium, contribute to the mechanism of action for antipsychotics via their effects on neuronal electrical activity and neurotransmission. In two other gene expression studies with animal models, the effect of electroconvulsive therapy (ECT) on the expression levels of voltage-gated potassium channel subunits in rat brain was examined ([@B61]; [@B40]). ECT is a pro-convulsive treatment in which electrical currents are passed through the brain in order to induce a brief seizure ([@B49]). However, over time, ECT actually has anticonvulsive properties ([@B75]). Its primary indication is for the treatment of major depression, although there is anecdotal evidence for its use in treating mania ([@B87]). The first study found that the expression levels of the potassium channel genes were affected by both chronic and acute ECT, and that these effects were specific to channel type, brain region, and timing ([@B61]). The second study also found that chronic ECT significantly (*p* \< 0.001) increased expression levels, again in a channel subtype and brain region specific manner ([@B40]). Since potassium channel genes are integral to the regulation of membrane potentials, the authors concluded that the increased levels of mRNA might be at least partially relevant to the seizure threshold elevating effect of ECT. ION CHANNELS AS THERAPEUTIC TARGETS =================================== Ion channels provide an attractive target for pharmacological intervention, given their diverse and ubiquitous roles in a broad range of physiological processes. Among all drugs with known targets, approximately 13.4% have their primary therapeutic action at ion channels. This ranks ion channels second as a target class, behind G protein-coupled receptors (GPCRs; [@B59]). Accumulating evidence suggests that existing treatments for BP may potentially exert their therapeutic action via ion channel regulation ([@B36]). Lithium, a simple monovalent cation, is a first-line treatment for BP ([@B67]). Although its mechanisms of action are not entirely clear ([@B57]), there is some evidence that it affects ion channel functioning. A recent investigation of *in vitro* lithium treatment in mouse brain tissue determined that, at therapeutic concentrations, lithium entered the cell through sodium channels and suppressed the outward membrane current, which directly affected membrane excitability ([@B14]). The authors concluded that ion channels that regulate neuronal excitability may be a common target among BP treatments. Lithium is also a direct and potent inhibitor of glycogen synthase kinase 3β (GSK3β; [@B41]; [@B67]). GSK3β has many functions, one of which is to phosphorylate the voltage-gated potassium channel KCNQ2. It has been shown *in vitro* that phosphorylation of KCNQ2 decreases the activity of the channel and subsequently reduces the M-channel current ([@B12]). It is intriguing to speculate that lithium may have therapeutic effect by blocking this downstream phosphorylation of KCNQ2. Antiepileptic drugs (AEDs) such as valproate, lamotrigine, and carbamazepine have a long history of successfully treating BP, and may similarly affect the regulation of neurotransmission and action potential firing via mediation of ion channel functioning. The AEDs that are effective against BP likely have multiple molecular targets with variable contributions to the drugs' efficacy ([@B62]). However, it is notable that they are all known to diminish the flow and accumulation of sodium ions in the cell ([@B26]). Voltage-gated potassium channels, specifically the Kv7 channels, are particularly attractive targets for novel therapeutics for BP. In addition to their functionally relevant roles within neurons, they offer the ability for selective intervention ([@B13]). For example, ezogabine (EZG) is a neuronal potassium channel enhancer that was recently approved by the FDA (Food and Drug Administration) to treat partial epilepsy ([@B47]). It has been shown ([@B91]) that EZG activates the KCNQ2/KCNQ3 hetero-tetramer ion channel complex, which in turn promotes the M-current and thereby stabilizes action potential firing. This action likely accounts for the drug's anticonvulsant properties ([@B12]). Several groups have studied the potential mood-stabilizing effects of EZG and other Kv7 channel enhancers in a series of animal studies. One study ([@B68]) used a mouse model of mania and determined that EZG (a non-selective opener) and ICA-27243 (a selective Kv7.2/3 opener), but not BMS-204352 (a selective Kv7.4--Kv7.5 and Kv7.7/3 heteromer opener) were able to reduce motility in the hyperactive mice, a paradigm for antimanic properties in drugs. In another study employing the [D]{.smallcaps}-amphetamine and chlordiazepoxide (AMPH + CDP) mouse model ([@B43]), EZG reduced cerebral glucose metabolic activity while increasing phosphor-serine-9 levels of GSK3β with a brain regional signature that mirrored lithium and valproate. Ezogabine has also been evaluated for the acute treatment for mania in a small (*N* = 10) open label pilot study of treatment resistant in patients with BP type I ([@B2]). Despite the limited sample, the brevity of the study design, and the severity of illness, improvement in mania scores was observed in four patients. The treatment was well tolerated with no depressogenic effects, indicating that further study may be warranted. NEURONAL CHANNELOPATHIES ======================== The accumulating evidence from molecular genetic studies provides a plausible case that BP may be due, at least in part, to a channelopathy. The term "channelopathy" is a relative newcomer to the medical lexicon, referring to disorders that result from defective ion channel functioning ([@B3]). Ion channels are complex, membrane-spanning glycoproteins composed of separate pore-forming and accessory subunits ([@B4]). These pores allow the selective, passive transport of charged ions through the hydrophobic cell membrane, which is otherwise relatively impermeable to ions. While the actual movement of the ions themselves is largely inconsequential, it is the resulting transmembrane current that elicits the electrical excitability that defines their physiological contributions ([@B32]). These currents allow ion channels to orchestrate electrical signals such as action potentials in neurons ([@B26]). Action potentials, which result from transient changes in the membrane's permeability to sodium and potassium ions, are electrical events that encode and transmit information, facilitating neuronal communication via neurotransmitters ([@B84]). Sodium (Na^+^) channels are responsible for the initial depolarization (or rising) phase of the action potential, and potassium (K^+^) channels are responsible for the repolarizing phase, which returns the membrane to its resting potential ([@B10]). Ion channel genes are quite prominent, representing 1--2% of the entire genome ([@B19]; [@B83]), and most display tissue-specific expression. Ion channels are extraordinarily diverse, suggesting a high degree of functional specificity ([@B72]). Their diversity is often amplified by the various combinations of heteromers between family members that co-assemble to form functional units of the channels. Additionally, variations in their different constituent subunits allow for modification of their functioning, through alternative splicing ([@B32]). Therefore, while ion channels play a role in a broad range of physiological processes in various tissues, genetic variation in the genes that code for them often result in tissue-specific diseases ([@B15]). Potassium channels in particular figure prominently in human channelopathies ([@B73]). Their structural and functional diversity, which is greater than any other group of ion channels ([@B42]), underscores the variety of physiological function that they modulate ([@B44]). In particular, variants within the KCNQ gene family (*KCNQ1--5*, also referred to as Kv7.1--5) have figured prominently in human channelopathies ([@B33]). Of the ten potassium channel genes that have proven associations with human disease, four are members of the KCNQ subfamily ([@B12]). Additionally, four of the five KCNQ gene family members have been associated with different hereditary diseases. The diversity, ubiquity, and functional significance of the KCNQ gene family imply that they are likely to play a role in the etiology of more diseases than have already been identified ([@B76]; [@B12]). As suggested above, two members of the KCNQ gene family (*KCNQ2* and *KCNQ3*) may be particularly relevant to BP etiology. *KCNQ2* and *KCNQ3* are both expressed in the brain and form hetero and homomeric voltage-gated potassium channels. ANK3 helps to direct the localization of the KCNQ2/KCNQ3 channels to the axonal initial segment (AIS) of neurons and nodes of Ranvier. Indeed, such localization is abolished in *ANK3* knock-out mice ([@B60]). The crucial role of KCNQ2 and KCNQ3 at the AIS is to form M-type channels that mediate sub-threshold M-currents. These M-currents stabilize the neuronal resting potential and inhibit repetitive firing of action potentials and thereby prevent neuronal hyperactivity ([@B24]). Thus, disruption of normal KCNQ2/KCNQ3 activity may be salient to the etiology of BP. Supporting this hypothesis are animal model studies which have shown that the suppression of the M-current by dominant-negative mutations in *KCNQ2* lead to hyperexcitability of neurons, morphological changes of the hippocampus, a notable decline in cognitive function, and behavioral changes corresponding to hyperactivity ([@B63]). Interestingly, mutations for benign familial convulsions, a rare autosomal dominant inherited form of epilepsy, have been identified in *KCNQ2* ([@B79]). BIPOLAR DISORDER AND EPILEPSY ============================= Bipolar disorder has many overlapping features with epilepsy ([@B1]; [@B5]), which has been referred to as the "signature channelopathy phenotype in the CNS" ([@B33]). Epilepsy is a brain disorder involving repeated and unprovoked seizures, which are periods of disturbed brain function caused by abnormally excited electrical signals ([@B88]). Epilepsy has a number of subtypes that are proven channelopathies ([@B39]), and as a result it can be used as a model to illustrate other central nervous system (CNS) conditions that may arise from disruptions of ion channel functioning ([@B32]). In particular, there are many intriguing parallels between BP and epilepsy that suggest common neurobiological basis, specifically implicating ion channel functioning ([@B50]). One of the most striking similarities between BP and epilepsy involves the course of illness for both disorders, which involves a predominantly normal phenotype punctuated by sporadic interruption of rhythmic functioning ([@B33]). The episodic course of both disorders often progresses to become chronic, and sometimes treatment resistant ([@B54]). The progression of these conditions known as "kindling," whereby repetitive, sub-threshold stimuli induce episodes until they begin to occur spontaneously may be a feature of both epilepsy and BP ([@B1]). This kindling phenomenon is thought to leave lasting physiological changes in the brain that predispose it to be more sensitive to further stressors. Further, AEDs that are used to treat both epilepsy and BP such as carbamazepine ([@B64]), valproate ([@B48]), and lamotrigine ([@B58]) have been shown to exhibit anti-kindling properties. Mood disorders are frequently comorbid with epilepsy ([@B55]). Earlier findings suggested that BP was rarely diagnosed in epilepsy patients. However, a large population-based survey in the United States found evidence of BP symptoms in 12.2% of epilepsy patients, and nearly half of these (6%) received a research-diagnosis of BP from a physician. This rate was higher than comorbidities of other medical disorders within the epileptic population ([@B28]). This comorbidity is puzzling given that epilepsy patients are treated with AEDs, which are potent mood stabilizers ([@B55]). Additionally, an affective syndrome, known as interictal dysphoric disorder (IDD), occurs in patients with epilepsy anywhere from 0.1 to 4.3% ([@B69]). In a study of 143 outpatients with epilepsy ([@B55]), 11.8% received a DSM-IV (Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition) diagnosis of BP, but only 1.4% were considered "pure" BP, unrelated to IDD, seizures, or AED treatment. Although IDD symptoms are thought to be physiologically distinct from the neurological symptoms of epilepsy ([@B34]), it is unclear how the etiology of these bipolar-like symptoms relates to BP itself. Regardless, it may be reasonable to speculate that the remarkable similarities in symptoms indicate some common mechanisms in the pathophysiology, potentially through ion channelopathies. Further epidemiological support for etiological overlap between these disorders comes from a population-based family study of the comorbidity and familiarity between epilepsy and psychosis ([@B20]). Among patients with epilepsy, the authors found an increased risk for broadly defined psychosis (5.5-fold increase in risk), schizophrenia (8.5-fold increase), and bipolar disorder (6.3-fold increase). Additionally, having a parent with epilepsy conferred a twofold increase risk for psychosis (compared to individuals without parental history of epilepsy), and conversely, a parental history of psychosis increased risk for epilepsy by 2.7-fold. One final intriguing parallel between BP and epilepsy is the overlap in their treatments. Remarkably, many AEDs, including carbamazepine, lamotrigine, and valproate are effective in the treatment of BP. In fact, valproate is considered a first-line treatment for acute mania as well as prophylaxis. And carbamazepine and lamotrigine are other leading alternative mood stabilizers, should first-line monotherapy with lithium or valproate fail. Lamotrigine, in particular, has been shown to be relatively more effective against bipolar depression ([@B74]). Although the mechanisms of action for these drugs in treating BP may not necessarily be the same as those in treating epilepsy ([@B70]), the similarities between the disorders coupled with overlapping treatment regimes strongly suggest for at least some common pathophysiology. DISCUSSION ========== Over the past few decades, countless effort and resources have been invested into researching the genetic contribution to bipolar disorder. While no genes have been conclusively shown to cause the disorder, each discovery has revealed a small part of the fascinatingly complex genetic architecture of BP. Evidence implicating various genes related to the structure and regulation of ion channels had peppered earlier linkage and association literature. This class of genes has garnered renewed support in the GWAS era, as both single SNP and pathways based analyses continue to implicate ion channel functioning. The course of illness of BP, specifically its episodic nature, fits well with the channelopathy paradigm. Thus, these ion channel genes, which profoundly impact the neuronal activity, are an interesting group as there is both statistical evidence as well as biological plausibility implicating their involvement. Much of the focus has been on the potential role of calcium ion channels in BP. Here, we review evidence that potassium ion channels may also contribute to the etiology and deserve further investigation. The identification of genes that contribute to BP susceptibility is important for a number of reasons. Elucidating the genetic underpinnings of BP may reveal new treatment targets, facilitating rational drug design that could produce more effective therapeutics with fewer side effects ([@B86]). It could also facilitate the development of diagnostic testing via genetic screening. This may have clinical implications, particularly in the case of disorders with genetic heterogeneity, whereby the course of treatment may depend on the precise etiology of the disorder ([@B3]). Conflict of Interest Statement ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. [^1]: Edited by: *Kathleen D. Askland, Butler Hospital/The Warren Alpert School of Medicine, Brown University, USA* [^2]: Reviewed by: *Radka Kaneva, Sofia Medical University, Bulgaria; Andrew McQuillin, University College London, UK* [^3]: This article was submitted to Frontiers in Behavioral and Psychiatric Genetics a specialty of Frontiers in Genetics.

Introduction ============ Typical methods for monitoring cultivation processes are offline analyses like cell counting, measurement of various substrates and products (e. g. glucose or lactate) as well as the online monitoring of several physical process parameters (temperature, pH-value or the concentration of dissolved oxygen). To improve cell cultivations detailed information about important analytes should be available online. Therefore new monitoring methods need to be established, preferably as *in-situ* methods to minimize the risk of contamination. Two different *in-situ* online-methods were used to monitor cultivations: *In-situ* microscopy and 2D fluorescence spectroscopy. Therefore CHO-K1 cells (provided by University of Bielefeld) were cultivated in a complex culture medium (TC 42, TeutoCell, Bielefeld, Germany) using a 2.5 L stainless steel reactor with a work volume of 2 L. A total of three cultivation runs were conducted. *In-situ* microscopy ==================== The *in-situ* microscope (ISM), developed at our institute, offers the possibility to determine the cell density and to obtain further information about certain cell characteristics such as size, compactness and excentricity. In addition, it is possible to extract other information like cell clusters or microbial contaminations from the recorded images. In this way a general comprehensive overview of a cultivation can be obtained. The *in-situ* microscope was immersed directly in the culture liquid using a 25 mm standard connection. During the cultivation process several images were obtained and subsequently evaluated by a self-developed algorithm which detects cells on the basis of grey scale values. It was necessary to correlate the cells per picture with offline data determined by a Neubauer counting chamber. This correlation gives a regression coefficient (R^2^) of 0.988. By means of this correlation the cells per picture were calibrated in reference to the counted cells per mL (root mean square error of calibration; RMSEC: 0.989). As a result it is possible to detect the cell density via ISM with an average standard error of 0.027 (pictures per cycle: 300) during further cultivations. 2D fluorescence spectroscopy ============================ The 2D fluorescence spectra were collected using a Bioview® sensor (Delta, Denmark). The fluorescence spectroscopy was used to monitor glucose, lactate und glutamate during cultivation. At first it was necessary to build a calibration model for each compound. Therefore two process runs were observed by collecting 2D fluorescence spectra every 15 min. Approximately every 6 hours a sample was taken and analyzed offline for glucose, lactate and glutamate. The offline reference data and the corresponding spectra were used to build a chemometric model for prediction of the most interesting variables during a further cultivation run. Utilizing multivariate analyzing software (Unscrambler X, vers. 10.1) a calibration model was built by partial least square regression (PLS1) for every variable. The number of factors necessary to built a model as well as the results for the regression coefficient (R^2^), the root mean square error of calibration and validation (RMSEC and RMSEV) for every variable are shown in Tab. [1](#T1){ref-type="table"}. ###### PLS-model results for glucose, lactate and glutamate. Compound factors **R**^**2**^ RMSEC \[g/L\] RMSEV \[g/L\] RMSEP \[g/L\] --------------- --------- -------------- --------------- --------------- --------------- **Glucose** 3 0.967 0.381 0.468 0.524 **Lactate** 3 0.972 0.368 0.461 0.494 **Glutamate** 4 0.983 0.0036 0.0059 0.0155 The PLS-models were used to predict all three compounds in a third cultivation run. The standard errors of prediction (RMSEP) are also shown in Tab. 1. For glutamate the RMSEP is about 5 % compared to a maximum glutamate concentration of 0.27 g/L. By applying the PLS-models to the fluorescence data the concentration gradients for every single compound can be predicted (Fig. [1](#F1){ref-type="fig"}). {#F1} Conclusions =========== In biotechnology, especially in pharmaceutical biotechnology it is important to closely monitor a cultivation process. The *in-situ* microscopy together with digital image processing and the 2D fluorescence spectroscopy in combination with chemometric tools complement one another as online monitoring methods. It was demonstrated that the ISM can be used to monitor the cell density during a bioprocess with the same accuracy in comparison to a Neubauer counting chamber. After a successful calibration the cell density can be determined without the need for taking samples. This can reduce the risk of contamination, particularly when the cells are cultivated without the addition of antibiotics. The Bioview® sensor could be utilized to observe three important analytes during a cultivation run with an error of prediction under 10 %. By monitoring these analytes in real time it is possible to observe sudden or unexpected changes during a cultivation and if necessary to react accordingly. Both online methods were successfully applied to monitor the cell density as well as the glucose, lactate and glutamate concentration during cultivation. Outlook ======= Concerning the ISM further research is focused on a new prototype with precise linear stages making it possible to adjust to a certain sample volume. This should make a calibration with offline data unnecessary. The robustness of the calibration models for the 2D fluorescence spectroscopy will be tested by adding glucose during cultivation in order to determine if the model can be applied to a fed batch cultivation. Acknowledgement =============== This project was funded by the Federal Ministry of Education and Research within the project "SpektroSens" (spectroscopic sensor systems for the area of biotechnology and food).

Single-molecule observation techniques using visible-wavelength fluorescent probes are valuable for answering many basic and important questions in biology and biophysics because they provide the positional information of single molecules with accuracy far beyond the half-wavelength optical diffraction limit. They can provide the center positional information of labeled molecules with an accuracy of approximately λ/100[@b1][@b2]. Using similar tracking techniques with X-ray (λ ≈ 0.1 nm) probes, we have previously observed the picometer-scale (λ/100) Brownian motions of individual protein molecules linked to gold nanocrystals[@b3]. This system for single-molecule observation is called diffracted X-ray tracking (DXT)[@b4]. In this study, instead of using X-rays, we used electron probes to improve this technology to develop a laboratory-scale single-molecule detection system. Wet scanning electron microscopy (wet SEM) has recently been used to probe the cytoplasm of whole cells using a thin membranous partition that protects the sample from high-vacuum conditions. This method allows images of living cells to be captured with little loss of resolution compared to standard SEM. Wet SEM can also detect the internal motion of protein molecule domains in aqueous solutions on the sub-Angstrom scale at high rates of image acquisition[@b5][@b6]. Because gold nanoparticles are highly electron dense, they are ideal for both transmission electron microscopy (TEM) and SEM. Colloidal gold has been recognized as an excellent marker for use in cell biology and materials science studies[@b7]. DXT allows direct picometer-scale observation of individual biomolecules in real time and in real space[@b8][@b9][@b10]. [Figure 1(a)](#f1){ref-type="fig"} shows the arrangement of our new dynamic single-molecule detection system. Our system uses wet SEM and time-resolved electron backscatter diffraction (EBSD) from labeled gold nanoparticles. We refer to our new imaging technique as diffracted electron tracking (DET). A cross-sectional view of the wet cell used for DET is shown in [Figure 1(b)](#f1){ref-type="fig"}. We can measure the dynamics of both the top-layer of the adsorbed polymer reagent and the labeled gold nanoparticle. As a preliminary experiment, gold nanocrystals were directly adsorbed to the carbon surface for measurement by DET. As a result, we did not observe changes in the gold nanocrystal DET data over time. Rather, in this case we measured the movement of the coating polymer, which dictated the nanocrystal movement. [Figure S1](#s1){ref-type="supplementary-material"} shows an enlarged image of a sample cross-section. Because gold nanocrystals are slightly larger (diameter = 40 nm), their movements do not represent individual single molecule movements, but rather, indicate the average movements of the nanoscale domains of the coating polymer. Because DXT uses X-rays, it requires the use of a perfect crystal. Therefore, for DXT, it is important to use gold nanocrystals instead of gold nanoparticles (colloidal gold). However, we can use commercially available gold nanoparticles for DET because the electron probe is sensitive to their surface structure. We confirmed from the electron diffraction data that the surface crystallinity of the gold nanoparticles was very good. As shown in [Figure 1(c)](#f1){ref-type="fig"}, EBSD patterns for the gold nanoparticles can be obtained with an integration time of only 60 ms. DET data collection is a simple process. First, we check the positions of the adsorbed gold nanoparticle using SEM, as shown [Figure 2](#f2){ref-type="fig"}. Next, we collect time-resolved EBSD measurements. The direction of each EBSD measurement is determined in three dimensions for each time point. The orientation displacements at each time point can be obtained by comparing differences in the EBSD from the first crystallographic orientation pattern. Results ======= The 3D (or 3-axis) motions of individual adsorbed gold nanoparticles within a vacuum are shown in [Figure 3(a)](#f3){ref-type="fig"}. The rotational motions around the β-axis were slightly smaller than the rotational motions around the α- and γ-axes. In the vacuum, the rotational motion velocities around the α- and γ-axes were approximately 0.05 rad/s. The 3D motion data for aqueous solutions are shown in [Figure 3(b)](#f3){ref-type="fig"}. The Brownian motion of the adsorbed gold nanoparticles in water was much more energetic than under vacuum. The rotational motions around each of the 3 axes were on the same order. In water, the rotational motion velocities around the α- and β-axes were approximately 0.55 rad/s. To understand the motion under vacuum and in solution, we analyzed and characterized the random displacements of individual single gold nanoparticles. The distribution of the displacement measurements observed using a 600 ms (ten 60-ms frames) time interval is shown in [Figure 4(a)](#f4){ref-type="fig"}. As shown in [Figure 4(b)](#f4){ref-type="fig"}, the log-normal distributions can be characterized by the average (μ) and standard deviation (σ) of the symmetrical normal Gaussian probability obtained when the displacements are plotted on logarithmic scales[@b11]. We confirmed that the log-normal distributions were similar at different time intervals (from 2 to 20 frames). We determined the μ and σ values from our distributions by fitting the statistical parameters with a Gaussian function as shown in [Figure 4(b)](#f4){ref-type="fig"}. The minimum value of μ under vacuum was assigned to the β-axis. This means that under vacuum conditions, the overall motions of the gold nanoparticles were rotational, thus defining the central axis as the LD axis ([Fig. 1(a)](#f1){ref-type="fig"}). The nanoparticle motions in aqueous solutions were 5--7 times larger than those in vacuum. We can hypothesize that this enhanced motion was due to collisions between the water molecules and the gold nanoparticles. The minimum value of μ under aqueous conditions was assigned to the γ-axis. This means that under aqueous conditions, the overall gold nanoparticle motions were rotational, thus defining the central axis as the ND axis ([Fig. 1(a)](#f1){ref-type="fig"}). Because we observed rotation of the β-axis around the LD axis under vacuum, the dynamical data from the observed EBSD patterns were counted as small β motions. The movement correlations between the other two axes (α and γ) produce two-dimensional histograms ([Fig. 5](#f5){ref-type="fig"}). In [Figure 5(a)](#f5){ref-type="fig"}, we can see from the three correlation coefficients that under vacuum conditions the β-axis is independent and has lower mobility. The movement characteristics in aqueous solutions can be inferred from the two-dimensional histograms shown in [Figure 5(b)](#f5){ref-type="fig"}. Although it is not clear for aqueous conditions, the three correlation coefficients imply that the independent lower mobility is present in the rotational movement around the γ-axis. As shown in [Figure 5(b)](#f5){ref-type="fig"}, it is clear that both γ-direction coefficients are relatively small. Discussion ========== To characterize the Brownian motion, we used plots of the mean-square displacement (MSD) against the time interval. [Figures 6(a) and 6(b)](#f6){ref-type="fig"} show the MSD curves for the observed gold nanoparticles as a function of the time interval Δ*t* under vacuum and in solution. For simple Brownian diffusion, the MSD plots are linear with a slope of 4*D*, where *D* is the diffusion coefficient. These parabolic MSD plots, however, are indicative of directed diffusion[@b2]. These data demonstrate that gold nanoparticles move at a constant drift velocity *v* with a diffusion coefficient *D*[@b12]. However, we are only concerned with the size of *D*. Under vacuum the smallest rotational motions occur around the β-axis ([Fig. 6(a)](#f6){ref-type="fig"}). Under aqueous conditions the smallest rotational motions occur around the γ axis ([Fig. 6(b)](#f6){ref-type="fig"}). We also considered why the dynamical movements differed between the vacuum and aqueous conditions. [Figure 1(b)](#f1){ref-type="fig"} shows the arrangement of the electron beam position and an adsorbed gold nanoparticle. Under vacuum, slight anisotropic movements of the gold nanoparticles were induced by the difference between the position of the primary electron beam and the center of the gold nanoparticle. These anisotropic movements are obscured by motions that occur due to collisions between water molecules and gold nanoparticles. These collisions also explain why the adsorbed gold nanoparticle motions in aqueous solutions are 5--7 times larger than those under vacuum. In addition, it is clear that the stiffness of the adsorbed polymer layer is completely different depending on whether it is under vacuum or in an aqueous solution. As a result, 3D isotropic movements of the adsorbed nanoparticles were observed in aqueous solutions. The rotational movement observed by DET can be converted to translational motion by making certain assumptions, as shown in [Figure 7](#f7){ref-type="fig"}. [Figure 1(b)](#f1){ref-type="fig"} shows the cross-sectional view of the DET sample. The rotational gold nanoparticle motions are directly reflected in the EBSD data. If the adsorbed polymer layer on the thin carbon film (i.e., the layer from the silane coupling agent) is rigid, it can transfer the local translational movements from the upper surface of the polymer film, as shown by the yellow arrow in [Figure 1(b)](#f1){ref-type="fig"}. The transferred motion widths are shown in [Figures 5(a) and 5(b)](#f5){ref-type="fig"}. For example, the range of motion is between 50 picometers per frame and 5 nanometers per frame. Because the detection limit of DET is approximately 1.0 mrad, we estimate that the smallest detectable movements of the polymer and gold nanoparticles were between 10 and 50 picometers. DET could detect the anisotropic movement around the LD axis under vacuum conditions. The difference in the movement direction was confirmed from the anisotropic movement distribution in the 0.1- to 5.0-Å range, as shown in [Figure 5(a)](#f5){ref-type="fig"}. DET allowed us to measure for the first time the delicate movement of an adsorbed nanoparticle excited by an electron probe. This movement was not from sample damage by the electron beam. The total thickness of the silane coupling agent between the adsorbed gold nanoparticle and the surface of the carbon film was approximately 100 Å, therefore, the slight displacements in the 0.1- to 5.0-Å range cannot be attributed to damage. We were able to collect time-resolved, 3D, super-accurate (picometer scale) motion tracking data on single nanoparticles under vacuum and in solution using wet SEM and EBSD patterns. We also measured slightly excited anisotropic motions from the interactions between individual gold nanoparticles (diameter = 40 nm) and the high-energy electron probe (30 keV). This unusual motion was so small that it was obscured by Brownian motion. Because the electron beam used for DET has a short wavelength, the 3D orientation of the labeled nanoparticles can be determined with super-high sensitivity, as shown in [Figure 8](#f8){ref-type="fig"}. When a gold nanoparticle is linked to a single protein molecule (as in DXT), a time-resolved, highly accurate, single-molecule detection system using an electron beam can be produced using normal laboratory-sized equipment (SEM equipment), without the need for a very strong X-ray source such as the SPring-8, ESRF, or APS beamlines. Methods ======= Sample preparations for wet -SEM (DET) -------------------------------------- An environmental cell (EC) for SEM was designed to observe samples in aqueous environments. The EC was assembled from a special three-slit grid (3 mm^2^, 0.2 mm thick) and an O-ring (inner diameter 2.0 mm, thickness 0.45 mm) at the top of the EC holder. The grid was designed to be used for DET at a 70 degree angle, as shown in [Figure 1(b)](#f1){ref-type="fig"}. The slits (0.1 × 0.3 mm) of the grid were covered with a thin carbon sealing film (15--20 nm thick) to prevent liquid or gas leakage from the EC. The surface of the carbon sealing film on the grid was coated with mercaptosilane reagent (KBM-803, Shinetsu silicone, Japan) using a vapor deposition method ([Fig. 1(b)](#f1){ref-type="fig"}). A solution containing 40 nm colloidal gold (Polyscience, USA) was incubated on the carbon sealing film for 1 hour before washing with distilled water. For aqueous observations, the grid was made hydrophilic through a glow discharge treatment (HDT-400, JEOL, Japan). A 200-nl volume of degassed distilled water was then placed on the carbon sealing film supporting the colloidal gold. Excess water was removed by wicking with filter paper. Finally, approximately 50 to 100 nanoliters (nl) of degassed water was enclosed in the EC. Measurement conditions for DET ------------------------------ For SEM studies, we used a JSM-7001F system (JEOL, Japan) equipped with a Schottky-type field emission electron gun. To detect the EBSP signal, we used a 30 keV accelerating voltage, an 87 pA beam current, and 17 mm working distance at a 70-degree angle relative to the EC sample stage. The EBSD detector on the JSM-7001F system was custom designed by TSL Solutions (DVC1412-FW-T1-EX, TSL, Japan). The detector included an image intensifier (V8070U-74, Hamamatsu Photonics, Japan), as shown in [Figure S2](#s1){ref-type="supplementary-material"}, providing ten times higher intensity than a normal detector. We could not obtain clear EBSD data without the image intensifier. [Figure 2](#f2){ref-type="fig"} shows tracking of the labeled gold nanoparticles on polymer substrates under vacuum and in solution. To measure the crystal orientation of the gold nanoparticles, EBSD patterns were obtained from 1.5 sec of electron beam irradiation at a fixed observation point. The EBSD pattern was obtained on the phosphor screen of the detector, intensified, and recorded with a CCD camera using a 60 ms shutter speed. The Euler angles were obtained from the EBSD patterns using OIM-data collection software (EDAX, USA). The Euler angles indicate the rotational relationship between the sample stage coordinate system and the crystal lattice coordinate system of the colloidal gold. The α, β, and γ rotational angles of the nanoparticle principal lattice vectors were calculated between adjacent frames from the Euler angles, as shown in [Figure 1(a)](#f1){ref-type="fig"}. Author Contributions ==================== Y.C.S. wrote the main manuscript text. K.H., H.S., K.I. and Y.M. analyzed DET data and prepared figures 1--6. N.O., Y.H., S.S., A.I. did main DET experiments. All authors reviewed the manuscript. Supplementary Material {#s1} ====================== ###### Supplementary Information supplementary information ###### Supplementary Information movie1 ###### Supplementary Information movie2 {#f1} {ref-type="fig"}), or a statistical treatment is carried out as shown in [Fig. 4](#f4){ref-type="fig"}, [5](#f5){ref-type="fig"}, and [6](#f6){ref-type="fig"}).](srep02201-f2){#f2} {#f3} {#f4} {#f5} {#f6} {#f7} {ref-type="fig"} is equal to the projections to the plane of this 3D histogram.](srep02201-f8){#f8}

Infectious agents have a long history of causing disease in humans and are a major contributor of childhood mortality worldwide (Table [1](#T1){ref-type="table"}). For many infectious agents, a complex interaction between the host and pathogen has developed over time as a result of an evolutionary co-existence and adaptation that is now reflected in both human and pathogen genomes. Initial studies have looked at these human-pathogen interactions on a broad scale, asking what components of the (typically human) genome are associated with disease susceptibility \[[@B1]\]. However, infectious diseases, unlike many other diseases, offer the chance to dissect this more precisely as the non-host or \'environmental\' component (the pathogen) has a genome that can be assayed just as accurately as the human genome. Today, we have the appropriate tools and platforms in place to begin examining these adaptations from a whole-genome perspective, using samples directly from clinical studies. By directly observing this human-pathogen interaction *in natura*, we can avoid the requirement for artificial models, which tend to be tedious to construct and may not reconstitute human physiology accurately \[[@B2]\]. This generates a powerful approach, as it integrates two complementary (but previously thought to be distinct) areas of infectious disease research - host genetic susceptibility to infection outcomes and pathogen genetic variation leading to differential virulence and individual susceptibility. ###### Global causes of childhood deaths Cause \% ----------------------------- ---- Pneumonia 18 Malaria 8 Sepsis 6 AIDS 2 Pertussis 2 Meningitis 2 Measles 1 Tetanus 1 Other infections 9 Preterm birth complications 12 Birth asphyxia 9 Congenital abnormalities 3 Diarrhea 15 Injury 3 Others 9 Infections contribute to 49% of childhood deaths globally. Causes that led to less than 1% of deaths are not presented. Adapted from \[[@B61]\]. On the basis of data obtained using biological candidate approaches, it has been assumed that broad-based responses mediated by B and T cells (both CD4^+^and CD8^+^) were the primary defenses against infectious agents and that these were sufficient for protection against most infecting pathogens \[[@B3]\]. Defects in Interleukin 2 (IL-2) signaling resulted in failed clonal expansion of both B and T cells, leading to severe combined immune deficiency \[[@B3]\]. Other examples include X-linked agammaglobulinemia - caused by mutations in Bruton\'s tyrosine kinase - where no mature antibody-producing B cells are produced, resulting in no antibody production \[[@B4]\]. However, in the majority of common infectious diseases studied so far from a genomic and genome-wide association perspective, the evidence has tended to point away from a picture of high-penetrance, generalized susceptibility caused by critical defects in single genes, and instead towards a more complex picture of multiple lower penetrance effects implicating individual genetic components for each invading pathogen. There are some notable exceptions to this, such as the shared human leukocyte antigen (HLA) associations (suggesting T-cell involvement) in leprosy susceptibility \[[@B5]\] and HIV viral set point \[[@B6]\]. As HLA is crucial for recognition of processed pathogen molecules and the initiation of the CD4^+^and CD8^+^T-cell response, it was unsurprising that some degree of shared susceptibility between pathogens is observed. Recent evidence has suggested an important role for innate immune interactions in disease susceptibility \[[@B5],[@B7]\] and specific evolution and adaptation of both host and pathogen genomes to the current state of mutual co-existence. Genome-wide association studies (GWASs) necessarily assay common genetic variation of mostly low penetrance, and this approach is now being used to study host genetic susceptibility to infectious diseases (Table [2](#T2){ref-type="table"}). Some of these recent GWAS data on infectious diseases have indeed pointed clearly to surprisingly specific pathogen-receptor interactions rather than broad-based susceptibilities. Similar advancements have been made from the pathogen perspective, in which GWASs on the pathogen genome have also yielded unexpected insights. The example in *Plasmodium falciparum*malaria showcases the adaptability of the pathogen genome when confronted by the human immune response and anti-malarial pharmacotherapy \[[@B8]-[@B10]\]. ###### Host susceptibility genes identified by GWAS studies Pathogen Phenotype Gene N cases/controls Effect size *P*-value References ---------------------- ----------------------- -------------------- ------------------ ------------- ---------------- ------------------- Dengue virus Severe dengue *MICB* 3, 742/4, 952 1.34 4.41 × 10^-11^ \[[@B20]\] Severe dengue *PLCE1* 3, 742/4, 952 0.80 3.08 × 10^-10^ \[[@B20]\] *N. meningitidis*^a^ Meningococcal disease *CFH* 1, 443/6, 079 0.64 2.2 × 10^-11^ \[[@B7]\] *P. falciparum*^a^ Severe malaria *ABO* 3, 906/5, 489 1.20 2.0 × 10^-7^ \[[@B49],[@B62]\] Severe malaria *HBB* 2, 045/3, 758 0.63 3.7 × 10^-11^ \[[@B62]\] *M. leprae* Leprosy *NOD2* 3, 960/7, 180 1.59 3.77 × 10^-40^ \[[@B5]\] *RIPK2* 3, 960/7, 180 0.76 1.38 × 10^-16^ \[[@B5]\] *HLA-DR* 3, 960/7, 180 0.67 5.35 × 10^-27^ \[[@B5]\] *TNFSF15* 3, 960/7, 180 1.37 3.39 × 10^-21^ \[[@B5]\] *C13orf31-CCDC122* 3, 960/7, 180 1.68 3.72 × 10^-54^ \[[@B5]\] *M. tuberculosis* Tuberculosis *-* 3, 632/7, 501 1.19 6.8 × 10^-9^ \[[@B63]\] HIV Viral set point^a^ *HLA-B* 486 9.6% 9.36 × 10^-12^ \[[@B6]\] Viral set point^a^ *HLA-C* 486 6.5% 3.77 × 10^-9^ \[[@B6]\] HBV Chronic infection *HLA-DP* 2, 086/4, 301 1.75 6.34 × 10^-39^ \[[@B64],[@B65]\] ^a^Analyzed as a quantitative trait. In contrast to highly penetrant Mendelian genetic defects causing broad-based susceptibilities with severe clinical outcomes, more common host genetic determinants of susceptibility of lower penetrance and modest effect size could be confined to specific pathogen species \[[@B11]\] or even serotypes (and possibly genotypes). A better understanding of host-pathogen interactions will allow more thorough dissection of the host immune response generated by specific pathogen invasion. Because pathogens adapt to survival pressures by evading drugs, vaccines, and the host immune response and have a higher intrinsic mutation rate than humans, this improved understanding will potentially have an impact on vaccine design and novel therapy discovery. In this review, we focus on three examples of this - dengue, Malaria, and meningococcal disease - to reflect the literature currently available. Host-pathogen interactions with dengue virus ============================================ Dengue fever is caused by infection from one of four serotypes of the dengue (DEN) virus (DEN-1 to DEN-4), which belong to the Flaviviridae family (other members of which include the yellow fever and Hepatitis C viruses) \[[@B12],[@B13]\]. It is an acute systemic viral infection with a wide spectrum of disease manifestations, ranging from subclinical infection to severe and fatal disease. The commonest severe complications are a transient increase in vascular permeability and altered hemostasis, which could lead to life-threatening hypovolemic circulatory shock (called dengue shock syndrome, DSS). It is the most common mosquito-borne infection after malaria, and its recent resurgence is largely attributed to a combination of factors, including exponential population growth, rapid urbanization, air travel and lack of proper vector control \[[@B14]\]. Previous studies have provided clues regarding the importance of pathogen genetic variation and severity of the infection. Amino acid variants determined from full-genome sequencing of the dengue virus revealed different potentials for causing severe dengue as opposed to uncomplicated dengue \[[@B15]\]. Current evidence also suggests that different serotypes of dengue virus are associated with more or less severe disease, following infections with DEN-2 \[[@B16]-[@B19]\]. From the human perspective, previous studies have pointed to several branches of the immune system in the pathogenesis of dengue \[[@B15]-[@B19]\], including the HLA system and dendritic cells, although there are few convincing genetic validation studies. However, a recent large-scale GWAS and replication study showed very strong statistical evidence of association between genetic variants at two distinct loci (*MICB*and *PLCE1*) and increased susceptibility to DSS (Table [2](#T2){ref-type="table"}) \[[@B20]\]. *MICB*encodes MHC class I polypeptide-related sequence B, which is an inducible activating ligand for the NKG2D type II receptor on natural killer (NK) cells. NK cells are distinct from B or T cells and are crucial for the early response to viral infections and in shaping the subsequent adaptive immune response to viral infection \[[@B21]-[@B23]\]. Interestingly another recent GWAS found that a genetic variant in the closely related *MICA*gene is strongly associated with Hepatitis C virus-induced hepatocellular carcinoma, suggesting a pivotal role for MIC proteins in the pathogenesis of these Flaviviridae infections \[[@B24]\]. *PLCE1*encodes phospholipase C epsilon gamma, and missense mutations in *PLCE1*cause nephrotic syndrome \[[@B25]\], a kidney disorder in which dysfunction of the glomerulus basement membrane results in proteinuria and hypoproteinemia that, when severe, leads to reduced vascular oncotic pressure and edema. These elements of nephrotic syndrome have striking similarities with the hypovolemic shock in severe dengue and suggest an important role for *PLCE1*in maintaining normal vascular endothelial cell barrier function. These associations with *MICB*and *PLCE1*are not serotype specific, but instead are applicable across all four dengue virus serotypes \[[@B20]\]. The GWAS on severe dengue has revealed unexpected insights into its pathogenesis, as it points specifically to NK cell pathology, shared susceptibility pathways with shock, and possibly shared pathology with other Flaviviridae \[[@B20]\]. However, the locations of the causal genetic variants responsible for the observed associations are still unknown and await elucidation through fine-scale mapping. What is notable is that these observations suggest that subtle genomic differences in members of one viral family affect related, yet distinct, components of the human immune response. But is this molecular pathway of susceptibility the only route to Flaviviridae infection or is this shared pathway the only one for a wide variety of viruses? The genetic variation at both *MICB*and *PLCE1*explained no more than 2.5% of the overall heritable variance in susceptibility to DSS, whereas very little evidence of disease association was observed elsewhere in the genome. Will a finer dissection of the infecting pathogen genome reveal more and discover some of the \'missing heritability\'? Whole-genome sequencing of the dengue virus isolated from patients with clinical data has been initiated on a large scale \[[@B19],[@B26]-[@B29]\], and identification of mutations in dengue virus genes and their systematic reconciliation with human genotype and clinical phenotype will be possible in the near future. The impact of host and pathogen genetic variations on individual susceptibility to meningococcal disease ======================================================================================================== *Neisseria meningitidis*is a Gram-negative, polysaccharide-encapsulated bacterium and is the cause of meningococcal meningitis and sepsis, which are potentially fatal infections without antibiotic treatment. Five out of the twelve identified *N. meningitidis*serotypes have been reported to cause epidemics (A, B, C, W135, and Y) \[[@B30]\]. Subgroup analysis with specific meningococcal serotypes has revealed differential virulence capabilities of each serotype, with serotypes B and C causing more adverse infection outcomes than the others \[[@B31]-[@B33]\]. Unlike for meningococcal serotypes A, C, Y, and W135, a vaccine is yet to be developed for serotype B *N. meningitidis*, thus highlighting the need for improved understanding of the host-pathogen interactions involved. Earlier genetic studies examining human susceptibility to *N. meningitidis*sepsis were limited to known genes whose biological functions have been well characterized \[[@B34]-[@B36]\], and many were likely to be false-positive findings as they subsequently were not successfully validated by independent studies \[[@B37]\]. It is hoped that by better defining the critical elements at the host-pathogen interface using unbiased whole-genome approaches, more definitive answers on the susceptibility determinants of overall meningococcal infection and the extent of the immunogenic differences between serotype B *N. meningitidis*and the other serotypes will emerge. A recent case-control GWAS study \[[@B7]\] on *N. meningitidis*sepsis (which included serotypes A, B, C, Y, and W-135) in a UK collection, with validation in sample collections from Austria, the Netherlands, and Spain, showed significant evidence of association between genetic markers in Factor H (*CFH*) and *CFH*-related genes (*CFHR3*, *CFHR1*) on chromosome 1 and decreased susceptibility to meningococcal disease (Table [2](#T2){ref-type="table"}). CFH and CFHR3 are atypical members of the complement family in that they are negative regulators of complement signaling, protecting host cells from destruction through complement activation in response to infection \[[@B38]\]. Previous human mutation studies had shown complement deficiency to be a susceptibility factor for many *Neisseria*species \[[@B39]\], so perhaps this study \[[@B38]\] may not seem surprising. However, when examined, a new mechanism was revealed. It is not the absence of complement that is important to the occurrence of widespread disease, but rather the presence of the factor H complement regulator. This means that the addition of complement to patients who were deficient in complement pathway components might not prevent them from suffering repeated disease, whereas the removal of the complement regulator might, because CFH protects the bacteria through a complex interaction of the host protein and bacteria (Figure [1](#F1){ref-type="fig"}). This new mechanistic understanding of the disease may have other implications as well, because the associations observed with *CFH*and *CFHR*genetic variants were significantly stronger (per-allele increased risk of disease between 1.6-fold and 1.8-fold compared with baseline; *P*-value about 10^-5^to 10^-8^) in collections that had a variety of *N. meningitidis*serogroup infections (enrolled before wide-spread serotype C conjugate vaccination), compared with those collections collected after widespread serotype C vaccination was adopted (Spain; per-allele increase in risk about 1.3- to 1.4-fold, *P*-value about 10^-2^to 10^-3^). The substantially weaker association at the *CFH*-*CFHR3*-*CFHR1*locus in the predominantly serotype B sample collection implies that serotype B *N. meningitidis*has subtly distinct mechanisms of host interaction from those of other serotypes and may explain why serotype B *N. meningitidis*has been refractory to vaccine development attempts. ![**The complement cascade, factor H (encoded by *CFH*), and its interactions with *N. meningitidis*factor H binding protein (fHBP)**. The milieux shown here are the host endothelial cells, the interstitium, and invading *N. meningitidis*cells. Factor H normally binds to glycosaminoglycan sugars on host cells. This same region in Factor H is bound by the *N. meningitidis*through factor H binding protein (fHBP). There is significant diversity in fHBP, which is more variable in *N. meningitidis*serogroup B than in other serogroups. The serum concentration of CFH also varies with *CFH*genotype. Adapted from Tan *et al*. \[[@B66]\].](gb-2011-12-10-229-1){#F1} *N. meningitidis*defends itself from host complement-mediated killing through sequestration of human CFH by factor H binding protein (fHBP), a surface lipoprotein present on all strains of *N. meningitidis*\[[@B40]\]. Owing to its antigenic potential and ability to induce bactericidal antibodies, fHBP is one of the antigens incorporated in the recombinant vaccine for group B *N. meningitidis*in vaccine trials. This GWAS observation \[[@B7]\] thereby confirmed an interaction between *N. meningitidis*and human CFH, mediated in a complex manner through fHBP, which was not previously detected in mutation studies of patients suffering from recurrent meningococcal infections due to complement deficiency (Figure [1](#F1){ref-type="fig"}). As different allelic variants of CFH itself resulted in easily discernable differences in actual human susceptibility to meningococcal disease, therapeutic designs based on synergistic action targeting both fHBP and CFH remain a possibility. Such an approach could be more effective than manipulation of either molecule alone. The fHBP gene has recently been sequenced in various serotypes, revealing significantly higher sequence diversity in serotype-B than non-serotype-B *N. meningitidis*. This makes serotype-B *N. meningitides*more challenging to target from the point of view of both the host and the vaccinologist \[[@B41]\]. This could in part explain the much weaker association signal seen in the GWAS and implies a correlation between increased fHBP variability and virulence, which needs to be confirmed by future work. *Plasmodium falciparum* ======================= Malaria is a predominantly tropical disease vectored by the *Anopheles*mosquito, and up to 40% of the worldwide population is at risk of infection. Although infection by four different species of *Plasmodium*(*falciparum*, *malariae*, *ovale*, and *vivax*) is causative of malaria, *P. falciparum*is the predominant form globally. It also results in higher complication and mortality rates than the other three species \[[@B42]\]. Susceptibility to and severity as a result of *Plasmodium*infection are well studied examples of how host genetic factors affect the disease process. Natural genetic variation resulting in the sickle cell trait and Duffy blood group antigens are two clear examples whereby the human genome adapts specifically in response to an infectious agent. The parasite also exerts considerable selective pressure on the human genome \[[@B43]\], underlying yet again the importance of understanding the host-pathogen interface from both perspectives. Although the immune responses to malaria have been extensively documented in both animal and human models \[[@B44],[@B45]\], robust evidence on the specific response that is critical for protective immunity remains elusive \[[@B46],[@B47]\]; often, the induction of a high degree of immunogenicity does not correlate with protective immunity \[[@B48]\]. GWASs have now made it possible to search systematically for strong association between functional variation in a given immune gene and susceptibility to (or severity of) infection to identify the specific immune responses responsible for protective immunity. All genetic studies for human susceptibility to malaria have yielded two consistent results: the sickle hemoglobin trait is associated with a 5- to 10-fold reduced susceptibility to severe malaria, and the ABO blood group is associated with a more modest reduction (a per-allele odds ratio of 1.2-fold increased risk for blood groups A, B, and AB compared with blood group O) \[[@B49]\]. Both findings have been confirmed by a subsequent GWAS on severe malaria in West Africa (Table [2](#T2){ref-type="table"}), revealing that the major interaction point of interest from the host perspective is the red blood cell. Indeed, it is this interaction between *P. falciparum*and the red blood cell that gives rise to all the clinical symptoms of malaria, and normally functioning erythrocyte physiology has been shown to be crucial to parasite survival \[[@B50],[@B51]\]. However, what is surprising is that genetic variation in all immune-related genes (such as the HLA, tumor necrosis factor (TNF) and lymphotoxin alpha (LTA) families), the very genes that have been thought to have critical roles in the immune response and clearance of *P. falciparum*infections \[[@B52],[@B53]\] do not consistently show association with susceptibility in large collections of malaria patients. The paucity of findings from the viewpoint of the host beyond the red blood cell suggests that for malaria, genetic variations in *P. falciparum*itself might account for substantially more disease variance. The genomic approach adopted from the perspective of *P. falciparum*has been two-pronged: using genome-wide association and detection of natural selection to identify molecules crucial at the host-pathogen interface. It has been observed that the strong selective pressure exerted by *P. falciparum*on the erythrocyte has led to increased incidence of otherwise rare hematological disorders (such as sickle-cell disease, hemoglobin C, and possibly thalassemia) \[[@B54],[@B55]\]. A better understanding of the selective pressure that malaria has exerted on the immune system would thus yield considerable insight into the process of malaria pathogenesis and severity. Indeed, surveys of the *P. falciparum*genome using sequencing and genotyping have revealed it to be highly variable. Genes encoding *Plasmodium*proteins interacting with the host immune system, such as *P. falciparum*erythrocyte membrane protein 1 (PfEMP1), are often under balancing selection resulting from pressures from the host immune system on one hand and the need to maintain diversity on the other \[[@B8]\]. The sequencing of *P. falciparum*virulence genes that interact with the red blood cell (var/PfEMP1) and transporters (such as *pfcrt*for chloroquine, *pfsurfin*for dihydroartemisinin and *dfhr*for cycloguanil) responsible for resistance to anti-malarial drugs revealed greater than average genomic diversity and the presence of positive selection signatures in these *Plasmodium*genes, indicating the continued evolution of the parasite in response to survival pressures from both the human immune response and to medical intervention. Indeed, a GWAS of *P. falciparum*genetic variation using an array of over 17, 500 markers obtained by parasite genome sequencing revealed very strong associations between genetic variations in many drug transporter genes, including the *P. falciparum*chloroquine transporter gene (PfCRT), *P. falciparum*dihydrofolate reductase (PfDHFR) and resistance to anti-malarial chemotherapy \[[@B9]\]. Strong evidence of association with drug resistance and positive selection was also observed at *pfsurfin*, a gene that had until then not been implicated in drug resistance. Curiously, PfSURFIN was reported to be co-transported with PfEMP1 to the infected surface of the red blood cell and is thought to be part of a protein complex involved in binding or transport of chemical compounds. The extension of this *P. falciparum*GWAS to severe malaria susceptibility in humans (as opposed to uncomplicated infection) will be extremely informative \[[@B56]\]. Future challenges ================= The outcome of a specific episode of infection is strongly suspected to depend in part on specific interactions between host and pathogen genotypes \[[@B57]\]. The cumulative effect size from the contribution from the host and pathogen genomes is likely to be larger than each genome alone, and clinical outcomes of infection are unlikely to be explained by a reductionist approach that studies disparate individual components \[[@B58]\]. Even in less complex organisms than humans, host resistance to pathogens has been shown to vary dramatically across different combinations of host and pathogen genotypes \[[@B59]\], with an even more complicated picture if the host is susceptible to infection by multiple different strains of the same pathogen species. To this end, GWASs and pathogen sequencing have revealed fresh and often unexpected insights for host-pathogen interactions in revealing hitherto unsuspected molecules (Table [2](#T2){ref-type="table"}). In cases of both dengue and meningococcal disease, GWASs revealed surprising interacting host molecules (the NK cell pathway for dengue and complement inhibition by factor H for meningococcal disease) that are important in disease pathogenesis. In addition, the GWAS on meningococcal disease revealed that pathogen genetic variation (in terms of different infecting strains) also contributes substantially to overall inter-individual susceptibility to disease - and this may also be case with dengue and other diseases. Although the GWAS effort from the human perspective has not produced the novel insights for malaria seen for other infectious diseases, it is clear that this is more than compensated for by approaches from the parasite perspective, which have already begun to bear fruit. Var genes identified from analysis of the sequenced *Plasmodium*genome are already turning up potential antigenic sites of interaction with the human genome, with the potential for being vaccine targets \[[@B8]\], and combined selection and association analysis have revealed PfCRT, PfSURFIN, PfMDR and *P. falciparum*apical membrane antigen 1 (PfAMA-1) as crucial molecules interacting with both anti-malarial medications and the host immune system, respectively, as well as *PFE1445c*, which encodes a *Plasmodium*conserved protein \[[@B9]\]. So, in this case, more detailed work stemming from analysis of *P. falciparum*resequencing data could reveal crucial interaction points between host and parasite. The new challenges in defining host-pathogen interactions are starting to become computational, because to comprehensively analyze and integrate whole genome data for both humans and pathogens, complex bioinformatics toolsets are required, as the order of human-pathogen genotype-phenotype combinations increases exponentially. The search is further complicated by the well known observations that pathogen genomes are not as conserved as the human genome; for example, the per-nucleotide mutation rate is 5 × 10^-4^per generation \[[@B8]\] for the *P. falciparum*genome, compared with about 2.5 × 10^-8^for humans \[[@B60]\]. To aid in these challenges, high-throughput analysis pipelines can now be used for alignment and mapping of pathogen genomes generated by next-generation sequencing experiments, thus allowing a comprehensive catalog documenting all pathogen genetic polymorphisms (not unlike the human HapMap project). This effort will enable in-depth characterization of pathogen phylogeny and measurement of correlations between pathogen genetic mutations and virulence, together with key human phenotypes of the infection (such as cytokine profile and disease severity). In most cases, the integration of genomic data obtained from GWASs and resequencing efforts from both the host and pathogen should reveal a complete catalog of the inter-individual susceptibility to infection.

Introduction {#Sec1} ============ Instrumental records provide clear evidence that long-term observations are required to establish the processes controlling AMOC variations^[@CR1]^. The only way to examine AMOC processes over longer time periods is to use paleoceanographic evidence from times when there were clear changes in the Earth system^[@CR2],[@CR3]^. Indeed, there have been considerable efforts to reconstruct AMOC strength during the deglacial period to examine its relationship with pronounced changes in climate^[@CR2]^, the atmospheric-oceanic carbon budget^[@CR3],[@CR4]^ and ice sheet volume^[@CR2],[@CR5]^ associated with millennial-scale climate events---Heinrich Stadial 1 (HS1, \~19--15 thousand years ago/ka), the Bølling-Allerød (BA, \~15--13 ka), and the Younger Dryas (YD, \~13--11 ka). However, the nature of the interplay between AMOC, climate and ice sheet systems during this period remains an outstanding scientific question, in large part because of the complications of interpreting circulation proxy records from individual locations that might not be representative of basin-wide AMOC strength. Existing approaches for reconstructing past AMOC strength have included radiocarbon (^14^C)^[@CR3]^, grain size^[@CR6]^, paleo-geostrophy^[@CR7]^ and ^231^Pa/^230^Th^[@CR2]^. The use of sedimentary ^231^Pa~ex,0~/^230^Th~ex,0~ (the activity ratio of ^231^Pa to ^230^Th unsupported by the decay of U in the sediment and corrected for post-depositional decay, hereafter ^231^Pa/^230^Th) as a geochemical proxy for ocean circulation rate is based on the difference in oceanic residence time between the two rapidly scavenged isotopes: 50--200 years for ^231^Pa and 10--40 years for ^230^Th^[@CR8]^. Lateral transport, including advection and eddy diffusion, allows greater movement of ^231^Pa, which stays dissolved in seawater longer than ^230^Th, giving rise to deviations of ^231^Pa/^230^Th from the production ratio (\~0.093), which are subsequently preserved by particles deposited on the seafloor^[@CR8]^. Observations of ^230^Th and ^231^Pa in the modern ocean suggest reversible exchange between the nuclides adsorbed onto sinking particulate matter and those dissolved in seawater^[@CR9]^, and the resulting ^231^Pa/^230^Th is proposed to reflect zonal integration of signal across large distances (\>1000 km)^[@CR10]^ and mainly \~1 km of the overlying water column when there is active water-mass advection^[@CR11]^. Therefore, in principle, sedimentary ^231^Pa/^230^Th may record a depth-integrated, large-scale lateral transport signature that reflects aspects of the overall circulation of the overlying water column^[@CR2]^. At present, the active southward advection of North Atlantic Deep Water (NADW), the deep component of AMOC, causes greater export of seawater ^231^Pa than ^230^Th from the mid-latitude and low-latitude Atlantic^[@CR10]^, giving rise to sedimentary ^231^Pa/^230^Th ratios lower than the production ratio (\<0.093)^[@CR12]^. Deglacial changes in AMOC strength have the potential to modify the rate of ^231^Pa export (relative to ^230^Th), a shift that should be recorded in the ^231^Pa/^230^Th of Atlantic sediments. The proposed link between sedimentary ^231^Pa/^230^Th and AMOC rate is supported by a number of ocean transport models^[@CR13],[@CR14]^, with recent modelling work indicating that this relationship is particularly strong in the deep western North Atlantic^[@CR14]^. A number of temporally well-resolved sedimentary ^231^Pa/^230^Th time-series point towards millennial-scale changes in AMOC strength during the most recent deglacial period^[@CR2],[@CR15]--[@CR17]^. However, this conclusion is not obviously supported by some other Atlantic ^231^Pa/^230^Th records from the equatorial^[@CR18]^, marginal^[@CR19]^, eastern intermediate-depth^[@CR20]--[@CR22]^, and southern sites^[@CR23]^. Such discrepancies are most likely related to the geochemical behaviour of ^231^Pa and ^230^Th in the ocean. In particular, seawater Pa is more effectively scavenged in areas of high opal and particulate flux^[@CR8]^. Previous down-core studies have shown strong positive correlations between ^231^Pa/^230^Th and both opal and bulk sediment fluxes, providing evidence that scavenging by opal and other particles has the potential to significantly influence sedimentary ^231^Pa/^230^Th at some locations, including the tropical^[@CR18]^ and South Atlantic^[@CR23]^. Recent observations^[@CR24]^ also reveal enhanced scavenging of both ^231^Pa and ^230^Th near margins (boundary scavenging), nepheloid regions, and the Mid-Atlantic Ridge (MAR), due to elevated particle fluxes, high concentrations of resuspended sediment, and Fe-rich hydrothermal fluxes respectively^[@CR24]^. Therefore, it is important to establish the role of scavenging when interpreting sedimentary ^231^Pa/^230^Th records in the context of circulation. In addition, the apparent timing of the observed changes in ^231^Pa/^230^Th reported in previous work suggests that deglacial changes are not in phase across the Atlantic^[@CR23]^ -- further challenging the interpretation of ^231^Pa/^230^Th as an integrative signal of overturning circulation and hampering the use of this approach as a tool for testing climate models and hypotheses. The aim of this study is to constrain a coherent picture of AMOC strength over the deglacial period by examining the observational and interpretive discrepancies between ^231^Pa/^230^Th records. This approach requires an evaluation of the location of the records, sediment chronologies, and geochemical controls on sedimentary ^231^Pa/^230^Th. We compiled all thirty-three of the available deglacial sedimentary ^231^Pa/^230^Th time-series including four new records (Supplementary Figs. [1](#MOESM1){ref-type="media"}, [2](#MOESM1){ref-type="media"}) within the Atlantic Ocean on a common chronology. We then tested the correlation of the ^231^Pa/^230^Th with both particle and opal fluxes (Supplementary Method, Supplementary Figs. [3](#MOESM1){ref-type="media"}--[5](#MOESM1){ref-type="media"}), and found that nineteen records show no strong relationship to either of these two variables. Six of these records are from the eastern intermediate depths (Supplementary Figs. [1,](#MOESM1){ref-type="media"} [2](#MOESM1){ref-type="media"}). These sites are most likely to be subjected to the influence of shallow circulation, although a consensus is yet to emerge on the changes of these shallower water masses and their relationship to the overall AMOC during the last deglaciation^[@CR25]^. Therefore, we focus on the remaining thirteen records (including four new records) from the deep basin (50° N--3° S, 2.7--4.6 km) and Brazil margin (2° S, 2.25 km) (Fig. [1](#Fig1){ref-type="fig"}, Table [1](#Tab1){ref-type="table"})---these parts of the Atlantic are found to be more responsive to changes in AMOC strength associated with the waxing and waning of NADW formation^[@CR14],[@CR26]^. The sites of these thirteen records are located both near and far from continental margins, within potential nepheloid regions and at the MAR (Fig. [1](#Fig1){ref-type="fig"}). Assessment of these combined thirteen records therefore minimises potential signals associated with scavenging, and provides an overview of past AMOC changes. Coherent changes in the West and deep high-latitude North Atlantic sites represent the best-constrained picture of the timing of AMOC shift on the millennial timescale, and therefore its relationship with the timing of ice sheet and climate changes during the last glacial termination.Fig. 1Location map of paleo-proxy records presented in the main discussion. These include sedimentary ^231^Pa/^230^Th records (square and circle symbols indicate new and previously published records respectively), sortable silt fraction (10--63 μm) grain size record (cross symbol), ice-rafted debris and meltwater discharge records ('plus' symbols). Numbers and letters in brackets denote the identity of the sediment records plotted in Figs. [2](#Fig2){ref-type="fig"}--[4](#Fig4){ref-type="fig"}, with references listed in Table [1](#Tab1){ref-type="table"}. The map was generated using the Ocean Data View programme^[@CR79]^Table 1Summary of sediment cores used to discuss deglacial changes in AMOC and ice sheet systemSiteRecordCore nameLatitude (° N)Longitude (° E)Water depth (km)ReferencesNotation on map/legendRockall Basin^231^Pa/^230^ThSU90-4450.02−17.104.279Gherardi et al.^[@CR16]^\[1\]Newfoundland margin^231^Pa/^230^ThMD95-202741.73−47.734.112Gherardi et al.^[@CR16]^\[2\]Mid-Atlantic Ridge^231^Pa/^230^ThIODP U131341.00−32.963.426Lippold et al.^[@CR23]^\[3\]Iberian margin^231^Pa/^230^ThSU81-1837.77−10.183.135Gherardi et al.^[@CR15]^\[4\]Bermuda Rise^231^Pa/^230^ThOCE326-GGC533.70−57.584.550McManus et al.^[@CR2]^\[5\]Bermuda Rise^231^Pa/^230^ThODP 106333.68−57.624.584Lippold et al.^[@CR77]^\[6\]African margin--off Mauritania^231^Pa/^230^ThMD03-270518.08−21.153.085Meckler et al.^[@CR78]^\[7\]Researchers Ridge^231^Pa/^230^ThJC094-GVY1415.4643−50.99152.714This study\[8\]Sierra Leone Rise^231^Pa/^230^ThJC094-GVY017.435−21.79633.426This study\[9\]Ceara Rise (northern)^231^Pa/^230^ThEW9209-1JPC5.907−44.1954.056This study\[10\]Ceara Rise (northern)^231^Pa/^230^ThEW9209-3JPC5.313−44.263.288This study\[11\]Brazil margin^231^Pa/^230^ThGeoB16202-2−1.9083−41.59172.248Mulitza et al.^[@CR17]^\[12\]Equatorial Atlantic^231^Pa/^230^ThRC24-12−3.01−11.4173.486Bradtmiller et al.^[@CR18]^\[13\]South of IcelandSortable siltODP 98461.42−24.071.650Praetorius et al.^[@CR6]^\[I\]Feni Drift, Rockall BasinIRDODP 98055.48−14.702.179McManus et al.^[@CR35]^; Benway et al.^[@CR36]^\[A\]Off NewfoundlandIRDEW9302-2JPC48.7950−45.08481.251Marcott et al.^[@CR27]^\[B\]Bay of BiscayFluvial dischargeMD95-200247.452−8.5342.174Menot et al.^[@CR32]^\[C\] Results {#Sec2} ======= Coherent signal in the West and deep high-latitude North {#Sec3} -------------------------------------------------------- Our new dataset reveals that sediment cores from the West and deep high-latitude North Atlantic exhibit remarkably consistent ^231^Pa/^230^Th changes both in timing and amplitude over the last 25 thousand years (kyr) (Fig. [2a](#Fig2){ref-type="fig"}). The ^231^Pa/^230^Th observed during the Last Glacial Maximum (LGM, \~22--19 ka) range from 0.059 to 0.083 and are higher overall than the Holocene (\<10 ka) values, which range from 0.041 to 0.065. High ^231^Pa/^230^Th values close to or above the production ratio (\~0.093) are observed during HS1 (\~19--15 ka). When data are available, these sites display a marked decrease in ^231^Pa/^230^Th from HS1 to BA (\~15--13 ka), followed by a distinct increase during the YD (\~13--11 ka). The cores that contain data through the Holocene exhibit a gradual decrease in ^231^Pa/^230^Th from around 11 to 8 ka (Fig. [2a](#Fig2){ref-type="fig"}).Fig. 2Atlantic sedimentary ^231^Pa/^230^Th records that reflect past changes in circulation rate. **a** West Atlantic and deep high-latitude (\>50° N) North Atlantic records, **b** deep subtropical North-East Atlantic and Mid-Atlantic Ridge (MAR) records, and **c** deep low-latitude East Atlantic records. Error bars represent 2 s.e.m. Red, green and blue shading categorise the ^231^Pa/^230^Th ratios into high, middle and low values, respectively. Triangle and diamond symbols respectively signify ^14^C and non-^14^C chronological tie-points. Bracketed numbers denote the identity of the sediment cores marked in Fig. [1](#Fig1){ref-type="fig"}, with references listed in Table [1](#Tab1){ref-type="table"}. Bold characters in the figure legend and the square symbols indicate new ^231^Pa/^230^Th reconstructions from this study. Annotations of key climate events: LGM Last Glacial Maximum, HS1 Heinrich Stadial 1 (purple shading), BA Bølling-Allerød, YD Younger Dryas (purple shading) Two-phase ^231^Pa/^230^Th transition during HS1 {#Sec4} ----------------------------------------------- An obvious feature of the newly combined West and deep high-latitude North Atlantic observations (Figs. [2](#Fig2){ref-type="fig"}a and [3b](#Fig3){ref-type="fig"}) is the pronounced increase in ^231^Pa/^230^Th towards or above the production ratio during HS1. Our compilation indicates that there may have been at least two phases in the observed HS1 transition. The first phase consists of a distinct early rise in ^231^Pa/^230^Th at the western sites from around 19 to 16.5 ka (Figs. [2](#Fig2){ref-type="fig"}a and [3b](#Fig3){ref-type="fig"}). In the second phase, widespread maximum values of ^231^Pa/^230^Th are observed in the West and deep North-East Atlantic from \~16.5 to 15 ka (Figs. [2](#Fig2){ref-type="fig"}a, b and [3b](#Fig3){ref-type="fig"}).Fig. 3Paleo-proxy records of ice sheet, Atlantic circulation and climate changes from 21 to 10 ka. **a** North Atlantic ice-rafted debris^[@CR27],[@CR35],[@CR36]^ records and a proxy (terrestrial organic matter isoprenoid tetraether (BIT) index) record of Eurasian fluvial discharge^[@CR32]^, **b** West and high-latitude North Atlantic ^231^Pa/^230^Th records that contain data over the Holocene, deglacial and LGM (see also Fig. [2a](#Fig2){ref-type="fig"}), **c** northern tropical Atlantic (5--15^o^ N) coral proxy records of ocean radiocarbon (B-Atm ^14^C ages with ellipsoid error bars)^[@CR3]^, **d** sortable silt mean grain size record of Iceland-Scotland Overflow Water strength (an important component of NADW)^[@CR6]^, **e** atmospheric CO~2~ record from the West Antarctic Ice Sheet Divide ice core (WDC)^[@CR51]^ and atmospheric Δ^14^C record from the IntCal13 compilation^[@CR50]^, and **f** Northern Greenland ice core temperature proxy (δ^18^O) record^[@CR55]^. Triangle and diamond symbols respectively signify ^14^C and non-^14^C chronological tie-points. Numbers and letters in brackets denote the identity of the sediment cores marked in Fig. [1](#Fig1){ref-type="fig"}, with references listed in Table [1](#Tab1){ref-type="table"}. Bold characters in the figure legend and the square symbols indicate new ^231^Pa/^230^Th reconstructions from this study. Yellow shading---early HS1, blue shading---late HS1, green shading---YD, grey shading mark the HS1--BA transition and the YD-early Holocene transition LGM-deglacial contrast between the East and West Atlantic {#Sec5} --------------------------------------------------------- Our dataset also indicates contrasting trends in ^231^Pa/^230^Th between the East and West Atlantic over the last 25 kyr. All the western sites exhibit higher LGM ^231^Pa/^230^Th values than the late Holocene (\<5 ka) values (Figs. [2](#Fig2){ref-type="fig"}a and [4](#Fig4){ref-type="fig"}). However, the eastern sites display different trends with water depths and latitudes. We observe lower ^231^Pa/^230^Th at the Iberian margin (38° N, 3.14 km, Figs. [2](#Fig2){ref-type="fig"}b and [4](#Fig4){ref-type="fig"}) and off Mauritania (18° N, 3.09 km, Figs. [2](#Fig2){ref-type="fig"}b and [4](#Fig4){ref-type="fig"}), similar ^231^Pa/^230^Th at the MAR (41° N, 3.43 km, Figs. [2](#Fig2){ref-type="fig"}b and [4](#Fig4){ref-type="fig"}) and Sierra Leone Rise (7° N, 3.43 km, Figs. [2](#Fig2){ref-type="fig"}c and [4](#Fig4){ref-type="fig"}), and higher LGM ^231^Pa/^230^Th in the Equatorial Atlantic (3° S, 3.49 km, Figs. [2](#Fig2){ref-type="fig"}c and [4](#Fig4){ref-type="fig"}) and Rockall Basin (50° N, 4.28 km, Figs. [2](#Fig2){ref-type="fig"}a and [4](#Fig4){ref-type="fig"}). This inter-basin contrast continued over the ensuing deglaciation at the low latitudes, with persistently low ^231^Pa/^230^Th observed at Sierra Leone Rise (7° N, 3.43 km, Fig. [2c](#Fig2){ref-type="fig"}), even during HS1 when there was a substantial ^231^Pa/^230^Th rise in the western basin (Fig. [2a](#Fig2){ref-type="fig"}).Fig. 4Differences in averaged ^231^Pa/^230^Th values between the LGM (\~22--19 ka) and late Holocene (≤5 ka) (∆Pa/Th~LGM-Hol~). **a** ∆Pa/Th~LGM-Hol~ versus water depth and **b** ∆Pa/Th~LGM-Hol~ versus latitude. The calculation was carried out based on the thirteen Atlantic ^231^Pa/^230^Th time-series selected for interpretation of deep Atlantic circulation (Fig. [2](#Fig2){ref-type="fig"}). The ∆Pa/Th~LGM-Hol~ plots have eleven data points as one Bermuda Rise core (34° N, 4.58 km) and the Newfoundland margin core (42° N, 4.11 km) do not have Holocene data. Purple shading indicates an East-West contrast in the ∆Pa/Th~LGM-Hol~ at mid-latitude and low latitudes of the North Atlantic. Error bars represent 95% confidence interval of the averaged ^231^Pa/^230^Th values, and represent 2 s.e.m when there is only one data point in each site at each time period. The red and blue lines connect the sites from the West Atlantic and the sites from the East Atlantic and MAR, respectively Discussion {#Sec6} ========== Our new integrated ^231^Pa/^230^Th dataset (Fig. [2](#Fig2){ref-type="fig"}) reveals large-scale patterns of AMOC change that are not dependent on any single-core location and includes cores from diverse oceanographic and sedimentary settings (Fig. [1](#Fig1){ref-type="fig"}). Most importantly, the coherent ^231^Pa/^230^Th signal in the West and deep high-latitude North Atlantic (Fig. [2a, Supplementary Data 3](#Fig2){ref-type="fig"}) provides robust evidence for millennial-scale changes in deep Atlantic circulation strength during the deglacial period. The temporal consistency of the integrated records differs from some prior observations, which found offsets in the timing of the deglacial ^231^Pa/^230^Th shift in the deep West Atlantic^[@CR23]^. In particular, the widespread ^231^Pa/^230^Th increase from the LGM to HS1 (Fig. [2a](#Fig2){ref-type="fig"}) indicates that there was an extensive, basin-scale control reducing ^231^Pa export during HS1. Our dataset strengthens the proposal^[@CR2]^ that this control was a reduction in deep-water transport in the North Atlantic. One working hypothesis for millennial-scale Heinrich stadials such as HS1 relies on significant freshwater addition to the North Atlantic^[@CR27],[@CR28]^. For example, meltwater associated with large ice-rafting events might alter surface water buoyancy leading to reduced deep-water production and AMOC slowdown. In turn, this would result in less northward heat transport, which would act to cool the Northern Hemisphere^[@CR28]^. This hypothesis has sparked a long-standing debate over the mechanisms linking the ocean, climate and ice-sheet systems during Heinrich stadials. For example, a recent study from the North Atlantic emphasised that climate cooling began prior to the arrival of ice-rafted debris (IRD) layers, indicating large ice-rafting events may be a consequence of, rather than a trigger for Heinrich stadials^[@CR29]^. This late arrival of the IRD layers is supported by other well-dated records from the North Atlantic^[@CR27],[@CR30]^. In contrast, there is evidence for early meltwater discharge^[@CR31],[@CR32]^ and iceberg rafting from the Eurasian ice sheets^[@CR33],[@CR34]^ that may have initiated an AMOC slowdown leading to the development of HS1. Clearly, the timing and duration of AMOC slowdown is key to establishing potential triggers and forcing mechanisms linking such climatic events with ocean behaviour. To examine the phasing of the ocean circulation and ice sheet systems during HS1, we compare our ^231^Pa/^230^Th compilation (Fig. [3b](#Fig3){ref-type="fig"}) with Eurasian fluvial discharge and IRD records from the North Atlantic (Fig. [3a](#Fig3){ref-type="fig"}). Fluvial discharge from the Bay of Biscay (47° N)^[@CR32]^ shows peaks of meltwater from the Eurasian glaciers and ice sheets from \~19 to 16.5 ka. During this same period, the eastern IRD record from Feni Drift (55° N)^[@CR35],[@CR36]^, which is expected to be sensitive to ice rafting from the Eurasian ice sheets, displays peaks of IRD grain counts. In contrast, the western record off Newfoundland (49° N)^[@CR27]^ shows pronounced IRD CaCO~3~ grain counts, indicative of ice rafting from the Laurentide ice sheet, from \~16.5 to 15 ka (Fig. [3a](#Fig3){ref-type="fig"}). These combined observations indicate that episodes of Eurasian meltwater discharge and iceberg calving occurred during early HS1 (\~19--16.5 ka), whereas Laurentide iceberg calving is evident in late HS1 (\~16.5--15 ka). The observed ^231^Pa/^230^Th increase beginning \~19 ka (Fig. [3b](#Fig3){ref-type="fig"}, Supplementary Figs. [6](#MOESM1){ref-type="media"}, [7](#MOESM1){ref-type="media"}) is aligned with enhanced fluvial draining of the Eurasian ice sheets (Fig. [3a](#Fig3){ref-type="fig"}), supporting an early AMOC decline that was linked to Eurasian-sourced freshwater forcing^[@CR37],[@CR38]^. Weakening of AMOC could cause subsurface warming of the Nordic seas that could further destabilise the Eurasian ice sheets^[@CR39]^, leading to increased iceberg calving activity as recorded in the Feni Drift core (Fig. [3a](#Fig3){ref-type="fig"}). Melting of these Eurasian icebergs potentially (Fig. [3a](#Fig3){ref-type="fig"}) contribute to progressive weakening of AMOC during early HS1 (Fig. [3b](#Fig3){ref-type="fig"}). This first phase of AMOC reduction (\~19--16.5 ka) might eventually lead to more widespread subsurface warming that could reach the Labrador Sea^[@CR27]^, triggering ice-rafting events from the Laurentide ice sheet as recorded in the Newfoundland core (Fig. [3a](#Fig3){ref-type="fig"}). The timing of this Laurentide iceberg flux coincides with widespread maximum ^231^Pa/^230^Th values observed in the West and deep North-East Atlantic (Figs. [2](#Fig2){ref-type="fig"}a, b and [3b](#Fig3){ref-type="fig"}, Supplementary Figs. [6](#MOESM1){ref-type="media"}, [7](#MOESM1){ref-type="media"}), indicating that Laurentide iceberg calving might have sustained the second phase of maximum AMOC reduction during late HS1 (\~16.5--15 ka). The phasing of AMOC, meltwater discharge and iceberg flux observed here suggests a two-phase mechanism for AMOC reduction at the beginning of the deglacial. The two-phase AMOC reduction has important implications for the HS1 climate, as it broadly coincides in timing with the two-phase changes in global hydrological cycle observed during this period^[@CR40]^. The widespread decrease in ^231^Pa/^230^Th in the West and deep North-East Atlantic sediment cores after peak values at around 15 ka (Figs. [2](#Fig2){ref-type="fig"}a, b and [3b](#Fig3){ref-type="fig"}, Supplementary Figs. [6](#MOESM1){ref-type="media"}, [7](#MOESM1){ref-type="media"}) signifies a recovery in the strength of deep-water transport at the onset of the BA. The inferred resumption in deep circulation and deep-water formation in the North Atlantic coincides in timing with abrupt Greenland warming (Fig. [3f](#Fig3){ref-type="fig"}) and major ^14^C ventilation of the Atlantic Ocean^[@CR3],[@CR41]^ (e.g. Fig. [3c](#Fig3){ref-type="fig"}). These proxy records are consistent with recent modelling studies^[@CR42]--[@CR45]^, which confirm that AMOC is a key player in abrupt changes in climate and the ocean-atmospheric carbon budget^[@CR43]^. The AMOC reinvigoration during the BA (\~15--13 ka), evident from our ^231^Pa/^230^Th dataset (Fig. [3b](#Fig3){ref-type="fig"}) and ^14^C record^[@CR3]^ (Fig. [3c](#Fig3){ref-type="fig"}), coincides with an interval of diminished IRD and Eurasian meltwater discharge (Fig. [3a](#Fig3){ref-type="fig"}). However, sea level records^[@CR46]^ and ice sheet modelling^[@CR47]^ suggest a considerable amount of ice melt from the Laurentide ice sheet entering the Atlantic Ocean during this period. If a significant amount of this freshwater flux reached the deep-water formation sites in the North Atlantic, this would be inconsistent with the hypothesis that AMOC recovery during the BA was due to reduced freshwater forcing from the northern ice sheets. In fact, rather than a driving factor of circulation change, the major Laurentide ice sheet melting and sea level rise might have been a consequence of reinvigorated AMOC during this interval, as previously inferred^[@CR2],[@CR47]^. Alternatively, a recent coupled atmosphere-ocean model^[@CR4]^ suggests that the overall increase of atmospheric CO~2~ from HS1 leading up to BA (Fig. [3e](#Fig3){ref-type="fig"}) could be the cause of AMOC reinvigoration, even in the event of increased freshwater forcing. The model shows that the atmospheric CO~2~ increase could alter the hydrological cycle, leading to northward transport of more saline surface current in favour of deep-water formation in the North Atlantic^[@CR4]^. A subsequent ^231^Pa/^230^Th increase is observed during the YD, but overall it is of a smaller magnitude and duration compared to HS1 (Fig. [3b](#Fig3){ref-type="fig"}, Supplementary Figs. [6](#MOESM1){ref-type="media"}, [7](#MOESM1){ref-type="media"}). This difference suggests a reduction in Atlantic deep-water transport during the YD that was less intense or shorter than the reduction during HS1^[@CR2]^, or that sedimentation rates in some of these cores are not sufficient to prevent smoothing of the YD signal^[@CR2],[@CR16]^ (Supplementary Fig. [11](#MOESM1){ref-type="media"}). Notably, The YD peak of ^231^Pa/^230^Th observed at the Researchers Ridge (15° N, 2.71 km) appears to be short-lived relative to the other sites (Fig. [2](#Fig2){ref-type="fig"}), potentially implying some spatial differences in the timing and duration of the reduction in water-mass transport during the YD decline of AMOC. Establishing the existence of such spatial differences needs to be constrained using additional cores with high sedimentation rates. Other strong evidence of YD weakening of AMOC have included ^14^C records^[@CR3],[@CR41]^ (Fig. [3c](#Fig3){ref-type="fig"}), sortable silt grain size^[@CR6]^ (Fig. [3d](#Fig3){ref-type="fig"}) and geostrophic reconstruction of Gulf Stream strength^[@CR7]^. Modelling studies have highlighted the role of AMOC reduction in causing widespread cooling of the northern hemisphere during the YD^[@CR45],[@CR48]^, such as that shown in the Greenland temperature record (Fig. [3f](#Fig3){ref-type="fig"}). The mechanism that could sustain a reduced AMOC during this stadial period is not well constrained. There is no evidence for meltwater discharge to the Bay of Biscay (Fig. [3a](#Fig3){ref-type="fig"}) suggesting limited Eurasian freshwater forcing. By contrast, there is some evidence of rerouting of Laurentide-sourced meltwater discharge to the Arctic Ocean at the onset of YD, but it was rather short-lived^[@CR49]^. The Feni Drift IRD record indicates increased iceberg discharge from \~13 to 11 ka, but the magnitude and duration of these events were much smaller than those in HS1 (Fig. [3a](#Fig3){ref-type="fig"}). A potential explanation for these observations links to the stability of AMOC when northern ice sheets reached intermediate heights^[@CR5],[@CR32],[@CR42],[@CR43]^ such as those during the YD. Under an intermediate ice volume like YD, climate has been shown to be more variable^[@CR35]^ and AMOC has been found to be more sensitive to climate forcing^[@CR5]^, so freshwater forcing associated with the meltwater discharge and the melting icebergs described above might be sufficient to sustain the AMOC decline during this stadial period. The compiled North Atlantic dataset shows a strong decrease in ^231^Pa/^230^Th towards present-day values during the early Holocene (Fig. [2](#Fig2){ref-type="fig"}). The inferred rebound of deep Atlantic circulation from the YD to early Holocene is associated with Greenland warming (Fig. [3f](#Fig3){ref-type="fig"}), and with the ventilation of deep-ocean carbon to the atmosphere as indicated by Atlantic ^14^C observations^[@CR3]^ (Fig. [3c](#Fig3){ref-type="fig"}) and atmospheric ^14^C^[@CR50]^ and CO~2~ records^[@CR51]^ (Fig. [3e](#Fig3){ref-type="fig"}). Our ^231^Pa/^230^Th dataset suggests that deep-water transport in the Atlantic Ocean first reached its modern strength around 8 ka (Fig. [2](#Fig2){ref-type="fig"}), in agreement with water-mass proxy records^[@CR52],[@CR53]^, and consistent with the commencement of deep-water formation in the Labrador Sea^[@CR54]^ and the onset of stable present-day temperatures in the north^[@CR55]^. An additional observation from the compiled dataset is an indication of contrasting behaviour between the East and West Atlantic over the last 25 kyr. The higher LGM ^231^Pa/^230^Th compared to late Holocene ^231^Pa/^230^Th observed in all western cores (Fig. [4](#Fig4){ref-type="fig"}) most likely indicates a weaker deep-water transport, consistent with sortable silt (10--63 μm) grain size data from the deep western boundary^[@CR56]^. Supporting evidence for reduced rates of deep transport in the LGM Atlantic come from ^14^C records^[@CR41]^, stable isotope modelling of oceanic vertical mixing^[@CR57]^, and geostrophic reconstruction of the Gulf Stream^[@CR7]^. The picture emerging from the ^231^Pa/^230^Th observations at the Iberian margin (38° N, 3.14 km), off Mauritania (18° N, 3.09 km), and Sierra Leone Rise (7° N, 3.43 km) (Fig. [4](#Fig4){ref-type="fig"}) hints at a difference in deep-water transport at the mid-low latitudes of East Atlantic, with potential for stronger LGM transport in these deep eastern sites compared to the west. Brine rejection during ice formation in the Nordic seas has the potential to form deep waters that are not found at present^[@CR58]^, and this mechanism might provide a route for deep water overflows through the Iceland-Scotland Ridge into the eastern basin instead of via the shallower Denmark Strait into the western basin. This hypothesis is supported by a recent LGM ocean transport model, which simulates a deep-water layer (\~2.5--3.5 km depth) which has a higher proportion of northern-sourced water in the eastern basin than the western basin^[@CR59]^. Supporting evidence for a deep Nordic water mass in the North-East Atlantic comes from B/Ca and δ^13^C ratios in the depth range of 2--3.5 km, between glacial northern-sourced intermediate waters and AABW^[@CR60]^. The ^231^Pa/^230^Th observations at the eastern sites (3.1--4.3 km) potentially reflect the transition from the deep Nordic water to an increased influence of glacial AABW with depth (Fig. [4a](#Fig4){ref-type="fig"}). The persistently low ^231^Pa/^230^Th (\<0.058) observed in the east (Sierra Leone Rise, 7° N, 3.43 km) (Fig. [2c](#Fig2){ref-type="fig"}) supports the continual presence of deep-water transport exporting ^231^Pa from the low latitudes of the East Atlantic during the deglaciation, including HS1. It will be important to further examine the East-West difference suggested above, and proposed future work includes obtaining additional well-resolved ^231^Pa/^230^Th time-series from the mid-altitude and low latitudes of the eastern basin. In conclusion, we found that sedimentary ^231^Pa/^230^Th time-series from the West and deep high-latitude North Atlantic provide compelling evidence of coherent changes in deep Atlantic circulation strength coincident in timing with millennial-scale climate changes during the deglacial period (Fig. [2](#Fig2){ref-type="fig"}). This finding is consistent with previous reconstructions of the strengths of Iceland-Scotland Overflow Water^[@CR6]^ (major component of NADW) and Florida Current^[@CR7]^ (part of the Gulf Stream: shallow northward flow with a component that compensates for Atlantic deep-water transport southward). The ^14^C ventilation records of deep and intermediate water masses in the North Atlantic^[@CR3],[@CR41]^ are also in agreement with our reconstructions. Together, these data support an overall reduced AMOC rate during HS1 and YD, with resumption during the BA, as previously proposed^[@CR2]^. The remarkable consistency of the combined proxy records of ocean circulation rate has allowed us to provide a more in-depth analysis of the LGM-deglacial phasing of ocean circulation, ice sheet and climate changes. In particular, we have identified two phases of AMOC reduction during HS1 (Fig. [3](#Fig3){ref-type="fig"}, Supplementary Figs. [6](#MOESM1){ref-type="media"}, [7](#MOESM1){ref-type="media"}). The sequence of events proposed here is one that supports the mechanisms whereby an early AMOC decline (\~19--16.5 ka) was initiated and sustained by freshwater forcing from the Eurasian ice sheets^[@CR37],[@CR38]^ and icebergs. Subsurface ocean warming in the Labrador Sea^[@CR27]^ caused by this AMOC slowdown triggered episodes of iceberg rafting from the Laurentide ice sheet from \~16.5 to 15 ka, leading to the second phase of maximum AMOC reduction (Fig. [3](#Fig3){ref-type="fig"}). This scenario explains the late arrival of the main ice-rafting event from the Laurentide ice sheet (\~16.5--15 ka), characterised as the classic (sensu stricto) Heinrich Event 1^[@CR29],[@CR30]^. The dataset presented in this study also hints at a different circulation history between the east and west basins of North Atlantic, with potential for stronger LGM deep-water transport in the east. This interpretation is in accord with previous reconstructions suggesting a weaker deep western boundary current^[@CR56]^ and the active southward transport of a Nordic deep-water mass (2--3.5 km) in the eastern basin^[@CR60]^ during the LGM. This East-West contrast potentially persisted over the ensuing deglaciation at the low latitudes, providing some evidence that deep circulation in the Atlantic Ocean did not cease during the cold stadial, despite an overall weakened AMOC. The compilation of sedimentary ^231^Pa/^230^Th records presented in this study also provides a refined dataset for future ocean transport models^[@CR13],[@CR14]^ to further constrain the magnitudes and rates of AMOC changes during millennial-scale climate events. Together with transient modelling efforts, a detailed picture of AMOC forcings and feedbacks during abrupt climate changes is emerging, providing insights into future long-term changes in AMOC behaviour associated with global climate change. Methods {#Sec7} ======= The four new sedimentary ^231^Pa/^230^Th records {#Sec8} ------------------------------------------------ New sedimentary ^231^Pa/^230^Th data were collected on four deep sediment cores from the tropical Atlantic. Three of these cores are located to the west of the MAR: JC094-GVY14 (15.464° N, 50.992° W, 2.714 km water depth), EW9209-1JPC (5.907° N, 44.195° W, 4.056 km), and EW9209-3JPC (5.313° N, 44.26° W, 3.288 km), while the fourth is located to the east of the MAR: JC094-GVY01 (7.435° N, 21.7963° W, 3.426 km) (Supplementary Fig. [8](#MOESM1){ref-type="media"}). Core 1JPC and 3JPC were retrieved at Ceara Rise in 1992^[@CR61]^ and have been stored in the core repository of Woods Hole Oceanographic Institution (WHOI). GVY14 and GVY01 are gravity cores recently retrieved at Researchers Ridge and Sierra Leone Rise, respectively, during the JC094 cruise which took place in October--November of 2013^[@CR62]^. The ^231^Pa/^230^Th analysis of these four cores provides new information on deep circulation (2.5--4 km) at the low latitudes of the Atlantic and enables a comparison of lateral transport at \~3 km depth in the east and west basins. The two new cores from Researchers Ridge and Sierra Leone Rise GVY14 and GVY01 were put in stratigraphic context using planktonic δ^18^O data (Supplementary Fig. [9](#MOESM1){ref-type="media"}), and their age models for the last 25 kyr were developed using eleven and six ^14^C dates, respectively, (Supplementary Fig. [10, Supplementary Data 1](#MOESM1){ref-type="media"}). The existing age models for the Ceara Rise cores 1JPC and 3JPC^[@CR63],[@CR64]^ were also improved for this study, with five and nine new ^14^C dates, respectively (Supplementary Fig. [10, Supplementary Data 1](#MOESM1){ref-type="media"}). Core 1JPC has HS1 ^231^Pa/^230^Th measurements which are consistent with the picture painted by other western cores, but there is some chronological uncertainty over the HS1 interval (Supplementary Fig. [10](#MOESM1){ref-type="media"}), and therefore those data (Supplementary Fig. [12](#MOESM1){ref-type="media"}) are not included in the main results. Core GVY14 has low sedimentation rate (\<1 cm ka^−1^) over 20--15 ka (Supplementary Fig. [11](#MOESM1){ref-type="media"}) and thus lacks sufficient data resolution to document the HS1 shift in ^231^Pa/^230^Th. Sediment chronology {#Sec9} ------------------- Errors associated with sediment chronology are critical when comparing the new and previously published ^231^Pa/^230^Th time-series (Fig. [2](#Fig2){ref-type="fig"}, Supplementary Fig. [2](#MOESM1){ref-type="media"}). Therefore, we established the sediment chronologies of these cores using a consistent approach to provide the best possible age comparisons. All ^14^C ages were calibrated or re-calibrated against the atmospheric ^14^C curve after applying surface reservoir correction using OxCal version 4.2^[@CR65]^ and MarineCal13 calibration curve^[@CR50]^ to calculate the calendar ages of the samples. For the published records, we employed surface reservoir ages recommended by the original authors for the respective core sites. For the new cores, we used a constant surface reservoir age of 400 years (modern day global mean). We did not make changes to the non-^14^C chronological tie-points (δ^18^O, IRD count and correlation of sediment composition) employed in published age models (Supplementary Table [1](#MOESM1){ref-type="media"}). Age models for the sediment cores were developed or re-developed by interpolating ages between both the ^14^C and non-^14^C chronological tie-points. The resulting age models are dependent on the methods employed for age interpolation^[@CR66]^. Here, we briefly compared two such methods: OxCal Poisson deposition method which assumes random sedimentation rate with time defined by Poisson distribution^[@CR67]^, and a method assuming linear interpolation between the chronological tie-points^[@CR66]^. The two methods yield some differences in the age models that do not appear to be significant for our study (Supplementary Fig. [13](#MOESM1){ref-type="media"}). The age models displayed in the main figures are those derived with OxCal Poisson deposition method. Surface reservoir age variability is a significant source of uncertainty in ^14^C-based age models for core sites affected by sea ice melt during the stadials (HS1 and YD) and the LGM^[@CR68]^. This surface reservoir uncertainty needs to be taken into account when considering the phasing of AMOC, meltwater discharge and iceberg calving during the last glacial termination (Fig. [3](#Fig3){ref-type="fig"}). Among the six ^231^Pa/^230^Th time-series (Fig. [3b](#Fig3){ref-type="fig"}) used to interpret the phasing of AMOC and ice system changes, the age model for the Rockall basin core (50° N, 4.28 km) was not developed with ^14^C-derived chronological tie-points^[@CR16]^. The other five ^231^Pa/^230^Th records with ^14^C-derived chronological tie-points are located at lower latitudes (\<35° N/S) (Fig. [3b](#Fig3){ref-type="fig"}), with mean surface reservoir ages of 400 ± 400 years over the last 25 kyr^[@CR69]^. Sediment records of fluvial discharge and IRD discussed in this study are located at higher latitudes (\>45° N) (Fig. [3a](#Fig3){ref-type="fig"}), and substantial increases in surface reservoir ages during HS1 and YD (\>1000 years) have been found at these locations^[@CR69]^. Here, we test the significance of reservoir age uncertainty by computing several sets of sediment core age models using maximum and minimum surface reservoir values expected for high latitudes (\>45° N) and low latitudes (\<35° N/S) over the last 25 kyr^[@CR69]^. In the different scenarios, we found shifts in the absolute timing of AMOC, meltwater discharge and IRD events (Supplementary Fig. [14](#MOESM1){ref-type="media"}), but these shifts do not affect our main interpretations. In particular, the beginning of AMOC decline at early deglacial still coincides with intensified Eurasian meltwater discharge, and Laurentide-sourced IRD is only evident later at the peak of AMOC reduction (Supplementary Fig. [14](#MOESM1){ref-type="media"}). The age models displayed in the main figures are those developed using mean surface reservoir values derived for high latitudes (\>45° N) and low latitudes (\<35° N/S) over the last 25 kyr^[@CR69]^. Analytical techniques {#Sec10} --------------------- Sedimentary ^231^Pa/^230^Th analysis was carried out by measuring U, Th and Pa isotopes in the bulk sediment (Supplementary Data 2) using isotope dilution^[@CR70]^. Some of the U-series measurements for the two Ceara Rise cores 1JPC and 3JPC were carried out at the Woods Hole Oceanographic Institution (WHOI) on a Thermo-Finnigan Element 2 single collector, inductively coupled plasma-mass spectrometer (ICP-MS)^[@CR2]^, and at Lamont-Doherty Earth Observatory (LDEO) using a Thermo Scientific Element XR, single collector, inductively coupled plasma-mass spectrometer (ICP-MS)^[@CR2]^; and the remainder of the measurements were made using a new protocol set up at the Bristol Isotope Group lab of University of Bristol (UoB) (Supplementary Fig. [12](#MOESM1){ref-type="media"}). The new UoB protocol had been demonstrated to yield U-series isotope measurements of good precision and accuracy for sedimentary ^231^Pa/^230^Th analysis by producing 10 replicate measurements (average ^231^Pa: 1.24 ± 0.04 dpm g^−1^, 230Th: 5.48 ± 0.09 dpm g^−1^) of a homogenised Southern Ocean siliceous ooze which agree with the measurements made in GEOTRACES Th and Pa intercalibration study (reference value ^231^Pa: 1.25 dpm g^−1^, ^230^Th: 5.53 dpm g^−1^)^[@CR71]^. The UoB procedures for sample preparation consists of ^236^U, ^229^Th and ^233^Pa spike addition, sediment digestion, co-precipitation of actinide elements with Fe hydroxide and chemical separation of the U, Th and Pa via ion-exchange chromatography. Sample analyses at UoB were carried out using a Thermo-Finnigan Neptune, multicollector, ICP-MS. The methods employed followed those in Burke and Robinson (2012)^[@CR72]^ for U and Auro et al. (2012)^[@CR73]^ for Th and Pa, with ^229^Th and ^230^Th measured alternately on the Secondary Electron Multiplier (SEM), and ^231^Pa and ^233^Pa measured on a multi-ion counter (MIC) array to avoid significant build-up of dark noise on the SEM. Sedimentary ^231^Pa~ex~/^230^Th~ex~ was calculated by correcting the ^231^Pa and ^230^Th measurements for fractions that are supported by the decay of lithogenic and authigenic U by assuming a lithogenic ^238^U/^232^Th activity ratio of 0.6. We found that the choice of lithogenic ^238^U/^232^Th within the uncertainty of the derived Atlantic average value (0.6 ± 0.1, or more recently, 0.55 ± 0.16)^[@CR8],[@CR74]^ made no significant difference to the ^231^Pa/^230^Th values for our four new cores (example in Supplementary Fig. [12](#MOESM1){ref-type="media"}). The ^231^Pa and ^230^Th measurements were also corrected for radioactive decay to the age of sediment deposition. The disequilibria that might result from alpha recoil^[@CR74]^ was not explicitly taken into account in the ^231^Pa/^230^Th calculations, as we did not have an independent measure of this effect. Core 3JPC had four ^231^Pa/^230^Th measurements published in a previous study^[@CR12]^, and the values were re-calculated (Supplementary Fig. [12](#MOESM1){ref-type="media"}) to correct for radioactive decay using the revised sediment ages. Uncertainty in ^231^Pa/^230^Th measurements was propagated from internal errors associated with weighing, spike calibration and ICP-MS measurements using a Monte-Carlo method. Sediment biogenic opal content was analysed using the alkaline extraction method and molybdate-blue spectrophotometry following the procedures in Mortlock and Froelich^[@CR75]^. The vertical flux of bulk sediment was determined using the ^230^Th-normalisation method, which assumes that the rate of removal of ^230^Th onto particle surfaces is equal to the production rate of ^230^Th in the water column^[@CR76]^. The vertical opal flux was calculated by multiplying ^230^Th-normalised bulk sediment flux by the fraction of opal in the sediment. Data availability {#Sec11} ----------------- The data reported in this paper are listed in the Supplementary Information and archived in Pangaea database (10.1594/PANGAEA.890942). Electronic supplementary material ================================= {#Sec13} Supplementary Information Peer Review File Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 **Electronic supplementary material** **Supplementary Information** accompanies this paper at 10.1038/s41467-018-05312-3. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We would like to express our gratitude to the European Research Council and NERC Radiocarbon Facility for providing support to this study. The contribution of Jerry McManus was supported in part by the US-National Science Foundation and the Comer Science and Education Foundation. Ruza Ivanovic was supported by NERC grant NE/K008536/1. We want to thank Susan Brown-Leger and Marti Jeglinski for helping with ^231^Pa/^230^Th sample preparation and analysis at WHOI, and Marty Fleisher and L. Gene Henry for assistance at LDEO. We appreciate the help and advice from Tim Elliot, Chris Coath, Katherine Adena and Carolyn Taylor from the Bristol Isotope Group in setting up the sediment ^231^Pa/^230^Th analytical protocol at the UoB. We thank Ellen Roosen for help with sampling of sediment cores 3JPC and 1JPC at the WHOI core repository. Great thanks to the scientists and crew of JC094 for their assistance in collection of GVY01 and GVY14 sediment core from the tropical Atlantic Ocean. We also thank the undergraduate volunteer students/alumni from the UoB (Sam Lucas, Bryony Essex, Kasia Clarke) for their assistance in sample preparation. H.C.N., L.F.R., K.J.M. and A.W.J retrieved GVY01 and GVY14 sediment cores during the JC094 cruise and developed the age models for the two new cores. J.F.M provided some of the U-series measurements for the EW9209-1JPC and EW9209-3JPC sediment cores at the LDEO, and the remaining measurements were carried out at the UoB by H.C.N. All authors (H.C.N., L.F.R., J.F.M., K.J.M., A.W.J., R.F.I., L.J.G. and T.C.) contributed to the data interpretation and participated in the manuscript preparation. Competing interests {#FPar1} =================== The authors declare no competing interests.

Introduction ============ Impaired energy metabolism in neurons is integral to many central nervous system diseases. An acute, intense metabolic challenge during conditions such as stroke and brain trauma can induce irreversible neuronal damage.^[@bib1]^ By contrast, age-related chronic metabolic impairment is thought to contribute to the evolution of neurodegenerative diseases such as Alzheimer\'s disease, Parkinson\'s disease, and Huntington\'s disease.^[@bib2],\ [@bib3]^ Neurochemical cascades initiated by both acute and prolonged metabolic challenges can induce cell death via an elevation of proapoptotic proteins. These metabolic challenges can also activate genomic and proteomic responses promoting cell survival in the immediate phase as well as preparing the cell to deal with similar challenges in the future (conditioning).^[@bib4]^ Subcellular organelles display complex proteomic responses to impaired energy metabolism, with these responses ultimately determining the death or survival of the cell. In mitochondria, metabolic stress can disrupt energy generation, provoke mitochondrial swelling, and initiate apoptosis, all with potentially lethal consequences.^[@bib5]^ In response to cell stress, the endoplasmic reticulum (ER) detects and reconfigures misfolded proteins in a process known as the unfolded protein response (UPR). The UPR is integral to cell survival as excessive accumulation of these damaged proteins in the ER lumen can result in the suppression of protein synthesis and irreversible pathology.^[@bib6]^ In the nucleus and ribosomes, metabolic stress leads to a genomic response and the synthesis of proteins such as heat-shock proteins and trophic factors, which contribute to protein repair and neuronal survival.^[@bib7]^ The genomic response to metabolic challenge has been well characterized both *in vitro* and *in vivo*.^[@bib8],\ [@bib9],\ [@bib10]^ In the present study, we provide a comprehensive cellular proteomic response and explore the extent to which proteins are altered after a metabolic challenge. Oxygen glucose deprivation (OGD) induces metabolic stress in neurons in culture.^[@bib11],\ [@bib12]^ The intensity of the metabolic challenge can be controlled by adjusting experimental conditions such as the choice of cell line and alterations in the duration of OGD. Mass spectrometry-based proteomics provides an excellent platform to quantify OGD-dependent protein changes and to gain a fresh perspective on adaptations that occur within the proteome in response to metabolic stress.^[@bib13]^ We have used human SH-SY5Y neuroblastoma cells and label-free liquid chromatography mass spectrometry (LC-MS) to define the integrated proteomic response to metabolic stress that induces severe mitochondrial dysfunction but only limited cell death. The majority of proteins analyzed (800 out of 1,000) were not significantly altered by OGD. Protein increases after OGD was more than twice as frequent as was decreases, consistent with an integrated cellular survival response. Protein interactomes highlight coordinated proteomic responses after OGD in relation to mitochondrial energy production, ER stress, and ribosomal dysfunction. Materials and methods ===================== Reagents -------- All standard growth media used was obtained from Invitrogen (Paisley, UK), unless otherwise stated, and all procedures were conducted in compliance with all relevant legislation in the United Kingdom. Cells were routinely maintained in standard 75 cm^2^ cell culture-treated flasks (Greiner Bio One, Stonehouse, UK) within a continuous flow incubator at 37°C, with 5% CO~2~ in fully humidified air. All growth media and solutions were prewarmed to 37°C before contact with cells, unless otherwise stated. The acetonitrile and water used in the LC-MS buffer solutions were of HPLC grade (Fisher, Loughborough, UK). Formic acid was of 98% to 100% purity (Merck, Darmstadt, Germany) and trifluoroacetic acid was 99% sequencing grade (Sigma-Aldrich, Gillingham, UK). All other chemicals used in the processing of the protein samples were of reagent grade or better. The trypsin used for protein digestion was from Promega (Southampton, UK). Cell Culture and Oxygen Glucose Deprivation ------------------------------------------- SH-SY5Y cells (ATCC, Manassas, VA, USA) were cultured in Dulbecco\'s Modified Eagle Medium: Nutrient Mixture F12 (DMEM/F12) supplemented with 2 mmol/L ℒ-Glutamine, 100 U/mL Penicillin, 100 mg/mL Streptomycin, and fetal bovine serum. Cells were routinely maintained at 37°C, 5% CO~2~, with culture medium replaced biweekly. Oxygen glucose deprivation (for 3, 6, 9, 12, and 18 hour durations) was achieved through incubation of the cells at 37°C in 95% N~2~:5% CO~2~. Aglycemia was induced by replacing standard culture medium with glucose-free DMEM supplemented with fetal bovine serum, after a glucose-free wash step. Control cells underwent the same glucose-free wash step, followed by incubation with glucose containing media (4,500 mg/L glucose). The osmolality of the DMEM solutions used was within the range of 300 to 340 mOsm/kg for the no glucose solutions, and 320 to 360 mOsm/kg for the glucose containing media. All media were equilibrated overnight before experimentation in either normoxic or anoxic chambers. Oxygen content in the media was recorded in preliminary studies using a fiber optic dual oxygen-temperature sensing probe (OxyLite 2000 E Series, Oxford Optronix, Oxford, UK). A rapid oxygen depletion phase was recorded, with oxygen partial pressure falling exponentially from 140 to 7 mm Hg over the initial 3 hours of OGD. Oxygen depletion plateaued and remained severely hypoxic between 4 and 18 hours with a recorded average partial pressure of 2 mm Hg. Assessment of Mitochondrial Function (MTS Assay) ------------------------------------------------ Cells were seeded onto poly-𝒟-lysine coated flat-bottomed 96-well plates at a density of 125,000 cells per well overnight. After OGD, glucose-free DMEM was replaced with 200 *μ*L glucose-containing DMEM and cell viability was assessed using 20 *μ*L of CellTiter 96 AQueous One Solution Cell Proliferation Assay (MTS) (Promega, Madison, USA) as per the manufacturer\'s guidelines. Cells were incubated for 2 hours in glucose containing, normoxic media (37°C, 5% CO~2~) to allow residual mitochondrial function after OGD to be assessed using a Dynex MRX plate reader (Dynex Technologies Ltd., Worthing, UK) at 490*λ*. Assessment of Global Cell Death (Trypan Blue) --------------------------------------------- Trypan blue staining was carried out to assess the extent of cell death occurring after each OGD time point. Cells were collected immediately at the end of each time point by trypsinization, centrifugation (200 *g*, 7 minutes) and resuspension in glucose-containing DMEM. Cells were incubated with 0.1% trypan blue as per the manufacturer\'s instructions (Invitrogen). A hemocytometer was used to count stained and unstained cells. The percentage of stained cells in each OGD and time-matched control sample was calculated. Preparation of Protein Samples for Liquid-Chromatography Mass Spectrometry -------------------------------------------------------------------------- SH-SY5Y cells were treated with 6 hours OGD (*n*=6 independent samples per group) and 18 hours OGD (*n*=6), with a time-matched control for each sample. After OGD or 'control\' incubations, SH-SY5Y cells were immediately scraped and centrifuged (200 *g*, 7 minutes). Cell pellets were washed in Hanks Buffered Saline Solution and recentrifuged. Washed cell pellets were resuspended in 1 ml Hanks Buffered Saline Solution and centrifuged (6,000 *g*, 3 minutes). Supernatants were removed and the pellets snap frozen for temporary storage at −80°C. Cell pellets were defrosted and lysed in 100 *μ*L 8 M urea using a cordless motor pellet pestle (Sigma-Aldrich, Dorset, UK). Samples were then centrifuged (14,000 *g*, 15 minutes) and supernatants recovered. Total protein content was determined using the Pierce BCA Protein Assay Kit (Thermo-Fisher, Loughborough, UK). Protein extract (140 *μ*g) from OGD and control samples were digested as previously described.^[@bib14]^ Liquid Chromatography Mass Spectrometry --------------------------------------- Label-free intensity-based LC-MS was used to identify proteins differentially expressed after 6 and 18 hours OGD relative to time-matched controls. Capillary LC-MS data were acquired on an online system consisting of a micro-pump (1200 binary HPLC system, Agilent, Edinburgh, UK) coupled to a hybrid LTQ-Orbitrap XL instrument controlled through Xcalibur 2.0.7 (Thermo-Fisher). Peptides were reconstituted in 20 *μ*L of loading buffer before injection and 8 *μ*L was loaded. The HPLC separation was provided by a gradient between buffer A (97.5% H~2~O, 2.5% acetonitrile, 0.1% formic acid) and buffer B (90% acetonitrile, 10% H~2~O, 0.025% trifluoroacetic acid, 0.1% formic acid). The peptide mixture was separated on a 140-minute gradient. The mass spectrometer was operated in 'data-dependent mode\', with a single MS scan (400 to 2,000 *m*/*z*) followed by MS/MS scans in the linear ion trap on the five most abundant ions and excluded for 120 seconds. Protein Identification and Quantification ----------------------------------------- All LC-MS runs were analyzed using a label-free intensity-based approach with Progenesis (version 4.0 Nonlinear Dynamics, Newcastle, UK), under randomized conditions. The MS/MS data were searched using MASCOT versions 2.3 (Matrix Science Ltd, London, UK) against a human subset of the NCBI database (downloaded on 12 January 2011) with 34,281 sequences using a maximum missed-cut of 2. Variable methionine oxidation, STY phosphorylation, protein N-terminal acetylation, and fixed cysteine carbamidomethylation were used in all searches. LC-MS label-free quantification was performed using Progenesis 4.0 (Nonlinear Dynamics). Protein conflict (peptides shared between different proteins) was solved as followed: conflict resulting from multiple sequence assignment to the same peak; we kept only the sequence having the highest score. Conflict resulting from same peptide sequences assigned to different proteins, the assignment was singly attributed to the protein that had the highest number of peptides. Regarding the label-free quantitation, the total number of Features (i.e., intensity signal at a given retention time and *m*/*z*) was reduced to MSMS peaks with charge of 2, 3, or 4+ and we only kept the five most intense MSMS spectra per 'Feature\'. The subset of multicharged ions (2+, 3+, and 4+) was extracted from each LC-MS run and the ions intensities summed for normalization. Protein quantitation was performed as follows; for a specific protein, the associated unique peptide ions were summed to generate an abundance value. The measured protein abundances were transformed using an ArcSinH function (as the method of detection can generate a significant amount of near zero measurement for which a log transform is not ideal). The within-group means were calculated to determine the fold change and the transformed data were then used to calculate the *P* values using one-way analysis of variance. Significant changes in protein expression (*P*\<0.01) between control and OGD experimental groups were determined for each time point (6 and 18 hours) using Student\'s *t*-test. Analysis of Protein Subcellular Localization -------------------------------------------- All 958 proteins identified by LC-MS by at least two or more peptides were uploaded to the Database for Annotation, Visualization and Integrated Discovery (DAVID) (<http://david.abcc.ncifcrf.gov>^[@bib15]^) to determine subcellular localization based on gene ontology (GO). Protein clusters founded on subcellular localization with DAVID scores \>2 were used. Proteins significantly altered (*P*\<0.01) after 18 hours OGD were also assessed by DAVID according to subcellular localization. Interactome Analysis of Proteins Altered with Oxygen Glucose Deprivation ------------------------------------------------------------------------ Protein identifiers of all the proteins significantly altered after 18 hours OGD (*P*\<0.01) were uploaded with their corresponding fold change values to Ingenuity Pathway Analysis software (IPA; <http://www.ingenuity.com>). Interactomes were algorithmically generated based on direct relationships (physical interactions and associations) between eligible proteins. The interactomes are color coded with green nodes representing proteins that were downregulated after 18 hours OGD and red protein nodes representing proteins that were upregulated after 18 hours OGD. The shading of the interactome nodes is positively correlated with the magnitude of fold change. The IPA generates a score for each interactome, which is a putative measure of probability (see Deighton *et al*^[@bib16]^ for critical analysis). Subcellular Morphology with Electron Microscopy ----------------------------------------------- SH-SY5Y cells were exposed to OGD using the same paradigm as the LC-MS study. Samples were collected immediately after 3, 9, and 18 hours of OGD and fixed for a minimum of 20 hours in aldehyde fixative (4% paraformaldehyde and 5% glutaradehyde in 0.08 M sodium cacodylate buffer, pH 7.2). Once fixed, cells were centrifuged at 100 *g*, the supernatant removed and replaced with a 3% solution of agar (type VII, low gelling temperature; Sigma-Aldrich, Gillingham, UK). The gels were set at 4°C for 5 minutes before being cut into 1 mm^3^ blocks and processed for electron microscopy (EM). Processing, embedding, and staining were performed, as described previously.^[@bib17]^ Briefly, the blocks were cut into 70 nm thick sections, collected on 300 mesh copper grids, stained with Reynold\'s lead citrate and contrasted with uranyl acetate. The EM sections were evaluated on a JEOL CX-100 II transmission electron microscope (Tokyo, Japan). The evaluator was blinded to the treatment condition of all samples. Cells were chosen from grid squares in a predetermined pattern. Fifty cells were assessed from each treatment. For each cell, the condition of mitochondria and rough ER was assessed. Mitochondria were graded as '*normal*\' (cristae clearly visible across \>50% of the matrix), '*swollen*\' (with a proportion of the matrix lost) and '*abnormal*\' (with most/ the entire matrix lost). The ER in each cell was graded similarly, i.e., '*normal*\' (where ER outer membranes ran parallel to each other), '*swollen*\' (where modest swelling of ER outer membrane was present), or '*abnormal*\' where ER displayed extensive membrane swelling and distortions. A Chi-Squared 3 × 2 matrix was used to test the null hypothesis that OGD would have no effect on the ultrastructure of organelles. Results ======= Mitochondrial Function, Cell Viability, and Ultrastructure after Oxygen Glucose Deprivation ------------------------------------------------------------------------------------------- Normal mitochondrial function was significantly reduced in SH-SY5Y cells after OGD relative to time-matched control cells (*P*\<0.0001, at each time point analyzed), with longer durations of OGD inducing more severe mitochondrial dysfunction ([Figure 1A](#fig1){ref-type="fig"}). Viability of SH-SY5Y cells treated with OGD was quantified using Trypan Blue staining ([Figure 1B](#fig1){ref-type="fig"}). Minimal cell death was detected after OGD, with 91% of cells remaining viable after 18 hours OGD. Ultrastructural morphologic changes in subcellular organelles were analyzed relative to time-matched controls by EM after OGD. With increasing OGD duration, the number of cells displaying abnormal ER was significantly increased (Chi-Square, *P*\<0.01) ([Figures 2A and 2C](#fig2){ref-type="fig"}). SH-SY5Y cells also showed changes in mitochondrial morphology with increasing OGD duration ([Figures 2B and 2D](#fig2){ref-type="fig"}). It should be noted that even after 18 hours OGD exposure, only 26% of cells displayed abnormal ER, and only 34% had abnormal mitochondria. Global Proteomic Response after Oxygen Glucose Deprivation (6 and 18 hours) --------------------------------------------------------------------------- A total of 958 proteins were identified across all samples with two or more peptides by LC-MS ([Supplementary Table 1](#sup1){ref-type="supplementary-material"}). Minimal protein changes were observed after 6 hours OGD, with only 14 proteins significantly altered in expression level (*P*\<0.01). After 18 hours OGD, 193 proteins were significantly altered in expression level (*P*\<0.01) with 130 proteins increased and 63 proteins decreased. For 765 proteins, no significant difference was observed between control and OGD samples after 18 hours OGD. Subcellular distribution of all the 958 proteins identified by LC-MS was examined using DAVID software. Proteins were detected from across many subcellular compartments, including the nuclear lumen, cytoskeleton, mitochondria, ER, golgi apparatus, ribosomes, and vesicles ([Figure 3A](#fig3){ref-type="fig"}). The ascribed subcellular locations of the significantly altered proteins with OGD (18 hours) indicate that organelles are not uniformly affected by the metabolic challenge. None of the 197 detected nuclear lumen proteins were significantly altered after OGD, while 41% of the 56 detected ER proteins and 32% of the 118 detected mitochondrial proteins were significantly altered ([Figure 3B](#fig3){ref-type="fig"}). The significantly increased and decreased proteins after OGD (18 hours) are listed by subcellular location in [Table 1](#tbl1){ref-type="table"} and [Table 2](#tbl2){ref-type="table"}, respectively. Functional protein--protein interactions for the 193 proteins altered with 18 hours OGD were defined using IPA software. IPA-generated 10 protein interactomes ([Supplementary Table 2](#sup1){ref-type="supplementary-material"}) and three high-scoring interactomes (IPA scores \>40) are discussed in context below. These interactomes highlight a shift in mitochondrial energy production, enhanced ER stress, and ribosomal dysfunction. Mitochondrial Protein Alterations after Oxygen Glucose Deprivation (18 hours) ----------------------------------------------------------------------------- In all, 38 mitochondrial proteins (out of 118 detected; assigned using the GO term *mitochondrion* GO:0005739) were significantly altered after OGD (18 hours): 29 were increased and 9 were decreased. VDAC1 and TOM22, proteins involved in mediating the movement of metabolites across the mitochondrial membrane, were significantly reduced after OGD. UQCRC1 and UQCRC2, integral components of complex III of the electron transport chain (ETC), were also significantly reduced ([Figure 4A](#fig4){ref-type="fig"}). Coordinated changes in other regions of the ETC were observed after 18 hours OGD treatment: ETC complex V proteins ATP5J and ATP5A1, and ETC complex I proteins NDUFS3 and NDUFA5 were all significantly increased (*P*\<0.01). Numerous mitochondrial proteins that have a role in lipid metabolism were also found to be altered after 18 hours OGD. ACADM, the enzyme involved in the first step of fatty acid beta oxidation was significantly decreased (*P*\<0.01); and ECH1, HSD17B10 and ACAA2, enzymes downstream in the beta-oxidation process were significantly increased. ACAT1, a mitochondrial protein involved in ketone metabolism was also significantly increased. The highest scoring protein interactome generated by IPA features some of the alterations in energy production described above, namely increases in Complex V proteins (including ATP5J and ATP5A1) and decreases in Complex III proteins (UQCRC1 and UQCRC2) ([Figure 4B](#fig4){ref-type="fig"}). This *Energy Production Interactome* also highlights interactions between altered components of the ETC and other cellular machinery. Both UQCRC1 and UQCRC2 interact with RTN4, a protein predominantly associated with regulating ER function that was significantly upregulated after OGD ([Figure 4A](#fig4){ref-type="fig"}). UQCRC2 is also shown to interact with upregulated RUVBL2, a cytoskeletal protein involved in numerous cellular activities including transcription and cellular transformation ([Figure 4A](#fig4){ref-type="fig"}). The *Energy Production Interactome* also contains a cluster of ribosomal proteins that are distinct but intersect with the ribosomal interactome described below. Endoplasmic Reticulum Protein Alterations after Oxygen Glucose Deprivation (18 hours) ------------------------------------------------------------------------------------- In all, 23 ER proteins (out of 56 detected; assigned using the GO term *Endoplasmic reticulum* GO:0005783) were significantly altered after 18 hours OGD: 19 were increased and 4 were decreased. A number of proteins centrally involved in the UPR were significantly increased, including, HSPA5, HSP90B1, PDIA3, and PDIA4 (*P*\<0.01) ([Figure 5A](#fig5){ref-type="fig"}). Other proteins including CALR, UGGT1, ERP29, ERP44, and VCP that are integral to ER stress response and protein folding were significantly increased. Multiple proteins that have a role in protein degradation were significantly altered after OGD (18 hours). Protein subunits of the 20S and 19S core structures of the 26S proteasome (PSMA5, PSMC3, PSMC4, PSMD1, PSMD2, PSMD6, and PSMD11) were increased. These proteasomal proteins, along with proteins involved in the UPR and heat-shock response after OGD dominate the high scoring, *Protein Folding and Degradation Interactome* generated by IPA ([Figure 5B](#fig5){ref-type="fig"}). Ribosomal Protein Alterations after Oxygen Glucose Deprivation (18 hours) ------------------------------------------------------------------------- In all, 11 ribosomal proteins (out of 82 detected; assigned using the GO term *Ribosome* GO:0005840) were significantly altered after 18 hours OGD: 1 was increased and 10 were decreased. The decreased ribosomal proteins were associated with both the 60S (RPL38, RPL23, RPL35, and RPL13) and 40S (RPS1, RPS14, RPS23, RPS9, and RPS17) ribosomal subunits ([Figure 5C](#fig5){ref-type="fig"}). These downregulated ribosomal proteins feature prominently in another of the high scoring protein interactomes generated by IPA ([Figure 5D](#fig5){ref-type="fig"}). An additionally important feature of this *Ribosomal Interactome* is the cluster of upregulated peroxidases (PRDX1, PRDX2, and PRDX5) ([Figure 5C](#fig5){ref-type="fig"}) involved in redox homeostasis and protecting neurons from oxidative insults (for a review, see Bell and Hardingham, 2011^[@bib18]^). Discussion ========== This study provides the first comprehensive analysis of the proteomic and morphologic response of human SH-SY5Y cells to prolonged metabolic challenge. The robustness of SH-SY5Y cells to external stressors has allowed the evolving proteomic changes occurring in response to OGD induced metabolic stress to be defined, in the absence of cell death. The observed protein changes provide insight into the complex molecular alterations underpinning cellular adaptation to metabolic stress. The lack of acute ischemic cell death in our *in vitro* model of neuronal metabolic stress makes the proteomic results particularly pertinent to mild, sustained ischemia challenges and the compromised energy metabolism associated with slowly evolving neurodegenerative disorders such as Alzheimer\'s disease.^[@bib3]^ Proteomics has showed alterations and interactions in mitochondrial, ER, and ribosomal proteins after OGD. The ETCproteins displayed bidirectional changes (Complex I and V proteins increased and Complex III proteins decreased) in response to OGD. Other mitochondrial proteins involved in lipid and ketone metabolism, including AD-related HSD17B10 (more commonly known as ABAD), were also increased after OGD (18 hours). The ER stress response proteins along with proteins involved in proteasomal/ubiquitination degradation also formed an important part of the upregulated proteomic response to the metabolic challenge of OGD. A significant mitochondrial proteomic response was observed after OGD. Two of the most markedly decreased proteins after 18 hours OGD were subunits 1 and 2 (UQCRC1 and UQCRC2) of ETC complex III. Complex III is the third component of the ETC located within the inner mitochondrial membrane. Complex III mediates the transfer of electrons from ubiquinone to cytochrome C, and in doing so translocates a proton into the inner-membrane space and contributes to the generation of the electrochemical gradient used for the production of ATP.^[@bib19]^ Complex III is also intrinsically involved in the generation of reactive oxygen species from the mitochondria during hypoxia, and these reactive oxygen species work to stabilize HIF1*α* and initiate protein transcription.^[@bib20]^ A significant decrease in UQCRC1 and UQCRC2 might be an attempt by the cells to reduce the production of reactive oxygen species from the mitochondria under stress conditions. Interestingly, UQCRC1 has also been recorded as significantly decreased in the brains of AD patients.^[@bib21]^ The similar decrease in UQCRC1 expression observed in AD patients and cells *in vitro* after OGD may provide a link between chronic oxidative challenge and the onset and progression of AD. The *Energy Production Interactome* ([Figure 4B](#fig4){ref-type="fig"}) also highlighted putative functional protein--protein interactions between UQCRC1, UQCRC2, and RTN4 (a protein increased after 18 hours OGD). RTN4 is increased in hippocampal neurons of AD patients^[@bib22]^ and is known to interact with BACE1, reducing AB40 and AB42 levels.^[@bib23]^ The dysregulation of RTN4 after a metabolic challenge (18 hours OGD), its interaction with complex III proteins, and its potential role in APP processing, highlights RTN4 protein as a compelling target for future investigation. Mitochondrial complex I and V proteins were significantly increased in expression after OGD in sharp contrast to the reduction in complex III proteins after OGD. Complex I is the first protein complex of the ETC and initiates the transfer of electrons required for oxidative phosphorylation. The process is initiated through binding and oxidation of NADH, with the concurrent transfer of two electrons from NADH to ubiquinone, and the translocation of four protons into the intermembrane space.^[@bib24]^ An increase in Complex I proteins after OGD may reflect a compensatory mechanism to counteract the impact of decreased complex III proteins on energy production. Similarly, the increase in Complex V proteins may reflect a compensation for downregulated Complex III proteins or a general remodeling of the ETC, as has previously been shown in studies investigating ischemic preconditioning.^[@bib25]^ Functional assays of the five complexes that constitute the ETC, in ischemic conditions will need to be carried out to confirm whether such remodeling occurs. Determining the regulatory mechanisms of these protein changes could provide valuable insight that would allow therapeutic intervention to alter the course of mitochondrial dysfunction after OGD and metabolic challenges *in vivo*. Concurrent with the dysregulation of the oxidative phosphorylation pathway, OGD (18 hours) induced an increase in mitochondrial proteins involved in lipid and ketone metabolism. The increase in these proteins (ACAT1 in ketone metabolism, and ACADM, ECH1, HSD17B10, and ACAA2 in lipid metabolism) indicates a shift toward using lipid and ketone bodies as alternative sources of energy for SH-SY5Y cells faced with glucose restriction. The availability of such alternative energy sources is likely to be important in preventing cells hitting the point where energy provision is insufficient for cell survival, leading to a downward spiral of metabolic and cognitive decline. A recent study using a triple transgenic model of AD showed an induction of ketogenesis after a 2-deoxy-𝒟-glucose diet was introduced. This correlated with an increase in nonamyloidogenic APP processing pathways, reduced AD pathology and reduced oxidative stress,^[@bib26]^ indicating a potential therapeutic role for enhanced ketone metabolism. To use ketones as an alternative energy source in the brain, acetoacetate and *β*-hydroxybutyrate (the main ketone bodies used for energy) need to be synthesized from acetyl-coA. Acetyl-coA is released after beta oxidation of fatty acids, a four step process within the mitochondrial matrix requiring beta-oxidation enzymes at each stage (namely, ACAD, ECH1, HSD17B10, and ACAA2). Each of these beta-oxidation enzymes, as noted above, was significantly altered after 18 hours OGD. Of note, the third beta-oxidative enzyme HSD17B10 (ABAD) is now emerging as an important player in AD. ABAD has been shown to directly interact with A*β* in the mitochondria of AD patients and AD transgenic mouse models. Inhibition of the ABAD-A*β* interaction suppresses induced apoptosis by increasing levels of A*β*, and conversely overexpression of ABAD in transgenic mice with high levels of A*β* has been shown to cause heightened neuronal oxidative stress and memory impairments.^[@bib27]^ The direct involvement of ABAD in AD pathophysiology makes ABAD an interesting candidate to investigate further in terms of its upregulation upon metabolic challenge. Pore forming proteins VDAC1 and TOM22 were significantly downregulated after OGD. VDAC1 is thought to be involved in the formation of the mPTP, which in turn allows calcium and solutes \<1.5 kDa to pass across the mitochondrial membranes, resulting in osmotic swelling, loss of the mitochondrial membrane potential, and damage to the electron transport respiratory chain.^[@bib28]^ Mitochondrial swelling after OGD was clearly showed in the human cells under the electron microscope ([Figure 2D](#fig2){ref-type="fig"}). The role that these mitochondrial pore forming proteins have in mediating mitochondrial swelling after OGD needs to be defined to facilitate the discovery of novel pathways that might be manipulated to restore mitochondrial normality. A major proteomic response in the ER was identified after OGD. The ER stress response (also known as the unfolded protein response, UPR) is a cellular reaction activated to clear misfolded proteins that accumulate upon ER stress, thus returning the cell to a state of proteostasis.^[@bib29]^ Two central molecular chaperones, HSPA5 and HSP90B1, involved in the early and late stages of the UPR respectively were upregulated after OGD in SH-SY5Y cells (as well as numerous other UPR proteins and ER stress proteins) ([Figure 5A](#fig5){ref-type="fig"}). These data are consistent with recent results from OGD in mixed primary cortical cultures where increases in both HSPA5 and HSP90B1 were detected.^[@bib30]^ The induction of the UPR in response to OGD was also implied in the electron microscopic data. The ER showed a progressive morphologic swelling that significantly increased after longer durations of OGD exposure compared with time-matched controls (*P*\<0.01, [Figure 2C](#fig2){ref-type="fig"}). An expanded ER forms an important part of the UPR as it enables larger numbers of misfolded proteins to be incorporated within the ER membrane, thus increasing cellular capacity to deal with abundant protein damage.^[@bib31]^ A larger ER volume also reduces the concentration of protein intermediates, therefore reducing the risk of protein aggregate formation.^[@bib32]^ The morphologic observation of enlarged ER along with the significant increase in UPR proteins HSPA5 and HSP90B1 detected in our proteomic data provides telling insights into how the human neuronal cells respond to the metabolic challenge of OGD. In accord with activation of the UPR, an upregulation of numerous proteins involved in proteasomal degradation was highlighted by the *Protein Folding and Degradation Interactome*. Subunits of the 20S and 19S (core and regulatory) particles that constitute the 26S proteasome were increased, suggesting 26S proteasomal degradation to be a fundamental mechanism for degrading the misfolded proteins that accumulate in response to metabolic challenge. There has been a long-standing hypothesis that protein aggregates in diseased brain impair the protein degradation function of the 26S proteasome.^[@bib33],\ [@bib34],\ [@bib35]^ Our proteomic data show proteasomal proteins to be upregulated after a metabolic challenge. It is therefore possible to hypothesize that any proteasomal response to metabolic challenge might be adversely affected in the presence of protein aggregates associated with neurodegenerative diseases such as AD. In this situation, the cellular capacity to cope with metabolic stress would be impaired, potentially leading to a deterioration in cellular function. A coordinated response after OGD (18 hours) was also observed in the ribosomal proteome. Numerous ribosomal proteins that constitute the core ribosomal machinery were downregulated, and in line with this, several eukaryotic initiation factors central to protein synthesis were also decreased. It is well established that cells undergoing severe metabolic stress experience a downregulation of protein translation, which can be reversible in more resistant brain areas, and irreversible in vulnerable regions.^[@bib36]^ Further work involving a recovery phase would need to be undertaken to understand whether the decrease in protein synthesis machinery is a survival mechanism to conserve energy, or whether it ultimately pushes the cells down a cell death route. The proteomic data does suggest that activation of the UPR correlates with a decrease in global protein synthesis, perhaps to deal with the accumulation of misfolded proteins.^[@bib37]^ Substrate deprivation both *in vivo* and *in vitro* has long been recognized to evoke a marked genomic response. Whether this genomic response translates into distinct alterations in protein expression has been a matter of debate. The present study clearly shows the considerable adaptive capacity of the SH-SY5Y proteome to a prolonged metabolic challenge *in vitro*, but caution must be exercised in the extrapolation of the present data to cerebral ischemia *in vivo*. No single cell culture system can model the complexity of the response of cerebral tissue to substrate deprivation. Astrocyte cultures and neuronal cells of neuroblastoma lineage (such as SH-SY5Y cells) can tolerate prolonged OGD, whereas primary neuronal and oligodendrocyte cultures are irreversibly damaged within a few hours.^[@bib38],\ [@bib39]^ The temporal susceptibility of these different cultures to metabolic challenge, their ability to generate energy from various sources (SH-SY5Y cells have considerable capacity to perform anaerobic glycolysis^[@bib40]^) and their intrinsic energy demands will all impact on the evoked proteomic response to OGD. Additional proteomic and biochemical analyses (not just LC-MS as in the present study) including protein--protein interaction assays, membrane potential recordings, and *in vitro* ATP levels across a range of different cell types would enhance our understanding of the true nature of the metabolic challenge posed by OGD. These studies would also provide more detail as to the subtleties of the proteomic changes, not only in terms of abundance, but also in cellular and subcellular localization, and the interacting partners of altered proteins. [Supplementary Information](#sup1){ref-type="supplementary-material"} accompanies the paper on the Journal of Cerebral Blood Flow & Metabolism website (http://www.nature.com/jcbfm) AGH is supported by the Medical Research Council. The research is supported by Age UK as part of the Disconnected Mind programme and performed under the aegis of the Centre for Cognitive Aging and Cognitive Epidemiology (part of the Lifelong Health and Wellbeing initiative of UK Research Councils). TLB and LEK are funded by SynthSys which is a Centre for Integrative Systems Biology (CISB) funded by BBSRC and EPSRC; reference BB/D019621/1. The authors declare no conflict of interest. Supplementary Material {#sup1} ====================== ###### Click here for additional data file. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} ###### Proteins significantly increased with OGD (18 hours) *Uniprot accn no.* *Gene name* *Protein name* P *value* *Fold change* *Protein function* ------------------------- ------------- --------------------------------------------------- ----------- --------------- ---------------------------- *Mitochondria*  P30048 *PRDX3* Peroxiredoxin III 0.0006 1.57 Cell redox homeostasis  P61604 *HSPE1* 10 kDa heat-shock protein 6.1E-06 1.51 Stress response  P18859 *ATP5J* ATP synthase-coupling factor 6 0.0016 1.44 ETC complex V  O75489 *NDUFS3* NADH-ubiquinone oxidoreductase 30 kDa subunit 0.0030 1.38 ETC complex I  P24752 *ACAT1* Acetyl-CoA acetyltransferase 9.1E-05 1.38 Ketone body metabolism  P09622 *DLD* Dihydrolipoyl dehydrogenase 0.0051 1.36 Gycine cleavage  Q9UJZ1 *STOML2* Stomatin-like protein 2 0.0052 1.36 Receptor binding  P09972 *ALDOC* Fructose-bisphosphare aldolase C 0.0061 1.34 Glycolysis  P38606 *ATP6V1A* V-type proton ATPase catalytic subunit A 0.0067 1.33 ATP hydrolysis transport  P13804 *ETFA* Electron transfer flavoprotein subunit alpha 0.0005 1.33 Respiratory ETC  P30084 *ECHS1* Enoyl-CoA hydratase 0.0043 1.33 Fatty acid metabolism  P40926 *MDH2* Malate dehydrogenase 0.0009 1.32 Citric acid cycle  P10809 *HSPD1* 60 kDa heat-shock protein 0.0012 1.31 Protein folding  P62258 *YWHAE* 14-3-3 protein epsilon 0.0056 1.30 Apoptosis  P04181 *OAT* Orthinine aminotransferase 0.0012 1.30 Amino acid biosynthesis  Q13011 *ECH1* Delta (3,5)-delta(2,4)-dienoyl-CoA isomerase 0.0048 1.28 Lipid metabolism  Q99714 *HSD17B10* 3-hydroxyacyl-CoA dehydrogenase type 2 0.0037 1.27 Lipid metabolism  P07954 *FH* Fumarate hydratase 0.0038 1.27 Citric acid cycle  P30044 *PRDX5* Peroxiredoxin 5 0.0014 1.26 Redox homeostasis  P42765 *ACAA2* 3-ketoacyl-CoA thiolase 0.0067 1.26 Lipid metabolism  P25325 *MPST* 3-mercaptopyruvate sulfurtransferase 0.0012 1.25 Response to Toxins  P30086 *PEBP1* Phosphatidylethanolamine binding protein 1 0.0067 1.24 Protease inhibition  P25705 *ATP5A1* ATP synthase subunit alpha 0.0019 1.21 ETC complex V  Q06830 *PRDX1* Peroxiredoxin 1 0.0009 1.20 Redox homeostasis  P43686 *PSMC4* 26S protease regulatory subunit 6B 0.0007 1.20 Protein degradation  Q16718 *NDUFA5* NADH-ubiquinone oxidoreductase 13 kDa subunit 0.0022 1.20 ETC complex I  O00154 *ACOT7* Cytosolic acyl-coenzyme A thioester hydrolase 0.0009 1.19 Acyl-CoA hydrolysis  Q00610 *CLTC* Clathrin heavy chain 1 4.9E-05 1.16 Endocytosis  P12277 *CKB* Creatine Kinase B type 0.0027 1.16 Creatine metabolism *Endoplasmic reticulum*  O60568 *PLOD3* Procollagen-lysine 2-oxoglutarate 5-dioxygenase 3 7.7E-05 1.87 Lysyl hydroxylation  P27797 *CALR* Calreticulin 1.0E-07 1.56 Protein folding  P11021 *HSPA5* Heat-shock protein 5 3.9E-05 1.53 Protein complex assembly  P14625 *HSP90B1* Heat-shock Protein 90 kDa beta member 1 5.5E-05 1.49 Protein transport  P30101 *PDIA3* Protein disulfide isomerase A3 2.7E-05 1.45 Protein folding  Q16799 *RTN1* Reticulon 1 0.0055 1.41 Neuron differentiation  Q8NBS9 *TXNDC5* Thioredoxin-domain containing protein 5 0.0017 1.40 Cell redox homeostasis  P13667 *PDIA4* Protein disulfide isomerase A4 1.4E-05 1.38 Cell redox homeostasis  Q9Y2B0 *CNPY2* Protein canopy homolog 2 0.0051 1.38 Protein binding  Q15084 *PDIA6* Protein disulfide isomerase A6 6.1E-06 1.37 Chaperone  P54920 *NAPA* Alpha soluble NSF attachment protein 0.0003 1.35 ER-golgi vesicle transport  P14314 *PRKCSH* Glucosidase 2 subunit beta 0.0016 1.33 Glycoprotein formation  Q9NQC3 *RTN4* Reticulon 4 0.0033 1.32 Neurogenesis/apoptosis  Q14697 *GANAB* Neutral alpha-glucosidase AB/GNAB protein 0.0018 1.27 Glycoprotein cleavage  P07237 *P4HB* Protein disulfide isomerase 0.0073 1.25 Redox homeostasis  Q9BS26 *ERP44* Endoplasmic reticulum resident protein 44 0.0007 1.23 ER stress response  Q9NYU2 *UGGT1* UDP glucose:glycoprotein glucosyltransferase 1 0.0074 1.23 ER protein folding  P30040 *ERP29* Endoplasmic reticulum resident protein 29 0.0003 1.23 ER protein folding  P55072 *VCP* Valosin-containing protein 0.0069 1.15 ER stress response *Ribosomes*  P69905 *HBA1* Hemoglobin subunit alpha 0.0003 1.97 O~2~ transport *Vesicle*  O15240 *VGF* Neurosecretory protein VGF 0.0035 1.39 Cell--cell interactions  P08133 *ANXA6* Annexin 6 0.0075 1.18 Ca2+ binding  P21281 *ATP6V1B2* V-type proton ATPase subunit B 0.0098 1.16 H+ ion transport  P07900 *HSP90AA1* Heat Shock Protein HSP90-alpha 0.0030 1.11 Stress response  Q9BTT0 *ANP32E* Acidic leucine-rich nuclear phosphoprotein 32E 0.0034 1.08 Phosphatase inhibitor *Cytoskeleton*  P52907 *CAPZA1* F-actin capping protein subunit alpha 1 0.0039 1.35 Protein complex assembly  P16949 *STMN1* Stathmin 0.0029 1.32 Microtubule disassembly  Q9Y230 *RUVBL2* RuvB like 2 0.0037 1.31 Transcription  P47756 *CAPZB* F-actin-capping protein subunit beta 3.2E-05 1.25 Protein complex assembly  Q03252 *LMNB2* Lamin B2 0.0076 1.24 Cytoskeleton regulation  Q16352 *INA* Alpha-internexin 4.9E-05 1.24 Neurogenesis  P07437 *TUBB* Tubulin beta chain 0.0016 1.23 Microtubule constituent  P36405 *ARL3* ADP-ribosylation factor-like protein 3 0.0071 1.23 Cell division  P61163 *ACTR1A* Alpha centractin 0.0029 1.23 Vesicle mediated transport  Q9UJU6 *DBNL* Debrin-like protein 0.0095 1.18 Endocytosis  Q16658 *FSCN1* Fascin 0.0018 1.18 Cytoskeleton organization  P62158 *CALM* Calmodulin 0.0059 1.17 Calcium modulation  Q9BZK7 *TBL1XR1* F-box like/WD repeat containing protein TBL1XR1 0.0046 1.17 Protein degradation  P49773 *HINT1* Histidine triad nucleotide-binding protein 1 0.0002 1.15 Signal transduction  P43487 *RANBP1* Ran-specific GTPase-activating protein 0.0029 1.15 Signal transduction *Spliceosome*  Q07955 *SRSF1* Serine/Arginine-rich splicing factor 1 1.1E-05 1.42 mRNA splicing  Q01081 *U2AF1* Splicing factor U2AF 35 kDa subunit 0.0062 1.23 mRNA splicing  Q15029 *EFTUD2* 116 kDa U5 small nuclear ribonucleoprotein 0.0069 1.19 mRNA splicing *Proteosome*  Q15008 *PSMD6* 26s proteasome non ATPase regulatory subunit 6 0.0053 1.61 Protein degradation  P17980 *PSMC3* 26S protease regulatory subunit 6A 0.0071 1.27 Protein degradation  Q13200 *PSMD2* 26S proteasome non-ATPase regulatory subunit 2 0.0005 1.22 Protein degradation  P54727 *RAD23B* UV excision protein RAD23 homolog B 0.0009 1.22 Protein degradation  Q99460 *PSMD1* 26S proteasome non-ATPase regulatory subunit 1 0.0002 1.18 Protein degradation  O00231 *PSMD11* 26S proteasome non-ATPase regulatory subunit 11 0.0008 1.18 Protein degradation  P28066 *PSMA5* Proteasome subunit alpha type 5 0.0036 1.16 Protein degradation ER, endoplasmic reticulum; ETC, electron transport chain; OGD, oxygen glucose deprivation. Proteins are grouped according to subcellular compartment based on gene ontology and ranked according to magnitude of fold change. The significance threshold was set *a priori* at *P*\<0.01. ###### Proteins Significantly Decreased with OGD (18 hours) *Uniprot accn no.* *Gene name* *Protein name* P*-value* *Fold change* *Protein function* ------------------------- ------------- ------------------------------------------------------- ----------- --------------- ------------------------------ *Mitochondria*  P21796 *VDAC1* Voltage-dependent anion selective channel protein 1 0.0079 0.26 Ion transport  P22695 *UQCRC2* Cytochrome b-c1 complex subunit 2 0.0017 0.31 ETC complex III  P31930 *UQCRC1* Cytochrome b-c1 complex subunit 1 0.0097 0.46 ETC complex III  Q9NS69 *TOM22* Translocase of outer membrane 22 kDa subunit 0.0027 0.46 Protein transport  Q9Y3E5 *BIT1* Peptidyl-tRNA hydrolase 2 0.0038 0.51 Apoptosis  P02786 *TFRC* Transferrin receptor protein 1 0.0002 0.58 Endocytosis  O75534 *CSDE1* Cold shock domain containing protein E1 1.0E-06 0.62 Transcription  Q9Y6H1 *CHCHD2* Coiled-coil helix--coiled-coil helix domain protein 2 0.0099 0.70 *unknown*  P11310 *ACADM* Medium chain specific acyl-CoA dehydrogenase 0.0084 0.79 Lipid metabolism *Endoplasmic reticulum*  Q81V08 *PLD3* Phospholipase D3 0.0016 0.61 Lipid degradation  Q96AG4 *LRRC59* Leucine rich repeat-containing protein 59 0.0048 0.64 ER membrane interactions  P13637 *ATP1A3* Na+/K+ ATPase alpha (III) subunit 0.0008 0.67 ATP hydrolysis catalyst  P04843 *RPN1* Ribophorin 1 0.0094 0.68 Glycosyltransferase activity *Ribosomes*  P63173 *RPL38* 60s ribosomal protein L38 0.0001 0.65 Protein biosynthesis  P62249 *RPS16* 40S ribosomal protein S16 7.9E-05 0.70 Protein biosynthesis  P35544 *FAU* Ubiquitin-like protein FUB 1 0.0012 0.70 Ubiquitination modulation  P62263 *RPS14* 40S ribosomal protein S14 0.0004 0.74 Protein biosynthesis  P62829 *RPL23* 60S ribosomal protein L23 5.3E-05 0.74 Protein biosynthesis  P62266 *RPS23* 40S ribosomal protein S23 0.0004 0.76 Protein biosynthesis  P46781 *RPS9* 40S ribosomal protein S9 0.0002 0.78 Protein biosynthesis  P08708 *RPS17* 40S ribosomal protein S17 0.0004 0.80 Protein biosynthesis  P42766 *RPL35* 60S ribosomal protein L35 0.0018 0.82 Protein biosynthesis  P26373 *RPL13* 60S ribosomal protein L13 0.0004 0.84 Protein biosynthesis *Vesicle*  P07355 *ANXA2* Annexin A2 0.0038 0.49 Ca^2+^ binding  P62491 *RAB11A* Ras-related protein Rab 11A 0.0026 0.76 Endocytic cycling  P07858 *CTSB* Cathepsin B 0.0053 0.85 Protein degradation *Cytoskeleton*  P06493 *CDK1* Cyclin-dependent kinase 1 0.0022 0.57 Cell-cycle control  P28289 *TMOD1* Tropomodulin 1 0.0005 0.64 Cytoskeleton organization  Q01518 *CAP1* Adenylyl cyclase-associated protein 1 4.7E-06 0.70 Signal transduction  Q9UQE7 *SMC3* Structural maintenance of chromosomes protein 3 0.0084 0.71 Cell cycle  P18206 *VCL* Viniculin 0.0067 0.78 Cell adhesion  Q01082 *SPTBN1* Spectrin beta chain brain 1 0.0092 0.81 Cytoskeleton movement *Spliceosome*  P17844 *DDX5* Probable ATP-dependent RNA helicase DDX5 2.9E-06 0.58 mRNA processing  Q9UKM9 *RALY* RNA-binding protein Raly 0.0014 0.66 mRNA splicing  Q86V81 *THOC4* THO complex subunit 4 0.0007 0.75 mRNA splicing  O43143 *DHX15* Pre-mRNA splicing factor ATP-dep. RNA helicase 5.5E-06 0.78 mRNA splicing  P07910 *HNRNPC* Heterogeneous nuclear ribonucleoproteins Ca/C2 0.0005 0.81 mRNA splicing ER, endoplasmic reticulum; ETC, electron transport chain; OGD, oxygen glucose deprivation. Proteins are grouped according to subcellular compartment based on gene ontology and ranked according to magnitude of fold change. The significance threshold was set a priori at *P*\<0.01.

{#GRAPHIC-d7e88-autogenerated} During the COVID-19 pandemic, our first priority as scientists has to be to support one another and to take the time we need to process this tragedy and the transitions it has necessitated. "Productivity" simply cannot be our goal during a global pandemic. We are all grappling with loss. Most profoundly, we have lost loved ones and cherished members of our communities. We have also suffered smaller professional and personal losses---the loss of research projects, of routines that have kept us grounded. We must be mindful of our collective suffering; certain individuals and communities are heavily burdened right now, while some are hardly affected. If you are overwhelmed with the crisis, the suggestions in this article may not be relevant for you. Take care of your needs! **But if the privileges of well-being, safety, and sufficient time permit, the COVID-19 crisis can provide the scientific community an opportunity to reexamine and strengthen our core values on three scales: personal, interpersonal, and societal.** Personal: Harnessing Adaptability and Rekindling Intrinsic Motivation {#sec2} ===================================================================== These are scary and uncertain times; every day is a dance of adaptation. But, as scientists, this ebb and flow is part of our nature: uncertainty and adaptation are at the core of scientific research. Conditions change, experiments get derailed, life transitions thwart progress, or results counter our expectations and leave us baffled. We have to plunge into the unknown, let go of what we thought was constant, and rechart our paths time and time again. Due to COVID, life has changed unavoidably---unprecedentedly for some. Much is beyond the realm of our control. Yet there are still many elements within our control: our outlook, our attention, our motivation, and our intrinsic goals. How can we harness what's within our control, and use this to grow our authentic selves during this time? During this transition, we have a unique opportunity to step back and consider: What do I *really* care about? How do I *really* want my research to advance society? This will not be the last time in our careers that a crisis shakes things up. My physics training has taught me to think in terms of constraints---how can these new constraints on our lives enable creativity to blossom? Perhaps we cannot do our normal experiments, but here's one experiment that I am trying out, grounded in the scientific method (ish!):1.*Hypothesis*. Ask yourself which goals drive you, what you value, and what kind of scientist you want to be.2.*Observation*. Notice your constraints, what you "default" to in times of high stress and panic, and where you are sending your time, energy, and attention.3.*Analysis*. Ask yourself: is your default behavior and use of energy serving your goals?4.*Conclusion*. With your constraints in mind, chart a path forward to align your attention and intention more closely with your goals. For me, giving up my work environment has been frustrating. I'm trained as an experimentalist, but COVID-19 does not give a hoot about that. But by abandoning the autopilot acceleration of the lab during quarantine and embracing the shift of pace, I've been compelled to zoom out and revamp my approach to my dissertation. My "hypothesis" is that I'm deeply motivated to find material solutions to climate change and renewable energy management challenges. However, my "observation" has made me aware of my *own* energy *mis*management (I tend to frantically context-switch between multiple research projects and get slurped into distractions like a zombie) and my "analysis" demonstrates that this has pulled me away from my grander goals. Rather than simply cranking out content, I want to strengthen my core understanding of the connection between energy and materials and society, using the incredible scientific resources I have access to (online trainings, lectures, courses), and focus more intentionally on what I value. To head toward my "conclusion," I am now revisiting computational material science fundamentals, a task I'd previously procrastinated profusely, and am challenging myself to link these fundamentals to my goals. Being trapped at home with a computer and an inquisitive mind encourages this exploration. Admittedly, this is a goal. Many days I am so anxious I can barely focus. I realize that is okay---in fact it is normal right now---so I'm practicing forgiving myself. How can we shift our outlook to prioritize mental and physical well-being before "productivity"? I know I must first pursue what gives me balance, joy, and energy; then I can better focus attention on doing effective and creative work. It is a delicate equilibrium. There will be productive days and unproductive days. It is important to acknowledge our emotions and give ourselves space and permission to grieve, but also to be open to and appreciative of "productivity" when it does come our way. **But can we let go of measuring self-worth based on "productivity" alone, and instead acknowledge that we are so much more than our output and we deserve to prioritize our well-being? Can we let go of what is not within our control, and instead harness what is within our control and redefine what inner growth means to us? Let us use this time to reconnect with these intrinsic qualities, and work toward becoming the scientists and people we truly want to be.** Interpersonal: Cultivating a Supportive and Sustainable Work Environment {#sec3} ======================================================================== Another element very much within our control is the way that we support each other. What if this is a collective opportunity for scientists to practice how to be there for one another as human beings? Academia can be an isolating and emotionally turbulent edifice; this has only been compounded by this crisis. Work contexts that could once be handled with neutrality may now give rise to new, confusing emotions. As scientists and supervisors, we do not have much formal training in how to manage human relations. We are or will be confronted with loss and grief; wherever you are in the world, you or your group members or your collaborators may lose loved ones or experience traumatic life disruptions. We need to be aware that groups of people typically underrepresented in academia may be facing compounded trauma: people of color are dealing with overt racism; immunocompromised and disabled individuals are dealing with the stress of increased health risk; members of low-income communities are struggling to meet their basic needs. The scientists and engineers deemed "essential" must grapple with the task of putting their lives in danger on a daily basis; others facing furloughs and firings are struggling with financial insecurity. For those of us like me lucky enough to be employed from home (a privilege of being a Ph.D. student during COVID), our professional work has collided with our personal lives. Some of us are home with children and increased responsibilities; some do not have stable or comfortable living conditions; some are quarantined completely alone. We may be struggling with increased irritability, dissipated focus, or swings in energy levels. Unfortunately, mental health issues are still commonplace in academia; those fighting disorders such as anxiety and depression are likely facing extra barriers. It is therefore more urgent than ever to express empathy and gratitude for one another. During a crisis, we may have to overcompensate in terms of the support we offer our colleagues. Of course it is important to maintain personal boundaries, but this sudden merging of our home and work lives is perhaps a chance to be there for each other in ways we have not before. It is time to establish strong support structures. If you are in a supervisor position, make it crystal clear to your students that it is okay to not be producing at maximum output, that it is okay to slow down and feel the feels. Pandemic or no pandemic, your students are a valuable member of your team for who they are as people, not just because of their research output. Please check in with your students regularly. Set up a platform where students can feel comfortable being honest. Ask yourself: how can we advocate for one another? Of course, at some level, this has always been the case; every day before COVID, each of us faced our own demons, many of which were invisible. Now there is a visible demon, but invisible ones are still lingering and likely growing stronger, hiding over Zoom as we put on a face to demonstrate that we're "fine." Just because your student shows up to Zoom does not mean that everything is fine. Some folks are struggling silently and grieving privately. Can we practice patience, assume that we're all trying our best, and give our colleagues the benefit of the doubt? Can we then continue this practice afterward? Unfortunately, rumors are spreading that some group leaders in the scientific community are unfairly chastising students for treating this time like "vacation," demanding the same output as before the pandemic, and threatening students who are unable to deliver it. There are even rumors that some are pressuring students into bringing lab equipment home! We need to make it emphatically clear that this is not acceptable behavior---especially in a modern workplace, especially with regards to structural power imbalances, and especially during a pandemic. Lastly, I could not pass up the opportunity to squeeze a corny science analogy into a science journal editorial. If you study solar cells like me, you know that the efficiency of a solar cell is only as good as its weakest component. Same with arrays: if just one cell in series breaks, the entire array is compromised. Our research community is like this too. We are as strong as our most vulnerable link. Can we connect now to lift each other up? Can we center the voices of our community that are heard the least, and raise them up too? **So, during this time let us ask ourselves: what extra steps can we take right now to create a strong and compassionate support structure for our colleagues and those who we interact with on a regular basis? How can we transform our research environments and institutions into sustainable, inclusive, and empowering systems that will outlive the crisis?** Societal: Practicing Communication and Strengthening Community {#sec4} ============================================================== As scientists, I believe we all want to create a better and more inclusive future. To this end, since we are all collectively struggling, let us pursue endeavors to connect with and strengthen our communities. Although "action" looks different these days, the way we interact with society is still very much within our control. First, let us again acknowledge that certain communities are heavily burdened, while some are virtually unaffected. Ask yourself where you individually lie on this spectrum, and which issues concern you most. Second, science is being challenged every day. Can we stand together as a scientific community and demand policy based on science and social justice rather than speculation and biased agendas? People in power are going to take advantage of this crisis to oppress, as we have already seen with the spreading of harmful and hateful misinformation, eschewing of science to reopen the economy too early, and [loosening of science-based environmental regulations](https://www.ucsusa.org/about/news/epa-particulate-matter-decision-puts-science-health-aside-protect-polluters). We must push back against this. We cannot individually do everything and help everyone, but each of us can do something. With this in mind, what unique strengths and assets do we each possess currently? What actions are within our own realm of possibility? Some of us have research skills directly applicable to COVID that we can offer (check out [Covid19Sci](https://covid19sci.org/)---some of you have already made amazing contributions!), but this route is not appropriate for scientists like me who are not experts in relevant fields. We can seek other ways to help. Here's just one example of a space where scientists can take tangible action. Overnight, over 3 million overworked and undervalued K-12 educators in the USA (and millions more outside the USA) have had to adjust to a new way of teaching, and fish up "virtual" curriculum content for over 55 million students. Meanwhile, we scientists are at home, many without experiments to work on, but many *with* science content buzzing through our brains (bzzzz!). I am mortified of public speaking, but I realize that to be an advocate for renewable energy I must overcome this fear. I cannot practice "public" speaking in the usual way, but I *can* practice translating my jargon-heavy projects into narratives digestible to nonscientists. In fact, this is critical right now. How can the public trust scientists if we cannot cohesively explain what we are working on? To this end, I challenged my tongue-tied self to give a five minute general-public pitch of my solar materials research as part of a ["TED\@Home" symposium in April](https://www.youtube.com/watch?v=yQ6VX_icj4c&feature=youtu.be). {#fx1} Additionally, with my own outreach organization [Cycle for Science](http://cycleforscience.org/) and Berkeley's [Community Resources for Science](https://www.crscience.org/), I am creating a hands-on video about how to make dye-sensitized solar cells from blackberries and other common household ingredients (here's the [lesson plan](http://cycleforscience.org/s/Cycle-for-Science-Dye-Sensitized-Solar-Cell-lesson.pdf)!). I aim to make this a resource for K-12 teachers and families, a demonstration of how solar energy conversion works, and an introduction to my research on solar cell contact materials. There are incredible STEM education organizations to volunteer with during COVID, including [Skype a Scientist](https://www.skypeascientist.com/), which connects you directly with students; you could also try recording a hands-on lesson, or a short video like mine explaining what you do as a scientist. If science education is not your cup of tea, there are plenty of other creative community-facing actions you can take. Here is one [resource with action ideas for scientists](https://www.the-scientist.com/news-opinion/how-scientists-can-volunteer-to-help-fight-covid-19-67412). You can write to your representatives, join a [Union of Concerned Scientists](https://www.ucsusa.org/) campaign aligned with your interests, or get involved with the [National Science Policy Network](https://scipolnetwork.org/). Another good place to start is with local mutual aid groups, where you can volunteer to exchange resources and services (for example, here is the [San Francisco Bay Area's mutual aid wiki](https://bayareamutualaid.org/wiki/Bay_Area_Mutual_Aid_Resources)). We have many options, and contributions of all scales are important: big and small, long-term and short-term, public-facing and behind-the-scenes, etc. But it is equally important to honor that if you are not in a headspace to focus on contributing---whether financially, physically, mentally, or emotionally---*that is absolutely okay*. In these times, it is critical to put your own self-care first, so please check in with yourself before taking action. **If you are in a place to give, I challenge you to ask yourself: What unique contribution can I bring to the table, in order to connect with and strengthen my community right now? How can I act as an ally? What kind of world do I want to enable with the science that I do? Consider these questions, then act accordingly.** Call to Action {#sec5} ============== Our integrity as scientific leaders and community members can emerge during this crisis. As the barriers of "normalcy" and "precedent" crumble, we can step forward with creative intention and remind ourselves of why we really do this work. This time is traumatic, weird, and uncertain. Stepping back and focusing on healing is important, too, if that is what you need. **But if we are able, let us harness what is within our control on the personal, interpersonal, and societal scales, and work together now to create the supportive and sustainable science community we aspire to become**. Views expressed in this editorial are those of the author and not necessarily the views of the ACS.

Related literature {#sec1} ================== The synthesis from 2-*tert*-butyl-4-methyl­phenol, 3-amino-1-propanol and formaldehyde is an example of carbon--carbon bond formation by the Mannich reaction. For another variation of the Mannich reaction involving 3-amino-1-propanol, see: Korepin *et al.* (2001[@bb3]). Experimental {#sec2} ============ {#sec2.1} ### Crystal data {#sec2.1.1} C~16~H~25~NO~2~*M* *~r~* = 263.37Orthorhombic,*a* = 6.4740 (7) Å*b* = 14.1928 (13) Å*c* = 16.7914 (16) Å*V* = 1542.9 (3) Å^3^*Z* = 4Mo *K*α radiationμ = 0.07 mm^−1^*T* = 293 K0.28 × 0.20 × 0.12 mm ### Data collection {#sec2.1.2} Rigaku R-AXIS Spider IP diffractometerAbsorption correction: multi-scan (*ABSCOR*; Higashi, 1995[@bb2]) *T* ~min~ = 0.980, *T* ~max~ = 0.99115222 measured reflections2044 independent reflections1664 reflections with *I* \> 2σ(*I*)*R* ~int~ = 0.022 ### Refinement {#sec2.1.3} *R*\[*F* ^2^ \> 2σ(*F* ^2^)\] = 0.035*wR*(*F* ^2^) = 0.117*S* = 1.112044 reflections180 parameters1 restraintH atoms treated by a mixture of independent and constrained refinementΔρ~max~ = 0.14 e Å^−3^Δρ~min~ = −0.12 e Å^−3^ {#d5e374} Data collection: *RAPID-AUTO* (Rigaku, 2002[@bb4]); cell refinement: *RAPID-AUTO*; data reduction: *CrystalClear* (Rigaku/MSC, 2002[@bb5]); program(s) used to solve structure: *SHELXS97* (Sheldrick, 2008[@bb6]); program(s) used to refine structure: *SHELXL97* (Sheldrick, 2008[@bb6]); molecular graphics: *X-SEED* (Barbour, 2001[@bb1]); software used to prepare material for publication: *publCIF* (Westrip, 2010[@bb7]). Supplementary Material ====================== Crystal structure: contains datablocks global, I. DOI: [10.1107/S1600536810012109/bt5237sup1.cif](http://dx.doi.org/10.1107/S1600536810012109/bt5237sup1.cif) Structure factors: contains datablocks I. DOI: [10.1107/S1600536810012109/bt5237Isup2.hkl](http://dx.doi.org/10.1107/S1600536810012109/bt5237Isup2.hkl) Additional supplementary materials: [crystallographic information](http://scripts.iucr.org/cgi-bin/sendsupfiles?bt5237&file=bt5237sup0.html&mime=text/html); [3D view](http://scripts.iucr.org/cgi-bin/sendcif?bt5237sup1&Qmime=cif); [checkCIF report](http://scripts.iucr.org/cgi-bin/paper?bt5237&checkcif=yes) Supplementary data and figures for this paper are available from the IUCr electronic archives (Reference: [BT5237](http://scripts.iucr.org/cgi-bin/sendsup?bt5237)). We thank South China University of Technology and the University of Malaya for supporting this study. Comment ======= Organic synthesis centers largely on stereoselective carbon--carbon and carbon--heteroatom bond-forming reactions; among such reactions is the class of Mannich reactions, which can be regarded as being the most important carbon--carbon bond-forming reaction. The reactions lead to β-aminocarbonyl compounds, which are important intermediates for pharmaceuticals. One variation of the Mannich reaction involves the catalytic addition of an amine, R~2~NH, to an alkene or alkyne, i. e., hydroamination. In the 2-*tert*-butyl-4-methylphenol reacts with 3-amino-1-propanol to yield a compound having a 1,3-oxazinyl ring (Scheme I, Fig. 1). Such a ring is difficult to synthesis by conventional routes. Experimental {#experimental} ============ 2-*tert*-Butyl-4-methylphenol (2.24 g, 12.3 mmol), 3-amino-1-propanol (0.93 g, 12.3 mmol), 37% aqueous formaldehyde (1.83 ml, 24.6 mmol) and triethylamine (2.49 g, 24.6 mmol) in ethanol (50 ml) were heated for 6 hours. Slow evaporation of the filtrate gave light-yellow crystals in 70% yield. Refinement {#refinement} ========== Carbon-bound H-atoms were allowed to ride on their parent atoms (C--H 0.93-- 0.97 Å) and their displacement parameters were set to 1.2--1.5U~eq~(C). The hydroxy H-atom was located in a difference Fourier map, and was refined isotropically with a distance restraint of O--H 0.84±0.01 Å. Due to the absence of anomalous scatterers, the absolute configuration could not be determined, and, therefore, 1488 Friedel pairs were merged. Figures ======= {#Fap1} Crystal data {#tablewrapcrystaldatalong} ============ ------------------------------- ---------------------------------------- C~16~H~25~NO~2~ *F*(000) = 576 *M~r~* = 263.37 *D*~x~ = 1.134 Mg m^−3^ Orthorhombic, *P*2~1~2~1~2~1~ Mo *K*α radiation, λ = 0.71073 Å Hall symbol: P 2ac 2ab Cell parameters from 12093 reflections *a* = 6.4740 (7) Å θ = 3.1--27.5° *b* = 14.1928 (13) Å µ = 0.07 mm^−1^ *c* = 16.7914 (16) Å *T* = 293 K *V* = 1542.9 (3) Å^3^ Block, yellow *Z* = 4 0.28 × 0.20 × 0.12 mm ------------------------------- ---------------------------------------- Data collection {#tablewrapdatacollectionlong} =============== ------------------------------------------------------------- -------------------------------------- Rigaku R-AXIS Spider IP diffractometer 2044 independent reflections Radiation source: fine-focus sealed tube 1664 reflections with *I* \> 2σ(*I*) graphite *R*~int~ = 0.022 ω scan θ~max~ = 27.5°, θ~min~ = 3.1° Absorption correction: multi-scan (*ABSCOR*; Higashi, 1995) *h* = −8→8 *T*~min~ = 0.980, *T*~max~ = 0.991 *k* = −18→18 15222 measured reflections *l* = −21→21 ------------------------------------------------------------- -------------------------------------- Refinement {#tablewraprefinementdatalong} ========== ------------------------------------- ------------------------------------------------------------------------------------------------- Refinement on *F*^2^ Primary atom site location: structure-invariant direct methods Least-squares matrix: full Secondary atom site location: difference Fourier map *R*\[*F*^2^ \> 2σ(*F*^2^)\] = 0.035 Hydrogen site location: inferred from neighbouring sites *wR*(*F*^2^) = 0.117 H atoms treated by a mixture of independent and constrained refinement *S* = 1.11 *w* = 1/\[σ^2^(*F*~o~^2^) + (0.0739*P*)^2^ + 0.0371*P*\] where *P* = (*F*~o~^2^ + 2*F*~c~^2^)/3 2044 reflections (Δ/σ)~max~ = 0.001 180 parameters Δρ~max~ = 0.14 e Å^−3^ 1 restraint Δρ~min~ = −0.12 e Å^−3^ ------------------------------------- ------------------------------------------------------------------------------------------------- Special details {#specialdetails} =============== ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å^2^) {#tablewrapcoords} ================================================================================================== ------ ------------ -------------- -------------- -------------------- -- *x* *y* *z* *U*~iso~\*/*U*~eq~ O1 0.8894 (2) 0.48173 (11) 0.62162 (8) 0.0654 (4) H1 0.972 (4) 0.5284 (14) 0.6242 (18) 0.095 (9)\* O2 0.9444 (3) 0.75355 (12) 0.59189 (10) 0.0816 (5) N1 1.0668 (3) 0.63695 (11) 0.67909 (9) 0.0550 (4) C1 0.7719 (3) 0.47819 (12) 0.68939 (10) 0.0494 (4) C2 0.8294 (3) 0.53321 (12) 0.75571 (10) 0.0506 (4) C3 0.7081 (3) 0.53034 (12) 0.82338 (10) 0.0536 (4) H3 0.7451 0.5670 0.8670 0.064\* C4 0.5332 (3) 0.47461 (12) 0.82833 (10) 0.0524 (4) C5 0.4816 (3) 0.42039 (12) 0.76175 (10) 0.0501 (4) H5 0.3656 0.3819 0.7646 0.060\* C6 0.5953 (3) 0.42121 (12) 0.69126 (10) 0.0468 (4) C7 1.0274 (3) 0.58874 (15) 0.75529 (11) 0.0594 (5) H7A 1.0222 0.6353 0.7975 0.071\* H7B 1.1415 0.5465 0.7666 0.071\* C8 0.4047 (4) 0.47229 (16) 0.90293 (11) 0.0731 (6) H8A 0.2749 0.4423 0.8919 0.110\* H8B 0.3806 0.5355 0.9211 0.110\* H8C 0.4765 0.4376 0.9434 0.110\* C9 0.5275 (3) 0.36320 (13) 0.61807 (10) 0.0539 (4) C10 0.4755 (5) 0.43096 (16) 0.54936 (12) 0.0754 (6) H10A 0.3555 0.4673 0.5630 0.113\* H10B 0.4484 0.3953 0.5019 0.113\* H10C 0.5901 0.4725 0.5403 0.113\* C11 0.3358 (4) 0.30415 (17) 0.63527 (14) 0.0745 (6) H11A 0.2244 0.3448 0.6510 0.112\* H11B 0.3651 0.2605 0.6774 0.112\* H11C 0.2971 0.2701 0.5882 0.112\* C12 0.7001 (4) 0.29608 (16) 0.59218 (13) 0.0731 (6) H12A 0.6504 0.2558 0.5504 0.110\* H12B 0.7423 0.2584 0.6368 0.110\* H12C 0.8157 0.3320 0.5731 0.110\* C13 0.9229 (3) 0.71409 (15) 0.66757 (13) 0.0668 (5) H13A 0.7828 0.6912 0.6742 0.080\* H13B 0.9476 0.7621 0.7076 0.080\* C14 1.1466 (4) 0.79381 (18) 0.58285 (17) 0.0839 (7) H14A 1.1654 0.8439 0.6215 0.101\* H14B 1.1600 0.8208 0.5300 0.101\* C15 1.3100 (4) 0.71942 (18) 0.59488 (14) 0.0743 (6) H15A 1.3017 0.6732 0.5524 0.089\* H15B 1.4457 0.7482 0.5930 0.089\* C16 1.2798 (3) 0.67164 (15) 0.67383 (13) 0.0633 (5) H16A 1.3757 0.6195 0.6790 0.076\* H16B 1.3065 0.7158 0.7167 0.076\* ------ ------------ -------------- -------------- -------------------- -- Atomic displacement parameters (Å^2^) {#tablewrapadps} ===================================== ----- ------------- ------------- ------------- -------------- -------------- -------------- *U*^11^ *U*^22^ *U*^33^ *U*^12^ *U*^13^ *U*^23^ O1 0.0583 (9) 0.0845 (9) 0.0533 (7) −0.0116 (8) 0.0156 (7) −0.0123 (7) O2 0.0656 (10) 0.0945 (10) 0.0847 (10) −0.0108 (9) −0.0202 (8) 0.0295 (9) N1 0.0417 (8) 0.0652 (8) 0.0579 (8) −0.0029 (7) −0.0030 (7) 0.0015 (7) C1 0.0461 (9) 0.0594 (8) 0.0427 (8) 0.0024 (8) 0.0025 (7) −0.0022 (7) C2 0.0480 (10) 0.0563 (9) 0.0474 (8) 0.0045 (8) −0.0048 (7) 0.0002 (8) C3 0.0618 (11) 0.0578 (9) 0.0412 (8) 0.0048 (9) −0.0042 (8) −0.0049 (7) C4 0.0547 (10) 0.0583 (8) 0.0443 (8) 0.0077 (8) 0.0048 (8) −0.0005 (8) C5 0.0474 (10) 0.0540 (8) 0.0491 (8) 0.0026 (8) 0.0026 (7) 0.0007 (7) C6 0.0456 (9) 0.0517 (8) 0.0431 (7) 0.0058 (7) −0.0011 (7) −0.0018 (7) C7 0.0526 (11) 0.0719 (10) 0.0537 (9) −0.0032 (10) −0.0084 (9) 0.0020 (9) C8 0.0795 (16) 0.0877 (13) 0.0519 (11) 0.0033 (13) 0.0197 (10) −0.0034 (10) C9 0.0553 (11) 0.0637 (10) 0.0428 (8) 0.0006 (9) −0.0051 (8) −0.0044 (8) C10 0.0868 (17) 0.0854 (13) 0.0541 (10) −0.0005 (13) −0.0201 (11) 0.0065 (10) C11 0.0764 (16) 0.0823 (13) 0.0649 (12) −0.0191 (12) −0.0065 (11) −0.0106 (11) C12 0.0834 (17) 0.0754 (12) 0.0606 (11) 0.0122 (13) 0.0005 (11) −0.0172 (10) C13 0.0492 (11) 0.0760 (11) 0.0752 (13) 0.0046 (10) −0.0018 (11) 0.0095 (11) C14 0.0771 (17) 0.0884 (14) 0.0861 (16) −0.0235 (14) −0.0163 (14) 0.0215 (13) C15 0.0625 (14) 0.0911 (14) 0.0694 (13) −0.0203 (12) 0.0040 (11) −0.0028 (11) C16 0.0458 (10) 0.0732 (11) 0.0708 (12) −0.0052 (9) −0.0033 (10) −0.0011 (10) ----- ------------- ------------- ------------- -------------- -------------- -------------- Geometric parameters (Å, °) {#tablewrapgeomlong} =========================== -------------------- -------------- ---------------------- -------------- O1---C1 1.370 (2) C9---C11 1.526 (3) O1---H1 0.852 (10) C9---C12 1.531 (3) O2---C13 1.396 (3) C9---C10 1.539 (3) O2---C14 1.436 (3) C10---H10A 0.9600 N1---C13 1.450 (3) C10---H10B 0.9600 N1---C16 1.467 (3) C10---H10C 0.9600 N1---C7 1.473 (2) C11---H11A 0.9600 C1---C6 1.401 (3) C11---H11B 0.9600 C1---C2 1.410 (2) C11---H11C 0.9600 C2---C3 1.382 (3) C12---H12A 0.9600 C2---C7 1.505 (3) C12---H12B 0.9600 C3---C4 1.384 (3) C12---H12C 0.9600 C3---H3 0.9300 C13---H13A 0.9700 C4---C5 1.398 (2) C13---H13B 0.9700 C4---C8 1.504 (2) C14---C15 1.508 (4) C5---C6 1.394 (2) C14---H14A 0.9700 C5---H5 0.9300 C14---H14B 0.9700 C6---C9 1.543 (2) C15---C16 1.502 (3) C7---H7A 0.9700 C15---H15A 0.9700 C7---H7B 0.9700 C15---H15B 0.9700 C8---H8A 0.9600 C16---H16A 0.9700 C8---H8B 0.9600 C16---H16B 0.9700 C8---H8C 0.9600 C1---O1---H1 110 (2) C9---C10---H10B 109.5 C13---O2---C14 110.29 (18) H10A---C10---H10B 109.5 C13---N1---C16 110.01 (15) C9---C10---H10C 109.5 C13---N1---C7 110.80 (16) H10A---C10---H10C 109.5 C16---N1---C7 111.79 (16) H10B---C10---H10C 109.5 O1---C1---C6 119.54 (15) C9---C11---H11A 109.5 O1---C1---C2 119.30 (17) C9---C11---H11B 109.5 C6---C1---C2 121.15 (16) H11A---C11---H11B 109.5 C3---C2---C1 118.89 (18) C9---C11---H11C 109.5 C3---C2---C7 120.23 (16) H11A---C11---H11C 109.5 C1---C2---C7 120.73 (17) H11B---C11---H11C 109.5 C2---C3---C4 122.09 (16) C9---C12---H12A 109.5 C2---C3---H3 119.0 C9---C12---H12B 109.5 C4---C3---H3 119.0 H12A---C12---H12B 109.5 C3---C4---C5 117.54 (16) C9---C12---H12C 109.5 C3---C4---C8 121.00 (16) H12A---C12---H12C 109.5 C5---C4---C8 121.45 (18) H12B---C12---H12C 109.5 C6---C5---C4 123.24 (18) O2---C13---N1 111.13 (18) C6---C5---H5 118.4 O2---C13---H13A 109.4 C4---C5---H5 118.4 N1---C13---H13A 109.4 C5---C6---C1 117.06 (15) O2---C13---H13B 109.4 C5---C6---C9 121.45 (17) N1---C13---H13B 109.4 C1---C6---C9 121.49 (15) H13A---C13---H13B 108.0 N1---C7---C2 113.26 (15) O2---C14---C15 110.27 (18) N1---C7---H7A 108.9 O2---C14---H14A 109.6 C2---C7---H7A 108.9 C15---C14---H14A 109.6 N1---C7---H7B 108.9 O2---C14---H14B 109.6 C2---C7---H7B 108.9 C15---C14---H14B 109.6 H7A---C7---H7B 107.7 H14A---C14---H14B 108.1 C4---C8---H8A 109.5 C16---C15---C14 110.1 (2) C4---C8---H8B 109.5 C16---C15---H15A 109.6 H8A---C8---H8B 109.5 C14---C15---H15A 109.6 C4---C8---H8C 109.5 C16---C15---H15B 109.6 H8A---C8---H8C 109.5 C14---C15---H15B 109.6 H8B---C8---H8C 109.5 H15A---C15---H15B 108.2 C11---C9---C12 107.79 (16) N1---C16---C15 109.08 (18) C11---C9---C10 107.87 (19) N1---C16---H16A 109.9 C12---C9---C10 109.61 (18) C15---C16---H16A 109.9 C11---C9---C6 111.95 (16) N1---C16---H16B 109.9 C12---C9---C6 110.53 (17) C15---C16---H16B 109.9 C10---C9---C6 109.03 (15) H16A---C16---H16B 108.3 C9---C10---H10A 109.5 O1---C1---C2---C3 178.97 (16) C16---N1---C7---C2 −166.73 (16) C6---C1---C2---C3 0.0 (3) C3---C2---C7---N1 −142.23 (18) O1---C1---C2---C7 −5.4 (3) C1---C2---C7---N1 42.2 (2) C6---C1---C2---C7 175.57 (16) C5---C6---C9---C11 −3.1 (2) C1---C2---C3---C4 0.6 (3) C1---C6---C9---C11 177.96 (17) C7---C2---C3---C4 −175.06 (16) C5---C6---C9---C12 −123.27 (19) C2---C3---C4---C5 −0.1 (2) C1---C6---C9---C12 57.8 (2) C2---C3---C4---C8 179.57 (19) C5---C6---C9---C10 116.2 (2) C3---C4---C5---C6 −1.0 (3) C1---C6---C9---C10 −62.8 (2) C8---C4---C5---C6 179.35 (17) C14---O2---C13---N1 −63.1 (2) C4---C5---C6---C1 1.5 (3) C16---N1---C13---O2 62.4 (2) C4---C5---C6---C9 −177.50 (16) C7---N1---C13---O2 −173.52 (16) O1---C1---C6---C5 −179.96 (17) C13---O2---C14---C15 59.0 (3) C2---C1---C6---C5 −0.9 (2) O2---C14---C15---C16 −54.6 (3) O1---C1---C6---C9 −1.0 (2) C13---N1---C16---C15 −56.7 (2) C2---C1---C6---C9 178.05 (16) C7---N1---C16---C15 179.72 (18) C13---N1---C7---C2 70.2 (2) C14---C15---C16---N1 53.5 (2) -------------------- -------------- ---------------------- -------------- Hydrogen-bond geometry (Å, °) {#tablewraphbondslong} ============================= --------------- ---------- ---------- ----------- --------------- *D*---H···*A* *D*---H H···*A* *D*···*A* *D*---H···*A* O1---H1···N1 0.85 (1) 1.90 (2) 2.665 (2) 149 (3) --------------- ---------- ---------- ----------- --------------- ###### Hydrogen-bond geometry (Å, °) *D*---H⋯*A* *D*---H H⋯*A* *D*⋯*A* *D*---H⋯*A* ------------- ---------- ---------- ----------- ------------- O1---H1⋯N1 0.85 (1) 1.90 (2) 2.665 (2) 149 (3)

{#sp1 .442} {#sp2 .443} {#sp3 .444} {#sp4 .445} {#sp5 .446} {#sp6 .447}

In [Table 1](#pone.0202143.t001){ref-type="table"}, the value in the row titled "All other genes vs. LRR^e^" and the column titled "Projection test *P*- value" should read as 1.9E-120. Please see the corrected [Table 1](#pone.0202143.t001){ref-type="table"} here. The publisher apologizes for this error. 10.1371/journal.pone.0202143.t001 ###### Summary of statistical tests used to determine genome-wide spatial correlations between extracellular receptor-like proteins (RLPs) and secreted peptides (SPs) versus intracellular nucleotide-binding leucine-rich repeat receptors (NLRs). {#pone.0202143.t001g}   Query Reference Relative Kolmogorov-Smirnov *P*-value^a^ Relative ECDF area correlation^b^ Relative ECDF deviation area *P*-value Jaccard measure *P*--value Projection test *P*--value ---------------------------- ------- ----------- ------------------------------------------ ----------------------------------- ---------------------------------------- ---------------------------- ---------------------------- -------- ------- ------- ---------- ------- ------- eLRR vs. NLR genes^e^ 256 366 3.2E-06 0.152 \<0.01 **✓** 0.43 **✓** 0.44 **✓** NLR vs. eLRR genes^e^ 366 256 0.862 -2.2E-04 0.72 **✓** 0.40 **✓** 0.34 **✓** RLP vs. NLR genes 152 366 1.2E-04 0.141 0.01 **✓** 0.44 **✓** 0.29 **✓** NLR vs. RLP genes 366 152 0.007 0.046 0.07 **✓** 0.45 **✓** 0.24 **✓** SP vs. NLR genes 104 366 0.009 0.167 \<0.01 **✓** 0.25 **✓** 0.21 **✓** NLR vs. SP genes 366 104 3.8E-04 0.064 0.02 **✓** 0.38 **✓** 0.14 **✓** SP vs. RLP genes 104 152 0.010 0.129 0.03 **✓** 0.12 **✓** 0.07 **✓** RLP vs. SP genes 152 104 8.2E-05 0.241 \<0.01 **✓**   0.05 **✓**   0.06 **✓**   LRR vs. ribosomal genes^e^ 622 1134 0.002 -0.054 \<0.01 **✓** \<0.01 **✓** 0.21 **✓** Ribosomal vs. LRR genes^e^ 1134 622 0.289 0.021 0.21 **✓** \<0.01 **✓** 0.02 **✓** LRR vs. all other genes^e^ 622 80305 0.000 -0.261 \<0.01 **✓** \<0.01 **✓** 8.9E-78 **✓** All other genes vs. LRR^e^ 80305 622 2.4E-04 0.009 \<0.01 **✓**   \<0.01   **✓** 1.9E-120   **✓** ^a^ *P* values are shown for all tests in both directions, using one dataset as a query and the other one as a reference. ^b^ ECDF: Empirical Distribution Cumulative Function ^c^ For the relative distance test, positive and negative relative ECDF area correlation values are labeled as close and far, respectively. ^d^ The outcome of Jaccard and projection tests is defined as overlapping or non-overlapping (N). ^e^ The combination of RLP and SP genes are referred to as eLRR genes. Control comparisons include those of all predicted leucine-rich repeat (LRR) genes, consisting of RLP, NLR and SP genes, to ribosomal genes or those to all other coding genes of the *Brassica napus* genome.

Introduction {#s1} ============ Mast cells are recognized as the major effector cells of the type I hypersensitivity reactions and they are known to play a pivotal role in allergic diseases, such as allergic rhinitis, asthma, and atopic dermatitis. Engagement of FcεRI by IgE, followed by the aggregation of multiple IgE-bearing FcεRI molecules by polyvalent antigens, leads to degranulation and release of histamine, LTC4, and other preformed chemical mediators [@pone.0025412-Galli1], [@pone.0025412-Galli2]. Additionally, multiple cytokine genes are transcribed and newly synthesized arachidonic acid metabolites are secreted [@pone.0025412-Galli1], [@pone.0025412-Galli2]. It is well established that mast cells promote the early phase of type I hypersensitivity reactions by releasing granule contents after FcεRI-crosslinking. Recently, we have demonstrated that IKK2, which is a central component of the intracellular signaling pathway mediating NF-κB activation [@pone.0025412-Hayden1]--[@pone.0025412-Perkins1], plays critical roles in IgE-mediated anaphylaxis in vivo and IgE-mediated degranulation of mast cells in vitro [@pone.0025412-Suzuki1]. Upon FcεRI stimulation, IKK2 phosphorylates SNAP-23, the target membrane soluble N-ethylmaleimide-sensitive fusion factor attachment protein receptor (SNARE), leading to degranulation and anaphylactic reactions [@pone.0025412-Suzuki1]. Moreover, accumulating evidence has shown that the kinase activity of IKK2 is regulated by post-translational modifications [@pone.0025412-Perkins1]. However, it is still unknown whether the post-translational modifications of IKK2 are involved in IKK2-mediated degranulation of mast cells and anaphylactic reactions. Transcription factors of the NF-κB family, which consists of NF-κB1, NF-κB2, p65 (also called RelA), c-Rel, and RelB, regulate the expression of hundreds of genes in the context of multiple important biological processes, such as apoptosis, proliferation, innate and adaptive immune responses, and inflammation [@pone.0025412-Hayden1], [@pone.0025412-Li1]. p65 has been shown to play critical roles in IKK2-mediated gene induction of proinflammatory cytokines [@pone.0025412-Hayden1], [@pone.0025412-Li1]. It has also been shown that the activity of p65 is modulated by several other molecules including CREB binding protein, glucocorticoid receptor, and SP1 transcription factor [@pone.0025412-Perkins1]. However, the regulatory mechanisms underling p65-mediated gene induction of proinflammatory cytokines in mast cells are still largely unknown. Tumor suppressor p53 is a sequence-specific transcription factor that is critical for maintaining genomic stability [@pone.0025412-Vogelstein1]. Without cellular stresses, protein levels of p53 in cells are maintained at low levels and the majority of p53 remains in the cytoplasm. Upon induction of various stresses such as apoptosis, cell cycle arrest, senescence, DNA repair, cell metabolism, and autophagy [@pone.0025412-Riley1]--[@pone.0025412-Kruse1], the half-life of p53 increases from minutes to hours. p53 then translocates into the nucleus and activates its target genes. Although the roles of p53 in stress-associated stimulation have been well studied, the roles of p53 in antigen receptor-mediated stimulation are poorly understood. In this study, we show that the protein levels of p53 are up-regulated upon IgE-mediated activation in mast cells and that the lack of p53 results in enhanced mast cell activation in vivo as well as in vitro. p53 inhibits not only the catalytic activity of IKK2 through the modulation of glycosylation but also p65-mediated transactivation. Our findings indicate that_p53 functions as a negative regulator of mast cell activation through the inhibition of NF-κB pathways. Methods {#s2} ======= Mice {#s2a} ---- p53-deficient (p53^−/−^) mice on a C57BL/6 background [@pone.0025412-Donehower1] were purchased from Taconic Farms (Hudson, NY). Mast cell-deficient WBB6F1-W/W^v^ mice (W/W^v^ mice) were purchased from SLC (Shizuoka, Japan). p65-heterozygous mice [@pone.0025412-Beg1] were kindly provided by Dr. Baltimore (California Institute of Technology, USA). Mice were bred and housed in the animal care facility at Salk Institute and Chiba University. Experimental procedures involving mice followed the guidelines from the National Institutes of Health and Chiba University and were approved by the Animal Use Committee (approval ID; 09-052) at the Salk Institute and at Chiba University (approval ID; 21-11). Flow cytometric analysis {#s2b} ------------------------ Cells were analyzed on FACSCalibur (Becton Dickinson, San Jose, CA) with CELLQuest software. The following antibodies were purchased: anti-CD117 (c-kit) allophycocyanin (APC) (2B8; BD Biosciences, San Diego, CA) and anti-FcεRIα (FITC) (MAR-1; eBioscience, San Diego, CA). Before staining, Fc receptors were blocked with anti-CD16/32 antibody (2.4G2; BD Biosciences). Negative controls consisted of isotype-matched directly conjugated, nonspecific antibodies (BD Biosciences). Histological quantification of mast cells {#s2c} ----------------------------------------- Peritoneal mast cells were stained with anti-CD117 (c-kit) APC and anti-FcεRIα (FITC) and analyzed by flow cytometry. The numbers of mast cells in the tissue were counted by light microscopy (x400) on Alcian blue-stained sections. Data were expressed as the number of mast cells per mm^2^ in ear dermis and the number of mast cells per villus crypt unit in jejunal mucosa. Cell culture {#s2d} ------------ Primary culture of IL-3-dependent bone marrow-derived mast cells (BMMCs) was prepared from 6- to 8-week-old WT and p53^−/−^ mice and maintained as previously described [@pone.0025412-Suzuki2]. Approximately 99% of cells recovered after 4 weeks of culture were morphologically mast cells. Reconstitution of mast cells in W/W^v^ mice with BMMCs {#s2e} ------------------------------------------------------ For reconstitution of skin mast cells in W/W^v^ mice, WT BMMCs or p53^−/−^ BMMCs (1×10^6^ cells in 20 µl of PBS) were injected intradermally into the right ear of W/W^v^ mice as described previously [@pone.0025412-Suzuki1]. As controls, PBS was injected to the left ear. Passive cutaneous anaphylaxis {#s2f} ----------------------------- To induce passive cutaneous anaphylaxis (PCA), 4 weeks after the transplantation of BMMCs, the mice were sensitized with monoclonal mouse dinitrophenol-specific (anti-DNP) IgE (100 ng in 20 µl of PBS, Sigma, St Louis, MO) by intradermal injection into the ear. Twenty-four hours later, 200 µg of DNP-HSA (human serum albumin; Sigma) diluted in sterile saline were injected intravenously to the mice. Ear swelling was quantified by three consecutive measurements of ear thickness using calipers before (base line) and 1, 2, 3, 4, and 6 hours after the antigen challenge. In some experiments, 0.5% Evans blue dye was injected together with DNP-HSA. Where indicated, 2-deoxy-D-glucose (2-DG, 50 µl of 0.2 mM/mouse) were given intravenously to the mice at 20 hours before the antigen challenge. Measurement of Evans blue dye extravasation {#s2g} ------------------------------------------- To extract Evans blue dye from the tissue, ear biopsy specimens were incubated in 0.3 ml of formamide at 60°C for 4 days. Absorption was measured at 620 nm as described previously [@pone.0025412-Suzuki1]. Late phase allergic reactions {#s2h} ----------------------------- Four weeks after the transplantation of BMMCs to W/W^v^ mice, the mice were sensitized with anti-DNP IgE (20 µg/mice) intravenously. Twenty-four hours later, mice were challenged with epicutaneous application of 10 µl of DNFB (0.2% wt/vol) in acetone/olive oil (4∶1) to both sides of the ears. Ear swelling was assessed before (base line) and 2, 4, 6, 8, 12, 24, and 48 hours after the challenge. IgE receptor engagement {#s2i} ----------------------- For stimulation of BMMCs via Fcε receptors, cells were incubated with mouse anti-trinitrophenol (TNP) IgE (1 µg/ml, BD Biosciences) at 37°C for 2 hours, washed twice with PBS, and then incubated with TNP^−34^-BSA (50 ng/ml, Bioresearch Technologies, Novato, CA) at 37°C for the indicated times in Tyrode\'s buffer (130 mmol/L NaCl, 5 mmol/L KCl, 1.4 mmol/L CaCl2, 1 mmol/L MgCl2, 5.6 mmol/L glucose, 10 mmol/L HEPES, and 0.1% BSA, PH 7.4). In some experiments, BMMCs were incubated with selective IKK2 inhibitor ML120B (10 µM) [@pone.0025412-Nagashima1] or vehicle (0.01% DMSO) for 1 hour prior to stimulation. For inhibition of glycolysis, BMMCs were incubated with 2-DG (4.5 mg/ml) for 2 hours prior to stimulation. For inhibition of O-GlcNAcylation, BMMCs were incubated with streptozotocin (STZ; 5 mM) for 3 hours prior to stimulation. β-hexosaminidase assay {#s2j} ---------------------- Enzyme activity of β-hexosaminidase was evaluated for both the supernatant and the cell lysate using p-nitrophenyl-N-acetyl β-D-glucosamine (Sigma) as a substrate. The percentage of specific β-hexosaminidase release was expressed as 100 x supernatant activity/(supernatant activity + cell lysate activity) as described previously [@pone.0025412-Suzuki1]. Measurement of cytokines {#s2k} ------------------------ BMMCs were cultured and stimulated with IgE receptor engagement as described above. The amounts of TNF-α and IL-6 in the culture supernatant were measured by TNF-α and IL-6 ELISA kit (BD Biosciences). The assay was performed in duplicate according to the manufacturer\'s instruction. The detection limit was 15 pg/ml for both cytokines. Intracellular staining of p53 {#s2l} ----------------------------- BMMCs were stimulated with IgE receptor engagement or Nutlin-3a (10 µM) for the indicated times. Cells were fixed, permeabilized with Perm/Wash buffer (BD Biosciences), stained with anti-p53-Alexa Fluor 647 (1C12, Cell Signaling Technology, Denvers, MA), and analyzed by flow cytometry. Annexin-V staining {#s2m} ------------------ To detect membrane fusion of BMMCs, cells were stained with Annexin-V-PE (BD Biosciences) in Annexin-V binding buffer (BD Biosciences) for 15 minutes and analyzed by flow cytometry. Reporter gene assay {#s2n} ------------------- Expression vectors for p65 [@pone.0025412-Tergaonkar1], IKK2SE [@pone.0025412-Suzuki1], and p53 [@pone.0025412-Xia1] and NF-κB-reporter-construct [@pone.0025412-Sigala1] were described previously. HEK293 cells were transiently transfected with these plasmids using Effectene transfection reagent (Qiagen, Valencia, CA). Cells were lysed and relative light units were assessed with a dual luciferase assay system (Promega Biotech Inc., Madison, WI). Firefly luciferase activity of reporter constructs was normalized to renilla luciferase activity of pRL-TK. RNA purification and quantitative PCR (Q-PCR) {#s2o} --------------------------------------------- BMMCs were stimulated with IgE receptor engagement as described above. Total RNA was extracted using TRIzol reagent (Invitrogen, Carlsbad, CA) and reverse-transcribed into cDNA using Super Script First-Strand Synthesis System for RT-PCR (Invitrogen). Q-PCR was performed by ABI Prism 7700 sequence detection system with SYBR Green PCR master mix (Applied Biosystems, Foster City, CA) using following primer pairs: p21^WAF1/CIP1^ (sense) 5′-CCTGGTGATGTCCGACCTG-3′; (antisense) 5′-CCATGAGCGCATCGCAATC-3′; BAX (sense) 5′-TGAAGACAGGGGCCTTTTTG-3′; (antisense) 5′-AATTCGCCGGAGACACTCG-3′; TIGAR (sense) 5′-CGCTTCGCCTTGACCGTTAT-3′; (antisense) 5′-ACCCAGTCTCCGAAAGGGG-3′; TNF-α (sense) 5′-ACAGAAAGCATGATCCGCG-3′; (antisense) 5′-GCCCCCCATCTTTTGGG-3′; IL-6 (sense) 5′-ACAACCACGGCCTTCCCTACTT-3′; (antisense) 5′-CACGATTTCCCAGAGAACATGTG-3′; Ifit2(sense) 5′-AGTACAACGAGTAAGGAGTCACT-3′; (antisense) 5′-AGGCCAGTATGTTGCACACATGG-3′; Ifit3 (sense) 5′-AGTGAGGTCAACCGGGAATCT-3′; (antisense) 5′-TCTAGGTGCTTTATGTAGGCCA-3′; The levels of each gene were normalized to the levels of cyclophylin A. Immunoprecipitation and immunoblotting {#s2p} -------------------------------------- Whole cell extracts were prepared and immunoblotting was performed as described previously [@pone.0025412-Suzuki3] using antisera to IKK2 (Cell Signaling Technology, Inc.), SNAP-23 (Novus Biologicals, Littleton, CO), O-GlcNAc (Covance, Princeton, NJ), p53 (Cell Signaling Technology, Inc.), and tubulin (Santa Cruz Biotechnology, Inc. Santa Cruz, CA). Antiserum to phospho-SNAP-23-Ser95 was a kind gift from Dr. Roche (National Institute of Health, Bethesda, MD). The specificity of these antibodies was described previously [@pone.0025412-Hepp1]. Immunoprecipitations were performed using anti-IKK2 antibodies (Cell Signaling Technology, Inc.) with Protein G Sepharose (Amersham Biosciences, Sweden). The aliquot of lysates was subjected to immunoblotting. Data analysis {#s2q} ------------- Data are summarized as means ± SD. The statistical analysis of the results was performed by the unpaired t-test. p values \<0.05 were considered significant. Results {#s3} ======= p53 expression is enhanced in mast cells upon IgE-mediated stimulation {#s3a} ---------------------------------------------------------------------- To determine the role of p53 in antigen receptor-mediated stimulation in mast cells, we first examined the kinetics of p53 expression upon IgE-mediated stimulation in bone marrow-derived mast cells (BMMCs). As shown in left panels of [Figure 1A](#pone-0025412-g001){ref-type="fig"}, IgE-mediated stimulation increased protein levels of p53 in BMMCs. Notably, the majority of BMMCs did not undergo apoptosis ([Figure 1A](#pone-0025412-g001){ref-type="fig"}, right panels) in spite of an increase in p53 expression. In contrast, Nutlin-3a [@pone.0025412-Liu1], a compound that induces p53 expression by blocking the interaction between p53 and Mdm2 [@pone.0025412-Vassilev1], strongly increased protein levels of p53 ([Figure 1B](#pone-0025412-g001){ref-type="fig"}, left panels) and induced apoptosis in mast cells ([Figure 1B](#pone-0025412-g001){ref-type="fig"}, right panels). Western blotting also showed that IgE-mediated stimulation increased the expression of p53 in wild type (WT) BMMCs but not in p53-deficient (p53^−/−^) BMMCs ([Figure 1C](#pone-0025412-g001){ref-type="fig"}). Not only Nutlin-3a but also IgE-mediated stimulation induced mRNA expression of p21^WAF1/CIP1^, one of direct targets of p53 [@pone.0025412-Polyak1], in BMMCs ([Figure 1D](#pone-0025412-g001){ref-type="fig"}). Importantly, whereas Nutlin-3a induced the expression of BAX, a pro-apoptotic target of p53 [@pone.0025412-Bensaad1], but not of TIGAR, an anti-apoptotic target of p53 [@pone.0025412-Bensaad1], in BMMCs, IgE-mediated stimulation induced the expression of TIGAR but not of BAX in BMMCs ([Figure 1E--F](#pone-0025412-g001){ref-type="fig"}). PUMA, another pro-apoptotic target of p53 [@pone.0025412-Villunger1], was not induced by IgE-mediated stimulation or Nutlin-3a (data not shown). These results suggest that IgE-mediated stimulation induces p53 expression without the induction of apoptosis in mast cells, presumably because of the induction of anti-apoptotic targets rather than pro-apoptotic targets. {ref-type="sec"}**. Cells were fixed, permeabilized, incubated with anti-p53-Alexa Fluor 647 (filled histograms) or isotype-matched IgG (dot line), and analyzed by flow cytometry. p53 expression at 24 hours after Nutlin-3a stimulation could not be analyzed because of apoptosis (ND; not determined). Apoptotic cells were assessed by dot plot analysis with Forward/Side Scattering analysis (Gate R1). Shown are representative data from 5 independent experiments. (**C**) WT BMMCs or p53^−/−^ BMMCs were stimulated with IgE receptor engagement as described in the **[Methods](#s2){ref-type="sec"}**. Before and 2, 6, and 24 hours after IgE receptor engagement, the expression of p53 and tubulin (as a control) was analyzed by immunoblotting. Shown are representative blots from 4 independent experiments. (**D--F**) BMMCs were stimulated with (black columns) or without (white columns) either IgE receptor engagement or Nutlin-3a. Four hours later, total RNA was isolated and Q-PCR analysis was performed. The expression of p21^WAF1/CIP1^ (**D**), BAX (**E**), or TIGAR (**F**) was normalized to the expression of cyclophylin A. Data are means ± SD of relative expression from 5 independent experiments. \*significantly different from the mean value of controls, \*p\<0.01.](pone.0025412.g001){#pone-0025412-g001} p53 expression is not required for the development of mast cells {#s3b} ---------------------------------------------------------------- We next examined whether the development of mast cells is affected by the lack of p53 using p53^−/−^ mice. The numbers of mast cells in the peritoneal cavity ([Figure S1A](#pone.0025412.s001){ref-type="supplementary-material"}), skin dermis (ear) ([Figure S1B](#pone.0025412.s001){ref-type="supplementary-material"}), or jejunal mucosa ([Figure S1C](#pone.0025412.s001){ref-type="supplementary-material"}) were indistinguishable between WT mice and p53^−/−^ mice. The development of BMMCs in p53^−/−^ mice was also similar to that in WT mice and more than 98% of cells recovered after 4 weeks of culture were c-kit positive (data not shown). The expression levels of c-kit and FcεRIα ([Figure S2A](#pone.0025412.s002){ref-type="supplementary-material"}) as well as morphological features ([Figure S2B](#pone.0025412.s002){ref-type="supplementary-material"}) were indistinguishable between WT and p53^−/−^ BMMCs. The lack of p53 expression in BMMCs from p53^−/−^ mice was confirmed by flow cytometry ([Figure S2C](#pone.0025412.s002){ref-type="supplementary-material"}). These results indicate that p53 is not required for the development of mast cells. Lack of p53 in mast cells results in enhanced anaphylaxis and degranulation {#s3c} --------------------------------------------------------------------------- To assess the physiological significance of p53 up-regulation upon IgE-mediated stimulation, we investigated the role of p53 in IgE-mediated anaphylaxis using a mast cell knock-in system [@pone.0025412-Galli1], [@pone.0025412-Galli2]. In this experiment, cultured p53^−/−^ BMMCs or WT BMMCs were intradermally transplanted into mast cell-deficient WBB6F1-Kit^w^/Kit^w-v^ (W/W^v^) mice. Four weeks after the transplantation, these mice were sensitized with DNP-specific IgE antibody and then challenged with DNP-HSA. Importantly, W/W^v^ mice reconstituted with p53^−/−^ BMMCs exhibited enhanced ear thickness ([Figure 2A](#pone-0025412-g002){ref-type="fig"}) and enhanced leakage of Evans blue dye ([Figure 2B--C](#pone-0025412-g002){ref-type="fig"}) as compared to W/W^v^ mice reconstituted with WT BMMCs. On the other hand, the numbers of mast cells in the ear were nearly identical between W/W^v^ mice reconstituted with p53^−/−^ BMMCs and WT BMMCs ([Figure S3](#pone.0025412.s003){ref-type="supplementary-material"}). Therefore, it is suggested that the enhanced anaphylaxis in W/W^v^ mice reconstituted with p53^−/−^ BMMCs is due to hyper-reactivity of p53^−/−^ mast cells to IgE-mediated stimulation. To address the cellular basis for the enhanced anaphylaxis in W/W^v^ mice reconstituted with p53^−/−^ BMMCs, we next compared IgE-mediated degranulation between p53^−/−^ BMMCs and WT BMMCs. Consistent with the enhanced anaphylaxis in W/W^v^ mice reconstituted with p53^−/−^ BMMCs, p53^−/−^ BMMCs showed a clear increase in IgE-induced degranulation assessed by the release of β-hexosaminidase ([Figure 2D](#pone-0025412-g002){ref-type="fig"}). p53^−/−^ BMMCs also showed an enhanced binding to Annexin-V, which specifically binds to phosphatidylserine that is exposed to outer surface by degranulation in mast cells, as compared to WT BMMCs ([Figure 2E--F](#pone-0025412-g002){ref-type="fig"}). These results suggest that p53 attenuates IgE-induced degranulation in mast cells. {ref-type="sec"}**. Ear swelling was quantified before (base line) and 1, 2, 4, and 6 hours after antigen challenge. n = 6 for each group. \*significantly different from the mean value of WT BMMCs, \*p\<0.01. (**B**) Representative photograph of leakage of injected Evans blue (EB) dye in the ear in W/W^v^ mice reconstituted with p53^−/−^ BMMCs or WT BMMCs. (**C**) Quantification of extravasated EB dye in the ear specimens. Data are means ± SD of absorbance at 620 nm, n = 6, each, \*p\<0.01. (**D**) WT BMMCs or p53^−/−^ BMMCs were stimulated with (black columns) or without (white columns) IgE receptor engagement. Degranulation was assessed by β-hexosaminidase assay. Data are means ± SD of the percent β-hexosaminidase release, n = 6, \*p\<0.01. (**E, F**) WT BMMCs or p53^−/−^ BMMCs were stimulated with or without IgE receptor engagement for 1 hour. Membrane fusion of BMMCs was assessed by Annexin-V binding by using flow cytometry. Shown are the representative histograms of Annexin-V binding (**E**) and means ± SD of mean fluorescence intensity (M.F.I.) of Annexin-V binding (**F**). n = 5, each, \*p\<0.01.](pone.0025412.g002){#pone-0025412-g002} Lack of p53 in mast cells results in enhanced kinase activity of IKK2 {#s3d} --------------------------------------------------------------------- Given that the phosphorylation of SNAP23 by IKK2 is a key biological process in mast cell degranulation and early phase allergic reactions [@pone.0025412-Suzuki1], we next examined whether p53 affects IKK2 functions. Importantly, we found a significant increase in IgE-induced phosphorylation of SNAP23 in p53^−/−^ BMMCs ([Figure 3A--B](#pone-0025412-g003){ref-type="fig"}). We also examined the role of IKK2 in the enhanced degranulation in p53^−/−^ BMMCs by using an IKK2-specific inhibitor ML120B [@pone.0025412-Nagashima1], [@pone.0025412-Greten1]. The efficacy of ML120B was confirmed by its effect on TNF-α-induced IκBα phosphorylation in WT BMMCs ([Figure S4](#pone.0025412.s004){ref-type="supplementary-material"}). Consistent with our previous report [@pone.0025412-Suzuki1], ML120B inhibited IgE-induced degranulation in WT BMMCs ([Figure 3C](#pone-0025412-g003){ref-type="fig"}). In addition, ML120B significantly inhibited IgE-induced degranulation in p53^−/−^ BMMCs ([Figure 3C](#pone-0025412-g003){ref-type="fig"}) to the level similar to that in ML120B-treated WT BMMCs ([Figure 3C](#pone-0025412-g003){ref-type="fig"}). Taken together, these results suggest that IKK2 activation is likely to be involved in the increase of IgE-induced degranulation in p53^−/−^ BMMCs. {ref-type="fig"}. Data are means ± SD of the percent β-hexosaminidase release, n = 5, \*p\<0.05. n.s. = not significant.](pone.0025412.g003){#pone-0025412-g003} IKK2 is O-GlcNAcylated in the absence of p53 in mast cells {#s3e} ---------------------------------------------------------- Recently, it has been reported that the absence of p53 results in the enhanced catalytic activity of IKK2 by causing O-GlcNAcylation at Ser733 in fibroblasts [@pone.0025412-Kawauchi1], [@pone.0025412-Kawauchi2]. We therefore examined O-GlcNAcylation of IKK2 in p53^−/−^ BMMCs and found that IKK2 was indeed O-GlcNAcylated in p53^−/−^ BMMCs but not in WT BMMCs ([Figure 4A](#pone-0025412-g004){ref-type="fig"}). In addition, the induction of TIGAR, which functions as a potent inhibitor of glycolysis [@pone.0025412-Bensaad1], was significantly increased in WT BMMCs but not in p53^−/−^ BMMCs upon IgE-mediated stimulation ([Figure 4B](#pone-0025412-g004){ref-type="fig"}). Therefore, it is possible that p53 reduces the levels of O-GlcNAcylation of IKK2 via the induction of TIGAR. {ref-type="fig"}. Data are means ± SD, n = 5, \*significantly different from the mean value of controls, \*p\<0.01. (**C**) 2-deoxy-D-glucose (2-DG) were given to W/W^v^ mice reconstituted with WT or p53^−/−^ BMMCs. Twenty hours later, PCA reaction was assessed as described in [**Figure 2C**](#pone-0025412-g002){ref-type="fig"}. Data are means ± SD, n = 6 for each, \*p\<0.05. (**D**) WT or p53^−/−^ BMMCs were incubated with 2-DG (4.5 mg/ml) for 2 hours. IgE-mediated degranulation was assessed as described in [**Figure 2D**](#pone-0025412-g002){ref-type="fig"}. Data are means ± SD of the percent β-hexosaminidase release, n = 5, \*p\<0.01. (**E**) WT BMMCs were incubated with streptozotocin (STZ; 5 mM) or vehicle for 3 hours. O-GlcNAcylation of IKK2 was detected as described in [**Figure 4A**](#pone-0025412-g004){ref-type="fig"}. Shown are representative blots from 5 independent experiments. (**F**) After incubation with STZ for 3 hours, WT BMMCs were subjected to IgE-mediated degranulation. Data are means ± SD of the percent β-hexosaminidase release, n = 5, \*significantly different from the mean value of vehicle, \*p\<0.05.](pone.0025412.g004){#pone-0025412-g004} To determine whether O-GlcNAcylation of IKK2 in p53^−/−^ BMMCs is involved in the enhanced anaphylaxis in W/W^v^ mice reconstituted with p53^−/−^ BMMCs ([Figure 2](#pone-0025412-g002){ref-type="fig"}), we investigated the effect of 2-deoxy-D-glucose (2-DG), a glycolytic inhibitor, on IgE-mediated anaphylaxis. As shown in [Figure 4C](#pone-0025412-g004){ref-type="fig"}, [2](#pone-0025412-g002){ref-type="fig"}-DG inhibited IgE-induced leakage of Evans blue dye not only in W/W^v^ mice reconstituted with WT BMMCs but also in W/W^v^ mice reconstituted with p53^−/−^ BMMCs. 2-DG also inhibited IgE-induced degranulation of p53^−/−^ BMMCs ([Figure 4D](#pone-0025412-g004){ref-type="fig"}). Because O-GlcNAcylation of IKK2 has been shown to enhance its kinase activity [@pone.0025412-Kawauchi2], we next examined the effect of streptozotocin (STZ), an O-GlcNAcase inhibitor, on IgE-induced degranulation. STZ enhanced O-GlcNAcylation of IKK2 ([Figure 4E](#pone-0025412-g004){ref-type="fig"}) and enhanced IgE-induced degranulation ([Figure 4F](#pone-0025412-g004){ref-type="fig"}) in WT BMMCs. Taken together, these results suggest that O-GlcNAcylation of IKK2 is a possible mechanism of the enhanced degranulation in p53^−/−^ BMMCs. p53 attenuates late phase allergic reactions and NF-κB-mediated cytokine production {#s3f} ----------------------------------------------------------------------------------- Late phase responses in IgE-mediated anaphylaxis are promoted by mast cell-derived pro-inflammatory cytokines such as TNF-α and IL-6 [@pone.0025412-Galli1]. We next examined late phase responses in IgE-mediated anaphylaxis in W/W^v^ mice reconstituted with p53^−/−^ BMMCs and WT BMMCs. As shown in [Figure 5A, W](#pone-0025412-g005){ref-type="fig"}/W^v^ mice reconstituted with p53^−/−^ BMMCs exhibited strong ear swelling as compared with W/W^v^ mice reconstituted with WT BMMCs. Consistent with strong ear swelling in W/W^v^ mice reconstituted with p53^−/−^ BMMCs, IgE-induced TNF-α and IL-6 secretion was significantly facilitated in p53^−/−^ BMMCs as compared with that in WT BMMCs ([Figure 5B--C](#pone-0025412-g005){ref-type="fig"}). IgE-induced mRNA induction of TNF-α and IL-6 was also facilitated in p53^−/−^ BMMCs ([Figure 5D--E](#pone-0025412-g005){ref-type="fig"}). These results suggest that p53 inhibits IgE-induced pro-inflammatory cytokine production in mast cells and inhibits subsequent late phase responses in IgE-mediated anaphylaxis. {ref-type="sec"}**. Data were means ± SD of IgE-mediated increase in ear thickness, n = 6 for each genotype, \*significantly different from the mean value of WT BMMCs, \*p\<0.01. (**B--E**) WT or p53^−/−^ BMMCs were stimulated with (black columns) or without (white columns) IgE receptor engagement. (**B, C**) The levels of TNF-α and IL-6 in the supernatants were quantified by ELISA. Data were means ± SD, n = 5, \*significantly different from the mean value of WT BMMCs, \*p\<0.01. ND = not detected. (**D, E**) Two hours after IgE receptor engagement, the levels of TNF-α and IL-6 mRNA were quantified by Q-PCR. Data were means ± SD of relative expression, n = 5, \*significantly different from the mean value of WT BMMCs, \*p\<0.05. (**F, G**) WT or p53^−/−^ BMMCs were incubated with ML120B or vehicle for 60 minutes and then stimulated with (black columns) or without (white columns) IgE receptor engagement. Sixty minutes later, the expression of Ifit2 (**F**) and Ifit3 (**G**) was assessed by Q-PCR analysis. n = 5, \*p\<0.01. (**H**) HEK293 cells were transfected with the expression vectors of IKK2SE, p65, and/or p53 in the presence of NF-κB-luciferase reporter construct. The luciferase activity of NF-κB reporter construct was quantified as described in the **[Methods](#s2){ref-type="sec"}.** Data are means ± SD of the percent luciferase activity, n = 5, \*p\<0.01.](pone.0025412.g005){#pone-0025412-g005} In the NF-κB pathways, p65 directly induces the expression of a number of target genes in a various cell types [@pone.0025412-Hayden1]. We next examined IgE-induced expression of Ifit2 and Ifit3, whose induction by LPS is shown to be dependent on p65 [@pone.0025412-Ogawa1], in WT BMMCs and p65-deficient (p65^−/−^) BMMCs. As expected, the induction of Ifit2 and Ifit3 by IgE-mediated stimulation was detected in WT BMMCs but not in p65^−/−^ BMMCs ([Figure S5A--B](#pone.0025412.s005){ref-type="supplementary-material"}), indicating that IgE-mediated induction of Ifit2 and Ifit3 is also dependent on p65. IKK2-specific inhibitor ML120B also inhibited IgE-mediated induction of Ifit2 and Ifit3 in WT BMMCs ([Figure 5F--G](#pone-0025412-g005){ref-type="fig"}). Consistent with the enhanced catalytic activity of IKK2 in p53^−/−^ BMMCs ([Figure 3A--B](#pone-0025412-g003){ref-type="fig"}), IgE-mediated induction of Ifit2 and Ifit3 was significantly enhanced in p53^−/−^ BMMCs and ML120B inhibited IgE-mediated induction of Ifit2 and Ifit3 in p53^−/−^ BMMCs ([Figure 5F--G](#pone-0025412-g005){ref-type="fig"}). However, importantly, even in the presence of ML120B, IgE-mediated induction of Ifit2 and Ifit3 was still observed in p53^−/−^ BMMCs ([Figure 5F--G](#pone-0025412-g005){ref-type="fig"}). The residual activity of p65-dependent transcription in ML120B-treated p53^−/−^ BMMCs but not in ML120B-treated WT BMMCs suggest that besides the inhibition of IKK2, p53 may directly regulate the activity of p65. To address the possibility that p53 directly inhibits the activity of p65, we finally investigated the effect of ectopic expression of p53 on NF-κB-dependent promoter activation induced by a phosphomimetic mutant IKK2, IKK2SE, or by p65. As shown in [Figure 5H](#pone-0025412-g005){ref-type="fig"}, p53 repressed not only IKK2SE-induced activation of NF-κB-dependent promoter but also p65-induced activation of NF-κB-dependent promoter, indicating that p53 directly represses p65 activity. Taken together, these results indicate that p53 attenuates IgE-mediated expression of pro-inflammatory cytokine genes not only by suppressing the catalytic activity of IKK2 through the modulation of glycosylation but also by suppressing the transcriptional activity of p65. Discussion {#s4} ========== In this study, we show that p53 functions as a negative regulator of mast cell activation through the inhibition of NF-κB pathways. We found that protein levels of p53 were up-regulated upon IgE-mediated stimulation in mast cells without the induction of apoptosis ([Figure 1](#pone-0025412-g001){ref-type="fig"}). We also found that the lack of p53 in mast cells results in enhanced IgE-induced degranulation ([Figure 2](#pone-0025412-g002){ref-type="fig"}) and cytokine production ([Figure 5](#pone-0025412-g005){ref-type="fig"}), leading to enhanced responses in both early phase ([Figure 2](#pone-0025412-g002){ref-type="fig"}) and late phase ([Figure 5](#pone-0025412-g005){ref-type="fig"}) of IgE-mediated anaphylaxis in vivo. These results suggest that p53 is involved in negative feedback regulation of IgE-madiated mast cell activation. It is well established that stress-associated stimulation, such as apoptosis, cell cycle arrest, and senescence, increases p53 expression, leading to apoptosis of various cell types [@pone.0025412-Riley1]--[@pone.0025412-Kruse1]. In contrast, we found that IgE-induced increase of p53 expression did not affect mast cell survival ([Figure 1A](#pone-0025412-g001){ref-type="fig"}). We also found that whereas Nutlin-3a, a compound that induces p53 expression [@pone.0025412-Vassilev1], induced the expression of a pro-apoptotic p53 target BAX but not of an anti-apoptotic p53 target TIGAR in mast cells, IgE-mediated stimulation induced the expression of TIGAR but not of BAX in mast cells ([Figure 1E--F](#pone-0025412-g001){ref-type="fig"}). These results suggest that the balance between a pro-apoptotic p53 target gene and an anti-apoptotic p53 target gene could be involved in the non-apoptotic feature of mast cells upon IgE-induced p53 expression. The reasons why IgE-mediated stimulation induces an anti-apoptotic p53 target TIGRR while p53 activation by Nutlin-3a induces a pro-apoptotic p53 target BAX in mast cells are still unclear. It is well established that Nutlin-3a induces p53 expression by blocking the interaction between p53 and Mdm2 [@pone.0025412-Vassilev1]. On the other hand, although a molecular machinery_of IgE-mediated up-regulation of p53 has not been clarified, IgE-mediated stimulation activates several signaling pathways including PKC pathway, calcineurin/NFAT pathway, Ras/RAF/MEK/ERK pathway, Rac/MEKK/JNK pathway, and IKK complex/NF-κB pathway in mast cells [@pone.0025412-Galli1]. Therefore, it is possible that the activation of these signaling pathways by IgE-mediated stimulation affects the balance between the expression of BAX and TIGAR in mast cells. Regarding the mechanism by which p53 attenuates IgE-mediated stimulation, we show that the lack of p53 results in a significant increase in IgE-induced phosphorylation of SNAP23 in mast cells ([Figure 3A--B](#pone-0025412-g003){ref-type="fig"}), a key biological process in mast cell degranulation and early phase allergic reactions [@pone.0025412-Suzuki1]. In addition, we found that the inhibition of IKK2, which phosphorylates SNAP23 upon IgE-mediated stimulation in mast cells [@pone.0025412-Suzuki1], cancelled the enhanced IgE-induced degranulation in p53^−/−^ BMMCs ([Figure 3C](#pone-0025412-g003){ref-type="fig"}), suggesting that IKK2-mediated phosphorylation of SNAP23 is involved in the enhanced IgE-induced degranulation in p53^−/−^ mast cells. Moreover, the expression of TIGAR, which functions as a potent inhibitor of glycolysis [@pone.0025412-Bensaad1], was significantly increased in WT BMMCs but not in p53^−/−^ BMMCs upon IgE-mediated stimulation ([Figure 4B](#pone-0025412-g004){ref-type="fig"}). Furthermore, Kawauchi et al. have recently shown that the absence of p53 results in the enhanced catalytic activity of IKK2 by causing O-GlcNAcylation in fibroblasts [@pone.0025412-Kawauchi2]. Taken together, these results raise the possibility that IgE-mediated stimulation induces the expression of TIGAR through the induction of p53 and subsequently TIGAR reduces the levels of O-GlcNAcylated IKK2, leading to the attenuation of IgE-mediated degranulation and cytokine production in mast cells. We found that O-GlcNAcylation of IKK2 was detected in p53^−/−^ BMMCs but not in WT BMMCs ([Figure 4A](#pone-0025412-g004){ref-type="fig"}). We also found that 2-DG inhibited IgE-induced degranulation in p53^−/−^ BMMCs but not in WT BMMCs ([Figure 4D](#pone-0025412-g004){ref-type="fig"}). These results suggest that O-GlcNAcylation of IKK2 is almost completely inhibited by the presence of p53 in WT BMMCs. In contrast to these in vitro studies, 2-DG treatment inhibited IgE-mediated leakage of Evans blue dye in W/W^v^ mice reconstituted not only with p53^−/−^ BMMCs but with WT BMMCs ([Figure 4C](#pone-0025412-g004){ref-type="fig"}). The discrepancy raises the possibility that the inhibition of O-GlcNAcylation of IKK2 by p53 is incomplete in mast cells in the in vivo studies. It is also possible that the glycolytic modification of some signaling molecule besides IKK2 is involved in IgE-mediated increase_of vascular permeability in the in vivo studies. Further experiments are required to address these possibilities. In conclusion, the evidence provided here suggests that p53 plays a crucial role in controlling the intensity and/or duration of mast cell responses to antigens by suppressing IgE receptor-IKK2-NF-κB axis at multiple steps. Supporting Information {#s5} ====================== ###### **The numbers of mast cells in the tissues are indistinguishable between WT mice and p53^−/−^ mice.** The numbers of mast cells in the peritoneal cavity (**A**), skin dermis (ear) (**B**), and jejunal mucosa (**C**) were evaluated as described in the **[Methods](#s2){ref-type="sec"}.** Data are means ± SD from 5 independent experiments. vcu = villus crypt unit. (TIF) ###### Click here for additional data file. ###### **Development of IL-3-dependent bone marrow-derived mast cells (BMMCs) is normal in p53^−/−^ mice.** (**A**) Bone marrow cells from WT mice or p53^−/−^ mice were cultured in the presence of IL-3 for 4 weeks. Cells were stained with anti-FcεRIα FITC and anti-c-kit APC and analyzed by flow cytometry. Shown are representative FACS profiles from five independent experiments. (**B**) BMMCs stained with Wright-Giemsa solution are shown. (**C**) p53 expression in WT BMMCs and p53^−/−^ BMMCs was evaluated by flow cytometry. (TIF) ###### Click here for additional data file. ###### **The numbers of mast cells in the ear skin dermis are indistinguishable between W/W^v^ mice reconstituted with WT BMMCs and p53^−/−^ BMMCs.** Four weeks after the transplantation of BMMCs, the numbers of mast cells in the ear skin dermis were assessed. (TIF) ###### Click here for additional data file. ###### **Selective IKK2 inhibitor ML120B inhibits phosporylation of IκBα.** (**A, B**) BMMCs were incubated with ML120B (10 µM) or vehicle (DMSO (0.01%)) as a control for 60 minutes and then stimulated with or without TNF-α (10 ng/ml). Fifteen minutes later, cell lysates were recovered and analyzed by immunoblot analysis using phosphorylation site-specific antibodies. Representative blots from five independent experiments **(A)** and means ± SD of the density of blots **(B)** were shown. \*significantly different from the mean value of vehicle, \*p\<0.01. d.u. = density unit. (TIF) ###### Click here for additional data file. ###### **IgE-induced Ifit2 and Ifit3 expression is p65 dependent.** (**A, B**) WT BMMCs or p65^−/−^ BMMCs were stimulated with or without IgE receptor engagement. Two hours later, total RNA was extracted and Q-PCR for Ifit2 (**A**) or Ifit3 (**B**) was performed. Data are means ± SD of relative expression from 5 independent experiments. \*significantly different from the mean value of controls, \*p\<0.01. (TIF) ###### Click here for additional data file. We are very grateful to Millennium Pharmaceuticals (Cambridge, MA) for a kind gift of ML120B. **Competing Interests:**The authors have declared that no competing interests exist. **Funding:**K.S. is supported in part by Grants-in Aids for Scientific Research from the Ministry of Education, Culture, Sports, Science, and Technology, the Japanese Government and by Global COE Program (Global Center for Education and Research in Immune System Regulation and Treatment), MEXT, Japan. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. [^1]: Analyzed the data: KS SHM YX MY DN. Wrote the paper: KS FL HN. Designed research: KS IMV HN. Performed research: KS SHM YX MY DN.

Introduction {#Sec1} ============ Artificial radioactive isotopes are present in the environment because of human activities, and these isotopes include (a) fission products generated in nuclear explosions in the atmosphere, which were performed by several countries until 1980, and (b) the products of the Chernobyl nuclear reactor accident in April 1986. The radionuclides ^137^Cs and ^90^Sr are present in the environment at higher concentrations than other artificial radionuclides. The systematic monitoring of radionuclides in the Polish environment has been performed by the Central Laboratory for Radiological Protection (CLOR) since the early 1960s. For two decades the monitoring study was focused on radionuclide contamination in surface waters (the Baltic Sea and Rivers), tap water, soil, air, agricultural products, and foodstuffs that was predominantly caused by nuclear explosions in the atmosphere. Short- and long-lived radionuclides were measured, and the main indicators of long-term pollution related to weapons were ^90^Sr (*t*~1/2~ = 28.8 year) and ^137^Cs (*t*~1/2~ = 30.17 year). The extent of the monitoring study was increased dramatically after the Chernobyl disaster. The CLOR and the network of stations that form the service for the measurement of radioactive contamination has been required to perform high-frequency analyses of the total β-particle activities in surface water (rivers and lakes) and tap water samples \[[@CR1]\]. The CLOR started a systematic study of radioactive contamination in water and bottom sediment samples from Polish rivers and lakes in 1994. At present, six lakes in different areas and the two main river systems in Poland, the Vistula River and two of its tributaries (the Bug and Narew Rivers) and the Oder River and its main eastern tributary (the Warta River), are regularly monitored. The activity concentrations of ^137^Cs and ^90^Sr are measured in the water samples, and the ^137^Cs and ^239,240^Pu activity concentrations are determined in the bottom sediment samples. Here, we present the results of the analyses of the water samples for ^137^Cs and ^90^Sr. Experimental {#Sec2} ============ Description of sampling sites {#Sec3} ----------------------------- The Vistula River is the longest river in Poland, with a total length of 1,022 km. The part of the Vistula River drainage basin that is in Poland measures 168,868 km^2^. The main tributaries of the Vistula River come from the east, and are the Bug River (590 km of which is within Poland) and the Narew River (443 km of which is within Poland). The sampling points on the Vistula River were in the upper, middle, and lower parts, in Kraków, Annopol, Warsaw, Płock, and Kiezmark. The sampling points on the Narew and Bug Rivers were in Pułtusk and Wyszków, respectively. The Oder River is 840 km long, and 726 km of the river is within Poland and along the Polish--German border. The part of the Oder River drainage basin that is within Poland measures 106,043 km^2^. The main tributary of the Oder River comes from the east, and is the Warta River, which is 795 km long. The sampling points on the Oder River were at Chałupki, Wrocław, Głogów, and Krajnik, and the sampling point on the Warta River was at Poznań. Six lakes, each located in a different part of Poland, were selected for the water monitoring program. The lakes were Drawsko Lake (in the West Pomeranian District), Niesłysz Lake (in the Lubuskie District), Wadąg and Partęczyny Lakes (in the Masurian District), Wigry Lake (in the Podlaskie District), and Rogóźno Lake (in the Lubelskie District). The largest of the lakes were Wigry Lake (which has an area of 21.2 km^2^ and a depth of 73 m) and Drawsko Lake (which has an area of 17.8 km^2^ and a depth of 79.7 m), and Rogóźno Lake (which has an area of 0.57 km^2^ and a depth of 25.5 m) was the smallest lake. The locations of the river and lake sampling sites are shown in Fig. [1](#Fig1){ref-type="fig"}.Fig. 1Sampling sites (*circles* indicate river sampling cites and*triangles* indicate lake sampling sites) Sample preparation {#Sec4} ------------------ Water samples were collected in the spring and autumn each year. The water samples were 20 L, and they were collected from the main stream of each river and from a platform extending from the shore into each lake. The water samples were put into polyethylene containers and acidified with HNO~3~. In the laboratory, each water sample was evaporated to 450 mL and filtered through a hard filter paper. The filter was then ashed in an oven at 450 °C, and the ash was dissolved in nitric acid and combined with the filtrate. Finally, each water sample was evaporated to 250 mL and the ^137^Cs and ^90^Sr concentrations in the sample were determined. Analytical method {#Sec5} ----------------- The ^137^Cs in a sample was separated from the solution by filtering the sample through a ^137^Cs-selective bed of ammonium molybdophosphate in a radiochemical funnel \[[@CR2]\]. The β-particle activity in the ammonium molybdophosphate bed was then measured using a low-level beta GM multicounter system (DTU Nutech, Roskilde, Denmark). After the cesium was removed, the ^90^Sr concentration in the filtrate was determined using the Volchok method \[[@CR3]\] with some modifications. The ^90^Sr concentration was determined from the β-radiation emitted by ^90^Y once equilibrium was reached in the ^90^Sr--^90^Y system. The same conditions were used for the ^90^Sr measurements as were used for the ^137^Cs measurements. The accuracy of each of the analytical methods was verified in national and international comparison exercises. Results and discussion {#Sec6} ====================== The ^137^Cs and ^90^Sr concentrations in the water samples from the Vistula and Oder Rivers and their tributaries and from the lakes are shown in Table [1](#Tab1){ref-type="table"}. The samples were collected in the same periods (spring and autumn) in 2012 and 2013, although the atmospheric and hydrologic conditions were different in the different years. There was a drought in 2012, and the spring samples were collected when there were low or very low water levels. This may have affected the overall results.Table 1^137^Cs and ^90^Sr activity concentrations (mBq L^−1^) in water samples from the Vistula and Oder Rivers, their tributaries, and six lakes in 2012 and 2013River/lakeSampling point^137^Cs activity concentration (mBq L^−1^)^90^Sr activity concentration (mBq L^−1^)2012201320122013SpringAutumnSpringAutumnSpringAutumnSpringAutumnVistula River and its tributaries VistulaKraków-Tyniec7.05 ± 0.433.40 ± 0.295.65 ± 0.634.18 ± 0.494.29 ± 0.337.24 ± 0.334.46 ± 0.872.56 ± 0.52Annopol12.33 ± 0.583.67 ± 0.302.63 ± 0.373.59 ± 0.435.32 ± 0.364.18 ± 0.234.38 ± 0.813.25 ± 0.66Warsaw9.71 ± 0.491.25 ± 0.182.97 ± 0.371.92 ± 0.189.61 ± 0.512.18 ± 0.164.62 ± 0.903.41 ± 0.70Płock2.64 ± 0.243.68 ± 0.292.99 ± 0.372.82 ± 0.264.00 ± 0.323.26 ± 0.275.05 ± 0.961.96 ± 0.40Kiezmark3.04 ± 0.272.07 ± 0.213.27 ± 0.402.64 ± 0.353.34 ± 0.272.71 ± 0.175.71 ± 1.201.59 ± 0.33 NarewPułtusk2.48 ± 0.241.29 ± 0.232.77 ± 0.352.43 ± 0.314.39 ± 0.381.59 ± 0.133.61 ± 0.692.46 ± 0.50 BugWyszków2.13 ± 0.222.33 ± 0.342.00 ± 0.274.30 ± 0.495.08 ± 0.373.32 ± 0.1911.59 ± 2.181.56 ± 0.32Oder River and its tributaries OderChałupki40.72 ± 1.044.95 ± 0.351.66 ± 0.247.25 ± 0.793.26 ± 0.284.86 ± 0.274.88 ± 0.883.01 ± 0.61Wrocław6.49 ± 0.416.00 ± 0.405.16 ± 0.585.30 ± 0.593.72 ± 0.304.03 ± 0.224.60 ± 0.833.70 ± 0.76Głogów3.30 ± 0.284.23 ± 0.324.42 ± 0.414.22 ± 0.492.15 ± 0.254.90 ± 0.244.65 ± 0.882.54 ± 0.52Krajnik3.73 ± 0.311.98 ± 0.222.41 ± 0.162.03 ± 0.152.65 ± 0.242.22 ± 0.164.83 ± 0.873.45 ± 0.70 WartaPoznań10.96 ± 0.542.41 ± 0.171.77 ± 0.252.18 ± 0.294.39 ± 0.353.18 ± 0.196.00 ± 1.166.22 ± 1.27Lakes WigryStary Folwark2.86 ± 0.261.97 ± 0.212.33 ± 0.302.94 ± 0.362.76 ± 0.262.86 ± 0.282.52 ± 0.481.42 ± 0.29 WadągMyki2.54 ± 0.241.59 ± 0.181.50 ± 0.221.88 ± 0.262.79 ± 0.273.12 ± 0.192.65 ± 0.572.26 ± 0.46 Wielkie  PartęczynyPartęczyny4.89 ± 0.344.34 ± 0.322.70 ± 0.341.06 ± 0.172.20 ± 0.233.18 ± 0.302.16 ± 0.451.69 ± 0.34 DrawskoStare Drawsko4.78 ± 0.347.25 ± 0.432.81 ± 0.300.76 ± 0.145.36 ± 0.394.34 ± 0.213.67 ± 0.724.68 ± 0.96 NiesłyszNiesulice3.60 ± 0.293.80 ± 0.315.63 ± 0.562.67 ± 0.342.49 ± 0.251.60 ± 0.132.49 ± 0.491.35 ± 0.28 RogóźnoRogóźno9.07 ± 0.505.38 ± 0.265.88 ± 0.655.30 ± 0.596.19 ± 0.419.10 ± 0.368.36 ± 1.637.56 ± 1.55 The highest ^137^Cs concentration (12.33 mBq L^−1^) in the water samples from the Vistula River was found in the sample that was collected in spring 2012 in Annopol. Relatively high ^137^Cs concentrations were also found in the samples collected from Warsaw (9.71 mBq L^−1^) and Kraków (7.05 mBq L^−1^) in spring 2012. The concentrations in the samples from Kraków and Annopol collected in autumn 2012 (3.40 mBq L^−1^ in Kraków and 3.67 mBq L^−1^ in Annopol) were lower by a factor of two to three than the concentrations in the samples collected in the spring. The concentrations in the samples from Warsaw were lower by a factor of eight (at 1.25 mBq L^−1^) in the autumn than in the spring. The differences between the concentrations found in the river water samples collected in the spring and autumn in 2013 were fairly small. Relatively high ^137^Cs concentrations were only found in the river water samples from Kraków in 2013 (5.65 mBq L^−1^ in the spring and 4.18 mBq L^−1^ in the autumn). The water samples collected from the lower Vistula River in 2012 and 2013 were characterized by low ^137^Cs concentrations and only small variations in the concentrations, and this agrees with data for previous years \[[@CR4]\]. The mean ^137^Cs concentrations in the samples collected from the Vistula River in Płock and Kiezmark in 2012 and 2013 were 3.03 ± 0.45 and 2.76 ± 0.53 mBq L^−1^, respectively. The ^137^Cs concentrations found in the water samples from the Narew and Bug Rivers were generally lower (2--3 mBq L^−1^ in both rivers) than the concentrations that were found in the samples from the Vistula River in the same sampling periods. The only exception to this being in the autumn of 2013, at which time the sample collected from the Bug River contained a ^137^Cs concentration of 4.3 mBq L^−1^. The ^137^Cs concentrations were distinctly different in the samples from the different sampling points on the Oder River, but the ^137^Cs concentrations were quite consistent over the entire sampling period at each sample site. The mean ^137^Cs concentrations in the samples collected in Wrocław, Głogów, and Krajnik were 5.74 ± 0.62, 4.04 ± 0.50, and 2.54 ± 0.82 mBq L^−1^, respectively, and it can be seen that the concentration clearly decreased moving downstream. The largest temporal differences in the ^137^Cs concentrations were found at the Chałupki site, where the ^137^Cs concentrations were 40.72 mBq L^−1^ in spring 2012 and 1.66 mBq L^−1^ in spring 2013. The Chałupki site is in the region of Poland that suffered the most contamination from the Chernobyl accident. Local meteorological conditions could, therefore, have affected the ^137^Cs concentrations found in the samples that we analyzed. Similar to the difference in the ^137^Cs concentrations in the water collected in Chałupki, the ^137^Cs concentrations in the samples collected from the Warta River were five times higher in spring 2012 than in spring 2013. The ^137^Cs concentrations in the samples from the lakes varied strongly between sample sites and between sampling periods. The highest ^137^Cs concentration was found in water from the small Rogóźno Lake, and significantly lower concentrations were found in the samples from the larger and deeper lakes (Wadąg and Wigry Lakes) in northeastern Poland. The mean ^137^Cs concentrations found in the samples collected in 2012 and 2013 were 6.41 ± 1.79 mBq L^−1^ for Rogóźno Lake, 1.88 ± 0.49 mBq L^−1^ for Wadąg Lake, and 2.53 ± 0.46 mBq L^−1^ for Wigry Lake. Similar differences between ^137^Cs concentrations in water samples from the lakes were found in previous studies \[[@CR4], [@CR5]\]. The ^90^Sr concentration in the river water samples varied less between sample sites and temporally than did the ^137^Cs concentrations. The ^90^Sr concentration ranged from 1.59 to 9.61 mBq L^−1^ over the study period, and the means (over 2012 and 2013) were 4.32 ± 1.20 mBq L^−1^ for the samples from the Vistula River, 4.02 ± 1.17 mBq L^−1^ for the samples from the tributaries of the Vistula River, 3.54 ± 0.64 mBq L^−1^ for the samples from the Oder River, and 4.39 ± 1.00 mBq L^−1^ for the samples from the tributaries of the Oder River. The highest ^90^Sr concentration in the river water samples (9.61 mBq L^−1^) was found in a sample from the Vistula River collected in Warsaw in spring 2012, but a relatively high concentration (7.24 mBq L^−1^) was found in a sample from the Vistula River collected in Kraków in autumn 2012. In 2012, the ^90^Sr concentration in the samples from the Oder (collected at Głogów) ranged from 2.15 mBq L^−1^ in the spring to 4.9 mBq L^−l^ in the autumn, and these samples were collected when the water levels were low. The ^90^Sr concentration in the sample collected in Chałupki in spring 2013 was 4.88 mBq L^−1^, and this decreased to 3.01 mBq L^−1^ in the sample collected in autumn 2013. We concluded that the mean ^90^Sr concentrations in the samples from both rivers were very similar. However, the ^90^Sr concentrations in the samples from the Warta River were somewhat more variable, and the mean ^90^Sr concentration in 2013 (6.11 mBq L^−1^) was higher than the mean ^90^Sr concentration in 2012 (3.79 mBq L^−1^). As was the case for ^137^Cs, the ^90^Sr concentrations were higher in the samples from the small Rogóźno Lake (in southeastern Poland), ranging from 6.19 mBq L^−1^ in spring 2012 to 9.2 mBq L^−1^ in autumn 2012, than the ^90^Sr concentrations in the samples from the other lakes. The ^90^Sr concentrations in the samples from Drawsko Lake were also relatively high, ranging from 3.67 mBq L^−1^ in spring 2013 to 5.36 mBq L^−1^ in spring 2012. Lower ^90^Sr concentrations were found in Niesłysz Lake (1.35 and 1.60 mBq L^−1^ in the autumn samples and 2.49 mBq L^−1^ in the spring samples). The mean ^90^Sr concentrations in the samples from the lakes were 3.84 ± 2.14 mBq L^−1^ in 2012 and 3.40 ± 2.33 mBq L^−1^ in 2013. As is stated in the introduction, the radioactive contamination in water was monitored using the total β-particle activity before 1986. The mean β-particle activity found in the Polish water samples that were analyzed in 1986 was 1.5 Bq L^−1^, and the highest total β-particle activity reached 417 Bq L^−1^ in May 1986, when the Chernobyl disaster occurred. The total β-particle activities found in water samples in 2001 were 24--386 mBq L^−1^. Since 1986, the ^137^Cs in Polish water samples has predominantly originated from the fallout from the Chernobyl accident, which caused severe contamination in some regions, especially in eastern and southwestern Poland. The ^137^Cs concentrations in soil show that the amount of ^137^Cs contamination in the Polish environment has decreased by a factor of 2.5 since 1988, but the geographical variations in the contamination have not changed \[[@CR6]\]. There is still a clear relationship between the geographical location and the ^137^Cs concentrations in soil and water samples. The main source of ^137^Cs to rivers is it being washed out of the soil by water. This partly explains why there are higher ^137^Cs concentrations in the upper parts of the Vistula and Oder Rivers, which are in southern Poland, than in the rivers near to their Baltic Sea estuaries. There was a correlation between the locations of the lakes and the radioactive fallout distribution. A particular case is Rogóźno Lake (in eastern Poland) which has no inflowing or outflowing rivers, so no water exchange can occur, and the only water supplies to the lake are springs and precipitation. All of the other lakes that were studied belong to water systems that include rivers. The impact of contamination caused by the Chernobyl disaster and meteorological conditions on the ^137^Cs concentrations in the river water samples can be identified by inspecting the long-term trends in the ^137^Cs concentrations. The plots shown in Fig. [2](#Fig2){ref-type="fig"} (see Table [1](#Tab1){ref-type="table"} for the detailed results) represent the mean annual ^137^Cs concentrations over 20 y (1994--2013) at selected sampling points in the upper and lower parts of the Vistula River (Kraków and Kiezmark) and the Oder River (Chałupki and Krajnik) \[[@CR4], [@CR5], [@CR7], [@CR8]\].Fig. 2Annual average ^137^Cs activity concentrations in water samples from four sampling sites The levels of radioactive contamination were higher in the samples collected in the upper parts of the rivers, at Kraków on the Vistula River and at Chałupki on the Oder River, than in the samples collected in the lower parts of the rivers. This agrees with the fact that southern Poland was contaminated more than northern Poland with radioactive fallout from the Chernobyl accident. The plots in Fig. [3](#Fig3){ref-type="fig"} show the ^137^Cs concentrations found in the lakes that were studied between 1994 and 2005 and between 2007 and 2013 \[[@CR4], [@CR7], [@CR8]\]. The ^137^Cs concentration remained more constant over the years in the lakes than it did in the main rivers. Again, the exception was Rogóźno Lake, in which the ^137^Cs concentration decreased significantly over the study period.Fig. 3Annual average ^137^Cs activity concentration in water samples from the lakes that were studied It should be stressed that the first measurements of ^90^Sr contamination of water in Poland were performed in 1961. A number of nuclear explosions, especially the intense period of nuclear tests in 1951--1958 and 1961--1962, caused the ^90^Sr concentrations in the water samples collected in 1961 to have a very broad range, from 4 to 104 mBq L^−1^. The ^90^Sr concentrations in 1962 were 17--60 mBq L^−1^ . The mean ^90^Sr concentration in water in the Vistula estuary on the Baltic Sea was 10.2 ± 1.5 mBq L^−1^ in 1983 and the ^137^Cs concentration was only 1.0 ± 0.1 mBq L^−1^. Similar concentrations have been found in lakes in northern Poland, e.g., the ^90^Sr and ^137^Cs concentrations in water samples from Żarnowieckie Lake in the late 1970s were 9.58 ± 1.39 and 2.68 ± 0.21 mBq L^−1^, respectively \[[@CR9]\]. Fallout connected with nuclear bomb explosions around the world between 1945 and 1980 is still the main source of ^90^Sr in the environment. Comparable amounts of ^90^Sr and ^137^Cs were present in the bomb-related fallout. Smaller amounts of ^90^Sr than ^137^Cs were released by the Chernobyl accident because Sr is not very volatile, and this means that the environment was affected by much less Sr contamination than Cs contamination. Therefore, the presence of ^90^Sr in the environment is mainly related to nuclear weapons testing in the atmosphere, and the reasons for the ^90^Sr variations that are observed in water bodies are rather complex \[[@CR10]\]. The ^90^Sr and ^137^Cs concentrations in the samples from the rivers belonging to the Vistula and Oder drainage basins are shown in Fig. [4](#Fig4){ref-type="fig"}.Fig. 4Annual average ^90^Sr and ^137^Cs activity concentrations in the Vistula and Oder drainage basins Large floods in the spring or early summer in 1997, 2007, and 2010 affected the ^137^Cs activity concentrations in the samples from the Vistula and Oder drainage basins. The surface water was diluted by rainwater particularly strongly in 1997. Large amounts of rainwater entered one or both of the drainage basins in those years (e.g., in 2010 the average water flow in the upper Oder River was 124 m^3^ s^−1^, while the average water flow in the upper Vistula River was only 43 m^3^ s^−1^) \[[@CR11]\]. The ^137^Cs and ^90^Sr concentrations found in 1997, 2007, and 2010 were, therefore, lower than the concentrations that were found in the following years. This is particularly clear when the concentrations found in 1998 and 2012 are inspected (the river levels were low in 2012, and the average water flows in the upper Oder River and the upper Vistula River were 43.5 m^3^ s^−1^ and 14.9 m^3^  s^−1^, respectively). The average ^137^Cs concentration in the Vistula drainage basin in 1997 was 3.78 ± 0.31 mBq L^−1^, and the concentration had increased to 4.82 ± 0.42 mBq L^−1^ in 1998. The average ^137^Cs concentrations in the Oder drainage basin were 5.16 ± 0.49 mBq L^−1^ in 1997 and 8.26 ± 0.78 mBq L^−1^ in 1998. The ^137^Cs concentrations were 4.08 ± 0.21 mBq L^−1^ in the Vistula River and 8.47 ± 0.68 mBq L^−1^ in the Oder River in 2012, and these concentrations were very close to the concentrations found in 1998. This proves that the meteorological conditions have a direct impact on the concentrations of radioactive elements in water bodies. The mean ^90^Sr activity concentrations found in samples from the Vistula and Oder Rivers have been consistent since 2004. Despite the relatively small input of ^90^Sr to the environment from the Chernobyl accident, the ^90^Sr concentrations were up to two times higher than the ^137^Cs concentrations in the river water samples from 2004 to 2011. The ^90^Sr and ^137^Cs concentrations in the river water samples were similar in 2013, indicating that ^90^Sr is still being washed out from the soil. The ^90^Sr concentrations found in the lake water samples are presented in Fig. [5](#Fig5){ref-type="fig"}. As stated earlier, the highest concentrations were always found in the Rogóźno Lake samples. The concentrations in the other lakes were quite constant between 2007 and 2013. The ^90^Sr and ^137^Cs concentrations in the lake water samples were comparable.Fig. 5Annual average ^90^Sr activity concentration in the lakes that were studied The radionuclide concentrations in the Polish water bodies were fairly similar to the concentrations that have been found in rivers and lakes in other regions of Europe. For instance, the ^137^Cs concentrations found in Lugano Lake and the rivers that feed into the lake were around 1 mBq L^−1^ in 2003--2004 \[[@CR12]\]. In 2012, the ^137^Cs concentrations found in four rivers in Finland were in the range 1.5 to 22 mBq L^−1^, and the ^90^Sr concentrations were in the range 2.5 to 6.4 mBq L^−1^ \[[@CR13]\]. ^90^Sr concentrations of \<1 to 6 mBq L^−1^ were found in the Netherlands in 2011 \[[@CR14]\]. ^90^Sr concentrations of 1.2--4.3 mBq L^−1^ and ^137^Cs concentration of \<1.5 mBq L^−1^ were found in three rivers in Portugal in 2011 \[[@CR15]\]. Conclusions {#Sec7} =========== The ^137^Cs and ^90^Sr concentrations found in surface water samples from Poland in 2012 and 2013 were compared with the concentrations found in previous years, with the aim of identifying temporal changes. The ^137^Cs and ^90^Sr concentrations have clearly but rather slowly decreased over the years. The ^137^Cs mainly originated from the Chernobyl disaster, while the ^90^Sr is to a large extent the residue of nuclear weapons explosions in the atmosphere that were performed in the 1950s and 1960s. Despite the differences in the times that the ^137^Cs and ^90^Sr were released into the atmosphere, the ^137^Cs and ^90^Sr concentrations in the surface water samples were quite similar. This study was conducted in the framework of the Polish Program of the National Monitoring of the Environment, which is based on a contract with the Chief Inspectorate for Environmental Protection, financed through the Polish National Fund for Environmental Protection and Water Management.

Published ahead of print at <http://diabetes.diabetesjournals.org> on 28 April 2008. T.N. and J.-P.D.V.H. contributed equally to this work. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. Diabetes has recently assumed an epidemic proportion. Estimates suggest that worldwide, 30 million women of reproductive age will suffer from diabetes by 2030. Data from birth certificates indicate that some form of maternal diabetes complicates ∼3% of pregnancies in the U.S. ([@r1],[@r2]). Preexisting diabetes complicates pregnancy at a rate of 1--3 per 1,000 births and increases the rates of obstetric complications, stillbirth, perinatal mortality, congenital malformations, and macrosomia compared with the background population ([@r3],[@r4]). Gestational diabetes is also associated with substantial rates of maternal and perinatal complications. Diabetic-associated malformations result from developmental defects occurring in early organogenesis ([@r5]). They include caudal regression syndrome and urogenital abnormalities, which can be as severe as renal agenesis ([@r6]). Besides these numerous epidemiological data concerning perinatal outcome of pregnancy of diabetic women, little is known about the long-term consequences of in utero exposure to maternal diabetes in adulthood ([@r7]--[@r9]). Fetal programming refers to the observation that an adverse environmental stimulus experienced in utero during the critical period of development of organogenesis can induce long-term effects on developing organism ([@r10],[@r11]). These structural and functional effects predispose the offspring to several diseases in adulthood, i.e., hypertension and cardiovascular diseases. Many epidemiological studies have clearly confirmed the seminal works of Barker and Bagby ([@r10]), which evidenced the inverse relationship between a low birth weight as a marker of intrauterine stress and the risk of developing cardiovascular disease and hypertension. Brenner and colleagues ([@r12]) proposed that the inborn nephron deficit associated with this low birth weight predisposes offspring to impaired renal sodium excretion and to increased susceptibility to hypertension. More recently, several studies have also suggested a similar association between low birth weight and chronic kidney disease ([@r12]). We have previously shown that offspring of streptozotocin-induced diabetic rats have an impaired nephrogenesis with a reduction of 30% of nephron number ([@r13]--[@r15]). Our model is characterized by moderate levels of maternal hyperglycemia and by normal gestation and delivery with healthy pups without intrauterine growth retardation or congenital malformation. Therefore, the first aim of this work was to determine in our rat model whether maternal diabetes programs adult hypertension in the offspring and to address the implication of impaired renal sodium excretion in this process. The second aim of this study was to determine whether inborn nephron deficit alters renal function and to identify glomerular hypertrophy and injury as factors of progression of renal diseases. RESEARCH DESIGN AND METHODS =========================== Animals and nephron counting. ----------------------------- Pregnant Sprague-Dawley rats, weighing 250--300 g, were made diabetic on day 0 of gestation by a single intraperitoneal injection of streptozotocin (Sigma, Saint Quentin-Fallavier, France) (35 mg/kg body weight in 0.4 mol/l citrate buffer, pH 4.5). The diabetic state was checked by measuring the plasma glucose concentration (Accuchek, Roche, France). Only pregnant females whose plasma glucose ranged between 15 and 20 mmol/l were included in the study. This diabetic status was confirmed every 2 days until delivery. As previously shown, this protocol results in a 30% reduction of the nephron number ([@r13]). On the day of delivery, the newborn rats were weighed. Each litter was then reduced to 10 pups. The present study was restricted to male offspring. All the animals were maintained in a temperature- and light-controlled room at 21°C with a 12-h light cycle. They had free access to food (UAR Laboratory, Villemoison sur Orge, France) and tap water. Seventy-six rats issued from 16 control mothers (control mother offspring) and 74 rats issued from 16 diabetic mothers (diabetic mother offspring) were used in this study. Six 1-month-old animals issued from three different litters of each group were removed, weighed, and prepared for nephron counting by acid maceration method as previously described ([@r13]). The remaining animals were followed from 1 to 18 months of life for blood pressure or for functional or histological renal evaluation. Some of them, seven and six rats issued from two different litters of control mother offspring and diabetic mother offspring groups, respectively, were followed after 18 months of life to determine the effect of in utero exposure to maternal diabetes on long-term survival. All experiments were conducted in accordance with the institutional guidelines and the recommendations for the care and use of laboratory animals put forward by the French Ministry of Agriculture. Design of the experimental protocol from 1 to 18 months. -------------------------------------------------------- In each group, we have assessed the following in rats of 1 to 18 months of age: systolic blood pressure, renal function and proteinuria, plasma renin activity (PRA), renal renin expression (RRE), glomerular histomorphometry, and renal histology. Systolic blood pressure was measured during 3 successive days in conscious rats by a noninvasive method, tail-cuff plethysmography, using a blood pressure monitor (blood pressure recorder 8005 Narco BioSystems). Before the effective measurements, rats were trained by placing them in restrainers and pressing their tail several times. Blood pressure records were made several consecutives times on quiet animals in a silent ambience at constant temperature under a heating apparatus. Measurements were always performed by the same person. Renal function was estimated by creatinine clearance and proteinuria. For that purpose, rats were individually housed in metabolic cages (Techniplast, Exton, PA) with free access to food and demineralized water. After 3 days of adaptation, food and water intakes were measured every 24-h period, and the urine was collected throughout the protocol during 3 other consecutive days. Urine that was spontaneously excreted during each 24-h period was collected to determine daily urinary creatinine and sodium and protein excretions. At the end of the 3-day experimental study, blood samples were collected for creatinine and PRA measurements. Kidneys and hearts from 1- to 18-month-old control mother offspring and diabetic mother offspring rats were removed, weighed separately, and fixed with formalin for either histology or histomorphometry. High-salt diet protocol. ------------------------ Six 3-month-old rats issued from two different litters from control mother offspring and diabetic mother offspring groups were individually housed in metabolic cages and fed with normal-salt diet (0.3% NaCl). Blood pressure and urinary sodium excretion were measured. The rats were then moved on a high-salt diet (3% NaCl), and the urinary sodium excretions were measured daily during 3 days. Food intake was measured every 24-h period. After 7 days, systolic blood pressure was determined, and kidneys were then removed for sodium transporter abundances evaluation. Analytical procedures. ---------------------- Plasma and urinary analyses were performed by standard methods using a Konelab 20 analyzer (Thermo Electron, Courtaboeuf, France) to determine sodium, creatinine, and protein concentrations. Urinary creatinine concentration was measured by high-phase liquid chromatography (Dionex DX-500; Dionex, Voisins le Bretonneux, France). Analysis of PRA was performed from aortic blood for 1-, 6-, and 18-month-old rats by radioimmunoassay (GammaCoat Plasma Renin Activity ^125^I RIA kit; DiaSorin, Stillwater, MN). Plasma glucose concentrations were determined immediately after sampling by the glucose-oxidase method using a glucose analyzer (Beckman Instruments, Fullerton, CA). Histology. ---------- Histological analysis was performed on kidney samples taken at 1, 3, 6, 12, 18, and 23 months of age for both groups. Tissues were fixed in formalin, embedded in paraffin, and cut in 4-μm-thick transversal sections. Kidney sections were routinely stained with hematein-eosin, Masson\'s trichrome, and silver staining. Semiquantitative evaluations of glomerulosclerosis, interstitial fibrosis, tubular atrophy, and vascular lesions were performed in kidney sections from both groups in a blinded fashion as previously described ([@r16]). Computer-assisted morphometric analysis. ---------------------------------------- Computer assisted-morphometric analysis was performed on kidney samples obtained at 1, 3, 6, 12, and 18 months for both groups, as previously described ([@r17],[@r18]). For each animal, the total glomerular surface area (TGA) was expressed as the mean of the values measured in 50 randomly sampled glomeruli (25 in the superficial cortex and 25 in the juxtamedullary cortex). The TGA, limited by the internal edge of Bowman\'s capsule, was determined in 1- to 18-month-old rats. In addition, the total area of capillary lumens (TCL) and total mesangial surface area (TMA) were measured in both groups at 6- and 18-month-old rats. The TCL-to-TGA and TMA-to-TGA ratios were then calculated for each glomerulus and expressed as the mean of 50 measured glomeruli for each animal. Immunohistochemistry. --------------------- Immunohistochemistry with anti-renin polyclonal antibody was performed on kidney sections with a three-step avidin-biotin immunoperoxidase method with prior antigen unmasking procedure as previously described ([@r19]). Renin immunoreactivity was then evaluated in a blinded fashion using a semiquantitative scoring system as previously described ([@r19]). Preparation of membrane fractions and Western blot analysis. ------------------------------------------------------------ Kidneys of both groups were removed from the anesthetized rats and cut into 5-mm slices. To identify Na transporters, membrane fractions from the cortex and the outer medulla (inner stripe) were prepared, and Western blot analysis was performed as previously described ([@r20],[@r21]) (primary antibodies used were anti-NHE3, 1:1,000; anti-NCC, 1:50,000; anti-BSC1, 1:5,000; anti-α ENaC, 1:3,000; anti-βENaC, 1:20,000; anti-γ ENaC, 1:2,000; and anti-Na^+^/K^+^ ATPase, 1:20,000). After incubation with the appropriate peroxidase-conjugated secondary antibodies, blots were washed, and luminol-enhanced chemiluminescence (ECL; Perkin Elmer Life Science Products, Boston, MA) was used to visualize bound antibodies before exposure to Hyperfilm ECL (Amersham, Arlington Heights, IL). The autoradiography was digitized with the use of a laser scanner (Epson Perfection 1650; Epson), and quantification of each band was performed by densitometry using NIH Image software. Densitometric values were normalized to the mean for the control group that was defined as 100%. Statistical analysis. --------------------- All results are expressed as the mean ± SE values. Statistical analysis was performed by unpaired and paired Student\'s *t* test or by ANOVA when appropriate. Statistical significance was defined as *P* \< 0.05. RESULTS ======= Gestational and delivery conditions and follow-up of rats exposed in utero to maternal diabetes. ------------------------------------------------------------------------------------------------ Glycemia of the 16 diabetic mothers was constant throughout gestation and fourfold higher than in the 16 controls (5.27 ± 0.34 vs. 21.6 ± 0.12 mmol/l). Gestation occurred normally, and the rats delivered spontaneously at term (21 days of gestation). The number of pups per litter was similar in both groups. At birth, pups appeared healthy with similar birth weight compared with controls. As previously described ([@r13]), in our model, in utero exposure to maternal diabetes induced nephron deficit in the offspring. At 1 month, the reduction of nephron number was ∼30% in rats issued from diabetic mothers compared with controls (35,133 ± 507 vs. 25,600 ± 570, respectively, *P* \< 0.0001, *n* = 6 in each group). As shown in [Table 1](#t1){ref-type="table"}, mean body weight progressively increased with age and was similar in both groups from the newborn period until 3 months. However, at 6 and 12 months of age, it was significantly higher in the diabetic mother offspring group than in the control mother offspring group. Kidney and heart weights (relative to body weight) decreased as a function of age but were not significantly different in both groups. The same normal blood glucose levels were observed in control mother offspring and diabetic mother offspring groups (5.90 ± 0.23 vs. 5.61 ± 0.70 mmol/l in 6-month-old rats from both groups, *n* = 6 from three litters). Effects of in utero exposure to maternal diabetes on blood pressure ------------------------------------------------------------------- ### Systolic blood pressure. Rats exposed in utero to maternal diabetes demonstrated higher blood pressure in adulthood ([Fig. 1](#f1){ref-type="fig"}). Although the systolic blood pressure of 1- and 3-month-old rats was similar in both groups, it was significantly increased from 6 months of age in diabetic mother offspring compared with control mother offspring. Hypertension lasted and further increased during the observation period. ### PRA and RRE. PRA and RRE are given in [Table 2](#t2){ref-type="table"}. At the prehypertensive stage (1- and 3-month-old rats), PRA and RRE were not significantly different in the two groups. From 6 months of age (hypertensive stage), a significant decrease of PRA was observed in diabetic mother offspring, when compared with their respective age-matched control mother offspring. With aging, the reduction of PRA and RRE was present in both groups. ### Effect of high-salt diet on blood pressure and urinary sodium excretion. To study the response to a chronic alteration in sodium balance, both groups of rats were given a high-salt diet (3% NaCl diet instead of 0.3%) ([Fig. 2](#f2){ref-type="fig"}). Means of the 24-h diet intake for the 3 days of sodium excretion measurement were similar in the two groups (28.32 ± 0.83 vs. 28.03 ± 1.1 g in control mother offspring and diabetic mother offspring group, respectively). Systolic blood pressure was similar in the two groups on normal-sodium diet (0.3%). High-salt diet (3%) induced a raise of systolic blood pressure in the diabetic mother offspring group (respectively, 131.7 ± 0.8 vs. 154.5 ± 3.2 mmHg; paired *t* test, *P* \< 0.01) and had no effect on blood pressure in the control mother offspring group. As expected, high-sodium diet led to a significant increase of urinary sodium excretion in both groups. However, this increase was significantly delayed in diabetic mother offspring compared with control mother offspring, accounting for a larger positive sodium balance. ### Effect of high-salt diet on relative abundance of renal sodium transporters. We assessed the effects of a high-salt diet on sodium transporter protein by semiquantitative immunoblots of membrane fractions from the cortex and outer medulla (inner stripe) obtained from the diabetic mother offspring and control mother offspring kidney of 3-month-old rats ([Fig. 3](#f3){ref-type="fig"}). In the cortex, both β- and γ-ENaC subunits were significantly upregulated in diabetic mother offspring compared with control mother offspring, whereas α-ENaC protein abundance was unchanged. Na/K ATPase protein abundance was also significantly upregulated in diabetic mother offspring. In the medulla, the high-salt diet led to a decrease in BSC1 protein abundance in the diabetic mother offspring group compared with the control mother offspring group. Protein levels of α-, β-, and γ-ENaC were unaffected. Finally, the protein levels of NHE3 and NCC were not different in diabetic mother offspring receiving a high-salt diet compared with control mother offspring either in the cortex or in the medulla. Renal effects of in utero exposure to maternal diabetes ------------------------------------------------------- ### Renal function and proteinuria. [Table 3](#t3){ref-type="table"} reports the follow-up of proteinuria and glomerular filtration rate (GFR) estimated by the creatinine clearance, in 1- to 18-month-old rats of both groups. The creatinine clearance progressively increased with age in both groups and was significantly lower in diabetic mother offspring compared with control mother offspring. GFR was reduced by ∼10% in 3-month-old rats and by 30% from the 6- to 18-month period. Proteinuria levels increased in both groups with aging and were significantly higher in diabetic mother offspring. ### Glomerulus histomorphometry. To address the question of a structural glomerular adaptation to the reduction of nephron number, a glomerular histomorphometry study was performed in kidneys taken from 1- to 18-month-old control mother offspring and diabetic mother offspring rats ([Fig. 4](#f4){ref-type="fig"}). In each age-group, no significant differences were observed between glomeruli measured in either superficial or deep cortex areas between control mother offspring and diabetic mother offspring kidney rats (data not shown). TGA significantly increased with advancing age. TGA was not significantly different between the two groups, although it was transitorily increased in 6-month-old diabetic mother offspring rats. TCL was significantly increased in the 6-month-old diabetic mother offspring compared with the control mother offspring. TCL/TGA was similar in both groups. At 18 months of age, TMA and TMA/TGA were similar in diabetic mother offspring and control mother offspring groups. ### Renal histopathology. Renal histopathology was assessed in 1- to 18-month-old control mother offspring and diabetic mother offspring rats. Before 18 months of age, all kidneys were normal and devoid of glomerulosclerosis and interstitial fibrosis (data not shown). At 18 months of age, renal lesions were limited to focal areas of interstitial fibrosis with minimal tubular atrophy and to very few glomeruli with segmental glomerulosclerosis ([Fig. 5*A* and *B*](#f5){ref-type="fig"}). Semiquantitative analysis showed that the extent of both glomerulosclerosis (glomerulosclerosis index: control mother offspring, 10.08 ± 1.62, *n* = 13 rats from three different litters vs. diabetic mother offspring, 8.49 ± 1.91, *n* = 9 rats from two different litters; unpaired *t* test; NS, arbitrary units) and tubulointerstitial lesions (Interstitial fibrosis index: control mother offspring, 5.12 ± 1.69, *n* = 8 rats from two different litters vs. diabetic mother offspring, 7.80 ± 4.25, *n* = 5 rats from three different litters; unpaired *t* test; NS, arbitrary units) were not different in the two groups. Survival study. --------------- Survival study showed an increased mortality after 18 months of age in diabetic mother offspring. At 23 months, long-term survival was markedly reduced in the diabetic mother offspring group, 33.3% compared with 85.7% in the control mother offspring group. We therefore decided to kill the remaining rats to address late kidney histopathology in rats of the two groups. A widespread interstitial fibrosis with tubular atrophy and dilatation was present in the kidneys of diabetic mother offspring rats, associated with glomerulosclerosis and glomerular cysts. A scarce interstitial fibrosis and tubular atrophy was present in control mother offspring rats ([Fig. 5*C*](#f5){ref-type="fig"}--*F*). Glomerulosclerosis index is 7.25 ± 3.50, *n* = 6, in control mother offspring versus 87 and 66 in two diabetic mother offspring rats, and interstitial fibrosis index is 27.5 ± 24.3, *n* = 6, in control mother offspring versus 230 and 300 in two diabetic mother offspring rats. No structural changes in the intra-renal vessels were observed at any age. DISCUSSION ========== The present study identifies the long-term consequences of in utero exposure to maternal diabetes in the rat and shows that in utero exposure is associated with the development of a salt-sensitive systolic hypertension and with a decrease in renal function in adulthood. A mild to moderate increase in systolic arterial blood pressure is observed in the offspring of diabetic mothers from 6 months of age, which progressively went worse with the age. This rise in blood pressure is associated with low PRA, suggesting a salt-sensitive hypertension. We further confirmed the salt sensitivity with high-sodium diet experiments performed at prehypertensive stage: in diabetic mother offspring rats, a high-sodium diet induced an increase of systolic blood pressure and led to a shift to the right of the urinary sodium excretion curve, indicating a delayed sodium excretion. A similar salt-sensitive hypertension has been reported in a neonatal uni-nephrectomized rat model by Woods et al. ([@r22]). Together, these results are in accordance with the hypothesis of Brenner et al. ([@r23]), which states that inborn nephron deficit predisposes reduced renal sodium excretion, leading to hypertension susceptibility, especially in the setting of dietary sodium excess. Interestingly, in inbred rat models of hypertension, the relationship between nephron number and blood pressure is still a matter of debate ([@r24]--[@r26]). In addition, recently, Rocha et al. ([@r8]) and Magaton et al. ([@r9]) reported hypertension in rats issued from diabetic mothers without inborn nephron deficit. However, their model slightly differs from ours: *1*) The level of maternal hyperglycemia is more pronounced in our model (we have previously shown that inborn nephron deficit correlated with the level of maternal hyperglycemia \[[@r13]\]); and *2*) we performed direct glomerular counting with the gold-standard acid-maceration method, whereas they used an histological-derived method that is more appropriate to evaluate the density of glomeruli. Because one mechanism involved in such renal sub-optimal sodium handling may traduce a resetting of tubular sodium cotransporter expression, we then evaluated the protein abundance of Na transporters and channels in the kidney at the prehypertensive stage and under a high-salt diet. In the cortex of diabetic mother offspring rats, we found an increase of three different sodium transporter proteins: βENaC, γENaC, and Na-K ATPase. Such relative increases without decreases in other sodium transporters would be predicted to result in enhanced tubular Na reabsorption and might play a role in the development or maintenance of elevated blood pressure in these animals ([@r27]). The decrease in BSC1 in the medulla may reflect a medullary compensatory effect linked to the increase of cortical sodium transporters. An abnormal expression of Na transporters has also been reported by Manning et al. ([@r28]). In their model, maternal protein restriction is associated with low birth weight, development of hypertension at 8 weeks of age, and significant increased expression in the sodium cotransporters BSC1 and TSC ([@r28]). Together, these results strongly substantiate a perinatal reprogramming of tubular function regulation in association with or as a result of inborn nephron deficit ([@r29]). In the present study, we also show that maternal diabetes is associated with a decreased renal function in adulthood, as assessed by creatinine clearance and proteinuria measurements. According to Brenner and colleagues ([@r12],[@r23]), a compensatory glomerular hypertrophy and hyperfiltration occur in response to inborn nephron deficit to sustain adequate renal function. Such glomerular adaptation made at the expense of intraglomerular hypertension accelerates injury to the remaining functional glomeruli and perpetuates the ongoing glomerulosclerosis, leading to renal failure. However, in our model in which inborn nephron deficit is observed, systematic analysis of renal histology shows that both glomerulosclerosis and interstitial fibrosis are absent in diabetic mother offspring and control mother offspring groups at all stages before 18 months of age. At 18 months, because of aging ([@r17]), mild renal structural lesions occur but at the same extent in the two groups. In addition, glomerular computer-assisted histomorphometry shows a mild glomerular hypertrophy only at 6 months, at the onset of hypertension in diabetic mother offspring rats. In our model, absence of early increase of glomerular size and of sustained glomeruli hypertrophy with aging does not support a major glomerular adaptation to the inborn nephron deficit. Together, these two sets of results seem to weight against a compensatory glomerular hypertrophy and hyperfiltration as factors of alteration and worsening of the renal function until 18 months of age. Concerning the study of Rocha et al. ([@r8]), one must note that although decreased nephron number was not observed in young animals, the number of nephrons was reduced at 12 months in diabetic mother offspring rats. However, because no histological data were provided in their study to evaluate the extent of glomerulosclerosis, the mechanisms of ongoing nephron loss and its implication in the decline of renal function has not been elucidated in their model. In the other hand, a dramatic impairment of renal histology with widespread glomerular and tubulo-interstitial lesions was observed in kidneys issued from 23-month-old diabetic mother offspring rats. Such late alterations of kidney structures compared with the early impairment of renal function could hardly be explained by the consequences of a congenital nephron deficit alone. The link between the congenital nephron deficit and the adverse effect of compensatory hyperfiltration on the remaining glomeruli leading to glomerulosclerosis has been unambiguously demonstrated (and quantitatively assessed) in models with severe reduction in the nephron number or in models with toxic renal impairment that associates both in utero glomerular reduction and tubulo-interstitial lesions ([@r30],[@r31]). In our present model, the mild (30%) congenital nephron deficit might not be sufficient per se to induce glomerular lesions, at least during the first 18 months of life. Another issue to consider is the kidney as a target organ of hypertension. In our model, although hypertension is present from 6 months of age, high level of systolic blood pressure is only achieved at 18 months. It is well known that the increase of blood pressure is limited and slowly increases with age in the majority of the model of perinatal programming of hypertension compared with other rat models of hypertension ([@r32],[@r33]). This may explain why we observe the histopathological renal consequences of hypertension only in the 23-month-old rats. However, the implication of hypertension alone in the renal lesions of our model is questionable, because even at 23 months, intra-renal vascular hypertensive lesions are nearly absent at histology, and myocardial (other target organ of hypertension) histology is similar in both groups (data not shown). This study thus identifies maternal diabetes as a novel risk factor for fetal programming of adult hypertension and impairment of renal function. Alteration of renal sodium handling, observed in our model, may be explained by a fetal resetting of tubular functions, as a consequence or in association with congenital nephron deficit. This work was supported by a grant from the Institut National de la Recherche Medicale. We acknowledge the technical assistance of Marie-France Belair and Martine Douheret (Institut National de la Santé et de la Recherche Médicale, Unite Mixte de Recherche S872, Centre de Recherche des Cordeliers). We also acknowledge the technical assistance of Michele Smirnoff for animal nursing and of Nelly Knobloch for secretarial assistance. Parts of this work were presented in abstract form at the 38th annual meeting of the American Society of Nephrology, Philadelphia, Pennsylvania, 8--13 November 2005. {#f1} {#f2} {#f3} {#f4} {#f5} ###### Body, kidney, and heart weights at death Age (months) ANOVA ------------------------------------ ----------- -------------- ------------------------------------------- ------------------------------------------ ----------- --------------- -------------- -------------- -- Body weight (g) CMO 164 ± 5 435 ± 16 518 ± 10 596 ± 17 730 ± 22 *n* 8 6 9 9 9 DMO 161 ± 24 441 ± 18 568 ± 20[\*](#t1fn1){ref-type="table-fn"} 723 ± 17[†](#t1fn2){ref-type="table-fn"} 731 ± 22 *P* \< 0.0001 *P* = 0.0002 *P* = 0.0009 *n* 8 6 9 9 Kidney weight (mg)/body weight (g) CMO 5.2 ± 0.3 3.1 ± 0.1 2.4 ± 0.4 2.3 ± 0.3 2.9 ± 0.1 *n* 8 6 9 9 9 DMO 4.4 ± 0.2 2.7 ± 0.3 3.3 ± 0.1 2.2 ± 0.1 2.9 ± 0.1 *P* \< 0.0001 *P* = 0.5673 *P* = 0.0021 *n* 8 6 9 9 9 Heart weight (mg)/body weight (g) CMO 5.6 ± 0.3 3.9 ± 0.2 2.8 ± 0.4 3.9 ± 0.1 3.2 ± 0.9 *n* 8 6 9 9 9 DMO 5.5 ± 0.4 3.2 ± 0.3 3.5 ± 0.1 3.4 ± 0.1 3.0 ± 0.2 *P* \< 0.0001 *P* = 0.6038 *P* = 0.0919 *n* 8 6 9 9 9 Data are means ± SE for *n* animals issued from two or three litters. CMO, control mother offspring; DMO, diabetic mother offspring. Two-way ANOVA with age and group effects. Comparisons are made between CMO and DMO at the same age (unpaired *t* test, *P* \< 0.05; *P* \< 0.001). ###### Plasmatic renin activity and kidney renin immunoreactivity index Age (months) ANOVA --------------------------------------------- ------------ -------------- --------------------------------------------- ------------ ------------- --------------- ------------- -------------- -- Plasma renin activity (μg · ml^−1^ · h^−1^) CMO 215 ± 16.6 175 ± 34 163 ± 20.7 162 ± 12.6 104 ± 5.2 *n* 7 7 7 9 5 DMO 216 ± 31.3 107 ± 13.4 104 ± 11.8[\*](#t2fn1){ref-type="table-fn"} 140 ± 8.3 91.1 ± 23.3 *P* \< 0.0001 *P* = 0.021 *P* = 0.4022 *n* 8 5 8 8 4 Renal renin expression (arbitrary unit) CMO 51.9 ± 2.4 54.4 ± 6.1 63.7 ± 6.3 50.2 ± 5.1 29 ± 3 *n* 8 7 7 9 5 DMO 58.9 ± 4.2 48.8 ± 2.9 37.8 ± 2.9[†](#t2fn2){ref-type="table-fn"} 38.2 ± 4.6 25.7 ± 2.3 *P* \< 0.0001 *P* = 0.083 *P* = 0.021 *n* 8 5 9 8 4 Data are means ± SE for *n* animals issued from two or three litters. CMO, control mother offspring; DMO, diabetic mother offspring. Two-way ANOVA with age and group effects. Comparisons are made between CMO and DMO at the same age (unpaired *t* test, *P* \< 0.05; *P* \< 0.01). ###### Creatinine clearance and proteinuria Age (months) ANOVA ------------------------------- ----- -------------- ---------------------------------------------- ---------------------------------------------- --------------------------------------------- --------------------------------------------- ------------- --------------- -------------- -------------- Creatinine clearance (ml/min) CMO 0.76 ± 0.03 1.33 ± 0.04 2.21 ± 0.13 2.72 ± 0.20 3.02 ± 0.23 2.99 ± 0.27 *n* 5 5 6 6 6 4 DMO 0.76 ± 0.12 1.21 ± 0.10 1.66 ± 0.08[\*](#t3fn1){ref-type="table-fn"} 1.75 ± 0.06[†](#t3fn2){ref-type="table-fn"} 2.20 ± 0.18[‡](#t3fn3){ref-type="table-fn"} 2.04 ± 0.22 *P* \< 0.0001 *P* \< 0.001 *P* = 0.0069 *n* 6 4 6 4 5 4 Proteinuria (mg/day) CMO 4.52 ± 0.31 11.0 ± 0.45 12.5 ± 1.20 14.8 ± 2.08 18.3 ± 2.75 26.5 ± 2.53 *n* 5 5 6 6 6 4 DMO 4.77 ± 0.82 14.7 ± 0.04[\*](#t3fn1){ref-type="table-fn"} 18.4 ± 1.34[‡](#t3fn3){ref-type="table-fn"} 23.3 ± 3.92 15.3 ± 1.55 57.2 ± 24.4 *P* \< 0.0001 *P* = 0.0112 *P* = 0.054 *n* 6 4 6 4 5 4 Data are means ± SE of *n* rats issued from two or three litters. CMO, control mother offspring; DMO, diabetic mother offspring. Two-way ANOVA with age and group effects. Comparisons are made between CMO and DMO at the same age (unpaired *t* test, *P* \< 0.001; *P* \< 0.01; *P* \< 0.05). [^1]: Corresponding author: Martine Lelièvre-Pégorier, <martine.lelievre-pegorier@crc.jussieu.fr>

Background {#Sec1} ========== Phosphate (Pi) is one of the indispensable macronutrients to plants for growth, development and reproduction. Pi deficiency (−Pi) is one of the main limiting factors for increasing crop yield and improving quality because of low bioavailability of Pi in soil \[[@CR1]--[@CR4]\]. To cope with Pi deficiency, plants develop a series of tightly controlled adaptive responses including external developmental alterations of increasing Pi absorption and internal metabolic, physiological and biochemical alterations of reducing Pi usage \[[@CR4]--[@CR6]\]. To increase Pi uptake under Pi depletion stress, plants adapt to modulate the root architecture bearing more and longer lateral roots as well as denser root hairs, which enable the roots to explore the Pi resources in soil \[[@CR7]--[@CR12]\]. To reduce Pi consumption under Pi deficiency condition, plants often modulate metabolisms to maintain intracellular Pi homeostasis by reducing metabolic consumption of Pi \[[@CR1], [@CR3], [@CR13]--[@CR15]\] and degrading compounds to release Pi \[[@CR16]\]. Furthermore, a series of physiological and biochemical adaptions including the induction and secretion of phosphatases and organic acids and accumulation of protective metabolites such as anthocyanin help augment the availability of both endogenous and exogenous Pi \[[@CR1], [@CR4], [@CR17], [@CR18]\]. Genetic, physiological and biochemical studies in Arabidopsis have demonstrated that the acquisition, allocation, and metabolism of Pi are highly regulated processes and require the concerted action of multiple membrane Pi transport systems \[[@CR19]--[@CR23]\]. Among five distinct classes of proteins possessing Pi transport activity, the plastidic Pi translocator group function as antiport systems, whereas the other four Pi transporter families, named the PHOSPHATE TRANSPORTER1 (Pht1), Pht2, Pht3 and Pht4, contribute to the acquisition, allocation and remobilization of Pi \[[@CR24]--[@CR27]\]. Most of the plasma membrane-localized high affinity transporters in the Pht1 family mediate Pi acquisition from external environment \[[@CR1], [@CR24], [@CR28]--[@CR30]\]. Once Pi is transported into root epidemic cells, translocation and allocation of Pi within the plants and cells are key steps in maintaining Pi homeostasis at cell and whole plant levels. The low affinity transporters in the Pht2 and Pht4 families are thought to participate mainly in Pi transfer across internal cellular membranes and thus allocate Pi in different compartments of the cells \[[@CR25], [@CR27], [@CR31]--[@CR33]\]. Pht1;5, a Pht1 family member, PHO1, Pht1;8 and Pht1;9 were shown to play critical roles in systemic regulation of Pi homeostasis, e.g. mobilization of Pi from source to sink organs in accordance with the Pi status of the plant \[[@CR34]--[@CR36]\]. Functional characterization of genes in a number of mutants with altered response to Pi depletion have led to the identification of several different regulatory mechanisms controlling the adaptive responses to Pi starvation. These regulatory mechanisms include transcriptional regulation by transcription factors such as PHR1, WRKY6, WRKY42, WRKY45, WRKY75, ZAT6, bHLH32, ERF070 and MYB62 \[[@CR37]--[@CR47]\], posttranscriptional regulation by microRNAs including miRNA399 \[[@CR48]--[@CR53]\], posttranslational regulation by protein modifications such as sumoylation of SIZ1 \[[@CR54]\], phosphorylation of Pht1.1 and activation of MKK9-MPK3/MPK6 module \[[@CR55], [@CR56]\], deubiquitination of UBP14 \[[@CR57]\] and chromatin histone modification and epigenetic \[[@CR46], [@CR58]\]. Furthermore, it was also demonstrated that Pi starvation response cross-talks signaling mediated by different hormones such as auxin \[[@CR10], [@CR15], [@CR59]--[@CR61]\], cytokinin \[[@CR10], [@CR15], [@CR62], [@CR63]\], ethylene \[[@CR9], [@CR10], [@CR64]--[@CR66]\] and gibberellin \[[@CR67]\]. Although great advance on the adaptive response to Pi starvation has been made during the last decade, the molecular mechanism that regulates these adaptive processes has yet to be elucidated in detail. The *Arabidopsis thaliana BOTRYTIS-INDUCED KINASE1* (*BIK1*) encodes a plasma membrane-localized receptor-like protein kinase and plays critical roles in *Botrytis cinerea* resistance \[[@CR68]\]. Recent studies have shown that BIK1 interacts with receptors for pathogen- or damage-associated molecular patterns such as FLS2 and PEPRs to regulate immune response against different types of pathogens \[[@CR69]--[@CR73]\] and defense response to insect pests \[[@CR74]\]. The *bik1* mutant plants showed an altered root architecture \[[@CR68]\], similar to morphological phenotypes often seen in mutants with Pi starvation response \[[@CR75]\], indicating a possible involvement of *BIK1* in Pi starvation response in Arabidopsis. Therefore, we investigated whether BIK1 functions in Pi starvation response and our results demonstrate that BIK1 plays a role in regulation of Pi homeostasis in Arabidopsis. Results {#Sec2} ======= Responsiveness of *BIK1* to Pi starvation {#Sec3} ----------------------------------------- When grown on MS medium under normal conditions, the *bik1* plants produced shorter primary roots and longer and significantly more root hairs and lateral roots than WT plants \[[@CR68]\], which is reminiscent of the mutants with defects in Pi nutrition \[[@CR75]\]. These observations led us to examine whether BIK1 has a function in Pi starvation response. We first examined whether *BIK1* is responsive to Pi starvation stress by analyzing the expression patterns of *BIK1* in seedlings grown under + Pi and --Pi conditions. As shown in Fig. [1a](#Fig1){ref-type="fig"}, expression of *BIK1* was detected in roots, shoots and leaves of seedlings grown under + Pi condition and no significant changes in *BIK1* expression was observed during the experiment period. However, expression level of *BIK1* was markedly induced with similar patterns in roots, shoots and leaves of seedlings after transferring to medium without Pi supplement (Fig. [1a](#Fig1){ref-type="fig"}). The transcript levels of *BIK1* in seedlings grown under --Pi condition increased at 12 h and peaked at 24 and 48 h after transferring, leading to 7.5 \~ 11.2 folds of increases over those in seedlings grown under + Pi condition (Fig. [1a](#Fig1){ref-type="fig"}). To gain further information on spatial expression of the *BIK1* gene in response to Pi starvation, we generated *BIK1*~*pro*~::GUS transgenic lines and compared the GUS staining patterns in T2 seedlings during Pi starvation stress. Slight GUS staining was observed in leaves and roots of the *BIK1*~*pro*~::GUS seedlings at 0 h after transferred to medium without Pi supplement (Fig. [1b](#Fig1){ref-type="fig"}), indicating a basal expression of *BIK1* in leaf and root tissues, similar to the results from RT-PCR. At 1 day after transferred to medium without Pi supplement, significant GUS staining was easily seen in roots, shoot and leaves of the *BIK1*~*pro*~::GUS seedlings (Fig. [1b](#Fig1){ref-type="fig"}). qRT-PCR analysis showed 7.3 and 9.7 folds of increases in the expression levels of *GUS* gene in root and leaf tissues of the *BIK1*~*pro*~::GUS seedlings at 1 day after transferring to --Pi condition (Fig. [1c](#Fig1){ref-type="fig"}). Notably, GUS staining was clearly observed in vascular tissues of roots, shoots and leaves of the *BIK1*~*pro*~::GUS seedlings subjected to Pi starvation (Fig. [1b](#Fig1){ref-type="fig"}). Together, data from RT-PCR and GUS staining demonstrate that *BIK1* is responsive to Pi starvation.Fig. 1Responsiveness of *BIK1* to Pi starvation. **a** Expression changes of *BIK1* in different tissues of WT plants under + Pi and --Pi conditions. Seven-day-old seedlings grown hydroponically under normal Pi condition were transferred to medium supplemented with Pi (+Pi, 250 μM) or without Pi (−Pi). Roots, shoots and leaves were collected for analysis of *BIK1* expression by qRT-PCR at indicated time points after transferring. **b** Detection of *BIK1* promoter activity in *BIK1* ~*pro*~::GUS seedlings by GUS staining. **c** Expression changes of *GUS* gene in root and lead tissues of *BIK1* ~*pro*~::GUS seedlings under --Pi condition. Seven-day-old *BIK1* ~*pro*~::GUS seedlings grown in normal medium were transferred to medium without Pi supplement and samples were collected at 0 and 1 day after transferring for GUS staining and analysis of gene expression. Data were normalized with the transcript level of *UBQ10* and relative expression levels were shown as folds of the *UBQ10* level. Data presented are the means ± SD from three independent experiments and \* above the error bars indicate significant differences at *p* \< 0.05 level between the + Pi and --Pi conditions Increased Pi concentration in *bik1* plants {#Sec4} ------------------------------------------- We next examined whether loss of *BIK1* function affects Pi homeostasis in *bik1* plants. Total Pi contents in leaves and roots of WT and *bik1* plants grown hydroponically under + Pi (250 μM) and --Pi conditions for 27 days were measured. As shown in Fig. [2](#Fig2){ref-type="fig"}, a significant increase in total Pi contents was observed in leaves and roots of *bik1* plants as compared to WT plants under both + Pi and --Pi conditions. When grown under + Pi condition, total Pi contents in roots and leaves of *bik1* plants were 0.43 and 1.12 times higher over those in WT plants, respectively (Fig. [2a](#Fig2){ref-type="fig"} and [b](#Fig2){ref-type="fig"}). Similarly, when grown under --Pi condition, total Pi contents in roots and leaves of *bik1* plants were 0.55 and 1.05 times higher than those of WT plants, respectively (Fig. [2a](#Fig2){ref-type="fig"} and [b](#Fig2){ref-type="fig"}). These data indicate that BIK1 has a function either in Pi uptake or in the transfer of Pi from the roots to the leaves.Fig. 2Increased total Pi content in *bik1* plants. Total Pi contents in roots (**a**) and leaves (**b**) of WT and *bik1* plants grown under + Pi (250 μM) and --Pi condition for 27 days were estimated. Data presented are the means ± SD from three independent experiments and different letters above the columns indicate significant differences at *p* \< 0.05 level between same tissues of WT and *bik1* plants grown under same condition Altered root architecture in *bik1* plants {#Sec5} ------------------------------------------ Root architecture plays important roles in maintaining Pi homeostasisn in plants \[[@CR75]\]. We compared the root architecture of WT and *bik1* seedlings grown under Pi normal (+Pi) and starvation (−Pi) conditions. When compared with WT seedlings, the *bik1* seedlings grown under + Pi condition showed shorter primary root and more lateral roots (Fig. [3a](#Fig3){ref-type="fig"}, [c](#Fig3){ref-type="fig"} and [d](#Fig3){ref-type="fig"}), and this trend was much evident in the *bik1* seedlings grown under --Pi condition (Fig. [3a](#Fig3){ref-type="fig"}, [c](#Fig3){ref-type="fig"} and [d](#Fig3){ref-type="fig"}). When grown under + Pi and --Pi conditions, the elongation rate of the primary and lateral roots of the *bik1* seedlings was markedly reduced as compared with WT seedlings (Fig. [3e](#Fig3){ref-type="fig"} and [f](#Fig3){ref-type="fig"}). Furthermore, the *bik1* seedlings grown under + Pi and --Pi conditions showed more and longer root hairs than WT seedlings (Fig. [3b](#Fig3){ref-type="fig"}, [g](#Fig3){ref-type="fig"} and [h](#Fig3){ref-type="fig"}). These data indicate that BIK1 plays an important role in regulating development of root architecture.Fig. 3Altered root architecture in *bik1* seedlings grown under + Pi and --Pi conditions. Seven-day-old seedlings grown under normal condition were transferred into fresh medium supplemented with (+Pi, 250 μM) or without Pi (−Pi) and grown for another 7 days. **a** Root architecture in WT and *bik1* seedlings. **b** Root hairs in WT and *bik1* seedlings. Fragments of 5 mm from root tips were presented. **c** to **h** Comparative parameters for length of primary roots (**c**), numbers of lateral roots (**d**), elongation of primary roots (**e**) and lateral roots (**f**), and numbers (**g**) and length (**h**) of root hairs of WT and *bik1* seedlings grown under + Pi and --Pi condition. Data presented are the means ± SD from three independent experiments and different letters above the columns indicate significant differences at *p* \< 0.05 level Increased anthocyanin accumulation and acid phosphatase activity in *bik1* plants {#Sec6} --------------------------------------------------------------------------------- Accumulation of anthocyanin and increased secretion of acid phosphatases are characteristic symptoms in plants under Pi starvation conditions \[[@CR3]\]. We thus examined and compared the accumulation of anthocyanin and activity of acid phosphatase between WT and *bik1* plants grown under + Pi and --Pi conditions. When grown on medium under --Pi condition for 5 days, the leaves of the *bik1* seedlings turned to be purple while WT seedlings still kept green (Fig. [4a](#Fig4){ref-type="fig"}). The anthocyanin contents in *bik1* seedlings grown under + Pi and --Pi conditions were much higher than those in WT seedlings, showing 32 % and 105 % increase, respectively (Fig. [4b](#Fig4){ref-type="fig"}). No visible staining and significant changes in acid phosphatase activity were detected in WT and *bik1* seedlings grown + Pi condition (Fig. [4c and d](#Fig4){ref-type="fig"}). When grown under --Pi condition, roots of the *bik1* seedlings are stained deep blue while the staining of roots of WT seedlings is much lighter (Fig. [4c](#Fig4){ref-type="fig"}). Similarly, activity of acid phosphatase in roots of *bik1* seedlings is much higher than that in WT seedlings, grown under --Pi condition, showing an increase of \~65 % (Fig. [4d](#Fig4){ref-type="fig"}). These results indicate that the *bik1* seedlings secrete higher levels of acid phosphatase than WT seedlings under Pi starvation stress. Collectively, these results suggest that loss of *BIK1* function results in typical Pi starvation responses as revealed by the increased anthocyanin contents and acid phosphatase activity.Fig. 4Increased levels of anthocyanin accumulation and acid phosphatases activity in *bik1* plants. **a** and **b** Measurements of anthocyanin contents. Seven-day-old WT and *bik1* seedlings grown on 1/2 MS were transferred into MS medium supplemented with Pi (+Pi, 1 mM) or without Pi (−Pi). Photos were taken and anthocyanin contents were measured at 5 days after treatment. **c** and **d** Detection of acid phosphatase activity. Ten-day-old seedlings grown in liquid medium were transferred to fresh medium supplemented with Pi (+Pi, 1 mM) or without Pi (−Pi) and covered with a layer of 0.008 % BCIP-containing agarose. Photos were taken and the activity of acid phosphatase in roots was estimated at 5 days after treatment. Data presented are the means ± SD from three independent experiments and different letters above the columns indicate significant differences at *p* \< 0.05 level between WT and *bik1* plants grwon under same condition Reduced ROS accumulation in *bik1* plants under Pi starvation condition {#Sec7} ----------------------------------------------------------------------- ROS, generated and accumulated in plant response to different types of stresses, has been implicated in many biological processes including biotic and abiotic responses \[[@CR76], [@CR77]\]. Therefore, we examined whether mutation in *BIK1* gene affects the balance of ROS and hence leads to accumulation of ROS in plants during Pi starvation stress. In NBT staining of superoxide anion, no significant difference was detected between WT and *bik1* plants grown under + Pi condition; however, significant accumulation of superoxide anion in leaves, stem and roots of WT and *bik1* plants grown under --Pi condition was observed (Fig. [5a](#Fig5){ref-type="fig"}). Notably, accumulation of superoxide anion in WT plants was much higher than that in *bik1* plants grown under --Pi condition (Fig. [5a](#Fig5){ref-type="fig"}). Quantification of H~2~O~2~ contents revealed a significant lower level of H~2~O~2~ accumulated in *bik1* plants than that in WT plants when grown under + Pi condition (Fig. [5b](#Fig5){ref-type="fig"}). Accumulation of H~2~O~2~ in WT and *bik1* plants grown under --Pi condition was markedly increased, leading to 95 and 78 % of increases as compared with those in plants grown under + Pi condition, respectively (Fig. [5b](#Fig5){ref-type="fig"}). Under --Pi condition, the content of H~2~O~2~ in *bik1* plants was approximately 54 % of that in WT plants (Fig. [5b](#Fig5){ref-type="fig"}). These results indicate that loss of *BIK1* function led to reduced accumulation of ROS in *bik1* plants under --Pi starvation stress.Fig. 5Reduced accumulation of ROS in *bik1* plants. Four-week-old seedlings grown under + Pi (250 μM) and --Pi conditions were collected for detection of superoxide anion and measurement of H~2~O~2~. **a** NBT staining of superoxide anion in WT and *bik1* plants grown under + Pi and --Pi conditions. **b** Quantification of H~2~O~2~ contents in WT and *bik1* plants grown under + Pi and --Pi conditions. Data presented are the means ± SD from three independent experiments and different letters above the columns indicate significant differences at *p* \< 0.05 level between WT and *bik1* plants grwon under same condition Altered expression of Pi starvation-induced genes and miRNA399 in *bik1* plants {#Sec8} ------------------------------------------------------------------------------- Expression of some well-characterized Pi starvation-induced genes was analyzed and compared between WT and *bik1* seedlings that had been transferred into medium with Pi (+Pi) or without Pi (−Pi) for 5 days. When grown under + Pi condition, the expression levels of *AT4* \[[@CR78]\], *WRKY75* \[[@CR39]\], *PHT1;2* and *PHT1;4* \[[@CR24], [@CR28]\] in *bik1* seedlings were comparable to those in WT seedlings, while the expression of *PHR1* \[[@CR37]\] in *bik1* seedlings was reduced (Fig. [6a](#Fig6){ref-type="fig"}). The expression levels of the Pi starvation-induced genes tested were markedly increased in WT and *bik1* plants grown under --Pi condition (Fig. [6a](#Fig6){ref-type="fig"}). However, significant reductions in the expression levels of these Pi starvation-induced genes was observed in *bik1* plants, as compared with those in WT plants, under --Pi condition (Fig. [6a](#Fig6){ref-type="fig"}). These results indicate that mutation in *BIK1* leads to reduced expression of the Pi starvation-induced genes in plants grown under --Pi condition.Fig. 6Down-regulated expression of Pi starvation-induced genes in *bik1* plants. **a** Expression patterns of Pi starvation-induced genes in WT and *bik1* seedlings grown under + Pi and --Pi conditions. Seven-day-old seedlings grown under normal Pi condition were treated with Pi (+Pi, 250 μM) or without Pi (−Pi) for 5 days and total RNA were extracted from root tissues at 5 days after treatment. **b** Expression patterns of *PHO2* and *miRNA399* in WT and *bik1* seedlings grown under + Pi and --Pi conditions. Seven-day-old seedlings grown under normal Pi condition were transferred to medium supplemented with Pi (+Pi, 250 μM) or without Pi (−Pi) and samples were collected at 48 h after treatment. Expression of Pi starvation-responsive genes and miR399s was analyzed by qRT-PCR using specific primers and data were normalized with the transcript level of *UBQ10* as an internal control. Data presented are the means ± SD from three independent experiments and different letters above the columns indicate significant differences at *p* \< 0.05 level It was found that expression of *PHO2*, encoding a ubiquitin-conjugating E2 enzyme, was suppressed by *miRNA399*, which controls Pi homeostasis in plants and whose expression is up-regulated by Pi starvation \[[@CR48]--[@CR52]\]. To examine whether *PHO2*/*miRNA399* links to increased accumulation of Pi in *bik1* plants, we analyzed the changes in levels of *PHO2* expression and two miRNA399 primary transcripts, *miRNA399a* and *miRNA399d*, in *bik1* seedlings grown under + Pi and --Pi conditions. The expression level of *PHO2* and the transcript levels of *miRNA399a* and *miRNA399d* were comparable in WT and *bik1* seedlings grown under + Pi condition. However, the expression levels of *PHO* in WT and *bik1* seedlings under --Pi condition were significantly reduced whereas the transcript levels of *miRNA399a* and *miRNA399d* in WT and *bik1* seedlings grown under --Pi condition were markedly increased, as compared with the corresponding seedlings under + Pi condition (Fig. [6b](#Fig6){ref-type="fig"}). A further reduction in the expression of *PHO2* and a further increase in the transcripts of *miRNA399a* and *miRNA399d* in *bik1* seedlings was observed as compared with those in WT seedlings grown under --Pi condition (Fig. [6b](#Fig6){ref-type="fig"}). Together, these data indicate that the *PHO2*/*miRNA399* plays a role in regulating Pi accumulation in *bik1* plants under --Pi condition. Discussion {#Sec9} ========== The maintenance of cellular Pi homeostasis in plants involves complicated regulatory mechanisms. Several studies have demonstrated that posttranslational modifications such as phosphorylation, sumoylation and ubiquitination of regulatory proteins play critical roles in Pi starvation responses \[[@CR54], [@CR55], [@CR57]\]. Recently, genome-wide co-expression analysis identifies dozens of protein kinases as important regulators of Pi deficiency-induced root hair remodeling \[[@CR79], [@CR80]\]. It was also found that *in vivo* phosphorylation activation of PPC1 (phosphoenolpyruvate carboxylase 1) is involved in the metabolic adaptations of Pi starvation \[[@CR81]\] and activation of MKK9-MPK3/MPK6 enhances phosphate acquisition in Arabidopsis \[[@CR56]\]. These findings indicate an important role for protein kinases in regulating Pi starvation response. In this study, we found that BIK1, a plasma membrane-localized receptor-like protein kinase \[[@CR68]\], plays an important role in modulating Pi starvation responses and functions as a negative regulator of Pi homeostasis in Arabidopsis plants. Thus, evidence presented in the current study renders BIK1 for a novel function in Pi starvation response, in addition to its previously reported functions in immunity \[[@CR68]--[@CR73]\]. Several lines of evidence presented in this study support that BIK1 functions in Pi starvation response. Earlier study has shown that expression of *BIK1* could be induced by infection with *B. cinerea* as well as treatments with some well-known defense signaling molecules \[[@CR68]\]. Our qRT-PCR analysis of *BIK1* expression and determination of *BIK1* promoter activity in stable *BIK1*~*pro*~::GUS transgenic seedlings demonstrate that *BIK1* can be induced by Pi starvation and Pi starvation-induced expression of *BIK1* in root and shoot tissues initiated earlier than in leaf tissues (Fig. [1a](#Fig1){ref-type="fig"}). The pattern of *BIK1* expression induced by Pi starvation is similar to most of the Pi starvation-induced genes identified so far, like *WKRY75*, *ERF070* and *ZAT6* \[[@CR39], [@CR40], [@CR44]\]. Importantly, promoter activity analysis revealed that GUS staining driven by the *BIK1* promoter was initiated in vascular tissues of root (Fig. [1b](#Fig1){ref-type="fig"}) that is the primary organ that responds to Pi starvation, further confirming the responsiveness of *BIK1* to Pi starvation. Thus, it is likely that *BIK1* can respond rapidly to the altered Pi status of plants under Pi starvation condition. It was particularly noteworthy that BIK1 is a plasma membrane-localized receptor-like protein kinase and has been shown to be phosphorylated by other kinases (e.g. BAK1 and FLS2) for its full activity in plant immune responses \[[@CR69], [@CR70]\]. In this context, it is thus possible that BIK1 is phosphorylated by unknown kinase during Pi starvation stress and phosphorylated BIK1 acts in Pi starvation response, in addition to its transcriptional regulation upon Pi starvation. The primary root, lateral root, and root hairs are three main components of the root architecture that is critical to absorption of Pi from soil. Low P availability can drastically alter the root architecture by switching the indeterminate growth to determinate growth to promote lateral root growth \[[@CR75]\]. Generally, not only the primary roots but also almost all mature lateral roots enter into the determinate developmental program under low Pi condition \[[@CR82]\]. The *bik1* seedlings showed a significant change in root architecture and the primary root length of *bik1* plants was significantly decreased (Fig. [6](#Fig6){ref-type="fig"}). This is similar to a common phenomenon observed that the primary root length is decreased significantly due to the Pi starvation-induced determinate growth in primary root \[[@CR82]\]. In addition to the change of the primary root, the *bik1* plants showed increased number and length of lateral roots and root hairs (Fig. [6](#Fig6){ref-type="fig"}), which increase root surface contacting an increased soil volume to explore Pi availability in soil \[[@CR75]\]. Similar root architectures were observed in the *bik1* plants under + Pi and --Pi conditions (Fig. [6](#Fig6){ref-type="fig"}). Thus, it is likely that the effect of BIK1 on the development of root architecture is independent of the Pi status in plants and Pi availability in soil. Nevertheless, the characteristic root architecture observed in *bik1* plants suggests that BIK1 is a negative regulator of lateral root and root hair development. This is in agreement with previous observations that the *bik1* plants showed some defects in growth and development, e.g. weak stem strength, early flowering and less seed setting \[[@CR68]\]. Together, these data further demonstrate that BIK1 is required for normal plant growth and development \[[@CR68]\]. It will be interesting to investigate whether the growth and developmental defects in *bik1* plants is due to an activated Pi starvation response and if this is the case, the results obtained will further support a cross-talk between Pi starvation response and plant growth/development. It was previously found that overexpression of *ZAT6* retards growth and results in typical Pi starvation responses \[[@CR40]\]. Accompanied with high contents of total Pi in roots and leaves of *bik1* plants under + Pi and --Pi conditions (Fig. [2](#Fig2){ref-type="fig"}) are the significant accumulation of anthocyanin (Fig. [4a](#Fig4){ref-type="fig"} and [b](#Fig4){ref-type="fig"}), increased activity of acid phosphatase (Fig. [4a](#Fig4){ref-type="fig"} and [b](#Fig4){ref-type="fig"}) and alterations in expression of Pi starvation-induced genes (Fig. [6](#Fig6){ref-type="fig"}). Collectively, these data imply that, whatever grown under + Pi condition or under --Pi condition, a Pi starvation response including those of physiological, molecular and metabolic changes is activated in *bik1* plants. Increased activity of acid phosphatases in *bik1* plants grown under --Pi condition (Fig. [3b](#Fig3){ref-type="fig"}) not only represents a characteristic Pi starvation response \[[@CR83]\] but also can release more Pi available for absorption by the roots. This is supported by the findings that mutations in genes encoding for purple acid phosphatases affect markedly the uptake of Pi from exogenous sources \[[@CR84]--[@CR86]\]. Because expression of *BIK1* was rapidly induced by Pi starvation (Fig. [1](#Fig1){ref-type="fig"}), it seems possible that BIK1 is involved in Pi uptake by regulating the activity of acid phosphatase. Similar observations were also obtained for some Pi starvation response regulators such as PHR1 and ZAT6, which have been shown to affect activity of acid phosphatase and Pi uptake in transgenic plants with altered expression of these genes \[[@CR40], [@CR85], [@CR87], [@CR88]\]. Interestingly, expression of *Pht1;2* and *Pht1;4*, encoding for high-affinity Pi transporters \[[@CR24], [@CR28]\], *WRKY75*, encoding a WRKY transcriptional factor involved in Pi acquisition \[[@CR39]\], *At4*, a member of the Mt4/TPSI1 gene family involved in Pi distribution \[[@CR78]\], and *PHR1*, encoding a MYB transcriptional factor involved in Pi starvation response signaling \[[@CR37]\], was suppressed significantly in roots of *bik1* plants under --Pi condition (Fig. [6a](#Fig6){ref-type="fig"}). These data demonstrate that BIK1 has global effect on a set of Pi starvation-induced genes that are involved in Pi acquisition and mobilization. Notably, the expression levels of the tested Pi starvation-responsive gene except *PHR1* in *bik1* plants were slightly reduced as compared to those in WT plants under + Pi condition (Fig. [6a](#Fig6){ref-type="fig"}), indicating a possibility that these Pi starvation-responsive genes may be also affected by mutation of *BIK1* itself. Particularly, the expression of *PHR1* in *bik1* plants grown under + Pi condition was significantly reduced as compared to that in WT plants (Fig. [6a](#Fig6){ref-type="fig"}), leading to an open question whether BIK1 acts upstream of PHR1 in regulating the Pi starvation signaling. This can be clarified by phenotyping and analysis of *PHR1* expression in *BIK1*-overexpressing plants and/or detailed examination of biochemical and genetic requirement of BIK1 for Pi starvation response. It is currently difficult to link the function of BIK1 to Pi uptake or its root-to-leaf translocation as Pi uptake and transportation in WT plants is also affected under --Pi condition (Fig. [2](#Fig2){ref-type="fig"}). This can further be explained by the reduced expression of these tested Pi starvation-induced genes in root of *bik1* plants as compared to those in WT plants under --Pi condition (Fig. [6](#Fig6){ref-type="fig"}). Furthermore, altered expression patterns of *PHO2* and *miRNA399a/d*, which are thought to be involved in systemic signaling of Pi starvation response and Pi distribution in plants \[[@CR49], [@CR52]\], were also observed in *bik1* plants grown under + Pi and --Pi conditions (Fig. [6b](#Fig6){ref-type="fig"}). Thus, it is likely that BIK1 is involved in maintaining Pi homeostasis in whole plants. Taken together, these data support an idea that BIK1 is a negative regulator of Pi starvation responses, probably through affecting development of root architecture and a series of physiological and biochemical events related to Pi acquisition, mobilization and translocation. It is reasonable to speculate that, under Pi starvation stress, the *bik1* plants may experience a reduced Pi content as the WT plants but they can uptake efficiently Pi from growth environment with their significantly increased root surface area, leading to an increased Pi content in root and leaf tissues of *bik1* plants (Fig. [2](#Fig2){ref-type="fig"}). However, the detailed mechanism that BIK1 functions in Pi starvation response remains to be explored further. Particularly, characterization of the targets that are phosphorylated by BIK1 will be greatly helpful in elucidating the early signaling events that determine Pi starvation response. ROS not only are toxic compounds produced in plant response to various stresses but also plays an integral role as signaling molecules in regulation of numerous biological processes such as growth, development, and responses to biotic and/or abiotic stress \[[@CR89]--[@CR91]\]. The involvement of ROS in Pi starvation response has been established recently. In this study, we found that both WT and *bik1* plants grown under --Pi condition accumulated more ROS, represented by H~2~O~2~ and superoxide anion, than those in plants grown under + Pi condition (Fig. [5](#Fig5){ref-type="fig"}). This is similar to the previous observations that H~2~O~2~ concentrations in roots increased upon Pi deprivation \[[@CR92]\]. Notably, the accumulation of superoxide anion in WT and *bik1* plants under + Pi condition was comparable, the accumulation of superoxide anion in *bik1* plants was significantly reduced as compared to that in WT plants under --Pi condition (Fig. [5](#Fig5){ref-type="fig"}). By contrast, the *bik1* plants accumulated lower level of H~2~O~2~ than WT plants grown under + Pi condition (Fig. [5](#Fig5){ref-type="fig"}), which is similar to the histochemical staining in soil-grown *bik*1 plants \[[@CR68]\]. Thus, it is likely that BIK1 affects the accumulation of superoxide anion and H~2~O~2~ in different ways: loss of *BIK1* function suppressed, at least partially, the Pi starvation-induced accumulation of superoxide anion while BIK1 regulates directly the accumulation of H~2~O~2~ in plants under normal condition. This may also imply different functions of superoxide anion and H~2~O~2~ in Pi starvation response. On the other hand, increased accumulation of ROS in plants under --Pi condition might be one of the stress responses as those in response to other abiotic stress \[[@CR90]\]. It was recently shown that localization pattern of ROS accumulated in root during Pi starvation stress is critical to shape the root architecture \[[@CR93]--[@CR96]\]. The reduced accumulation of ROS in *bik1* plants in relative to those in WT plants (Fig. [5](#Fig5){ref-type="fig"}), which is similar to the observation that loss of *BIK1* function impaired the PAMP-induced ROS burst in immunity \[[@CR97], [@CR98]\], may attribute to the altered root architecture of *bik1* plants (Fig. [3](#Fig3){ref-type="fig"}). However, detailed analysis of ROS localization and the possible mechanism regulating ROS generation in root of *bik1* plants will provide new insights into the connection between BIK1 and ROS in development of root architecture. Conclusions {#Sec10} =========== In summary, this study demonstrates that BIK1 is a Pi starvation-responsive gene and functions as a negative regulator of Pi homeostasis in Arabidopsis. This not only renders a novel function for BIK1 but also strengthens our understanding of post-transcriptional regulation during Pi starvation responses in plants. Considering that BIK1 is a plasma membrane-localized receptor-like protein kinase, it thus may play a role in sensing and/or processing of the Pi starvation signal during early stage of Pi starvation stress. Methods {#Sec11} ======= Plant materials and growth condition {#Sec12} ------------------------------------ *Arabidopsis thaliana* ecotype Columbia (Col-0) and *bik1* (provided by Dr. Tesfaye Mengiste, Purdue University, USA) mutant \[[@CR68]\] were used in this study. Seeds were surface sterilized with 70 % ethanol, washed with sterilized distill water and vernalized at 4 °C for 2 days before germination. Hydroponic, solid medium and liquid medium cultivation were used for different purpose of experiments. For solid medium cultivation, the basic medium used contained 2.06 mM NH~4~NO~3~, 1.88 mM KNO~3~, 0.31 mM MgSO~4~, 0.1 mM MnSO~4~, 0.03 mM ZnSO~4~, 0.1 mM CuSO~4~, 0.3 mM CaCl~2~, 5.0 mM KI, 0.1 mM CoCl~2~, 0.1 mM FeSO~4~; 0.1 mM EDTA, 0.1 mM H~3~BO~3~, 1 mM Na~2~MoO~4~.2H~2~O, 3 g/L sucrose, 10 g/L agar, pH 5.8. When treated for Pi sufficiency (+Pi) and Pi deficiency (−Pi), the medium was supplemented with 1 mM KH~2~PO~4~, 10 μM KH~2~PO~4~ and 0.99 mM KCl. For hydroponic cultivation, seedlings at 5 \~ 7-leaf stage were transferred to 1/2 Hoagland solution for short adaption and then transferred to hydroponic solution containing 250 μM Pi or no Pi \[[@CR99]\]. For liquid medium cultivation, seeds were dispensed in 1/2 MS medium without agar and rinsed with sterilized distilled water at 7 days. Seedlings were then transferred in MS liquid medium with Pi (1 mM) or without Pi and allowed to grow with shaking at 85 rpm. Plants were grown in a growth room under a 16 h light (100 μmol · s^−1^ · m^−2^ photons m^−2^ sec^−1^ of intensity) and 8 h dark cycle at 22 ± 2 °C with 60 % relative humidity. Measurement of root system architecture {#Sec13} --------------------------------------- Seven-day-old seedlings grown on 1/2 MS medium were transferred to Petri dishes and treated for + Pi or --Pi for another 7 days. Length of main root, number and length of the lateral root were measured. Total numbers of root hairs in a 5 mm region from root tip were recorded. Data were recorded from 15 individual plants from each treatment. Anthocyanin analysis {#Sec14} -------------------- Measurement of anthocyanin was performed as described previously \[[@CR100]\]. Briefly, seedlings grown under + Pi or --Pi condition in solid medium were put into 20 mL extraction solution (propanol : HCl : H~2~O = 18:1:81, V/V/V) and boiled for 90 s. The mixtures were kept overnight in dark at room temperature, centrifuged for 40 min at 5000× g, and the absorbance (A) was measured at 535 nm and 650 nm. The A~535~ values were corrected with the A~650~ values using formula A~535~ = A~535~-2.2 × A~650~. Anthocyanin contents were calculated according to a standard curve prepared with the same protocol. Quantification of total Pi content {#Sec15} ---------------------------------- Total Pi content was quantified according to the U.S. Environmental Protection Agency Method 365.2 with minor modifications \[[@CR39]\]. Briefly, samples (\~50 mg/sample) were flamed to ash after recording of their dry weights and then 100 μL of concentrated HCl was added. Ten microliters of the mixture were drawn and diluted into 790 μL of water, followed by addition of 200 μL assay solution (4.8 mM NH~4~MoO~4~, 2.5 N H~2~SO~4~, and 35 μM ascorbic acid). The reactions were incubated at 45 °C for 20 min and the absorbance at 650 nm was measured spectrophotometrically. Contents of total Pi were calculated according to the Pi standard curve prepared with the same procedure. Quantification and staining of acid phosphatase activity {#Sec16} -------------------------------------------------------- Activity of acid phosphatase was quantified using the pNPP hydrolysis assay according to a previously described method \[[@CR101]\]. Briefly, 30 mg of samples were grounded, transferred to Eppendorf tubes and then spin for 10 min at 2000 × g. Reactions containing 100 μL supernatant, 100 μL *p*-nitrophenol sodium phosphate and 2.8 mL buffer were kept at 30 °C for 10 min, with shaking occasionally, and terminated by addition of 1 mL 0.5 M NaOH. The absorbance of the reactions was determined spectrophotometrically at 400 nm and the activity of acid phosphatase was calculated from the production of *p*-nitrophenol. Total protein was estimated using Bradford's reagent and the total acid phosphatase activity was expressed as mU/mg protein. Staining of acid phosphatase activity was performed as described by Tomscha et al. \[[@CR102]\] with minor modifications. Ten-day-old seedlings grown in liquid medium were rinsed in --Pi medium, transferred to fresh medium supplemented with Pi (1 mM) or without Pi and then covered with a layer of 0.05 % agarose solution containing 0.008 % 5-bromo-4-chloro-3-indolyl phosphate (BCIP) \[[@CR103]\]. Detection and quantification of H~2~O~2~ {#Sec17} ---------------------------------------- For quantification of H~2~O~2~, 30 mg samples from 4-week-old plants were completely ground, followed by addition of 200 μL 20 mM K~2~HPO~4~ (pH6.5) phosphate buffer. Quantification of H~2~O~2~ was performed using a commericial kit (Jiancheng Bioengineering Institute, Nanjing, China) according to the manufactruer's recommendation. Detection of superoxide anion was performed by the nitroblue tetrazolium (NBT) staining \[[@CR104]\]. Four-week-old seedlings were vacuum infiltrated with 2 ml of 10 mM potassium phosphate buffer (pH 7.5) containing 10 mM NaN~3~ and 0.1 % NBT for 30 min, cleared by boiling in 96 % ethanol, remained in 50 % ethanol before taking photos. qRT-PCR analysis of gene expression {#Sec18} ----------------------------------- Total RNA was extracted using TRIZOL reagent (Invitrogen, Shanghai, China) according to the manufacturer's instructions. First strand cDNA was synthesized from 500 ng of total RNA using SuperScript III Kit (Invitrogen, Shanghai, China). The qPCR reactions contained 12.5 μL SYBR Premix Ex TaqTM (TaKaRa, Dalian, China), 0.1 μg cDNA and 7.5 pmol of each of gene-specific primers in 25 μL and were conducted on a CFX96 real-time PCR system (BioRad, Hercules, CA, USA). Gene-specific primers used were as followings: BIK1-q-F, 5′-ACT TAT GGG TAC GCC GCG CCT GAG T-3′; BIK1-q-R, 5′-GGC ACG GAC CAC TTG GTC CA-3′; GUS-q-1F, 5′-AGG TGC ACG GGA ATG TTT CG-3′; GUS-q-1R, 5′-TGT GAG AGT CGC AGA ACA TT-3′; PHR1-q-1F, 5′-GTG ATT GGC ATG AAT GGG CTG AC-3′; PHR1-q-1R, 5′-CGC AAT TCC ACA GAC GGA GAA GG-3′; AT4-q-1F, 5′-GAT CGA AGT TGC CCA AAC GA -3′; AT4-q-1R, 5′-GAG CGA TGA AGA TTG CAT GAA G-3′; WRKY75-q-1F, 5′-GAG AAA TCC ACC GAA AAC TTC GAG CAT AT-3′; WRKY75-q-1R, 5′-GCA TGG TTT TTC TTT TCA ACA CAC GTA AAA TGT A-3′; PHT1.2-q-1F, 5′-AGG GCA AGT CCC TCG AAG AAC T-3′; PHT1.2-q-1R, 5′-ATC AAA CAA ACC ACA AAC AAC TCC ACA T-3′; PHT1.4-q-1F, 5′-TTG CTC CTA ATT TTC CTG ATG CT -3′; PHT1.4-q-1R, 5′-TGT GCC GGC CGA AAT CT-3′. Pri-miRNA399a-1F, 5′-TGG CAG GAA ACC ATT ACT TAG ATC T-3′; Pri-miRNA399a-1R, 5′-TCA CTA ATT AAA AGC AAT GCA TAA AGA GA-3′; Pri-miRNA399d-1F, 5′-TTA CTG GGC GAA TAC TCC TAT GG-3′; Pri-miRNA399d-1R, 5′-ATT TTA CTT GCA TAT CTA GCC AAT GC-3′; PHO2-q-1F, 5′-AGG TTT GAA GCT CCA CCC TCA-3′; PHO2-q-1R, 5′-CCC AAG ATG TGA TTG GAG TTC C-3′; UBQ10-q-1F, 5′-GGC CTT GTA TAA TCC CTG ATG AAT AAG-3′; UBQ10-q-1R, 5′-AAA GAG ATA ACA GGA ACG GAA ACA TAG T-3′. Relative gene expression levels were calculated using 2^--△△CT^ method with three independent biological replicates. Generation of *BIK1*~*pro*~::GUS transgenic line and GUS staining {#Sec19} ----------------------------------------------------------------- A 2 kb sequence upstream of the *BIK1* start codon was PCR amplified using primers AtBIK1-GUS-1F (5′-ATA **CTG CAG** CTT GTT GAT TGA TTA ATA GAT TAC C-3, a *Pst*I site in bold) and AtBIK1-GUS-1R (5′-GCC **GGA TCC**AGA ACT GAA GCA AGA ACC CAT C-3′, a *Bam*HI site in bold) and cloned into vector pCAMBIA1301. Transformation of wild-type Col-0 plants was performed using the floral dip infiltration method mediated by *Agrobacterium tumefaciens* strain GV3101. Plants of T2 generations from kanamycin-resistant transformants were used for GUS histochemical staining \[[@CR105]\]. Abbreviations {#Sec20} ============= +Pi, Pi sufficiency; *B. cinerea*, *Botrytis cinerea*; BCIP, 5-bromo-4-chloro-3-indolyl phosphate; *BIK1*, *Botrytis-induced kinase1*; MKK, mitogen-activated protein kinase kinase; MPK, mitogen-activated protein kinase; NBT, nitroblue tetrazolium; PAMP, pathogen associated molecular pattern; *Pht1*, *PHOSPHATE TRANSPORTER1*; Pi, phosphate; −Pi, Pi deficiency; pNPP, *p*-nitrophenol sodium phosphate; *PPC1*, *phosphoenolpyruvate carboxylase 1*; qRT-PCR, quantitative reverse transcription PCR; ROS, reactive oxygen species; WT, wild type We are grateful to Dr. Tesfaye Mengiste, Purdue University, USA, for providing the Arabidopsis WT and bik1 seeds used in this study. Funding {#FPar1} ======= This work was supported by the National Natural Foundation of Sciences (no. 31272028) and the Ph.D. Program Foundation of Ministry of Education of China (no. 20120101110070). Availability of supporting data {#FPar2} =============================== All data supporting the results of this article are included within the article. Authors' contributions {#FPar3} ====================== HZ and FS designed the study and all experiments; HZ, LH and YH conducted all the molecular biology and physiology experiments; HZ and FS analyzed the data; FS drafted the manuscript with HZ. All authors have read and approved the final version of the manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== Not applicable. No human or animals were involved in this study.

Background ========== There is strong evidence that a substantial proportion of new HIV infections in African countries are associated with herpes simplex virus type 2 (HSV-2) \[[@B1]-[@B6]\]. Infection with HSV-2 has a significant impact on the risk of HIV acquisition and the impact increases as the HIV epidemic progresses \[[@B7],[@B8]\]. One reason for this association could be the relative importance of HSV-2 in genital ulcer disease (GUD) in Africa has increased markedly. Infection with HSV-2 disrupts the genital mucosa and provides a portal of entry for HIV, leading to increased susceptibility of HIV in HIV-negative persons. In HIV-positive persons, infection with HSV-2 accelerates replication and genital shedding of the virus, thus such individuals are more likely to transmit HIV \[[@B9]\]. Studies from sub-Saharan countries have reported a high prevalence of HSV-2 in adults ranging from 30% to 80% in women and up to 50% in men \[[@B5],[@B10],[@B11]\]. In Tanzania, recent studies have shown HSV-2 represents 71% to 83% of all identified pathogens in genital ulcers \[[@B12],[@B13]\]. Thus, the magnitude of HSV-2 infection may suggest the expected course of the HIV epidemic. Tanzania is among the countries with a stabilised HIV epidemic. The estimated adult HIV prevalence was 8% by the end of 2004 based on data from the Tanzania HIV/AIDS indicator survey \[[@B14]\], closely matching HIV estimates (8.7%) reported from antenatal clinic surveillance in the same year \[[@B15]\]. The HIV prevalence (2%) in Manyara region by the end of 2004, matched with the estimate pregnant women attending local clinics within rural Manyara and Singida regions \[[@B14],[@B16]\]. In the antenatal group, young women (15--19 years) presented with the highest prevalence (3.0%) and in the general population, men aged 35--49 years \[[@B16],[@B17]\]. In comparison with other regions in Tanzania, the HIV prevalence is still low \[[@B14]\]. The magnitude of HSV-2 in the same area is largely unknown, while the prevalence of syphilis among pregnant women was 17% as reported by a previous study within the area \[[@B18]\]. Unlike genital herpes, studies exploring syphilis as a GUD enhancing HIV susceptibility and infectiousness in high-risk groups have revealed conflicting results \[[@B19],[@B20]\]. The limitations have been due to either a few number of syphilis cases or case ascertainment by self-reports rather than laboratory confirmed results. In Tanzania, 51% of all stillbirths are associated with maternal syphilis \[[@B21]\]. Analysis of maternal syphilis and HIV prevalence among pregnant women attending selected antenatal clinics in by the end of 2004 showed geographical variation in the prevalence \[[@B22]\]. Rural antenatal clinics had high prevalence of syphilis (9.5%) and low HIV prevalence (5.2%) compared with other sites (6.8% syphilis vs. 9.3% for HIV). The high prevalence was associated with lack of formal education, inadequate access to health care and poverty \[[@B23],[@B24]\]. Thus to determine the prevalence of syphilis and genital herpes in this rural area with low but increasing HIV prevalence, we analysed antibodies against HSV-2 and *Treponema pallidum*. We used data and serum samples collected from a previously reported antenatal survey \[[@B16]\]. Methods ======= Study area and population ------------------------- The current study was part of the epidemiological study conducted within the remote rural areas of Manyara and Singida in Tanzania. The area falls within the service catchment area of Haydom Lutheran Hospital (HLH). The main focus aimed towards understanding the epidemiological context of HIV in the area as part of the comprehensive prevention-intervention programme launched by the hospital to combat HIV and other sexually transmitted infections in the area \[[@B25]\]. Part of the interventions included voluntary counselling and testing for HIV (VCT), syphilis and HSV-2. Education provided for participants included HIV- related preventive education and benefits of highly active antiretroviral treatment (HAART) for prevention of mother to child transmission (PMTCT) explained, according to national guidelines for care and treatment of people living with HIV and AIDS. Sampling -------- The survey included six antenatal clinics within the three administrative divisions that comprised the above-defined catchment area, between November 2003 and April 2004, and the study population is described in detail \[[@B16]\]. In brief, it included all counselled and consenting pregnant women aged between 15--49 years attending six local antenatal clinics (ANC) for the first time in that particular pregnancy during the surveillance period. Two clinics from each administrative division were included. Based on previous estimates of HIV, HSV-2 and syphilis prevalence in the area, HIV prevalence was the lowest, so the sample size was based on that estimate. A sample size of 1000 pregnant women would allow us to calculate a 2% prevalence of HIV with 95% confidence intervals of 1 to 3%. Point prevalence estimates of HIV, HSV-2 and syphilis are presented as the point estimate with their 95% confidence intervals. Data collection tools and laboratory methods -------------------------------------------- The research team collaborated with the hospital PMTCT teams in data collection. The PMTCT team was responsible for registration and coding of participant\'s identities and providing follow up care. The research team used a standard structured questionnaire to obtain information regarding socio-demographic and sexual behaviour characteristics. Included in the questionnaires were marital status, self-reported history of sexually transmitted infections and genital ulcers, the use of condoms and other contraceptives, and sexual partners. Blood was collected using vacutainers, and prepared serum samples were stored in the refrigerator and tested for HIV antibodies and syphilis the next day. Following HIV and syphilis antibody testing, the samples were stored at -80°C until analysis for HSV-2 antibodies. Screening for anti-HIV IgG antibody was performed anonymously using Bionor HIV-1 & 2 assay^®^(BIONOR AS, Skien, Norway). Reactive samples were confirmed using Bionor HIV-1 & 2 assay^®^(BIONOR AS, Skien, Norway). Screening for syphilis was done using Rapid plasma reagin (RPR) by Omega Immutrep-RPR^®^, Omega Diagnostics, Alva, UK and the reactive samples were confirmed by TPHA (Immutrep TPHA^®^, Omega Diagnostics, Alva, UK). Positive RPR and TPHA tests were considered to indicate current or untreated syphilis. Specific antibodies for HSV-2 were detected by the non-commercial peptide-55 Enzyme-linked immunosorbent assay as described elsewhere \[[@B26]-[@B28]\]. The assay is specific for reactivity with an epitop within glycoprotein G of HSV-2 (gG2) and does not cross-react with Herpes simplex virus type 1. Sensitivity and specificity of 100% and 98.4%, respectively, have been reported \[[@B27]\]. The cut-off value was set at a mean absorbance of the negative controls plus five standard deviations. Tanzanian sera, initially analysed by the same assay and the results confirmed with a commercial HerpeSelect 1 and 2 Immunoblot IgG ^®^(IB0900G) from Focus Technologies Inc. (USA), were used as positive and negative controls \[[@B29]\]. The Regional Committee for Medical, Health Research, and Ethics in Western Norway (REK) and the Tanzanian National Institute of Medical Research (NIMR) gave ethical approval for research exploring HIV, HSV-2 and syphilis within the mentioned area. Regional and District level Ethical approval was sought and granted. Oral informed consent was sought from individuals and they where informed of their right to refuse to participate with out any consequences. Statistical analyses -------------------- Data were entered twice in Excel spreadsheets and analysed using Stata 9.2^®^(College station, Texas, USA). We assessed self-reported previous adverse pregnancy outcomes as a probable indicator of previous syphilitic infection for women with previous pregnancies. The age at first pregnancy was explored as an indirect measure of unprotected sexual intercourse. The prevalence of HSV-2 and syphilis was compared to the previously reported HIV prevalence in the same population of pregnant women to assess the coexistence of these infections. We define polygamous relationship as women with sexual relations with a man having more than one wife or sexual partner(s) at the same time and monogamous relation women in marriage or sexual relationship with a man who has no other but only one wife/sexual partner at the same time. Logistic regression was used to explore potential risk factors for HSV-2 and syphilis. Since HSV2 antibodies remain for life in infected people, older pregnant women would be more likely to have antibodies from previously acquired HSV2 infections. We adjusted for age in the analysis of HSV-2 and risk factors to avoid overestimation in older women and underestimation in younger women thus allowing comparison of the risk of HSV-2 infection. Variables with p-values less than 0.25 on age-adjusted analysis were included in the final multivariate model and the associations presented as Odds ratios (OR) with a 95% confidence interval (CI). Results ======= In a sample of 1377 consenting pregnant women, 1296 women completed interviews and provided serum samples for laboratory testing. Four per cent of women (58/1377) declined a blood test, and we discarded 1.7% (23/1319) haemolysed samples. The mean age of the studied population was 26.3 (SD ± 6.1) years and the mean age at first pregnancy was 19 (SD ± 3) years. Ninety-one per cent of the women were married (Table [1](#T1){ref-type="table"}) and 79% of them reported monogamous marriage while 28% of all women had no formal education. ###### The prevalence of genital herpes and associated factors among pregnant women **Characteristics** **N (%)** **HSV-2 prevalence % (95% CI)** **Age-adjusted OR (95% CI)** **Multivariate analysis OR (95% CI)** ------------------------------------------------------- ------------- --------------------------------- ------------------------------ --------------------------------------- **Total** **1296** **20.7 (18.53--23.00)** **[Clinic]{.ul}** Basotu 242 (18.7) 7.0 (4.27 -- 11.21) 1 1 Haydom 319 (24.6) 23.8 (19.33 -- 28.86) **4.2 (2.41 -- 7.31) \*** **4.2 (2.39 -- 7.41) \*** Muslur 192 (14.8) 32.8 (26.32 -- 40.00) **6.5 (3.63 -- 11.52) \*** **6.8 (3.76 -- 12.27) \*** Dangayda 113 (8.7) 32.9 (23.59 -- 41.38) **6.0 (3.19 -- 11.35) \*** **5.3 (2.78 -- 10.16) \*** Mwanga 249 (19.2) 20.1 (15.40 -- 25.71) **3.3 (1.86 -- 5.95) \*** **3.3 (1.84 -- 6.02) \*** Nduguti 181 (14.0) 14.4 (9.76 -- 20.53) **2.2 (1.16 -- 4.23) \*** **2.2 (1.15 -- 4.24) \*** **[Age groups (years)]{.ul}** 15--19 165 (12.7) 19.5 (13.91 -- 26.58) 1 20--24 417 (32.2) 19.4 (15.80 -- 23.62) 1.0 (0.63 -- 1.57) 0.8 (0.48 -- 1.28) 25--29 349 (26.9) 17.4 (13.69 -- 21.90) 0.9 (0.54 -- 1.40) 0.7 (0.41 -- 1.13) 30--34 224 (17.3) 24.5 (19.58 -- 31.30) 1.3 (0.82 -- 2.19) 0.9 (0.55 -- 1.61) ≥35 141 (10.9) 27.7 (20.62 -- 35.93) 1.6 (0.92 -- 2.69) 1.0 (0.56 -- 1.77) **[Marital status]{.ul}** Married 1183 (91.3) 20.9 (18.62 -- 23.33) 1 \_\_ Single/divorced/separated/widow 113(8.7) 18.6 (12.12 -- 27.23) 1.1 (0.68 -- 1.91) **[Education (years)]{.ul}** None 349 (28.2) 20.9 (16.85 -- 25.64) 1 \_\_ 1--6 years 174 (13.4) 17.8 (12.60 -- 24.49) 0.8 (0.51 -- 1.40) 7 years or more 773 (59.7) 21.2 (18.42 -- 24.30) 1.0 (0.74 -- 1.9) **[^a^Polygamous relationship]{.ul}** No 1024 (79.0) 17.5 (15.23--19.90) 1 Yes 272 (21.0) 32.7 (27.24--38.69) **2.2 (1.63 -- 3.01) \*** **2.2 (1.62 -- 2.99) \*** **[^b^Age at first pregnancy]{.ul}** \< 18 years 158 (14.0) 23.4 (17.21 -- 30.94) 1 ≥ 18 years 974 (86.0) 20.4 (17.97 -- 23.13) 1.2 (0.85--1.73) 1.2 (0.82 -- 1.70) **[Number of lifetime sexual partners]{.ul}** One only 1231 (95.0) 18.2 (16.14 -- 20.54) 1 More than one 65 (5.0) 35.3 (24.36 -- 47.90) **2.0 (1.21 -- 3.46) \*** **1.8 (1.04 -- 3.04)** **[Use of contraceptives other than condom]{.ul}** No 1088 (84.0) 19.2 (16.93--21.71) 1 **1** Yes 208 (16.0) 28.4 (22.45--35.09) **1.7 (1.20--2.37)\*** **1.7 (1.21--2.41) \*** **[History of sexually transmitted infections]{.ul}** No 1135 (87.6) 19.8 (17.56 -- 22.29) 1 Yes 161 (12.4) 27.0 (20.20 -- 34.35) **1.5 (1.01 -- 2.17)** 1.3 (0.87--1.92) **[Ever used condom]{.ul}** No 1233 (95.1) 20.6(18.47 -- 23.07) 1 \_\_ Yes 63 (4.9) 20.6 (11.85 -- 33.04) 1.1 (0.56 -- 1.97) ^[a]{.ul}^Relationship with a man having more than one female sexual partner regardless of marital status ^b^Analysed for those aged 20 ≥ years at the time of survey Total HSV-2 positive (n = 268/1296) \*[P-values: \< 0.01]{.ul} Table [1](#T1){ref-type="table"} summarizes results on HSV-2 prevalence and associated factors. The overall prevalence of HSV-2 was 20.7% (95% CI: 18.53--23.00), varying between 7.0% and 32.9%. HSV-2 prevalence increased with age, ranging from 19% in the 15--19 year age group to 27% among women aged 35 years or older. The prevalence was between 18% and 35% for all socio-demographic variables. Women in polygamous relationships, had high prevalence of HSV-2 (32.7%, 95% CI: 27.24--38.69) compared with women in monogamous relationships (17%, 95% CI: 15.23--19.90). HSV-2 prevalence did not differ with a history of condom use and only 4.9% of the women reported to have ever used condoms. In age-adjusted analysis, the likelihood of HSV-2 antibodies was associated with a self-reported history of sexually transmitted infections, polygamous relationship, using non-condom contraceptives and more than one lifetime sexual partner. On multivariate analysis, all clinics except Basotu had high odds ratios of HSV-2 infection. Higher odds were particularly seen in Muslur (OR 6.8 (95% CI: 3.76--12.27) and Dangayda clinics (OR 5.3 (95% CI: 2.78 -- 10.16), Table [1](#T1){ref-type="table"} right column. HSV-2 remained significantly associated with polygamous relationships, OR 2.2 (95% CI: 1.62 -- 2.99) or the use of non-condom contraceptives OR 1.7 (95% CI: 1.21--2.41) and weakly associated with more than one lifetime sexual partner (OR 1.8 (95% CI: 1.04--3.04). We found no dual infections with HIV for those with HSV-2 antibodies (data not shown). Age at first pregnancy was not associated with increased risk of HSV-2 antibodies (Table [1](#T1){ref-type="table"}). The overall prevalence of syphilis was 1.6%, but was higher in women aged 20--24 years and 30--34 years (2.4% and 3.1%) (Table [2](#T2){ref-type="table"}). The highest prevalence of syphilis (8.8%) was associated with more than one lifetime sexual partner, followed by self-reported spontaneous abortion 6.1%. Women in polygamous relationships and those with a self-reported history of sexually transmitted infection had a prevalence of 3.7% (95% CI: 1.88--6.87) and 3.1% (95% CI: 1.15--7.48), respectively. None of those who had ever used condoms had antibodies against syphilis. Prevalence of syphilis in women aged less than 18 at first pregnancy was 2.5% (Table [2](#T2){ref-type="table"}) and 1.7% for pregnancies at older age. ###### The prevalence of syphilis and associated factors among pregnant women **Characteristics** **N (%)** **Syphilis prevalence % (95% CI)** **Age-adjusted OR (95% CI) (p-value)** **Multivariate analysis OR (95% CI) (p-value)** ------------------------------------------------------- ------------- ------------------------------------ ---------------------------------------- ------------------------------------------------- **Total** **1296** **1.6 (1.03 -- 2.51)** **[Age groups (years)]{.ul}** 15--19 165 (12.7) 0 \_\_ \_\_ 20--24 417 (32.2) 2.4 (1.22 -- 4.51) 25--29 349 (26.9) 0.9 (0.22 -- 2.69) 30--34 224 (17.3) 3.1 (1.38 -- 6.60) ≥35 141 (10.9) 0.7 (0.03 -- 4.47) **[Marital status]{.ul}** Married 1183 (91.3) 1.8 (1.13 -- 2.75) Single/divorced/separated/widow 113 (8.7) 0 \_\_ \_\_ **[Education (years)]{.ul}** None 349 (28.2) 2.3 (1.07 -- 4.64) 1 1--6 years 174 (13.4) 1.7 (0.45 -- 5.36) 0.8 (0.20 -- 2.85) 7 years or more 773 (59.7) 1.3 (0.66 -- 2.48) 0.6 (0.22 -- 1.43) \_\_ **[Polygamous relationship]{.ul}** No 1024 (79.0) 1.1 (0.57 -- 1.98) 1 **1** Yes 272 (21.0) 3.7 (1.88 -- 6.87) **2.7 (1.08 -- 6.64) (p = 0.03)** **2.6 (1.06 -- 6.51) (p = 0.03)** **[^a^Age at first pregnancy]{.ul}** \< 18 years 158 (14.0) 2.5 (0.81 -- 6.76) \_\_ ≥ 18 years 974 (86.0) 1.7 (1.05 -- 2.84) **[Outcome of previous pregnancy (n = 1155)]{.ul}** Live child 916 (79.3) 1.3 (0.71 -- 2.34) 1 1 Stillbirth 140 (12.1) 2.1 (0.55 -- 6.62) 1.6 (0.45 -- 5.92) 2.4 (0.64 -- 8.95) Spontaneous abortion 99 (8.57) 6.1 (2.49 -- 13.24) **4.9 (1.78 -- 13.25) (p = 0.00)** **4.3 (1.52 -- 12.02) (p = 0.00)** **[Number of lifetime sexual partners]{.ul}** One only 1231 (95.0) 1.2 (0.71 -- 2.05) 1 **1** More than one 65 (5.0) 8.8 (3.64 -- 18.85) **7.8 (2.94 -- 20.86) (p = 0.00)** **5.4 (1.88 -- 15.76) (p = 0.00)** **[Use of contraceptives other than condom]{.ul}** No 1088 (84.0) 1.8 (1.16 -- 2.88) 1 1 Yes 208 (16.0) 0.5 (0.02 -- 3.06) 0.3(0.03 -- 1.93) 0.3 (0.04 -- 2.66) **[History of sexually transmitted infections]{.ul}** No 1135 (87.6) 1.4 (0.84 -- 2.33) 1 1 Yes 161 (12.4) 3.1 (1.15 -- 7.48) 2.2 (0.81 -- 6.20) 1.8 (0.60 -- 5.18) **[Ever used condom]{.ul}** No 1233 (95.1) 1.7 (1.08 -- 2.64) \_\_ \_\_ Yes 63 (4.9) 0 Total syphilis positive (n = 21/1296). Results for analysis of syphilis prevalence by clinic locations are not shown due to very small numbers or none-to give any substantial statistical inference. ^a^Analysed for those aged 20 ≥ years at the time of survey (n = 1132) In multivariate analysis, women who reported previous spontaneous abortion as an adverse pregnancy outcome, were more likely to have a positive syphilis serology (OR 4.3, 95% CI: 1.52--12.02) compared with those reporting stillbirth (OR 2.4, 95% CI: 0.64--8.95). Odds were also higher for women who had more than one lifetime sexual partner, OR 5.4 (95% CI: 1.88--15.76). Dual syphilis and HIV infection occurred in only one individual aged 30 years and this person had had more than one lifetime sexual partner. Eight of the 21 women with syphilis were infected with HSV-2 (OR 2.5, 95% CI: 0.32 -- 19.36), but the figures are too small to interpret the precise association. In addition, being less than 18 years at first pregnancy did not have a significantly increased risk of positive syphilis serology. Discussion ========== The overall prevalence of HSV-2 antibodies among pregnant women in the rural Manyara and Singida regions in Tanzania was 20.7%, varying from 7% to 33% between clinics and increasing with age. This is a slightly lower prevalence than observed by Obasi and colleagues in a different rural region in Tanzania (Mwanza) where 42% of the women had HSV-2 antibodies \[[@B30]\]. The prevalence of syphilis in the present study (1.6%) is much lower than previously observed (17%) in this region, the finding may partly reflect the effect of syndromic management of STI imposed in health services since 1999 \[[@B31]\]. Our findings though suggestive of the relatively low HSV-2 and syphilis prevalence in the area compared to other places in Tanzania, point out a potential opportunity for indirect HIV prevention by controlling prevalent STIs. Measurement of the HSV-2 antibodies was performed with a non-commercial method using a branched peptide with a sequence similar to a specific epitope in glycoprotein G of HSV-2 \[[@B26],[@B28]\]. Commercial assays use the entire glycoprotein G, and may sometimes have problems with false positive results. When comparing performance of peptide 55 method with Western blot, we did not detect any false positive results in Tanzanian sera from persons 13 years old or older \[[@B32]\]. Studies from rural areas in Gambia in West Africa, using the peptide 55 method, have shown infection in 30% of the studied women \[[@B33],[@B34]\]. Other prevalence studies in urban regions of Tanzania have shown that 35% of pregnant women in Dar es Salaam \[[@B35]\] and 39% of women attending primary health care clinics in Moshi were infected with HSV-2 \[[@B4]\]. Among STD patients in urban Mbeya and Dar es Salaam regions of Tanzania, HSV-2 was the most commonly (71%) identified agent in genital ulcer specimens \[[@B12]\]. In these patients, detection of HSV-2 was at significantly higher in HIV infected compared to those not infected \[[@B12]\]. Therefore, taken together, these results indicate a generalised HSV-2 infection in the Tanzanian population. Suggestions on the limited impact of HSV-2 on incident HIV infections in the early stages of the HIV epidemic may also hold true for this area \[[@B7]\]. Despite the generalised, stable HIV epidemic in Tanzania, the epidemiological patterns of STIs have shown regional variations on the level and magnitude, largely influenced by cultural, economical political and social factors. \[[@B14],[@B17]\]. The prevalence of HSV-2 was significantly higher in all clinics compared to Basotu clinic but this village reportedly, had the highest HIV prevalence \[[@B17]\]. We are uncertain to the reasons that might explain this distribution. Basotu presents a very mixed population compared to the other villages, Lake Basotu may attract fishermen from other regions and in-migration of HIV infected persons is possible. Fishing provides a substantial economical gain that is lacking in other villages. Our definition of positive syphilis serology would not only include those with infectious syphilis (primary and secondary), but also cases of untreated latent syphilis though persons in the latent syphilitic state are non-infectious. Reported estimates of the prevalence of syphilis in other rural antenatal sentinel sites in Tanzania in 2004 ranged from 14.9% in Kagera and 2.1% in Kigoma region. The corresponding prevalence of HIV was 4.7% and 5.1% respectively \[[@B15]\]. The later study performed RPR under field conditions, and lacked confirmatory tests for syphilis and likely to have overestimated prevalence of syphilis. Likewise, Kagera region is experiencing a different stage of the HIV epidemic compared to rural Manyara, Singida and Kigoma regions. Kagera was the first region in Tanzania to report AIDS cases and experienced a severe HIV epidemic (24.2%) since 1987 and is low levelling off as a result of excessive mortality and multiple interventions \[[@B36]\]. While rural Manyara, Singida and Kigoma regions had relatively lower HIV prevalence in the last decade but have recently shown an increasing potential. Thus, the interpreting compared prevalence of syphilis with rural Manyara and Singida region is with caution. The proportion of active syphilis in our study might be lower than our prevalence estimates by using TPHA, since antibodies detected by this method remain for life. While, the use of Venereal Disease Research Laboratory (VDRL) test alone in the previous local study in 1999 might have overestimated the prevalence \[[@B18]\]. The low prevalence of HIV in our study (2.0%) may explain why we did not detect an association between infections with HIV and HSV-2. The potential reason for the low prevalence was associated with remoteness and reduced mobility of the people within this rural area. High numbers of lifetime sexual partners have previously shown to increase the risk of acquiring infection with HSV-2 and syphilis \[[@B4],[@B35],[@B37]\]. These risk factors are also presented in this study, as shown in Tables [1](#T1){ref-type="table"} and [2](#T2){ref-type="table"}. Women in polygamous relationships presented with higher prevalence of HSV-2 (Table [2](#T2){ref-type="table"}), in agreement with a report from Gambia \[[@B11]\]. In the present work as well as in others syphilis was shown to be a high probable risk for spontaneous abortion (Table [2](#T2){ref-type="table"}) \[[@B38],[@B39]\]. However, we did not find an association with stillbirth, even though 25% of the pregnancies might result in stillbirth in cases of maternal syphilis \[[@B21]\]. The lack of association in our study could be, partly due to confounding effects of other causes of stillbirth, beyond the scope of this study and the lower number of syphilis cases. Although sexual transmission of syphilis is usually limited to the early stages, vertical transmission is still possible in latency. Conclusion ========== Though we cannot rule out the effect of other STIs on the level HIV infection presented, the identified risk factors for STI are similar to other areas in Tanzania that have experienced a higher level of HIV and related STIs. The low but changing prevalence of HIV infection offers a unique opportunity for strengthening HIV prevention in a cost-effective manner \[[@B40]\]. The identification and control of other prevalent curable STIs other than syphilis and specific intervention for HSV-2 in specific populations like pregnant women should be among approaches towards preventing incident HIV infections. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= All authors contributed substantially to the paper. All authors conceived the study. The first author conducted the study and performed HSV-2 serology, supervised by LH. BE-O and MIM supervised the fieldwork. KIY--M did the statistical analysis and interpretation. All authors helped to conceptualise ideas, interpret the findings and reviewed drafts of the manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2334/8/75/prepub> Acknowledgements ================ We would like to acknowledge the \"Gender Generation and Communication in these times of AIDS\" project (GeGCA-NUFU), Bergen, Norway and the Centre for International Health, University of Bergen, Norway, for funding this study. We thank Dr G. Somi of the Epidemiology unit at the National AIDS control Programme in Dar es Salaam, Tanzania for his ideas on STI surveillance. Further, we thank the following for their contribution in the preparation for laboratory work; Ms Basra Doulla (Dar es Salaam, Tanzania) and Sissel Slettjord (Kristiansand, Norway). In the same capacity, Reidun Haegland, Kjerstin Jakobsen and Mabula Kasubi of the Department of Microbiology and Immunology, Gades Institute. We thank Arvid Nilsen of the Department of Dermatology, Institute of Medicine, and Lindsey Moore of the Department of Biology, University of Bergen for their contribution to the manuscript. We thank Howard Marsden for his important suggestions for the laboratory work. Further, we thank the antenatal clinic authorities of Rural Manyara and Singida regions and Haydom Lutheran Hospital and the Haydom Voluntary AIDS Control Programme (HAVACOP). We would also like to acknowledge the research assistants and lastly, we thank the women who took part in the study.

1. Introduction {#sec1} =============== Resulting from a variety of factors, diabetes is a metabolic disorder mainly characterized by chronic high blood glucose. The number of patients diagnosed with diabetes has been increasing at a rapid rate worldwide. Among them, Chinese patients increase at the highest rate. The overall prevalence of diabetes in the adult population of China grew from 0.67% in 1980 to 11.6% by 2010 \[[@B1]\], and it greatly influences people\'s life. Traditional Chinese Medicine (TCM) diagnosis is based on the information obtained from four diagnostic processes, that is, looking, listening and smelling, asking, and touching. The most common tasks are taking the pulse and inspecting the tongue \[[@B2]\]. Studies demonstrate that tongue images have relatively strong correlation with diabetes in TCM \[[@B3]--[@B5]\]. Traditionally, doctors diagnose diseases and identify patterns by inspecting, describing, and experiences. Thus, the results are easily affected by doctors\' own professional skills and surrounding environment. Without objective assessment criteria, the precision of pattern identification and the repeatability of verification are unclear. Thus, a workable solution is to apply computer techniques and image processing, guided by TCM theory, to make tongue diagnosis standard, objective, and quantitative. Fortunately, computational methods for digital image processing techniques in tongue have been developed, which achieve promising results \[[@B6]--[@B8]\]. Based on the results above, various machine learning methods have been used in tongue manifestation recognition or classification, such as support vector machine (SVM) \[[@B9]--[@B11]\], *k* Nearest Neighbor (*k*-NN) \[[@B12], [@B13]\], Naive Bayes \[[@B11]\], Decision Tree \[[@B11]\], and Neural Network \[[@B9], [@B14]\]. Throughout all mentioned works on the inspection, the popular machine learning algorithms, such as *k*-NN and SVM, are still the first choice in current literature \[[@B15]\]. Though the identification and classification of tongue image have made certain achievements in the past researches, there still existed some issues, firstly, a standard lighting source environment is needed in the tongue image collection, and an effective method is necessary in the tongue image analysis. Moreover, for a successful SVM model, the selection of its kernel parameter and optimization of the data set is of great importance. In this study, tongue images were collected by the TDA-1 digital tongue instrument, which can create a stable light source environment for tongue image acquisition. Research on tongue image mainly focuses on two parts: tongue body and tongue coating. The segmentation of these two parts is an important step in tongue diagnosis since it provides a premise for analyzing the color and texture. This study adopted division-merging method and chrominance-threshold method to distinguish tongue body from tongue coating and then to obtain the parameters of each part \[[@B16]\]. Among commonly used data mining methods, SVM \[[@B17]\] is widely applied due to its excellency in generalization and nonlinear function fitting, and it also presents a lot of advantages in dealing with small sample studies \[[@B18]\]. For a successful SVM model, kernel parameters of SVM are the most important factors affecting the prediction accuracy. Therefore, with its kernel parameters optimized by genetic algorithm (GA), we designed a genetic-based SVM (GA-SVM) model, which was adopted to establish a diagnosis model for diabetes on the basis of tongue images. In our study, techniques for collecting and analyzing information were employed to achieve interdisciplinary research and application in TCM, which will promote the construction of assessment system supported by information acquired from TCM four diagnostic methods in the future for medical treatment in communities and individual health management. 2. Materials and Methods {#sec2} ======================== 2.1. Subjects {#sec2.1} ------------- This study included 827 subjects with informed consent from outpatients and physical examination centers at Shuguang Hospital Affiliated to Shanghai University of Traditional Chinese Medicine (SHUTCM) and TCM Hospital of Baoshan Area in Shanghai, from July 2013 to June 2015. Among them, 296 (159 males and 137 females, average age: 58.35 ± 12.99) were diagnosed with diabetes; 531 (191 males and 340 females, average age: 62.37 ± 8.13) had no diabetes. 2.2. Inclusion and Exclusion Criteria {#sec2.2} ------------------------------------- Inclusion criteria for diabetes group signed with informed consent were from World Health Organization (WHO) in 1999 \[[@B19]\], which include (1) symptoms of high blood sugar and random plasma glucose ≥11.1 mmol/L (200 mg/dL); (2) fasting plasma glucose level ≥7.0 mmol/L (126 mg/dL); (3) plasma glucose ≥ 11.1 mmol/L (200 mg/dL) two hours after a 75 g oral glucose load as in a glucose tolerance test. Exclusion criteria include (1) those diagnosed with other severe diseases such as tumor and diseases of immune and hematological systems; (2) those who cannot make a clear description or cooperate with the imaging collection due to mental disorders; (3) those who refuse to sign informed consent. 2.3. Tongue Image Collection and Analysis Methods {#sec2.3} ------------------------------------------------- ### 2.3.1. Collection Instrument {#sec2.3.1} Developed by the research team handling intelligent processing of TCM diagnosis information in SHUTCM, the TDA-1 digital tongue instrument was applied to collect images. This apparatus ([Figure 1](#fig1){ref-type="fig"}) is made up with an Eolane digital camera, a LED3 light, a removable collection ring, a circuit board, and a hand. The main technical parameters are Charge-coupled device: Eolane A12; light: cold white LED light; color temperature: 6466 K; luminance: 23541 ux \[[@B20]\]. ### 2.3.2. Collection Methods {#sec2.3.2} The TDA-1 tongue instrument was used to collect tongue images in the morning before breakfast. With build-in light on, parameters were set into a manual mode with an aperture of 13, a shutter speed of 1/60 s, microlens, and no flash. Subjects were asked to take a seat, look at the front horizontally, and extend tongue with tongue tip hanging naturally at a 60-degree angle to the horizontal line. The chin was supported by the inferior margin of the collection ring so that the face was closely attached to the ring. With flat surface, 1/2 to 2/3 part of the tongue was protruded. Then, the OK button was pressed. After closing the build-in light, the collected images were checked. If these images did not conform to the requirements mentioned above, collect it again. ### 2.3.3. Analysis Methods {#sec2.3.3} To obtain parameters of tongue images, tongue body and coating need to be separated, and the color and texture of both body and coating need to be analyzed and recognized \[[@B16]\]. The method has been developed into a Tongue Diagnosis Analysis System (TDAS) by the TCM diagnosis Intelligent Information Processing Laboratory of Shanghai University of TCM ([Figure 2](#fig2){ref-type="fig"}). The upper part of the system is the toolbar, including settings, projects, analysis, data, and print; the left side is the module of tongue manifestation analysis, including new data, tongue manifestation segmentation and results. The middle part is tongue body and tongue coating picture. Manual and automatic analyses can be used in color and texture. Because manual analysis can be based on the experience of TCM experts to choose different points and boundaries, we choose the combination of manual and automatic analyses according to experience. The data view window shows the color of the tongue body, the tongue coating, and texture parameters. The bottom right part shows the results of tongue manifestation analysis ([Figure 2](#fig2){ref-type="fig"}, light red tongue; light yellow coating). *Segmentation of Tongue Body and Coating*. The identification of tongue body and coating is an important procedure in tongue diagnosis since it is the premise for analyzing the color and texture features of body and coating. The biggest distinction between tongue body and coating lies in the color: with red as the dominant hue, the color of the tongue body can be presented as light white, light red, crimson red, and purple. For tongue coating, the color can be white, yellow, grey, and black. Due to different color attributes and value ranges, division-merging algorithm based on the color of body and coating was adopted. For images with typical color of body and coating, this method achieves good results. But when tongue coating is thin, the color of tongue body overlaps with that of tongue coating. In this case, this approach fails to separate two colors. Therefore, chrominance-threshold method was adopted too. In this study, the division-merging algorithm and chrominance-threshold method were combined to separate the tongue body and tongue coating. Detailed algorithm was referred to the relevant literature \[[@B16]\]. The tongue images after segmentation were shown in [Figure 3](#fig3){ref-type="fig"}. *Acquisition of Features in Color and Texture*. After separating the area of tongue body and coating, RGB color values of the pixels in tongue body and coating were calculated and then the values of total pixels were averaged. Considering the visualization of color and the feasibility and practicability of classification, we transformed RGB chroma space into LAB and HIS \[[@B21]\]. The texture of tongue contains rough tongue, tender tongue, greasy coating, and rough coating. Among them, tender tongue body and greasy coating are fine and smooth. The changes of texture mainly lie in the variations of gray level. Thus, this study applied gray scale differential algorithm to describe the texture of body and coating. The obtained parameters included contrast (CON), angle second moments (ASM), entropy (ENT), and mean \[[@B22]\]. 2.4. Study Design and Setting {#sec2.4} ----------------------------- The data, mainly including color feature and texture feature, were derived from tongue images of diabetes group and nondiabetes group. These feature parameters were input as independent variables in the preprocessing. Whether the subject has diabetes or not (dependent variable) was considered as the classified variable. 80% specimens were used for training, while 20% for test. The model is illustrated in [Figure 4](#fig4){ref-type="fig"}. ### 2.4.1. Sample Equalization {#sec2.4.1} The sample size of diabetes group and nondiabetes group differs a lot. To avoid the influences of unequal sample size on classification model, Synthetic Minority Oversampling Technique (SMOTE) was adopted to equalize samples. SMOTE, proposed by Chawla et al. in*Artificial Intelligence* in 2002 \[[@B23]\], is a solution based on oversampling. In this study, the equalization was achieved by DMwR package in R Language. ### 2.4.2. Feature Normalization {#sec2.4.2} Due to the differences of feature parameters in magnitude orders, to eliminate the negative effects of these differences, the value ranges were scaled and mapped into the range between −1 and 1. ### 2.4.3. Dimension Reduction of Features {#sec2.4.3} The increase in the number of variables will make the SVM more complicated. In addition, variables may have relevance between each other. Thus, in our study, principal component analysis (PCA), the classic approach to reduce dimensions, was adopted to process the acquired raw features of tongue images so that the information integrity can be maintained as much as possible in the process of dimension reduction. ### 2.4.4. Optimization of Kernel Parameters for SVM by GA {#sec2.4.4} GA, first proposed by American professor Holland in 1962 \[[@B24]\], is a computational model for optimization with parallel search that simulates genetic mechanism and biological evolution in nature. In the study, the penalty parameter *c* and kernel function parameter *g* were optimized by GA. The accuracy of training sample prediction was considered as the fitness function value of GA. The process of algorithm is shown in [Figure 5](#fig5){ref-type="fig"} \[[@B25], [@B26]\]. ### 2.4.5. Development Platform {#sec2.4.5} The study was performed in MATLAB platform by the toolbox of LIBSVM-FarutoUltimate \[[@B27]\] that adds some auxiliary functions on the basis of LIBSVM \[[@B28]\]. 3. Results {#sec3} ========== 3.1. Results of Sample Equalization {#sec3.1} ----------------------------------- In SMOTE, by confirming the frequency of sampling, the samples of both groups were equalized. The result is shown in [Table 1](#tab1){ref-type="table"}. 3.2. Results of Dimension Reduction of Features {#sec3.2} ----------------------------------------------- There were 23 input parameters, which include personal information (gender, age, and BIM) and parameters of tongue color and texture. PCA was applied to reduce the dimensions of raw data on the condition that the 95% information was maintained. The result is shown in [Figure 6](#fig6){ref-type="fig"}. 3.3. Optimized SVM Parameters {#sec3.3} ----------------------------- In the training process of SVM model, the penalty parameter *c* and kernel function parameter *g* were optimized by GA. With population size set as 20, evolutionary generations as 100, and other parameters of LIBSVM toolbox as the default, the precision of sample tests with 10-fold cross-validation was considered as fitness and the accuracy of cross-validation in the training process grew from 72% or so to 83.06%, which is shown in [Figure 7](#fig7){ref-type="fig"}. 3.4. Results of Prediction with GA-SVM Model {#sec3.4} -------------------------------------------- We established three GA-SVM classifiers on different datasets which are raw data, normalized data, and normalized data after PCA, respectively. The result shown in [Table 2](#tab2){ref-type="table"} demonstrates that the classifier on normalize data after PCA yields a better accuracy than other two datasets, which is 1.89% higher than that of raw data at 79.72%. A receiver operating characteristic (ROC) curve is a graphical plot that illustrates the performance of a binary classifiers system. The curve is created by plotting the true positive rate (TPR) against the false positive rate (FPR) at various threshold settings. Sensitivity is also known as TPR, which means the probability that true judgement is made for having diabetes. Specificity is equal to true negative rate, which means the probability that true judgement is made for not having the disease. The area under ROC curve (AUC) is most commonly used as precision index. When the sensitivity and specificity reached 1, the area under ROC curve is obtaining a perfect precision. The best possible prediction method would generate a point on the upper left corner (0, 1) in ROC space, representing 100% sensitivity (no false negatives) and 100% specificity (no false positives) \[[@B29]\]. In this study, we used sensitivity, specificity, ROC, and AUC to assess the performance of classifiers. As shown in Figures [8](#fig8){ref-type="fig"}[](#fig9){ref-type="fig"}--[10](#fig10){ref-type="fig"}, the ROC curves in three figures are for the classifiers using different datasets. Blue curves represent the ROC curve of the nondiabetic class, while other curves represent the ROC curve of the diabetic class. The AUC values for the classifiers using different datasets are 0.8773, 0.9065, and 0.9037, which indicates that the classifier is effective in distinguishing the two classes of objects. We can know from [Table 3](#tab3){ref-type="table"} that the performance of third classifier (normalized data with PCA) has higher sensitivity and specificity than that of the first two (raw data and normalized data). Although the AUC of the third classifier is lower than the second by 0.3%, it substantially saves the average training time as shown in [Table 2](#tab2){ref-type="table"}. 3.5. Comparison with Other Algorithms {#sec3.5} ------------------------------------- In order to evaluate efficacy for established GA-SVM model, three distinct prediction models, *k*-NN, Naive Bayes, and Backpropagation Neural Network (BP-NN), were employed to compare with GA-SVM model. The three classification model using *k*-NN, Naive Bayes, and BP-NN methods are established in MATLAB. As shown in [Table 4](#tab4){ref-type="table"}, accuracy, specificity, and AUC of GA-SVM models are higher than other algorithms, except for sensitivity a little lower than *k*-NN. Combined with the results above, it can be concluded that our algorithm has a better established classification model for tongue manifestation. 4. Discussion {#sec4} ============= Over the past 3000 years, tongue diagnosis has been proved to be one of the most valuable and the most extensively applied TCM diagnostic approaches in clinical practice. The color, moisture, size, shape, and texture of tongue reveal the overall health condition and dysfunctions of specific organs. Tongue color and coating have long been key parameters in differentiating diseases. In this research, we took photos of tongues with the TDA-1 tongue instrument, which segmented the tongue body and coating. After obtaining color parameters of RGB, HIS, and LAB and texture parameters of CON, ASM, ENT, and MEAN, with these data, a diabetes diagnosis model was established on the basis of SVM. Sample imbalance is a problem that must be faced in many computational problems in medical research. For unbalanced samples between two groups, SMOTE was adopted to equalize the data. Due to the differences of feature parameters in magnitude orders, to eliminate the negative effects of these differences, the value ranges of input variables were scaled and mapped into the range between −1 and 1. After processing, the accuracy of test samples was slightly increased from 77.83% to 78.77%, and the AUC was increased from 0.8773 to 0.9065. In addition, in the training process of SVM model, the penalty parameter *c* and kernel function parameter *g* in SVM model were optimized by GA. As shown in [Figure 7](#fig7){ref-type="fig"}, the accuracy of cross-validation in training process grew from 72% or so to 83.06%, indicating the significance of parameter optimization in improving the precision of classification. To handle multiple input variables, PCA was adopted to decrease the number of variables from 29 to 8 on the condition that 95% information was retained. The results show the accuracy of classification was not reduced and it substantially saved the training time. With standardized tongue image parameters, we developed a novel model to search the optimal values of SVM parameters, to increase the accuracy of prediction. In order to evaluate efficacy for the established GA-SVM model, *k*-NN, Naive Bayes, and BP-NN model were applied to our datasets. From [Table 4](#tab4){ref-type="table"}, the results show that the GA-SVM model performs the best, implying that the hybrid system has a high potential to dramatically increase the predictive accuracy when integrating GA with traditional SVM model. As shown in [Figure 7](#fig7){ref-type="fig"}, the accuracy of cross-validation in training process has greatly increased after SVM parameters optimized by GA. The diabetes is of varied TCM syndromes for its clinical manifestations, such as the deficiency of qi and yin syndrome, yin-deficiency and fire-hyperactivity syndrome, and different syndromes manifest with different tongue image features. However it is difficult for the traditional statistical methods to identify the diabetes through its tongue image automatically, so it is essential to find a method which is suitable for the diagnosis of diabetes via its tongue images. By comparison with other algorithms and internal validation of the model, it is indicated that the SVM classification model we established had a fair effect in this study. 5. Conclusion {#sec5} ============= In this paper, with tongue images, SVM was used to establish the classification model for diabetes, which achieves good classification results. It indicates the feasibility of using the information science method to carry out TCM diagnosis. Data preprocessing and parameter optimization directly impact the results. Feature dimension reduction is a double-edged sword. On the one hand, it can accelerate the training speed and avoid overfitting; on the other hand, it may cause the loss of useful information. GA can find the optimal option without going through the whole search space and can also be used for feature selection in other studies. With information collecting and analyzing techniques, this interdisciplinary study researches on the informatization of TCM and its application and provides a reference for designing more effective data analysis and processing algorithms. In future researches, other pieces of TCM diagnostic information can be integrated to improve the precision of classification. This work is partially supported by the National Natural Science Foundation of China (no. 81373556), National Science and Technology Support Program during the Twelfth Five-Year Plan Period (no. 2012BAI37B06), "Shu Guang" Project Supported by Shanghai Municipal Education Commission and Shanghai Education Development Foundation (no. 12SG36), and Basic Subject Innovation Fund Supported by Shanghai Municipal Education Commission (no. A2-P1600317). Competing Interests =================== The authors declare that there are no competing interests regarding the publication of this paper. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} {#fig6} {#fig7} {#fig8} {#fig9} {#fig10} ###### Samples before and after equalization. Samples Diabetes group Nondiabetes group ----------- ---------------- ------------------- Original 296 531 Equalized 531 531 ###### SVM classification before and after data processing. Datasets Accuracy of cross-validation (%) Accuracy of training samples (%) Accuracy of test samples (%) Running time (s) -------------------------- ---------------------------------- ---------------------------------- ------------------------------ ------------------ Raw data 81.65 100.00 77.83 817.86 Normalized data 84.35 99.53 78.77 747.40 Normalized data with PCA 83.06 99.88 79.72 465.52 ###### Specificity, sensitivity, and AUC of SVM classification before and after data processing. Datasets Specificity (%) Sensitivity (%) AUC -------------------------- ----------------- ----------------- -------- Raw data 81.05 75.21 0.8773 Normalized data 82.80 75.63 0.9065 Normalized data with PCA 83.16 76.92 0.9037 ###### Result compared with other algorithms. Algorithms Accuracy (%) Specificity (%) Sensitivity (%) AUC ------------- -------------- ----------------- ----------------- -------- *k*-NN 78.77 80.18 77.36 0.8471 Naive Bayes 75.94 78.30 73.58 0.8248 BP-NN 75.00 73.58 76.42 0.8285 GA-SVM 79.72 83.16 76.92 0.9037 [^1]: Academic Editor: Zexuan Ji

(J Am Heart Assoc. 2018;7:e007529 DOI: 10.1161/JAHA.117.007529.)29899018 Clinical PerspectiveWhat Is New?A J‐shaped relationship is observed between serum sodium levels at time of diagnosis and long‐term clinical outcomes in heart failure with preserved ejection fraction.What Are the Clinical Implications?Serum sodium could be an important clinical variable for identifying heart failure with preserved ejection fraction patients at increased risk of adverse events.Neurohormonal activation, a major determinant of serum sodium levels in heart failure, may play a role in the natural history of heart failure with preserved ejection fraction. {#jah33267-sec-0008} Epidemiologic studies estimate that the prevalence of heart failure with preserved ejection fraction (HFpEF), previously termed diastolic heart failure (HF), is 1.1% to 5.5% of the general population,[1](#jah33267-bib-0001){ref-type="ref"}, [2](#jah33267-bib-0002){ref-type="ref"} and ranges from 40% to 71% among HF patients.[1](#jah33267-bib-0001){ref-type="ref"}, [3](#jah33267-bib-0003){ref-type="ref"}, [4](#jah33267-bib-0004){ref-type="ref"} Over the past 3 decades, while the prevalence of HF with reduced ejection fraction (EF) or systolic HF has remained stable, HFpEF has increased commensurately with the aging of the population at a rate of 1% per year.[3](#jah33267-bib-0003){ref-type="ref"} Hospitalizations for HFpEF are increasing relative to HF with reduced ejection fraction,[1](#jah33267-bib-0001){ref-type="ref"} and the mortality rates from HFpEF have ranged from 10% to 30% per year in epidemiologic studies.[5](#jah33267-bib-0005){ref-type="ref"} The cost associated with this condition is high and poses a significant burden on the healthcare system.[6](#jah33267-bib-0006){ref-type="ref"} A major gap in treatment of HFpEF is that no evidence‐based therapy has yet been identified for HFpEF in spite of multiple clinical trials testing the efficacy of angiotensin‐converting enzyme inhibitors,[7](#jah33267-bib-0007){ref-type="ref"} angiotensin receptor blockers,[8](#jah33267-bib-0008){ref-type="ref"} β‐blockers,[9](#jah33267-bib-0009){ref-type="ref"} digoxin,[10](#jah33267-bib-0010){ref-type="ref"} and spironolactone.[11](#jah33267-bib-0011){ref-type="ref"} The major limiting factors in previously reported clinical trials and prospective observational studies include lack of uniformity in accurately defining the HFpEF cohort, and failure in distinguishing HFpEF from HF with intermediate EF and HF with recovered EF, as advocated in the clinical guidelines of the European Society of Cardiology and the American College of Cardiology Foundation/American Heart Association.[4](#jah33267-bib-0004){ref-type="ref"} HFpEF is extremely difficult to curate in a large national database given its lack of consistency in identifying and reporting by healthcare providers and the difficulty of extracting all known left ventricular ejection fraction (LVEF) values to ensure accurate diagnosis. Low serum sodium, defined as serum sodium \<135 mmol/L (hyponatremia), is a powerful prognostic marker in HF with reduced EF and indicates neurohormonal activation.[12](#jah33267-bib-0012){ref-type="ref"} However, studies on the relation between hyponatremia and long‐term clinical outcomes in HFpEF patients have yielded inconsistent results.[13](#jah33267-bib-0013){ref-type="ref"}, [14](#jah33267-bib-0014){ref-type="ref"}, [15](#jah33267-bib-0015){ref-type="ref"}, [16](#jah33267-bib-0016){ref-type="ref"} To our knowledge, previous studies have not addressed whether values of serum sodium \>135 mmol/L are associated with adverse clinical outcomes in HFpEF patients in a large cohort. Characterizing the relation between the whole spectrum of sodium values and outcomes in HFpEF would be important to understand the role of sodium homeostasis on the prognosis of patients with HFpEF. Accordingly, we curated a large cohort of HFpEF subjects from the national Veterans Affairs (VA) patient database following published European Society of Cardiology/American College of Cardiology Foundation/American Heart Association guidelines and tested the hypothesis that there is a J‐shaped relationship between serum sodium measured at the time of diagnosis of HFpEF and all‐cause mortality, heart failure hospitalizations, and all‐cause hospitalizations in subjects with HFpEF. Methods {#jah33267-sec-0009} ======= In accordance with current VA policies, we are not able to make our data publicly available, but we can provide the methods used in analysis to any researcher for purposes of replicating the procedure upon request from the senior author. All study procedures were approved by the Institutional Review Board of the Veterans Affairs Boston Healthcare System. Because this was a database study, the informed consent requirement was waived. Development of the HFpEF Algorithm {#jah33267-sec-0010} ---------------------------------- Using an iterative process, we arrived at the following algorithm to curate HFpEF cases from the national VA patient care databases. The inclusion criteria in the algorithm consisted of (1) *International Classification of Diseases, Ninth Revision* (*ICD‐9*) codes diagnosis of HF (any 428.xx); (2) presence of B‐type natriuretic peptide (≥100 pg/mL) or aminoterminal pro--B‐type natriuretic peptide (≥300 pg/mL), or diuretic usage within 1 month of HF diagnosis; (3) availability of echocardiogram in our system; (4) all recorded LVEF values ≥50% (during the entire study follow‐up); and (5) at least 1 serum sodium measurement within 30 days after the HF diagnosis. As will be described below, criterion 2 was arrived at by an iterative process to ensure selection of cases with definite clinical heart failure. We included subjects diagnosed with HF in either an inpatient or outpatient setting between January 1, 2002, and December 31, 2012. We defined the index date as this initial HF diagnosis date around which patients also displayed elevated B‐type natriuretic peptide/pro--B‐type natriuretic peptide laboratory values or diuretic usage. For outpatient diagnoses, the diagnosis date was used as the index date. If an outpatient HF diagnosis was followed by hospitalization within 48 hours or if the diagnosis was made as an inpatient, the discharge date was used as the index date. As validated before,[17](#jah33267-bib-0017){ref-type="ref"}, [18](#jah33267-bib-0018){ref-type="ref"}, [19](#jah33267-bib-0019){ref-type="ref"} a natural language processing tool was used to extract EF values from text integration utilities documents, including history and physical examination notes, progress notes, discharge summary notes, echocardiography reports, nuclear medicine reports, cardiac catheterization reports, and other cardiology notes. All recorded LVEF values were used for classifying cases as HFpEF. Patients with any diagnosis code for constrictive pericarditis (423.2), hypertrophic cardiomyopathy (425.1), restrictive cardiomyopathy (425.4), or amyloidosis (277.30, 277.39, 425.7) were excluded. To be included in the analysis cohort, patients were required to have a loop diuretic or thiazide diuretic prescription within 1 year before or after the index date. Validation of HFpEF Cohort {#jah33267-sec-0011} -------------------------- Three independent reviewers (Y.R.P., T.F.I., A.R.O., all physicians) completed chart review of 100 algorithm‐identified cases and 100 controls in a blinded fashion, with each chart reviewed by 2 reviewers. Validation was based on ascertaining the presence of definite HF (at least 1 symptom, at least 2 signs, and definite treatment for HF), and confirming that all recorded LVEF values were ≥50%.[20](#jah33267-bib-0020){ref-type="ref"} Chart review validation of the HFpEF algorithm demonstrated a positive predictive value of 96% and a negative predictive value of 87%. Exposure {#jah33267-sec-0012} -------- The primary exposure was serum sodium concentration measured within 30 days after the index date. For subjects who were hospitalized (inpatient diagnosis), we used the last serum sodium measurement in the series as the baseline value under the assumption that the most stable serum sodium value would be on the day of hospital discharge. By doing this, we reduced variability of serum sodium values that could occur with diuretic use and clinical status of subjects during their hospitalization. Outpatient diagnoses followed by hospitalization within 48 hours were counted as inpatient diagnoses, and sodium values were taken at discharge or later. We excluded subjects with any history of HF before the index date in order to ensure that we were analyzing sodium values obtained at the time of the initial diagnosis. Outcomes {#jah33267-sec-0013} -------- Our primary outcome was all‐cause mortality. Our secondary outcomes were the number of days hospitalized for HF and for all‐cause hospitalizations per year. HF hospitalizations were defined as having an associated primary diagnosis for heart failure (*ICD‐9* code of 428.xx). The outcomes were captured from the VA health system as well as from Centers for Medicare and Medicaid Services data to include care outside the VA system. We excluded people who stayed in the hospital for more than 180 days so as to accurately capture the hospitalization outcomes. Covariates {#jah33267-sec-0014} ---------- We obtained baseline anthropometric and laboratory variables falling on or within 30 days before the baseline sodium date, closest to the baseline sodium date. This was primarily done to extract information from when subjects were most stable after their acute HF event. Demographic and anthropometric variables including age, body mass index, heart rate, and race were obtained from electronic medical health records. Estimated glomerular filtration rate was calculated using the chronic kidney disease epidemiology collaboration formula.[21](#jah33267-bib-0021){ref-type="ref"} Two independent physicians adjudicated laboratory values of serum creatinine (used for estimated glomerular filtration rate calculation) and serum potassium, and any discrepancy was resolved by further discussion. We obtained medical comorbidities falling anytime before baseline sodium measurement using *ICD‐9* codes from inpatient and outpatient medical records. These include history of type 2 diabetes mellitus (250.x, 362.0, 357.2, 366.41, 249, 790.2--791.6, V45.85, V53.91, V65.46), coronary artery disease (including myocardial infarction and history of percutaneous coronary intervention or coronary artery bypass graft; 410, 411, 414.0x, 429.2x, 429.5x, 440.x, 429.7x, V45.82), hypertension (401--405.99, 437.2), hyperlipidemia (272.x), atrial fibrillation (427.31), chronic obstructive pulmonary disease (490--496, 510, 781.5), and anemia (280--285). Beta‐blockers, angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, and calcium channel blockers that were dispensed within 6 months before the baseline sodium date were identified from pharmacy records in the VA system and adjudicated by 2 independent physicians. Medications that were prescribed but not dispensed by the pharmacy and clinical trial medications were excluded from our analyses. Statistical Analysis {#jah33267-sec-0015} -------------------- We analyzed serum sodium both as a continuous and then as a categorical variable. To examine the nonlinear relationship between baseline sodium as a continuous variable and mortality and hospitalization outcomes, we developed macros to fit cubic splines with 6 equally distributed knots to crude and multivariable proportional hazards regression and generalized estimating equation--based negative binomial regression models with a sodium value of 139.00 mmol/dL as the reference. Because of instability and low frequency among extreme values, all sodium values \<125.00 mmol/dL and \>150.00 mmol/dL were winsorized. Five baseline serum sodium categories were determined based on the shape and inflection points of the cubic spline plots, the distribution and quantiles of serum sodium values, and generally established clinical ranges. These categories were used for all proceeding regression analyses. We also compared these results to an a priori decile‐based analysis. We built our models based on a priori knowledge of risk factors and examination of confounders, starting with crude models, then age‐adjusted, parsimonious (age, sex, race), and finally multivariable adjusted. Given that the results of the parsimonious model were similar to the age‐adjusted and multivariable models, we will not present those results. No automated variable selection process was used (eg, stepwise). Subjects entered the analysis at the time of their first sodium measurement, using a counting process approach[22](#jah33267-bib-0022){ref-type="ref"}, [23](#jah33267-bib-0023){ref-type="ref"} and person‐time accrued from baseline sodium measurement until the occurrence of death, last VA visit, or the end of follow‐up (December 31, 2014). As expected, the association between sodium and mortality did not meet the proportional hazards assumption, leading us to do a piecewise Cox proportional hazard analysis, with the first time interval covering the first 30 days and the second covering the post--30‐day period. Subjects entered the model at different times within the initial 30‐day period, and we expected them to be less clinically stable during that period, thus having different hazard rates than at later time points. Further assessment of nonproportionality of hazards showed no meaningful deviations. Hazard ratios (HRs) and 95% confidence intervals (CIs) were estimated for each period for serum sodium categories, relative to the referent group. We also examined whether the relationship between baseline serum sodium and mortality varied by age. We stratified the analysis by age, categorized as \<65, 65 to 75, and \>75 years. Generalized estimating equation--based negative binomial regression analyses were used to estimate the incidence density ratios for number of days of HF and all‐cause hospitalizations per year for sodium categories relative to the referent group. Incidence density was defined as the total number of days hospitalized per year divided by the total number of days observed per year. The number of days observed per year (on the log scale) was included in these models as an offset. Crude Poisson hospitalization rates were also estimated. Patients who had any hospital stay \>180 days were excluded from these analyses, as they were likely part of long‐term care. The multivariable models were adjusted for age; sex; race; body mass index; heart rate; estimated glomerular filtration rate; serum potassium; average LVEF; history of coronary artery disease, hypertension, hyperlipidemia, atrial fibrillation, chronic obstructive pulmonary disease, diabetes mellitus, and anemia; and history of medications including calcium channel blockers, β‐blockers, and angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers. Diuretics were not adjusted for in the multivariate analysis, as everyone in the analysis cohort was required to have a loop diuretic or thiazide diuretic prescription within 1 year before or after the diagnosis of HF. Medications including aldosterone antagonists and statins were not included in multivariate analysis, as they were not only not associated with baseline serum sodium measurements but also not clearly associated with clinical outcomes in HFpEF (*P*\>0.25). All analyses were completed using SAS Enterprise Guide 7.1 (SAS Institute Inc, Cary, NC). All *P* values were 2‐tailed, and significance level was set at an alpha of 0.05. Results {#jah33267-sec-0016} ======= For the whole cohort, mean age±SD was 70.8±11.5 years. During a median follow‐up of 3.6 (interquartile range, 1.7--6.4) years there were 4983 deaths. There were 107 397.5 person‐years of follow‐up to estimate incidence density ratio for number of days of HF and all‐cause hospitalizations per year. Table [1](#jah33267-tbl-0001){ref-type="table"} demonstrates baseline characteristics across serum sodium categories. Lower sodium categories tended to have younger and thinner subjects, fewer black subjects, higher average heart rate, and estimated glomerular filtration rate values, and among comorbidities a higher incidence of atrial fibrillation, anemia, diabetes mellitus, coronary artery disease, and chronic obstructive pulmonary disease and a lower incidence of hypertension. Mean LVEF was slightly but significantly higher in the lower sodium categories. Prescriptions for calcium channel blockers were less frequent in lower sodium categories. ###### Baseline Characteristics of Patients With Heart Failure With Preserved Ejection Fraction (n=27 975) According to Serum Sodium Categories Characteristics Serum Sodium Measurements (mmol/L) *P* Value[a](#jah33267-note-0003){ref-type="fn"} ------------------------------------------ ------------------------------------ -------------------------------------------------- ----------- ----------- ----------- ---------- Demographics Sex (% male) 96.3 96.1 96.2 96.0 96.6 0.7922 Black, % 10.3 12.7 13.7 15.0 17.8 \<0.0001 Age, y 70.0±11.6 69.9±11.6 70.6±11.5 71.7±11.2 72.5±11.0 \<0.0001 BMI (kg/m^2^) \<18.5, % 11.7 11.2 11.5 11.2 12.3 \<0.0001 18.5--24.9, % 22.3 15.2 12.2 12.1 14.2 25--29.9, % 25.1 23.5 23.8 22.5 23.3 30--34.9, % 18.3 21.4 21.7 22.7 21.8 ≥35.0, % 22.6 28.7 30.9 31.6 28.5 Clinical Heart rate, bpm 79.7±15.6 77.7±15.0 75.9±14.8 75.0±14.9 75.8±15.2 \<0.0001 Average LVEF, % 60.6±6.1 60.3±5.8 60.2±5.8 60.1±5.7 60.1±5.9 0.0013 Comorbidities Coronary artery disease, % 42.0 40.5 41.8 41.7 40.7 0.4708 Hypertension, % 86.3 87.2 88.1 89.3 90.1 \<0.0001 Hyperlipidemia, % 59.2 65.1 67.1 67.2 63.6 \<0.0001 Atrial fibrillation, % 24.1 21.0 22.1 21.7 21.4 0.0286 Chronic obstructive pulmonary disease, % 50.6 47.0 44.2 42.3 41.8 \<0.0001 Diabetes mellitus type 2, % 54.0 55.5 53.0 52.2 52.6 0.0065 Anemia, % 42.4 35.8 32.0 31.6 35.2 \<0.0001 Laboratory eGFR (mL/min per 1.73 m^2^) 67.1±26.8 64.7±25.1 63.9±23.9 61.5±23.6 57.2±23.8 \<0.0001 Serum potassium (mmol/L) 4.27±0.59 4.23±0.53 4.20±0.51 4.20±0.51 4.17±0.54 \<0.0001 Medications Calcium channel blocker, % 40.3 38.5 40.4 41.9 42.5 0.0027 ACE/ARB, % 60.9 60.9 61.1 62.2 61.3 0.5696 β‐Blocker, % 60.7 59.7 59.6 59.7 61.0 0.7754 Missing values: race=1378; heart rate=780; eGFR=440; serum postassium=36. ACE‐I/ARB indicates angiotensin‐converting enzyme inhibitor/angiotensin receptor blocker; BMI, body mass index; eGFR, estimated glomerular filtration rate; LVEF, left ventricular ejection fraction. *P* value for general association (ANOVA or chi‐square). The 147 patients who died or were censored on the same day as their baseline sodium measurement were excluded from all analyses. An additional 2662 patients were excluded from the mortality multivariable analyses because of missing covariates. Examining Sodium as a Continuous Measure {#jah33267-sec-0017} ---------------------------------------- As shown in Figure [1](#jah33267-fig-0001){ref-type="fig"}, the association between serum sodium and all‐cause mortality follows a J‐shaped hazard curve both during the first 30 days (test for nonlinear trend, *P*\<0.0001) and after 30 days of follow‐up (*P*\<0.0001). Considering the obvious J shape of these cubic spline plots, the distribution of sodium values, and established sodium clinical ranges, baseline serum sodium was categorized as 115.00 to 134.99 (low serum sodium), 135.00 to 137.99 (low‐normal serum sodium), 138.00 to 140.99 (referent group), 141.00 to 143.99 (high‐normal serum sodium), and 144.00 to 160.00 (high serum sodium) mmol/dL. Sodium values between 138.00 and 140.99 mmol/L were selected as the referent group for all categorical analyses because they were considered to be a clinically normal and the center of the distribution fell within this range. A similar pattern was observed for the all‐cause hospitalization and HF hospitalization outcomes, leading us to use the same sodium categories (Figure [2](#jah33267-fig-0002){ref-type="fig"}). We also looked at the association of serum sodium deciles and mortality and hospitalization outcomes, obtaining similar results. {#jah33267-fig-0001} {#jah33267-fig-0002} All‐Cause Mortality Within 30 Days of Index Diagnosis {#jah33267-sec-0018} ----------------------------------------------------- Compared with the referent group, low serum sodium and low‐normal serum sodium were associated with a 2.5‐fold (HR, 2.46; 95% CI, 1.99--3.05) and 56% (HR, 1.56; 95% CI, 1.26--1.93) higher risk of all‐cause mortality, respectively, during the first 30 days after the index diagnosis, in a multivariable model controlling for age; sex; race; body mass index; heart rate; estimated glomerular filtration rate; serum potassium; LVEF; history of coronary artery disease, hypertension, hyperlipidemia, atrial fibrillation, chronic obstructive pulmonary disease, diabetes mellitus, or anemia; and history of medications including calcium channel blockers, β‐blockers, and angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers (not shown in tables). Similarly, in this multivariable model, high serum sodium was associated with an almost 3‐fold (HR, 2.81; 95% CI, 2.19--3.59) higher risk of all‐cause mortality during the first 30 days of follow‐up compared with the referent group. The 30‐day predicted mortality rate for a patient with a baseline sodium value of 139.00 mmol/dL is 2.4%. All‐Cause Mortality After 30 Days of Index Diagnosis {#jah33267-sec-0019} ---------------------------------------------------- Multivariable analyses also demonstrated that compared with the referent group, low, low‐normal, and high serum sodium were associated with 36% (HR, 1.36; 95% CI, 1.28--1.44), 6% (HR, 1.06; 95% CI, 1.01--1.12), and 9% (HR, 1.09; 95% CI, 1.01--1.17) higher risk of all‐cause mortality, respectively (Table [2](#jah33267-tbl-0002){ref-type="table"}), after 30 days following the index date. Median survival time (95% CI) was 3.13 (2.86--3.42) years, 4.90 (4.70--5.10) years, 5.68 (5.49--5.88) years, 5.72 (5.49--5.89) years, and 4.80 (4.34--5.25) years for the baseline serum sodium categories 115.00 to 134.99, 135.00 to 137.99, 138.00 to 140.99, 141.00 to 143.99, and 144.00 to 160.00 mmol/dL, respectively. The 5‐year predicted mortality rate for a patient who survived the first 30 days after initial heart failure diagnosis and had a baseline sodium value of 139.00 mmol/dL is 44.4%. ###### Hazard Ratios for All‐Cause Mortality (Beginning 30 days After Initial Heart Failure Diagnosis) by Baseline Serum Sodium Category in 24 517 Subjects With Heart Failure With Preserved Ejection Fraction Who Survived Past Day 30 Serum Sodium (mmol/L) Number of Deaths/Total Number of Subjects in Group Person‐Time (y) Crude HR (95% CI) Age‐Adjusted HR (95% CI) Multivariable HR (95% CI) ----------------------- ---------------------------------------------------- ----------------- ------------------- -------------------------- --------------------------- 115.00--134.99 1797/2717 9655.95 1.55 (1.47--1.63) 1.59 (1.51--1.68) 1.36 (1.28--1.44) 135.00--137.99 2906/5046 21 831.45 1.12 (1.07--1.17) 1.14 (1.09--1.20) 1.06 (1.01--1.12) 138.00--140.99 4936/9053 41 477.88 REF REF REF 141.00--143.99 3362/6041 28 482.09 0.99 (0.95--1.04) 0.95 (0.91--0.99) 0.99 (0.95--1.04) 144.00--157.00 1009/1660 7535.83 1.13 (1.05--1.21) 1.07 (1.00--1.14) 1.09 (1.01--1.17) Multivariable model was adjusted for age; sex; race; body mass index; heart rate; estimated glomerular filtration rate; serum potassium; average LVEF; use of calcium channel blockers, angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, and β‐blockers; and history of coronary artery disease, hypertension, hyperlipidemia, atrial fibrillation, chronic obstructive pulmonary disease, diabetes mellitus, and anemia. CI indicates confidence interval; HR, hazard ratio; LVEF, left ventricular ejection fraction; REF, referent group. Number of Days of HF Hospitalizations per Year {#jah33267-sec-0020} ---------------------------------------------- Low serum sodium and low‐normal serum sodium were associated with 48% (incidence density ratio \[IDR\], 1.48; 95% CI, 1.10--2.00) and 38% (IDR, 1.38; 95% CI, 1.08--1.77) increases in the average number of days hospitalized for HF per year, respectively, compared with the referent group (Table [3](#jah33267-tbl-0003){ref-type="table"}) in the multivariable model. High serum sodium was not significantly associated with an increased risk of HF hospitalizations. ###### Incidence Density Ratios for Number of Days Hospitalized for Heart Failure per Year by Baseline Serum Sodium Category in 25 095[a](#jah33267-note-0006){ref-type="fn"} Subjects With Heart Failure With Preserved Ejection Fraction Serum Sodium (mmol/L) Average Number of Days Hospitalized for HF Per 100 Person‐Years (95% CI) Crude IDR (95% CI) Age‐Adjusted IDR (95% CI) Multivariable IDR (95% CI) ----------------------- -------------------------------------------------------------------------- -------------------- --------------------------- ---------------------------- 115.00--134.99 35.67 (34.49--36.89) 1.29 (1.00--1.66) 1.44 (1.08--1.92) 1.48 (1.10--2.00) 135.00--137.99 32.85 (32.09--33.62) 1.32 (1.08--1.63) 1.35 (1.09--1.66) 1.38 (1.08--1.77) 138.00--140.99 26.12 (25.63--26.62) REF REF REF 141.00--143.99 25.49 (24.90--26.08) 0.93 (0.77--1.12) 0.86 (0.72--1.03) 0.84 (0.70--1.01) 144.00--160.00 28.66 (27.47--29.90) 1.24 (0.94--1.63) 1.19 (0.91--1.54) 1.24 (0.94--1.64) Multivariable model was adjusted for age; sex; race; body mass index; heart rate; estimated glomerular filtration rate; serum potassium; average LVEF; use of calcium channel blockers, angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers, and β‐blockers; and history of coronary artery disease, hypertension, hyperlipidemia, atrial fibrillation, chronic obstructive pulmonary disease, diabetes mellitus, and anemia. CI indicates confidence interval; HF, heart failure; IDR, incidence density ratio; LVEF, left ventricular ejection fraction; REF, referent group. Excludes people who had a hospital stay \>180 days. Number of Days of All‐Cause Hospitalizations per Year {#jah33267-sec-0021} ----------------------------------------------------- Low serum sodium and low‐normal serum sodium were associated with 44% (IDR, 1.43; 95% CI, 1.32--1.56) and 18% (IDR, 1.18; 95% CI, 1.10--1.27) increases in the average number of days of all‐cause hospitalizations per year, respectively (Table [4](#jah33267-tbl-0004){ref-type="table"}), in the multivariable model. In this analysis, higher serum sodium was associated with 19% (IDR, 1.19; 95% CI, 1.06--1.33) higher risk of number of days for all‐cause hospitalizations in the multivariable mode. ###### Incidence Density Ratios for Number of Days Hospitalized for Any Cause per Year by Baseline Serum Sodium Category in 25 095[a](#jah33267-note-0008){ref-type="fn"} Subjects With Heart Failure With Preserved Ejection Fraction Serum Sodium (mmol/L) Average Number of Days Hospitalized for Any Cause per Person‐Year (95% CI) Crude IDR (95% CI) Age‐Adjusted IDR (95% CI) Multivariable IDR (95% CI) ----------------------- ---------------------------------------------------------------------------- -------------------- --------------------------- ---------------------------- 115.0--134.9 7.59 (6.98--8.26) 1.55 (1.44--1.67) 1.61 (1.50--1.74) 1.44 (1.32--1.56) 135.0--137.9 5.95 (5.59--6.34) 1.20 (1.13--1.28) 1.22 (1.14--1.31) 1.18 (1.10--1.27) 138.0--140.9 5.06 (4.82--5.32) REF REF REF 141.0--143.9 5.02 (4.73--5.33) 1.03 (0.96--1.09) 1.00 (0.94--1.07) 1.03 (0.96--1.10) 144.0--160.0 5.66 (5.08--6.32) 1.19 (1.08--1.31) 1.16 (1.05--1.28) 1.19 (1.06--1.33) Multivariable model was adjusted for age; sex; race; body mass index; heart rate; estimated glomerular filtration rate; serum potassium; average LVEF; use of calcium channel blockers, angiotensin‐converting enzyme inhibitors/angiotensin receptor blockers; and β‐blockers; and history of coronary artery disease, hypertension, hyperlipidemia, atrial fibrillation, chronic obstructive pulmonary disease, diabetes mellitus, and anemia. CI indicates confidence interval; HF, heart failure; IDR, incidence density ratio; LVEF, left ventricular ejection fraction; REF, referent group. Excludes people who had a hospital stay \>180 days. Age‐Stratified Analysis {#jah33267-sec-0022} ----------------------- There were 8750 patients \<65 years old, 6183 patients 65 to 75, and 10 507 people \>75 years old. Among patients \<65 years of age, compared with the referent group, low‐normal and low serum sodium were associated with 13% (HR, 1.13; 95% CI, 1.03--1.24) and 50% (HR, 1.50; 95% CI, 1.35--1.67) (not shown in tables) higher risk of all‐cause mortality, respectively, in the multivariable model. Similarly, low serum sodium was associated with 39% (HR, 1.39; 95% CI, 1.23--1.56) and 23% (HR, 1.23; 95% CI, 1.13--1.34) higher risk for all‐cause mortality among patients between 65 and 75 years and \>75 years of age, respectively (not shown in tables). Discussion {#jah33267-sec-0023} ========== In a well‐curated national cohort of HFpEF subjects, we found a J‐shaped relationship between serum sodium and all‐cause mortality and number of days of all‐cause hospitalizations per year. We also found that low sodium levels were associated with an increased number of days of hospitalizations for HF per year. In addition to low sodium values, values in the low‐normal range were also significantly associated with worse outcomes. Our study, the largest to our knowledge that has examined the relation of serum sodium to outcomes in HFpEF, suggests the utility of serum sodium to identify HFpEF patients who are at increased risk of adverse outcomes. In addition, our study also suggests that further investigations into the mechanisms of the relation of serum sodium to the natural history of HFpEF are warranted, especially whether there is a causative role for neurohormonal activation in the association between low serum sodium and adverse outcomes in HFpEF. In our study, we demonstrated that low serum sodium is associated with worse immediate (30‐day follow up) and long‐term adverse outcomes in HFpEF. Similar associations were reported by previous smaller studies.[13](#jah33267-bib-0013){ref-type="ref"}, [15](#jah33267-bib-0015){ref-type="ref"}, [16](#jah33267-bib-0016){ref-type="ref"} For example, Rosinaru and colleagues[15](#jah33267-bib-0015){ref-type="ref"} showed that among 358 patients with HFpEF who survived a first hospitalization for HF during 7‐year follow‐up, hyponatremia (defined as serum sodium \<136 mEq/L) was associated with an 87% higher risk of mortality (HR, 1.87; 95% CI, 1.40--2.52) compared with serum sodium \>136 mEq/L. A meta‐analysis done by the Meta‐Analysis Global Group in Chronic Heart Failure (MAGGIC) showed that hyponatremia (serum sodium \<135 mEq/L) at baseline is associated with a 35% higher risk of mortality during 3 years of follow‐up among 3737 patients with HFpEF (HR, 1.35; 95% CI, 1.06--1.72), compared with serum sodium \>135 mEq/L.[16](#jah33267-bib-0016){ref-type="ref"} On analyzing serum sodium as a continuous variable, the MAGGIC meta‐analysis found a linear increase in the risk of all‐cause mortality at 3 years at concentrations \<140 mEq/L, which plateau at concentrations ≥140 mEq/L.[16](#jah33267-bib-0016){ref-type="ref"} These studies examined the end point of all‐cause mortality alone and included patients with a known history of HF; hence, the relation of serum sodium to time after diagnosis of HF was variable. In our study, low serum sodium was independently associated with not only a higher risk of mortality but also a higher risk of recurrent all‐cause hospitalizations and HF hospitalizations in patients without any previous history of HF before assessment of serum sodium included in the study. By excluding previous diagnosis of HF, our study enabled us to examine the association of serum sodium at the time of initial diagnosis with long‐term outcomes and avoid confounding effects of previous therapy for HF on serum sodium levels. It also allows us to examine how these associations might differ in the immediate aftermath of a diagnosis as well as longer‐term effects. In addition, our study incorporated a comprehensive list of clinical and laboratory variables, as well as medications, as covariates in our analyses. Another study addressed the association including only outpatients identified by *ICD‐9* codes.[13](#jah33267-bib-0013){ref-type="ref"} Among 2704 veterans with HFpEF treated in VA ambulatory clinics, Bavishi et al[13](#jah33267-bib-0013){ref-type="ref"} showed that at 2‐year follow‐up, hyponatremia (defined as serum sodium ≤135 mEq/L) was associated with all‐cause mortality (HR, 1.40; 95% CI, 1.12--1.75) but not with all‐cause hospitalization. In contrast to this, in our study, which also included subjects identified during an index admission to ensure that serum sodium was captured at baseline, low serum sodium was associated with an increased number of days of all‐cause hospitalizations and HF hospitalizations per year. Our study was able to capture outcomes more accurately by not only including readmissions within the VA health system but also including outcomes from Centers for Medicare and Medicaid Services in case patients were admitted to hospitals outside of the VA health system. In another study that analyzed the relation of serum sodium as a continuous variable with outcomes in 407 patients with HFpEF, Kusaka et al[14](#jah33267-bib-0014){ref-type="ref"} showed that serum sodium predicted combined cardiovascular death and rehospitalization for HF. In this study, both hyponatremia (serum sodium \<135 mmol/L) and low‐normal serum sodium (between 135 and 140 mmol/L) were independently predictive of future combined cardiovascular death and rehospitalization for HF. This study had various limitations, including being a single‐center study with a small sample size of subjects included after the initiation of medical therapy, with only one third of enrolled subjects treated with diuretics. In previous studies, the prevalence of hyponatremia (serum sodium \<135 mmol/L) among patients with HFpEF ranged from 11% to 14%.[13](#jah33267-bib-0013){ref-type="ref"}, [16](#jah33267-bib-0016){ref-type="ref"} The exact prevalence of low‐normal serum sodium is unknown, but in a study conducted by Kusaka and colleagues[14](#jah33267-bib-0014){ref-type="ref"} the prevalence of low‐normal serum sodium between 135 and 140 mmol/L was 47%, which was associated with higher HF‐related events compared with serum sodium more than 140 mmol/L. In our study, the prevalence of low‐normal serum sodium (between 135 and 137.99 mmol/L) was 21% and was associated with worse 30‐day and long‐term outcomes in patients with HFpEF. This shows that the prevalence of low‐normal serum sodium is high in patients with HFpEF and is associated with a worse prognosis. There are several potential mechanisms that could explain the observed relation between low serum sodium and adverse clinical outcomes. Hyponatremia is associated with activation of vasoactive neurohormonal systems such as the sympathetic nervous system, renin‐angiotensin‐aldosterone system, and arginine vasopressin.[24](#jah33267-bib-0024){ref-type="ref"} Activation of the sympathetic nervous system leads to systemic vasoconstriction and an increase in heart rate and cardiac contractility, whereas activation of the renin‐angiotensin‐aldosterone system leads to direct systemic vasoconstriction and activation of secretion of aldosterone and arginine vasopressin.[25](#jah33267-bib-0025){ref-type="ref"} Arginine vasopressin activation leads to an increase in free‐water reabsorption in the renal collecting duct, thereby increasing the blood volume.[25](#jah33267-bib-0025){ref-type="ref"} These defense mechanisms are important compensatory mechanisms in emergent conditions including blood loss and low cardiac output; however, sustained activation of these systems can lead to worsening HF by multiple biologic mechanisms, including adverse effects on afterload and myocardial oxygen consumption, as well as promotion of myocyte loss and fibrosis.[26](#jah33267-bib-0026){ref-type="ref"}, [27](#jah33267-bib-0027){ref-type="ref"} In our analysis, we also found that higher serum sodium measurements are associated with a 19% increase in average number days of all‐cause hospitalizations per year in patients with HFpEF. Our inability to find this for heart failure hospitalizations is likely related to a paucity of events. This is consistent with the study by Kovesdy et al[28](#jah33267-bib-0028){ref-type="ref"} showing a U‐shaped relationship between serum sodium level and mortality in patients with chronic kidney disease with and without HF. Hypernatremia has also been shown to be associated with increased mortality in the general population[29](#jah33267-bib-0029){ref-type="ref"} and patients with HFpEF.[30](#jah33267-bib-0030){ref-type="ref"} The mean age in this analysis is 70 years; elderly patients are at higher risk for developing hypernatremia because of an inability to replace water losses from physical and mental limitations to access water and decreased thirst perception. Hypernatremia can have direct adverse effects on multiple organ systems, most notably on the central nervous system.[31](#jah33267-bib-0031){ref-type="ref"} Because high sodium values were significantly associated only with all‐cause hospitalization in our study, it is also possible that hypernatremia is merely a surrogate marker of the more severe multimorbid state. Limitations and Strengths {#jah33267-sec-0024} ------------------------- There were a few limitations in this analysis of a large cohort. Care received outside of VA medical centers may not have been completely captured in this database, although the inclusion of data derived from the Centers for Medicare and Medicaid Services enabled us to capture the vast majority of events in those \>65 years old. Events in veterans who were \<65 years of age and ineligible for Medicaid could have been missed. There is no expectation that serum sodium would be associated with the probability of having missing events, resulting in minimal bias. In this study, we were not able to determine cause‐specific mortality. The population is predominantly white males; however, because there are no sex differences reported in the association of serum sodium with mortality in HF with reduced EF, we do not consider our findings to be affected by sex. As with any observational study, residual confounding or confounding due to unmeasured factors cannot be entirely excluded. The current study has several strengths, this being the largest cohort of HFpEF with a long period of follow‐up compared with previously published studies. By limiting our cohort to incident heart failure, we were able to truly assess the influence of baseline sodium close to time of diagnosis. Finally, our algorithm, by ensuring that all recorded EFs were ≥50% at any time period in the VA system, prevented inclusion of heart failure with reduced, recovered, and intermediate EF in the cohort. Conclusions {#jah33267-sec-0025} =========== Our study, in a large cohort of patients with HFpEF rigorously curated from a large national database, demonstrates a J‐shaped relationship between serum sodium levels and long‐term clinical outcomes. Our data suggest that serum sodium could be an important clinical variable for identifying patients with HFpEF at increased risk of adverse events, healthcare utilization, and increased costs. Further investigation is needed to understand the role of neurohormonal activation in the relation between sodium levels and the natural history of HFpEF. Author Contributions {#jah33267-sec-0026} ==================== The authors responsibilities were as follows---research design: Patel, Joseph, Gagnon, Kurgansky, Djousse, McLean; data analysis: Gagnon, Kurgansky; Article draft: Patel, Joseph, Kurgansky; article revisions: Patel, Joseph, Gagnon, Kurgansky, Djousse, McLean, Imran, Orkaby, Ho, Cho, Gaziano. The authors thank Ms Constance Nelson for her assistance with management of this project. Sources of Funding {#jah33267-sec-0027} ================== This project was funded by a grant from Otsuka Pharmaceuticals to Dr Joseph. Disclosures {#jah33267-sec-0028} =========== Joseph reports research grant support from the National Heart, Lung, and Blood Institute; Novartis; Otsuka; Amgen; and consulting for Amgen. The remaining authors have no disclosures to report. [^1]: Dr Gagnon and Dr Joseph are co--senior authors.

1. Introduction {#sec1} =============== Diabetes mellitus (DM) is a major cardiovascular risk factor and is associated with increased cardiovascular events and mortality. Atherosclerosis (AS) is one of the most common vascular complications of diabetes mellitus and a leading cause of death and disabling cardiovascular disease \[[@B1], [@B2]\]. The underlying mechanisms of diabetic atherosclerosis can be attributed to a combination of factors, including oxidative stress, inflammation, increased expression of growth factors, and increased production of AGEs \[[@B3], [@B4]\]. AGEs are the products of nonenzymatic glycosylation and oxidation of a group of proteins and lipids after continuous contact with reducing sugars or short-chain aldehydes. They are common in the vasculature of diabetic patients, accelerating the development of AS. The interaction of AGEs with their receptors (RAGE) activates downstream molecular pathways leading to atherosclerosis \[[@B5]\]. A previous study reported that activated vascular smooth muscle cells (VSMCs) can effectively proliferate and migrate, promoting the repair of the vascular wall \[[@B6]\]. However, whether AGEs induce the proliferation and migration of VSMCs so as to promote AS is still unclear. The migration and proliferation of VSMCs are essential in the process of vascular remodeling \[[@B7]\]. Under normal circumstances, VSMCs are mainly found in the middle membrane of the arteries and are major cellular components that maintain the morphology and function of blood vessels. Nevertheless, when intima is stimulated by inflammation, injury, stress, and other harmful stimuli, VSMCs proliferate rapidly and migrate to the injured site, participating in intimal hyperplasia \[[@B8]\]. Intimal hyperplasia is a pathogenic marker of restenosis after stent implantation, including abnormal proliferation and migration of VSMCs and accumulation of extracellular matrix, which eventually leads to lumen stenosis and even occlusion \[[@B9]\]. In addition, excess AGEs in diabetic patients accelerate the progression of arteriosclerosis by affecting arterial wall thickening. Interestingly, this effect is particularly marked by the proliferation and migration of VSMCs \[[@B10]\], which implied that the proliferation and migration of VSMCs may play an important role in the development of atherosclerosis. Thus, this study is aimed at investigating the effects of AGEs on the proliferation and migration of HASMCs and revealing the molecular mechanisms underlying these effects. 2. Materials and Methods {#sec2} ======================== 2.1. Cell Culture and Identification {#sec2.1} ------------------------------------ The primary human aorta vascular smooth muscle cells (HASMCs, ScienCell, California, USA) were cultured in Dulbecco\'s Modified Eagle Medium (DMEM, Hyclone Laboratories, Utah, USA) supplemented with 10% fetal bovine serum (FBS, ScienCell, California, USA), penicillin (100U/mL), and streptomycin (100 *μ*g/mL ) and then incubated at 37°C in a humidified 5% CO~2~ atmosphere. Smooth muscle actin (*α*-SMA) expression of the cells was identified by immunofluorescence assay. Cells that have been passaged for 3 to 5 generations were used. 2.2. Cell Proliferation {#sec2.2} ----------------------- Cell proliferation was assessed using Cell Counting Kit-8 (CCK8, Dojindo, Shanghai, China). HASMCs were seeded into 96-well plates at 1000 cells/well in 200 *μ*L of medium for 24 h. The adherent cells were serum-starved for another 24 h and then treated with different concentrations of AGEs (5, 10, 20, and 40 mg/L; BioVision, Japan) or without any AGEs for 24 h in a humidified atmosphere with 5% CO~2~. Following the experimental treatment, supplemented with 10 *μ*L of CCK8 solution for each well, the cells were incubated at 37°C for 3 h. Finally, the absorbance value was measured at 450 nm using a microplate reader. 2.3. Cell Migration {#sec2.3} ------------------- Cell migration was assessed by transwell insert chambers (Corning, Maine, USA). HASMCs (5 × 10^3^ cells) were seeded in the serum-free medium into the upper chamber; the lower chamber was filled with DMEM with 10% FBS and supplemented with different interventions according to the experimental design. After 24 h, the cells from the upper surface of the filters were carefully wiped off with a cotton swab while those on the lower surface were washed with PBS, fixed for 30 min, air-dried at room temperature, and stained with 0.1% crystal violet (Beyotime Biotechnology, Shanghai, China) for 20 min. Cell migration was then visualized microscopically. 2.4. F-Actin Staining {#sec2.4} --------------------- HASMCs (5.0 × 10^3^ cells/mL) were inoculated uniformly on glass slides for 24 h. Synchronized cells were then subjected to one of the following treatments for 24 h: AGEs (0 mg/L) which represent the control group, AGEs (20 mg/L), or AGEs (20 mg/L) + LY294002 (a commonly used PI3K inhibitor, Selleck, USA). Cell slides were washed with PBS, fixed for 30 min, washed again, permeabilized in 0.2% Triton X-100 at room temperature for 5 min, treated with 1% bovine serum albumin for 30 min, stained with 10 *μ*g/mL FITC-rhodamine phalloidin (AmyJet, Wuhan, China) at room temperature for 30 min, washed again, treated with 50 *μ*g/mL RNase at 37°C for 30 min, stained with 10 *μ*g/mL propidium iodide (PI, BD, USA) for 30 min, washed again, and finally sealed with 50% buffer glycerin. F-Actin staining was visualized and analyzed by confocal microscopy. 2.5. RAGE Knockdown by RAGE-RNAi Lentivirus {#sec2.5} ------------------------------------------- RAGE-RNAi lentivirus vector construction and packaging was entrusted to Shanghai Genechem Co. Ltd. The siRNA target sequence was 5′-AGTCCGTGTCTACCAGATT-3′, and the contrast insertion sequence was 5′-TTCTCCGAACGTGTCACGT3′. The MOI of HASMCs was found to be 100 through a preliminary experiment following the manufacturer\'s instructions. Cells were seeded on 24-well plates (3 × 10^4^ cells/well) until they reached 70% confluence, at which time they were cultured with prediluted RAGE-RNAi lentivirus and 8 *μ*g/mL polybrene at 37°C under 5% CO~2~ for 8 h. Cell growth was observed, after which cells were maintained in fresh medium for 96 h at which time virus infection was visualized using inverted fluorescence microscopy. When the GFP expression exceeded 90%, the expression of RAGE protein was detected by western blotting. 2.6. Western Blotting {#sec2.6} --------------------- Briefly, whole cell lysates were extracted from cultured HASMCs with Radioimmunoprecipitation Assay (RIPA) buffer (Beyotime Biotechnology, Shanghai, China) and the protein concentration determined using the BCA method as recommended by the manufacturer (Beyotime Biotechnology, Shanghai, China). The collected proteins were boiled for 5 min and separated using 10% SDS-polyacrylamide gels (SDS-PAGE; Sigma, USA), then transferred to PVDF membranes (Millipore, Bedford, MA, USA) using 10% SDS-PAGE gel electrophoresis. Next, samples were sealed with 5% nonfat milk powder and TBST at room temperature for 1 h, after which membranes were incubated with the following primary antibodies: rabbit anti-human PCNA (1 : 1000), Cyclin-D1 (1 : 1000), MMP-9 (1 : 800), MMP-2 (1 : 500), RAGE (1 : 800), AKT (1 : 1000), phospho- (p-) AKT (1 : 2000), and antibody GAPDH (1 : 5000) (CST, Santa Cruz, Biotechnology, CA, USA) at 4°C overnight. Samples were then washed 3 times with 10 mL of TBST (15 min), after which the PVDF membranes were incubated with HRP-labeled goat anti-rabbit IgG (1 : 5000) (Abgent, Suzhou, China) at 37°C for 1 h. Finally, the specific protein bands were visualized with an ECL Plus kit (Beyotime Biotechnology, Shanghai, China) and quantified with the Quantity One software (Bio-Rad, USA). 2.7. Statistical Analysis {#sec2.7} ------------------------- All data were presented as means ± standard deviation (SD) and analyzed using statistical software SPSS19.0. Comparisons between the groups were performed by one-way analysis of variance (ANOVA). Student\'s *t*-tests were conducted to determine the differences between two groups. *P* \< 0.05 was considered as significant difference. 3. Results {#sec3} ========== 3.1. HASMC Identification {#sec3.1} ------------------------- The immunofluorescence imaging ([Figure 1](#fig1){ref-type="fig"}) revealed that the HASMCs expressed *α*-SMA, as the positive green fluorescence was emitted from the cytoplasm of the cells. 3.2. Effects of AGEs on Cell Proliferation and Migration {#sec3.2} -------------------------------------------------------- CCK8 assay showed that, after HASMCs were treated with AGEs, the proliferation ability of cells was significantly enhanced, compared with control cells, in a concentration-dependent manner, and cell proliferation ability was the strongest when AGE concentration was 10 mg/L ([Figure 2(a)](#fig2){ref-type="fig"}). AGE-induced HASMC migration was detected by the transwell assay; the results demonstrated that, in a certain range, AGEs promoted the migration of cells in a concentration-dependent manner, and the promigratory ability of AGEs increased as the AGE concentration increased, peaked at 20 mg/L, and then decreased gradually ([Figure 2(c)](#fig2){ref-type="fig"}), compared with those of the control cells. In addition, as shown in Figures [2(b)](#fig2){ref-type="fig"} and [2(d)](#fig2){ref-type="fig"}, the expression of proteins related to proliferation and migration showed concentration-dependent changes to AGEs. 3.3. AGEs Activated PI3K/AKT Signaling {#sec3.3} -------------------------------------- In order to verify whether the PI3K/AKT signal pathway is involved in the proliferation and migration of HASMCs induced by AGEs, we detected the expression of AKT and its phosphorylated proteins by western blotting. We found that p-AKT levels were notably increased, compared to the control group, after cells were treated with AGEs (20 mg/L) for 5-15 min ([Figure 3(a)](#fig3){ref-type="fig"}). Phosphorylation of AKT induced by AGEs was significantly inhibited by LY294002 which is a potent protein kinase inhibitor of phosphotidylinsitol-3-kinase (PI3K); however, the total AKT levels were not obviously decreased ([Figure 3(b)](#fig3){ref-type="fig"}). To further investigate this effect, HASMCs were pretreated with and without LY294002 and then treated with AGEs; the results showed that the proliferation and migration of HASMCs induced by AGEs were dramatically inhibited by LY294002 (Figures [4(a)](#fig4){ref-type="fig"}--[4(d)](#fig4){ref-type="fig"}). In addition, as shown in [Figure 4(e)](#fig4){ref-type="fig"}, LY294002 also markedly reduced the expression of filament actin (F-actin) in HASMCs compared to the control cells. 3.4. RAGE-Mediated Activation of PI3K/AKT Signaling {#sec3.4} --------------------------------------------------- RAGE is a transmembrane protein, which mainly mediates the transmission of the intracellular signal. [Figure 5(a)](#fig5){ref-type="fig"} shows that the expression of the RAGE protein in HASMCs is incrementally dependent on AGE concentration. To investigate whether RAGE mediates PI3K/AKT signaling, the RAGE gene in HASMCs was silenced through RAGE-RNAi lentivirus vector transfection. Compared with the control group, a large amount of green fluorescence was stably detected in both the empty vector lentivirus transfection group and the RAGE-RNAi lentivirus transfection group ([Figure 5(b)](#fig5){ref-type="fig"}), indicating successful transfection. Further, we found that the expression of RAGE was dramatically downregulated after the RAGE gene was knocked down ([Figure 5(c)](#fig5){ref-type="fig"}). Interestingly, the expression of phosphorylated AKT was also inhibited significantly, while the expression of total AKT was almost unchanged ([Figure 5(d)](#fig5){ref-type="fig"}). As shown in Figures [5(e)](#fig5){ref-type="fig"} and [5(f)](#fig5){ref-type="fig"}, compared with cells in the control group, the expression of proteins related to proliferation and migration of HASMCs was notably reduced after RAGE was silenced, revealing that the proliferation and migration of HASMCs was also significantly inhibited after RAGE was blocked. 4. Discussion {#sec4} ============= AS is a chronic and complicated process involving numerous types of cells and cell-to-cell interactions that ultimately lead to progression from the "fatty streak" to formation of more complex atherosclerotic plaques \[[@B11]\]. In the early stage, damage to the vessel intima can initiate a series of self-repair, but excessive proliferation and migration of VSMCs accelerate progression of AS \[[@B12]\]. Studies found that AGEs largely exist in the arterial lipid stripes, atherosclerotic plaques, and macrophages of DM patients, the quantity of which is positively related to the severity of atherosclerotic plaques \[[@B13]--[@B16]\], which means AGEs can accelerate the development of AS in diabetics. Therefore, the study of the effect of AGEs on the proliferation and migration of VSMCs and its related mechanism is significant for the prevention and treatment of diabetic arteriosclerosis. Although some studies suggested that AGEs can induce the proliferation of VSMCs and play an important role in the pathogenesis of AS in DM patients \[[@B17]--[@B20]\], the effect of different AGE concentrations on cell proliferation and migration remains unclear. When studying the effect of AGEs-BSA on the proliferation of rabbit VSMCs in vitro, Satoh et al. found that AGEs-BSA could significantly promote the proliferation of rabbit VSMCs in the range of 1-10 mg/L, and the proliferation was notably inhibited when the concentration exceeded 20 mg/L \[[@B21]\]. This study further confirmed the relationship between the proliferation and migration of HASMCs and the concentration of AGEs. It was found that cell proliferation and migration predominated when the AGE range was low (5-20 mg/L), but decreased significantly when the concentration increased to 40 mg/L (Figures [2(a)](#fig2){ref-type="fig"} and [2(c)](#fig2){ref-type="fig"}). In addition, our previous study revealed that AGEs could promote apoptosis in HASMCs within the range of 50\~200 mg/L and induce calcification in a concentration-dependent manner \[[@B22]\]. The mechanism of different cellular effects of different AGE concentrations is not clear at present, but we speculate that this may be partly due to the fact that high concentrations of AGEs induce apoptosis and inhibit the activity of cell proliferation and migration. A study showed that the expression of PCNA, Cyclin-D1, MMP-2, and MMP-9 genes can induce the formation of colorectal cancer and promote cancer cell proliferation, migration, and invasion \[[@B23]\], suggesting that these genes may be closely related to cell proliferation and migration. Some other studies reported that PCNA and Cyclin-D1 mainly regulate cell proliferation and that MMP-2 and MMP-9 play important roles in cell migration and invasion \[[@B24], [@B25]\]. In this study, we observed that AGEs can upregulate the expression of PCNA, Cyclin-D1, MMP-2, and MMP-9 proteins (Figures [2(b)](#fig2){ref-type="fig"} and [2(d)](#fig2){ref-type="fig"}), which further proved that AGEs can activate the expression of these genes and induce HASMC proliferation and migration. Now that it has been firmly confirmed that AGEs exert proproliferative and promigratory functions on VSMCs, it is necessary to identify the specific mechanism governing the process. Yoon et al. \[[@B26]\] confirmed that AGEs could initiate their proproliferative function on VSMCs by activating the pathways on which P38 and ERK are dependent. The PI3K/AKT pathway has been found having the function of regulating multiple target proteins to participate in cell proliferation and migration \[[@B9], [@B12], [@B27]\]. Shi et al. \[[@B28]\] reported that Irisin stimulates cell proliferation and invasion by targeting the PI3K/AKT pathway in human hepatocellular carcinoma. However, whether the PI3K/AKT pathway is involved in the proliferation and migration of VSMCs in patients with diabetes needs further verification. In this study, AGE-induced proliferation and migration of HASMCs were detected after the PI3K/AKT pathway was inhibited by LY294002. The results revealed that inhibition of the PI3K/AKT pathway not only reduced cell proliferation and migration activity (Figures [4(a)](#fig4){ref-type="fig"} and [4(c)](#fig4){ref-type="fig"}) but also downregulated the protein expression related to cell proliferation and migration (Figures [4(b)](#fig4){ref-type="fig"} and [4(d)](#fig4){ref-type="fig"}). In addition, this study also found that the expression of F-actin, the main cytoskeleton system involved in cell contraction, adhesion, and migration \[[@B29]\], decreased after inhibition of PI3K/AKT ([Figure 4(e)](#fig4){ref-type="fig"}). These results confirmed that the PI3K/AKT pathway plays an important role in the proliferation and migration of HASMCs induced by AGEs. RAGE is a multiligand receptor member of immunoglobulin superfamily cell surface molecules. As transmembrane proteins on the cell membrane, RAGE and AGEs can initiate intracellular signal transduction through receptor-ligand binding \[[@B30]\]. In this study, the expression of the RAGE protein was confirmed to be positively dependent on AGE concentrations ([Figure 5(a)](#fig5){ref-type="fig"}), suggesting that AGEs may play biological roles through specific binding to RAGE. A previous study reported that lysophosphatidic acid is involved in tumor and angiogenesis through the RAGE and the AKT signal \[[@B31]\]. However, it is unclear whether AGE/RAGE, as a classical ligand-receptor binding axis, promotes the proliferation and migration of HASMCs by activating the PI3K/AKT pathway. In our experiments, the RAGE-RNAi lentivirus vector was transfected into HASMCs to silence the RAGE gene (Figures [5(b)](#fig5){ref-type="fig"} and [5(c)](#fig5){ref-type="fig"}). Subsequently, western blotting assay detection showed that AGEs induced AKT phosphorylation by RAGE ([Figure 5(d)](#fig5){ref-type="fig"}), which accelerated cell proliferation and migration (Figures [5(e)](#fig5){ref-type="fig"} and [5(f)](#fig5){ref-type="fig"}). In conclusion, this study reveals that AGEs induce the proliferation and migration of HASMCs *in vitro*. It also confirms, at least in part, that the AGE/RAGE axis can activate the PI3K/AKT pathway and thus facilitates HASMC proliferation and migration. This implies that the PI3K/AKT signaling pathway plays a pivotal role in AGE-induced proliferation and migration of HASMCs. Therefore, reducing AGEs and the expression of RAGE or developing more effective drugs to inhibit the PI3K/AKT pathway may become a new strategy for the treatment of diabetic atherosclerosis. However, it has to be said that since this experiment only studied HASMCs cultured *in vitro*, it is necessary to conduct additional experiments *in vivo*. The study was supported by research grants from the National Natural Science Foundation of China (Fund No. 81270358). The authors thank Dr. Xiaoming Zhu (Macau University of Science and Technology) and Li Liu (Southwest Medical University) for their assistance. Data Availability ================= The data used to support the findings of the study are available from the corresponding author upon request. Conflicts of Interest ===================== The authors declare no conflicts of interest. Authors\' Contributions ======================= Gang Yuan and Guangyan Si contributed equally to this work. {#fig1} {#fig2} {#fig3} {#fig4} {#fig5} [^1]: Academic Editor: Francesco Onorati

Introduction {#s1} ============ Venous thromboembolism (VTE), which includes deep vein thrombosis (DVT) and pulmonary embolism (PE), is a significant health problem with an estimated annual incidence of approximately 1--2 per 1000 people among the general population.[@R1] Vitamin K antagonists (VKAs) with initial heparinisation have long been considered a mainstay for the management of VTE.[@R2] However, there is a high degree of interindividual variability in the dose--response relationship to VKAs, which necessitates frequent inconvenient laboratory monitoring of coagulation status.[@R3] Possibly due to this variability and the narrow therapeutic window of VKAs, their use is often associated with serious bleeding complications.[@R4] To overcome some of the limitations of VKAs, direct oral anticoagulants such as edoxaban, a potent factor Xa inhibitor, were developed. As a class, compared with VKAs, these agents are at least as efficacious, are associated with a lower risk of major bleeding, have more predictable pharmacodynamics and require less frequent laboratory monitoring in individuals with VTE.[@R5] The individual response to VKAs, such as warfarin, is influenced by demographics, clinical characteristics and genetic factors.[@R3] Genetic factors influencing the response to VKAs include polymorphisms in the vitamin K epoxide reductase (*VKORC1*) and cytochrome P450 2C9 (*CYP2C9*) genes.[@R6] The *CYP2C9* alleles that are the most associated with increased warfarin sensitivity are more common in individuals of European descent compared with African or Asian descent.[@R11] Conversely, the prevalence of the *VKORC1* allele that is associated with greatest increase in warfarin sensitivity is the highest in individuals of Asian descent.[@R11] Together, the polymorphisms in these two genes account for up to 50% of the variability in warfarin dose in patients of European descent.[@R8] In a recent subanalysis of the Effective Anticoagulation with Factor Xa Next Generation in Atrial Fibrillation--Thrombolysis in Myocardial Infarction 48 (ENGAGE AF-TIMI 48) study, a large double-blind, double-dummy, randomised study comparing edoxaban to warfarin for the prevention of stroke and systemic embolic events in patients with atrial fibrillation (AF), patients randomised to receive warfarin and classified as sensitive or highly sensitive warfarin responders, based on *CYP2C9* and *VKORC1* genotypes, experienced more warfarin-associated bleeding and spent more time overanticoagulated relative to normal responders.[@R13] This effect was observed during the first 90 days of warfarin therapy, but was not present afterwards, suggesting that the effect of genetics on warfarin-induced bleeding risk is strongest during the initiation of therapy when the warfarin dose is still being optimised. These results confirmed that for stroke prevention in patients with AF, compared with warfarin, edoxaban was associated with a greater reduction in bleeding risk in sensitive and highly sensitive responders than in normal responders in the first 90 days. However, it is not known whether similar genetically based elevated bleeding risks associated with warfarin also exist for patients with VTE when transitioning from heparin to oral anticoagulants. Similar to the ENGAGE AF-TIMI 48 study, DNA samples were collected in the Hokusai-venous thromboembolism (Hokusai VTE) trial, allowing for a comparison of clinical outcomes by genotype. The Hokusai VTE trial was a randomised, multinational, double-blind, non-inferiority study to evaluate the safety and efficacy of edoxaban versus warfarin (dosed to an international normalised ratio (INR) of 2.0 to 3.0) in patients with VTE initially treated with heparin. The trial demonstrated that, following initial parenteral anticoagulant therapy, edoxaban was non-inferior to warfarin for the treatment of VTE and the prevention of recurrent VTE and was also associated with significantly fewer bleeding events in patients with VTE.[@R14] The present analysis was designed to replicate the ENGAGE AF-TIMI 48 findings and examine whether the previous results could be extended to patients with VTE. Methods {#s2} ======= Study design and patient population {#s2a} ----------------------------------- A detailed description of the study design and patient population of Hokusai VTE (ClinicalTrials.gov: NCT00986154) has been previously published.[@R14] Briefly, Hokusai VTE was a randomised, double-blind trial comparing edoxaban with warfarin in patients with acute, symptomatic VTE (DVT, PE or both). All patients received open-label heparin treatment for at least 5 days. Patients were randomised 1:1 to receive either edoxaban 60 mg once daily or warfarin. The edoxaban dose was halved to 30 mg once daily in patients with a creatinine clearance of 30 to 50 mL/min, a body weight of 60 kg or less or who were receiving concurrent treatment with the P-glycoprotein (inhibitors verapamil or quinidine. Warfarin was started concurrently with heparin and the warfarin dose was adjusted to maintain an INR between 2.0 and 3.0. To maintain blinding, sham INR measures were provided to patients randomised to edoxaban. The institutional review board at each participating centre approved the protocol. All patients provided written informed consent. The Hokusai VTE trial enrolled 8292 patients from 439 centres worldwide. Eligible patients were ≥18 years of age with diagnosed acute, symptomatic DVT that involved the popliteal, femoral or iliac veins or with acute, symptomatic PE with or without DVT. Patients were excluded from the study if they presented contraindications to heparin or warfarin, had been treated for more than 48 hours with heparin at therapeutic doses, had taken more than 1 dose of a VKA, had cancer for which long-term treatment with low-molecular-weight heparin was anticipated, had another indication for warfarin therapy, had received treatment with aspirin (at more than 100 mg daily) or dual antiplatelet therapy or had a creatinine clearance of less than 30 mL/min. Major bleeding was defined as overt bleeding that was associated with a reduction in haemoglobin of 2 g/dL or greater, required a transfusion of two or more units of blood, occurred at a critical site or contributed to death. Clinically relevant non-major (CRNM) bleeding was defined as overt bleeding not meeting the criteria for major bleeding but that required medical intervention, unscheduled contact with a clinician, interruption of study drug or discomfort or impairment of daily activities. Total bleeding was the composite of major bleeding, CRNM bleeding and nuisance bleeding (ie, all bleeding events not classified as major or CRNM). All bleeding events were adjudicated by an independent, central, blinded committee. Genotypes {#s2b} --------- Genotypes were determined for *CYP2C9 \*2* and *CYP2C9 \*3* (rs1799853 and rs1057910, respectively) and *VKORC1* ---1639 G→A alleles (rs9923231) in accordance with Good Laboratory Practice guidelines by ILS Genomics (Morrisville, North Carolina, USA) with an analytically validated assay using Sequenom (San Diego, California, USA) technology. PCR primers were validated using a total of 165 independent samples of known genotype representing African, Asian, Caucasian and Hispanic populations. The genotyping assay was previously validated as part of the ENGAGE AF-TIMI 48 pharmacogenetic analysis.[@R13] On the basis of the *VKORC1* and *CYP2C9* genotypes, patients were divided into three warfarin sensitivity types ([figure 1](#F1){ref-type="fig"}), normal responders, sensitive responders and highly sensitive responders, that correspond with the US Food and Drug Administration (FDA)'s categories for warfarin response included in the warfarin label.[@R15] In addition, because the total number of patients included in this analysis was substantially smaller than in the ENGAGE AF-TIMI 48 pharmacogenetic analysis (3956 vs 14 348),[@R13] exploratory analyses were also conducted using a two-bin system in which sensitive and highly sensitive responders were pooled together (ie, pooled sensitive responders) and compared with normal responders to increase statistical power ([figure 1](#F1){ref-type="fig"}). {#F1} Statistical analysis and data presentation {#s2c} ------------------------------------------ All statistical analyses were conducted using SAS V.9.3 or higher by Quintiles (Durham, North Carolina, USA) and Daiichi Sankyo (Parsippany, New Jersey, USA). All patients who received at least one dose of the study drug consented to be included in the pharmacogenetic analysis and provided a pharmacogenetic sample were included in the analysis. All statistical significant conclusions are based on two-sided tests with a significance level of 0.05. The demographics and baseline clinical characteristics, as well as time until discontinuation of heparin therapy, were compared across genetically defined warfarin sensitivity groups using Kruskal-Wallis tests. In warfarin-treated patients, final weekly warfarin dose and time in therapeutic range (TTR; defined as the percentage of days over the first 90 days and over the entire duration of the study that patients spent within INR 2.0--3.0) were also compared using Kruskal-Wallis tests. Global p values are based on the Kruskal-Wallis tests. Multiple comparisons were performed using Dwass-Steel-Critchlow-Fligner multiple comparison analysis, which is based on pairwise two-sample Wilcoxon comparisons; only comparisons in which the global p value was significant are shown. All on-treatment bleeding events occurring within 90 days of treatment or over the entire duration of the study were included in the analysis. Cox proportional hazards models were used to compare the bleeding outcomes (ie, total bleeding, major or CRNM bleeding and major bleeding) associated with edoxaban relative to warfarin stratified by genetically defined warfarin sensitivity types. Covariates used in the hazards model included presenting diagnosis (PE with or without DVT vs DVT only), baseline risk factors (temporary risk vs other) and the need for edoxaban dose adjustment. Kaplan-Meier methodology was used to assess the time-to-bleeding event curves for patients treated with warfarin during the first 90 days of treatment; p values are based on the Cox proportional hazards model with genotype of responder as covariate. Results {#s3} ======= Genotyping {#s3a} ---------- All of the 4013 samples collected were genotyped; all genotypes were concordant across samples. There was a 100% call rate for *VKORC1* and *CYP2C9* genotypes. The *CYP2C9* and *VKORC1* polymorphisms were in Hardy-Weinberg equilibrium for all three race groups examined (ie, Caucasians, African Americans and East Asians), and the observed allele frequencies within each race group were consistent with previously published findings (online [supplementary table 1](#SP1){ref-type="supplementary-material"}).[@R11] 10.1136/heartjnl-2016-310901.supp1 ###### Supplementary data Out of the 8292 patients in Hokusai VTE, 3956 (47.7%) participants (1978 from each treatment group) received ≥1 dose of the study drug, consented to be included in the pharmacogenetic analysis and provided a pharmacogenetic sample and were therefore included in the present analysis. Among the 1978 patients randomised to the warfarin regimen, 63.0% (1247) were normal responders, 34.1% (675) were sensitive responders and 2.8% (56) were highly sensitive responders (online [supplementary figure 1](#SP1){ref-type="supplementary-material"}). Baseline demographics and clinical characteristics {#s3b} -------------------------------------------------- In general, the baseline and clinical characteristics were similar across the genotype categories. However, a larger proportion of sensitive and highly sensitive responders were of Asian ancestry relative to normal responders ([table 1](#T1){ref-type="table"}). This variation in warfarin sensitivity by race is expected based on the known differences in allele frequencies[@R11] and is consistent with the ENGAGE AF-TIMI 48 pharmacogenetic subanalysis.[@R13] ###### Demographics and baseline clinical characteristics across genotype bins ------------------------------------------------------------------------------------------------------------------------------ Normal responders\ Sensitive responders\ Highly sensitive responders\ (n=2459) (n=1366) (n=131) -------------------------------------------------- -------------------- ----------------------- ------------------------------ Age (year), mean±SD 55.3±15.92 56.5±15.58 56.3±17.47 Male, n (%) 1486 (60.4) 802 (58.7) 70 (53.4) Race, n (%)  Caucasian 2031 (82.6) 1122 (82.1) 108 (82.4)  Black 147 (6.0) 5 (0.4) 0  Asian 113 (4.6) 157 (11.5) 15 (11.5)  Native Hawaiian/Pacific Islander 0 1 (\<0.1) 0  Other 164 (6.7) 77 (5.6) 7 (5.3) Weight (kg), mean±SD 85.5±19.42 83.7±19.25 83.2±19.91 BMI (kg/m^2^), mean±SD 28.8±5.95 28.3±5.45 28.5±6.40 Creatinine clearance (mL/min), mean±SD 108.5±40.14 106.2±39.69 103.8±40.62 Edoxaban dose adjusted at randomisation, n (%)\* 295 (12.0) 207 (15.2) 17 (13.0) ------------------------------------------------------------------------------------------------------------------------------ \*Edoxaban patients meeting the dose reduction criteria (ie, creatinine clearance 30--50 mL/min, body weight ≤60 kg or receiving select concomitant P-gp inhibitor) had their edoxaban dose reduced to edoxaban 30 mg once daily. All data are from baseline unless otherwise noted. BMI, body mass index; P-gp, p-glycoprotein. Duration of heparin dose, final warfarin dose, TTR and bleeding risk with warfarin {#s3c} ---------------------------------------------------------------------------------- In warfarin-treated patients, but not edoxaban-treated patients, as genetically defined warfarin sensitivity increased, heparin therapy was discontinued sooner (global p\<0.001; p\<0.0001 vs normal responders for both sensitive and highly sensitive responders) and the final mean weekly warfarin doses decreased (global p\<0.001; p\<0.0001 vs normal responders for both sensitive and highly sensitive responders) ([figure 2A and B](#F2){ref-type="fig"}). Over the first 90 days of treatment and over the entire study duration, there were no differences observed in the proportion of time warfarin-treated patients spent in normal INR therapeutic range (global p=0.0593 and 0.6313, respectively) ([figure 3A and B](#F3){ref-type="fig"}). However, as genetically defined warfarin sensitivity increased, warfarin-treated patients tended to spend a higher percentage of time with supratherapeutic INRs during the first 90 days of treatment (global p\<0.0001) and over the entire study duration (global p=0.0004) ([figure 3](#F3){ref-type="fig"}). Among the warfarin-treated patients over the entire duration of the study, the median percentages of time with an INR value \>3 were 13.1% for normal responders, 15.2% for sensitive responders (p=0.0013 vs normal responders) and 16.8% for highly sensitive responders (p=0.0639 vs normal responders). In addition, among patients randomised to warfarin, sensitive and highly sensitive responders experienced significantly more bleeding events sooner than normal responders during the first 90 days of treatment (HR: sensitive responder, 1.38 \[95% CI 1.11 to 1.71\], p=0.004; highly sensitive responders, 1.79 \[1.09 to 2.99\]; p=0.03) ([figure 4](#F4){ref-type="fig"}). {#F2} {#F3} {#F4} Bleeding analysis for edoxaban versus warfarin by genotype categories {#s3d} --------------------------------------------------------------------- Over the first 90 days of treatment and over the entire duration of the study, there were no significant interactions observed across genotype categories for the risk of total bleeds or major and CRNM bleeds for edoxaban versus warfarin (online [supplementary figure 2A](#SP1){ref-type="supplementary-material"}). Edoxaban-treated patients had a numerically greater number of major bleeds relative to warfarin-treated individuals across genotype categories during the first 90 days and particularly during the first 28 days of treatment (online [supplementary table 2](#SP1){ref-type="supplementary-material"}, [supplementary figure 3](#SP1){ref-type="supplementary-material"}). However, because the size of the warfarin-sensitive and highly sensitive cohorts was small, there were too few major bleeds within these cohorts to reliably establish estimates of the major bleeding rate. In an exploratory analysis using a two-bin system in which sensitive and highly sensitive responders were pooled (ie, pooled sensitive responders) and compared with normal responders, a statistically higher proportion of pooled sensitive warfarin responders experienced any bleeding events compared with normal warfarin responders during the first 90 days of treatment (HR 1.41 \[95% CI 1.14 to 1.74\]; p=0.0013) (online [supplementary figure 4](#SP1){ref-type="supplementary-material"}). In addition, warfarin was associated a significantly higher risk of total bleeds relative to edoxaban in the first 90 days as a function of warfarin sensitivity (p~int~=0.0377; online [supplementary figure 2B](#SP1){ref-type="supplementary-material"}). Discussion {#s4} ========== The present pharmacogenetic subanalysis of the Hokusai VTE trial was designed to confirm and extend the results of the previously reported ENGAGE AF-TIMI 48 pharmacogenetic analysis in a distinct population of patients with VTE requiring a similar treatment.[@R13] Following an approach consistent with the previously reported ENGAGE AF-TIMI 48 pharmacogenetics subanalysis, the results of this new set of analyses confirmed that genetic polymorphisms in *CYP2C9* and *VKORC1* affect the pharmacology and the safety of warfarin. Specifically, as genetically defined warfarin sensitivity increased, heparin therapy was discontinued earlier and the final weekly warfarin dose decreased.[@R13] The proportion of time spent in a supratherapeutic INR also tended to be elevated in sensitive and highly sensitive warfarin responders relative to normal responders. In the Hokusai VTE trial, the overall time spent in therapeutic range was 63.5%,[@R14] which is higher than what is typically reported in registries of clinical practice but is similar to what was reported in the ENGAGE AF-TIMI 48 study.[@R16] It is therefore possible that the effect of genetic variants may be more pronounced outside of the clinical trial environment and that the rigour required in registrational clinical trials leads to a mitigation of the risks associated with the polymorphism through the increase in laboratory monitoring. During the first 90 days of warfarin treatment, the risk of any bleeding event was significantly increased by approximately 1.8 times in the highly sensitive responders and 1.4 times in the sensitive responders relative to normal responders (p\<0.05 for both). However, unlike the pharmacogenetic analysis of the ENGAGE AF-TIMI 48 study,[@R13] only a trend was observed towards a higher bleeding risk with warfarin versus edoxaban in sensitive and highly sensitive responders relative to normal responders (p~int~=0.0747). In addition, warfarin-associated major bleeding during the first 90 days was lower than what was observed in the pharmacogenetics subanalysis of ENGAGE AF-TIMI 48.[@R13] The lack of a significant interaction between the bleeding risk associated with warfarin versus edoxaban as a function of warfarin sensitivity may be due to the lower number of patients included in this analysis compared with the ENGAGE AF-TIMI 48 pharmacogenetic analysis.[@R13] Although the FDA recommends dose adjustments for warfarin in patients with warfarin-sensitive genotypes,[@R15] in clinical practice, *CYP2C9* and *VKORC1* genotypes are not routinely assessed, and it has never been demonstrated that genotyping can reduce the risk of bleeding. The present results and the results of the previous ENGAGE AF-TIMI 48 pharmacogenetic analysis suggest that further investigations are needed into the potentially beneficial effects of genotyping prior to prescribing anticoagulation therapy in patients with AF or VTE.[@R13] Genotype-informed dosing may be especially beneficial in those patients who are initiating therapy, as an individual's genetic sensitivity could interact with existing clinical factors to exacerbate overall bleeding risk.[@R13] This pharamcogenetic subanalysis of Hokusai VTE has a number of limitations which could constrain the generalisability of the results. Hokusai VTE did not stratify patients by warfarin sensitivity genotypes prior to randomisation. There were not enough major bleeding events to determine the interaction between treatment regimen and warfarin sensitivity. In addition, the study was underpowered for the three-bin analysis and the results of the exploratory two-bin analysis should only be considered hypothesis generating. There was also a low number of black patients included in this subanalysis, and therefore additional analyses will be needed to confirm the results in different racial groups. It should also be noted that genotyping was not used in guiding the starting dose of warfarin and that the high TTR observed in this study is not what is typically seen in clinics.[@R16] Finally, the results of this subanalysis could differ in other disease states besides VTE. Conclusion {#s5} ========== The results of this analysis extend the results of the previous ENGAGE AF-TIMI 48 pharmacogenetic analysis to demonstrate that patients with VTE who have a sensitive or highly sensitive warfarin genotype spend more time overanticoagulated, require a lower warfarin dose and have higher bleeding rates with warfarin therapy. In addition, this subanalysis suggests that compared with normal warfarin responders sensitive warfarin responders may have a higher risk of bleeding on warfarin relative to edoxaban. ###### Key messages What is already known on this subject? -------------------------------------- In a recent subanalysis of the Effective Anticoagulation with Factor Xa Next Generation in Atrial Fibrillation--Thrombolysis in Myocardial Infarction 48, a study comparing edoxaban with warfarin for the prevention of stroke and systemic embolic events in patients with atrial fibrillation, patients randomised to receive warfarin and classified as sensitive or highly sensitive warfarin responders---based on *CYP2C9* and *VKORC1* genotypes---experienced more warfarin-associated bleeding and spent more time overanticoagulated relative to normal responders. What might this study add? -------------------------- The present pharmacogenetic subanalysis of the Hokusai-venous thromboembolism (Hokusai VTE) trial, a trial evaluating edoxaban versus warfarin in patients with VTE initially treated with heparin, demonstrates that patients with VTE who have a sensitive or highly sensitive warfarin genotype spend more time overanticoagulated (p\<0.001), require a lower warfarin dose (p\<0.001) and have higher bleeding rates with warfarin therapy (sensitive responders HR 1.38 \[95% CI 1.11 to 1.71\], p=0.0035; highly sensitive responders 1.79 \[1.09 to 2.99\]; p=0.0252). How might this impact on clinical practice? ------------------------------------------- The results presented here may influence the practical management of warfarin therapy for the treatment of VTE in patients who are sensitive or highly sensitive responders to warfarin. Medical writing and editorial support was provided by Stefan Kolata, PhD, of AlphaBioCom, LLC (King of Prussia, Pennsylvania, USA) and funded by Daiichi Sankyo (Parsippany, New Jersey, USA). **Contributors:** AV, JW, KSB, GZ, ML, MAG and MFM made substantial contributions to the conception or design of the work and analysis or interpretation of data, drafted or critically revised the manuscript, and had final approval of the version to be published. **Funding:** The study was funded by Daiichi Sankyo, Inc (Parsippany, New Jersey, USA). **Competing interests:** AV, GZ, ML, MAG and MFM are employees of Daiichi Sankyo, Inc. JW and KSB were employees at Daiichi Sankyo, Inc at the time the study was conducted. **Patient consent:** This manuscript does not contain identifiable medical information. **Ethics approval:** The institutional review board at each participating centre approved the protocol. **Provenance and peer review:** Not commissioned; externally peer reviewed.

Introduction ============ Due to the aging of the population in industrialized nations, the percentage of the Japanese population aged ≥65 years is projected to increase from 22.1% in 2008 to 30.3% in 2025 as outlined in Japanese Government statistics. After smoking, age is an important risk factor in the development of bladder cancer.[@b1-cmar-10-3669],[@b2-cmar-10-3669] Therefore, as the population ages, the incidence of urothelial carcinoma (UC) of the urinary bladder is expected to increase in the coming years, and elderly patients will compose a larger number of metastatic UC (mUC) cases at diagnosis. The gold standard for the treatment of patients with mUC is systemic cisplatin-based chemotherapy. A combination regimen of methotrexate, vinblastine, doxorubicin, and cisplatin (MVAC) has been used for the past 2 decades.[@b3-cmar-10-3669] A more recent alternative standard treatment for mUC is combination chemotherapy with gemcitabine and cisplatin (GC).[@b4-cmar-10-3669],[@b5-cmar-10-3669] However, long-term survival rates were deemed disappointing in long-term follow-up studies. Therefore, the rescue of patients who develop relapse after first-line cisplatin-based chemotherapy is an important issue. There is no standard second-line treatment for mUC outside the United States, where atezolizumab, an anti-PDL1 antibody, was approved by the US Food and Drug Administration (FDA) in May 2016 for patients with prior first-line platinum-based chemotherapy.[@b6-cmar-10-3669],[@b7-cmar-10-3669] The use of gemcitabine and docetaxel (GD) combination therapy as a second-line treatment for patients with mUC after failure of first-line platinum-based chemotherapy has been reported in our previous study.[@b8-cmar-10-3669] However, the efficacy of second-line therapy for elderly patients with mUC is unclear. Several studies have described how elderly patients may be less tolerant to chemotherapy because of comorbidities and hematologic toxicities.[@b9-cmar-10-3669],[@b10-cmar-10-3669] Furthermore, elderly patients with mUC are underrepresented in clinical trials,[@b11-cmar-10-3669]--[@b13-cmar-10-3669] and data regarding outcomes and adverse events (AEs) for second-line therapy are limited for this age group. Therefore, in this study, we retrospectively examined the effectiveness of second-line GD therapy for elderly patients with mUC by analyses of our pooled cumulative data. Patients and methods ==================== Patients -------- We enrolled eligible patients with histologically confirmed mUC of the urinary bladder or upper urinary tract and who were admitted to Nagoya City University Hospital and four affiliated institutions between July 2010 and June 2016. Patients had previously been surgically treated or had undergone biopsies of their primary lesions, and staging had been performed by enhanced computed tomography (CT). Patients diagnosed with mUC had also undergone more than one cycle of chemotherapy with GC (1,000 mg/m^2^ gemcitabine on days 1, 8, and 15 and 70 mg/m^2^ cisplatin on day 2) or gemcitabine and carboplatin (1,000 mg/m^2^ gemcitabine on days 1 and 8, and carboplatin area under the curve 4--5 mg/mL per minute on day 1), which was completed a minimum of 4 weeks prior to enrollment. Gemcitabine in these first-line regimens was not administered on days 8 and 15 if grade 3 toxicities occurred. In the study period, 124 patients underwent first-line chemotherapy, of whom, two patients, aged 71 and 78 years, discontinued the treatment because of a poor performance status caused by chemotherapy or cancer progression. Therefore, in total, we studied 122 patients who received second-line GD therapy. Using pooled cumulative data, we divided patients into the following three groups based on age: \<64 years (Group A: n=45), from 65 to 74 years (Group B: n=56), and ≥75 years (Group C: n=21). Data were retrospectively analyzed. All patients were evaluated for the presence of any toxicity and were assessed at every cycle by imaging studies by enhanced CT. Patients were required to have an Eastern Cooperative Oncology Group performance status (ECOG-PS) of 1 or lower as per World Health Organization (WHO) criteria: an adequate bone marrow reserve (white blood cell \[WBC\] count \>3,500/µL, platelet count \>100,000/ µL, and hemoglobin \>10 g/dL). Other requirements included reasonable hepatic function (serum bilirubin ≤1.5 mg/dL) and an estimated life expectancy of ≥12 weeks. Prognostic comorbidity was estimated using the Charlson comorbidity index (CCI).[@b14-cmar-10-3669] Ineligible patients included those with non-malignant systemic diseases such as an active infection that precluded them from receiving therapy or those with any clinically significant cardiac arrhythmia and/or congestive heart failure. All patients provided written informed consent prior to this clinical trial. The institutional chemotherapy review boards (ethical committees) of Nagoya City University Hospital and Nagoya City University (\#1152) approved this study, which was conducted in accordance with the Declaration of Helsinki (according to the 2004 Tokyo revision). Treatment schedule ------------------ The 122 selected patients were given an intravenous infusion of gemcitabine (800 mg/m^2^) for 30 minutes and docetaxel (40 mg/m^2^) for 60 minutes on days 1 and 8 according to the study by Dreicer et al.[@b15-cmar-10-3669] This cycle was repeated every 21 days. Dexamethasone (6.6 mg administered intravenously for 30 minutes) was used as a premedication for docetaxel. The same GD doses were administered on days 1 and 8 of each cycle if patients displayed WBC and platelet counts of \>3,000 and \>75,000 µL/mL, respectively; treatment was discontinued in each cycle if the counts were lower. When grade 3 AEs occurred, a 10% dose reduction was performed in the next cycle, and GD treatment was continued until progression. The efficacy of the GD regimen as second-line chemotherapy was assessed in a follow-up analysis. Anti-emetics and analgesics for AEs were given as supportive care to patients. Treatment evaluation -------------------- The estimated glomerular filtration rate (eGFR) was measured and calculated prior to each chemotherapy course, and hematological status and serum chemistries were measured twice a week during treatment. Radiology was used to assess tumor sizes, and physical examinations were also conducted. After each chemotherapy cycle, tumor sizes were remeasured. At least 4 weeks after the administration of chemotherapy, each patient's response to treatment was evaluated. The cutoff for the relative dose intensity (RDI) in first-line chemotherapy was 90% in accordance with a previous study.[@b16-cmar-10-3669] Death from the start of second-line therapy was the end point for the measurement of overall survival (OS), and progression-free survival (PFS) rates, and the time to failure from the start of second-line therapy was measured until the discontinuation of treatment, death, or progression. The Response Evaluation Criteria in Solid Tumors guidelines, version 1.1, were used to classify responses.[@b17-cmar-10-3669] The National Cancer Institute Common Terminology Criteria for AEs, version 3.0,[@b18-cmar-10-3669] were used to classify AEs. Tumor response, PFS, and OS were considered as end points of this study. Statistics ---------- Differences in categorical parameters were assessed using a Fisher's exact test, ANOVA test, Tukey's post hoc methods, Kruskal--Wallis test, and a chi-squared test, whichever was appropriate. Cumulative rates were estimated using the Kaplan--Meier method, and the significance of differences between curves was tested by the log-rank test. Univariate and multivariate analyses were conducted using the Cox proportional hazard regression model. A *P* value of \<0.05 was considered statistically significant. All data were analyzed using EZR software (Saitama Medical Center, Jichi Medical University, Yakushiji, Japan). Results ======= Treatment responses and outcomes of second-line GD chemotherapy --------------------------------------------------------------- [Table 1](#t1-cmar-10-3669){ref-type="table"} lists patients' clinical characteristics. A statistical significance was not found between the three groups except for the median eGFR and RDI for first-line chemotherapy; a high ratio for the carboplatin-based regimen in Group C reflected the degradation of the eGFR. Patient's responses were assessed after they underwent one or more chemotherapy courses and are described in [Tables 1](#t1-cmar-10-3669){ref-type="table"} and [2](#t2-cmar-10-3669){ref-type="table"}. The CCI score for Group C was significantly higher compared to those for Groups A and B. Second-line GD chemotherapy was performed for a median of three cycles (range: 1--16) for Group A, three cycles (range: 1--15) for Group B, and two cycles (range: 1--11) for Group C; the three groups did not show any significant differences. The objective response rate (ORR% of complete response \[CR\]+partial response \[PR\] cases per total cases) after second-line GD treatment was 22.2% for Group A, 14.3% for Group B, and 14.3% for Group C. The relative response rate (% of PR and stable disease \[SD\] cases per total cases) was 46.7% for Group A, 55.4% for Group B, and 38.1% for Group C. The median PFS from the start of second-line GD therapy was 3.7 months (95% CI: 2.6--5.3), 3.5 months (95% CI: 2.5--5.0), and 2.6 months (95% CI: 2.0--3.0) for Groups A, B, and C, respectively. The median response duration in PR cases for Group C was 5.4 months (range: 2.4--42.2). The median OS from the start of second- line GD therapy was 8.1 months (95% CI: 5.0--11.7), 8.5 months (95% CI: 6.8--10.3), and 5.3 months (95% CI: 3.4--7.9) for Groups A, B, and C, respectively. The survival rate after 1 year of follow-up from the start of second-line GD therapy was 30.2%, 29.0%, and 17.9% for Groups A, B, and C, respectively ([Figure 1A and B](#f1-cmar-10-3669){ref-type="fig"}). The median OS for the GD regimen from the start of first-line chemotherapy was 17.7 months (95% CI: 12.4--24.7), 17.5 months (95% CI: 14.4--19.5), and 12.7 months (95% CI: 9.3--18.2) for Groups A, B, and C, respectively. Significant differences in survival periods between the three groups were not found. The median follow-up period was 8.2 months (range: 2.1--100). Univariate and multivariate analyses for prognostic factors ----------------------------------------------------------- Baseline parameters for the whole cohort were analyzed by univariate and multivariate analyses to elucidate predictive factors for PFS ([Table 3](#t3-cmar-10-3669){ref-type="table"}) and OS ([Table 4](#t4-cmar-10-3669){ref-type="table"}). We found that ECOG-PS 0 at the start of second-line GD therapy was the only prognostic factor for both PFS (95% CI: 2.55--6.08, HR: 3.94 and 95% CI: 2.49--6.00, HR: 3.86, respectively) and OS (95% CI: 2.48--5.84, HR: 3.80 and 95% CI: 2.34--5.68, HR: 3.64, respectively) from the start of second-line treatment. AEs in second-line GD chemotherapy ---------------------------------- [Table 5](#t5-cmar-10-3669){ref-type="table"} lists all AEs associated with second-line GD therapy in the three groups. Severe hematologic AEs that were the most frequently observed included decreased WBC (42.2% for Group A, 32.1% for Group B, and 42.9% for Group C) and neutropenia (42.2% for Group A, 35.7% for Group B, and 33.3% for Group C), followed by decreased platelets (24.4% for Group A, 30.4% for Group B, and 38.1% for Group C). Significant differences in the occurrence of these AEs between the three groups were not observed. However, the occurrence of anemia in Group C was found to be significantly increased when compared with Group A (19.0% and 6.7%, respectively, *P*\<0.05). The frequencies of non-hematologic AEs, including grade 1--2 fatigue and nausea, were significantly higher in Group C compared with those of Groups A and B. However, grade 2 alopecia showed a significantly higher incidence in Group A compared with Groups B and C. Deaths related to treatment were not noted. Discussion ========== With industrial countries experiencing rapid increases in aging populations, a critical emerging issue is the proper treatment of increasing numbers of elderly patients with cancer. The WHO criterion for elderly patients is those ≥65 years of age. The median age of patients with mUC treated in clinical trials has consistently been between 61 and 64 years.[@b19-cmar-10-3669]--[@b21-cmar-10-3669] However, the median age at the time of diagnosis of UC is \>70 years, witĥ4% of the patients showing metastatic disease.[@b2-cmar-10-3669] Thus, elderly patients with mUC are underrepresented in clinical trials. This underrepresentation has led to uncertainties in the expected level of tolerability of chemotherapy and outcomes following second-line chemotherapy. In this study, we therefore divided our patient cohort into three age groups to examine the efficacy of second-line GD chemotherapy between the groups. To date, only a few limited reports have described second-line chemotherapy in elderly patients with mUC. Noteworthy among these, Salah et al[@b22-cmar-10-3669] pooled the individual data of elderly patients with mUC (over 70 years of age) from 10 studies for the assessment of second-line chemotherapy regimens and demonstrated that combination chemotherapy was associated with greater toxicity without any improvement in OS. However, a report that evaluated the effectiveness of a single regimen of second-line chemotherapy between elderly and younger patients with mUC is lacking. Our pooled data analyses of the response rate and prognostic outcomes of second-line GD chemotherapy for patients with mUC over 75 years of age showed that the ORR and median OS from the start of second-line therapy were 14.3% and 5.3 months (95% CI: 3.4--7.9), respectively, which were similar to the results found in younger patients. In this study, gemcitabine was consistently used in two consecutive regimens as it was the main focus of our strategy of sequential chemotherapy. Recently, mechanisms of acquisition of chemoresistance to gemcitabine in urothelial cancer cells have been outlined.[@b23-cmar-10-3669],[@b24-cmar-10-3669] However, gemcitabine shows a synergistic effect when combined with different chemotherapeutic agents, and several studies have consistently used gemcitabine in sequential therapy,[@b25-cmar-10-3669],[@b26-cmar-10-3669] as in our chemotherapeutic strategies for mUC. As a result, GD therapy after the failure of gemcitabine and platinum-based combination therapy showed good anti-tumor effects: the total ORR waŝ20% and the median OS was 7.6 months for the entire cohort receiving second-line GD therapy. So far, the most extensively studied second-line combination chemotherapy regimen is gemcitabine and paclitaxel (GP). However, like other treatments, GP regimen has shown variable results. Our results are in line with recent reports demonstrating that in patients treated with a second-line combination regimen including GP regimen for mUC, the ORR was 8.6%--41.7% and median OS was 4.8--12.4 months.[@b8-cmar-10-3669],[@b19-cmar-10-3669],[@b21-cmar-10-3669],[@b27-cmar-10-3669]--[@b31-cmar-10-3669] Compared with these results, the oncological outcomes of our study in elderly patients were encouraging with regard to a durable response rate and survival, therefore, supporting the use of second-line GD therapy in elderly patients with a comorbidity. It is well known that elderly patients may be less tolerant to chemotherapy because of comorbidities as well as decreased organ and hematologic functions; therefore, special attention must be paid to the occurrence of AEs. For example, in lung cancer, although the response rates of older and younger patients were comparable (80% vs 88%, respectively; *P*=0.11), severe hematologic toxicity, defined as grade 4--5, was significantly greater in elderly patients (84% vs 61%, respectively; *P*\<0.01) following chemoradiation with a cisplatin and etoposide regimen.[@b32-cmar-10-3669] In patients with UC who received gemcitabine plus cisplatin in a neoadjuvant setting, Jan et al[@b33-cmar-10-3669] described that anemia, neutropenia, and thrombocytopenia were the main reasons for a dose reduction, with 7 out of 28 patients aged ≥65 years discontinuing chemotherapy because of decreased tolerability. In this study, the frequencies of grade 3--4 anemia as well as grade 1--2 fatigue and nausea were significantly higher in the ≥75-year-old group; however, they soon recovered after a transfusion and dose reduction. Furthermore, the GD regimen was well tolerated and could be safely used in patients who were previously treated for impaired renal function. In fact, patients with UC often have impaired renal function due to advanced age, prior platinum-containing chemotherapy, prior nephrectomy, and/or disease-related hydronephrosis. Despite the small patient sample of this study, in addition to an oncological outcome, our results show that second-line GD therapy may be a feasible and tolerable option for elderly patients even when renal function is insufficient and they have a comorbidity. Nowadays, five immune checkpoint inhibitors have been approved by the FDA for the treatment of mUC in a second-line setting after platinum-based chemotherapy: atezolizumab,[@b6-cmar-10-3669],[@b34-cmar-10-3669] durvalumab,[@b35-cmar-10-3669] avelumab,[@b36-cmar-10-3669] nivolumab,[@b37-cmar-10-3669] and pembrolizumab.[@b38-cmar-10-3669] These showed an ORR of 15%--21.1% and median OS of \~10 months. Approximately 10% of severe treatment-related AEs were directly linked to an early immune response within a month after the initiation of treatment despite the similar characteristics of study populations, with a similar median age of 67 years. In general, these new immunotherapies are well tolerated and effective in elderly patients with mUC; however, data are, as yet, insufficient to reach any firm conclusion regarding treatment outcomes. Another important issue is how to select the best predictive markers associated with a clinical response in order to choose appropriate immunotherapies or chemotherapies because of the limitation of the response rate as outlined earlier. Therefore, based on this study, a prospective randomized trial as second-line therapy in patients with mUC comparing a GD regimen and pembrolizumab after first-line platinum-based chemotherapy was recently initiated (UMIN ID: R000037420). It is hoped that the results of this trial will be reported in due course. In this study, several clinical parameters were measured in conjunction with PFS and OS, such as age, gender, eGFR, RDI, ECOG-PS, and the presence of visceral metastasis. Both univariate and multivariate analyses demonstrated that only ECOG-PS 0 at the start point of second-line GD therapy was a strong prognostic factor for PFS and OS. Thus, our findings suggest that rather than discontinuing chemotherapy entirely among those who are unable to tolerate full-dose therapy, elderly patients with mUC aged ≥75 years with a good ECOG-PS can still safely and adequately be treated by continuing with second-line GD chemotherapy and reducing doses. Bearing in mind the safety and other advantages of using second- line GD chemotherapy observed in this study, as well as its cost-effectiveness compared with the use of checkpoint inhibitors, further prospective trials, including this study, with the addition of biomarkers to help select patients are warranted to evaluate this strategic chemotherapeutic approach for elderly patients with mUC. Several limitations were evident in this study. First, the patient cohort used was relatively small, and second, the study itself was undertaken in a retrospective manner. In spite of this, for patients with mUC previously treated with platinum- based therapy, combination therapy with GD appears to be well tolerated and shows activity against disease even in selected elderly patients ≥75 years of age. Conclusion ========== This study concludes that for patients with mUC previously treated with platinum-based chemotherapy, combination therapy with GD appears to be well tolerated and shows activity against disease even in elderly patients. The authors wish to state that they did not receive any financial support for this study. **Author contributions** Taku Naiki made critical revisions of the manuscript. Toshiki Etani, Takashi Nagai, Yutaro Tanaka, RyosUke Ando, Shuzo Hamamoto, Rika Banno, Daisuke Nagata, Noriyasu Kawai, and Yosuke Sugiyama carried out the acquisition of data and coordinated and helped to draft the manuscript. Keitaro Iida conducted statistical analyses concerning this study. Takahiro Yasui supervised this manuscript. All authors contributed toward data analysis, drafting and critically revising the paper and agree to be accountable for all aspects of the work. **Disclosure** The authors report no conflicts of interest in this work. {#f1-cmar-10-3669} ###### Patients' characteristics and oncological outcomes in first-line chemotherapy in the three age groups --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Characteristics Group A (n=45), age \<65 years Group B (n=56), age from 65 to 74 years Group C (n=21), age ≥75 years *P*-value ----------------------------------------------------------- -------------------------------- ----------------------------------------- ------------------------------- ---------------------------------------------------- ------------------------------------------------------ ---- Median age, years (range) 59 (37--64) 69 (65--74) 79 (75--82) ns Originating organ of urothelial carcinoma, n (%) Upper urinary tract 5 (11.1) 6 (10.7) 6 (28.6) ns Bladder 40 (88.9) 50 (89.3) 15 (71.4) Gender, n (%) Male 35 (77.8) 45 (80.4) 18 (85.7) ns Female 10 (22.2) 11 (19.6) 3 (14.3) First-line Metastatic site, n (%) Lymph node alone 24 (53.3) 31 (55.4) 11 (52.3) ns Visceral metastasis 21 (46.7) 25 (44.6) 9 (47.7) ECOG-PS, n (%) 0 37 (82.2) 45 (80.4) 18 (85.7) ns 1 8 (17.8) 11 (19.6) 3 (14.3) Median eGFR (mL/min)/1.73 m^2^ (range) 63.1 (8.0--123.0) 55.5 (17.8--104.9) 42.6 (30.9--61.6) [\*\*\*](#tfn1-cmar-10-3669){ref-type="table-fn"}\ [\#](#tfn3-cmar-10-3669){ref-type="table-fn"} Chemotherapy regimen, n (%) Cisplatin-based 43 (95.6) 50 (89.3) 15 (71.4) \# Carboplatin-based 2 (0.4) 6 (10.7) 6 (28.6) RDI, n (%) ≥ 90% 38 (84.4) 25 (44.6) 4 (19.0) [\*\*\*\*](#tfn2-cmar-10-3669){ref-type="table-fn"}\ [\#\#](#tfn4-cmar-10-3669){ref-type="table-fn"}\ [\$](#tfn5-cmar-10-3669){ref-type="table-fn"} \<90% 7 (15.6) 31 (55.4) 17 (81.0) Response of first-line chemotherapy, n (%) PR 18 (40.0) 18 (32.1) 5 (23.8) ns SD 13 (28.9) 20 (35.7) 9 (42.9) PD 14 (31.1) 18 (32.1) 7 (33.3) Median total cycles of first-line chemotherapy, n (range) 4 (2--10) 3 (1--15) 4 (1--8) ns --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Notes:** *P*\<0.001, *P*\<0.0001, statistically significant comparing Group A vs Group C. *P*\<0.05, *P*\<0.01, statistically significant comparing Group B vs Group C. *P*\<0.05, statistically significant comparing Group A vs Group B. **Abbreviations:** ECOG-PS, Eastern Cooperative Oncology Group performance status; eGFR, estimated glomerular filtration rate; GD, gemcitabine and docetaxel; ns, not significant; PD, progressive disease; PR, partial response; RDI, relative dose intensity; SD, stable disease. ###### Patients' characteristics and oncological outcomes in second-line chemotherapy in the three age groups ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Characteristics Group A (n=45), age \<65 years Group B (n=56), age from 65 to 74 years Group C (n=21), age ≥75 years *P*-value ----------------------------------------------------------------------- -------------------------------- ----------------------------------------- ------------------------------- ----------------------------------------------- -------------------------------------------------- ---- Median CCI scores at the start of second-line GD chemotherapy (range) 0 (0--2) 0 (0--3) 1 (0--2) \*\*\ \# Second-line Metastatic site, n (%) Lymph node alone 16 (35.6) 21 (37.5) 8 (38.1) ns Visceral metastasis 29 (64.4) 35 (62.5) 13 (61.9) ECOG-PS, n (%) 0 25 (55.6) 37 (66.1) 11 (52.4) ns 1 20 (44.4) 19 (33.9) 10 (47.6) Median eGFR (mL/min)/1.73 m^2^ (range) 55 (8.0--101.5) 51 (26.5--85.0) 45 (21.4--78.1) [\*](#tfn7-cmar-10-3669){ref-type="table-fn"} RDI, n (%) ≥ 90% 22 (48.9) 13 (23.2) 2 (9.5) [\*\*](#tfn8-cmar-10-3669){ref-type="table-fn"}\ [\#\#](#tfn10-cmar-10-3669){ref-type="table-fn"} \<90% 23 (51.1) 43 (76.8) 19 (90.5) Response of second-line chemotherapy, n (%) PR 10 (22.2) 8 (14.3) 3 (14.3) ns SD 11 (24.4) 23 (41.1) 5 (23.8) PD 24 (53.3) 25 (44.6) 13 (61.9) Median total cycles of second-line GD chemotherapy, n (range) 3 (1--16) 3 (1--15) 2 (1--11) ns ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ **Notes:** *P*\<0.05, *P*\<0.01, statistically significant comparing Group A vs Group C. *P*\<0.05, *P*\<0.01, statistically significant comparing Group B vs Group C **Abbreviations:** CCI, Charlson comorbidity index; ECOG-PS, Eastern Cooperative Oncology Group performance status; eGFR, estimated glomerular filtration rate; GD, gemcitabine and docetaxel; ns, not significant; PD, progressive disease; PR, partial response; RDI, relative dose intensity; SD, stable disease. ###### Univariate and multivariate analyses of baseline parameters to find prognostic factors for progression-free survival in 122 patients treated with second-line GD chemotherapy Parameters Univariate Multivariate -------------------------------------------------- ------------ -------------- ------------------------------------------------------ ------ ------------ ------------------------------------------------------ Age, \<74 vs ≥75 years 1.29 0.79--2.12 0.31 1.16 0.69--1.97 0.70 Gender, male vs female 1.01 0.64--1.58 0.98 0.92 0.58--1.47 0.73 eGFR at the start of second-line GD, \<60 vs ≥60 0.92 0.64--1.33 0.68 1.03 0.70--1.52 0.36 RDI of second-line GD, \<90 vs ≥90 0.87 0.58--1.30 0.50 0.82 0.54--1.26 0.36 ECOG-PS at the start of second-line GD, 0 vs 1 3.94 2.55--6.08 [\*\*\*\*](#tfn12-cmar-10-3669){ref-type="table-fn"} 3.86 2.49--6.00 [\*\*\*\*](#tfn12-cmar-10-3669){ref-type="table-fn"} Visceral metastasis, yes vs no 1.44 0.98--2.11 0.64 1.26 0.84--1.89 0.09 **Note:** *p*\<0.0001 indicates a significant difference. **Abbreviations:** ECOG-PS, Eastern Cooperative Oncology Group Performance Status; eGFR, estimated glomerular filtration rate; GD, gemcitabine and docetaxel; HR, hazards ratio; RDI, relative dose intensity. ###### Univariate and multivariate analyses of baseline parameters to find prognostic factors for overall survival in 122 patients treated with second-line GD chemotherapy Parameters Univariate Multivariate -------------------------------------------------- ------------ -------------- ------------------------------------------------------ ------ ------------ ------------------------------------------------------ Age, \<74 vs ≥75 years 1.34 0.80--2.22 0.27 1.12 0.64--1.94 0.70 Gender, male vs female 0.86 0.53--1.39 0.54 0.76 0.46--1.23 0.26 eGFR at the start of second-line GD, \<60 vs ≥60 0.84 0.57--1.24 0.37 0.92 0.60--1.41 0.70 RDI of second-line GD, \<90 vs ≥90 0.90 0.59--1.36 0.61 0.82 0.53--1.28 0.39 ECOG-PS at the start of second-line GD, 0 vs 1 3.80 2.48--5.84 [\*\*\*\*](#tfn14-cmar-10-3669){ref-type="table-fn"} 3.64 2.34--5.68 [\*\*\*\*](#tfn14-cmar-10-3669){ref-type="table-fn"} Visceral metastasis, yes vs no 1.71 1.13--2.58 0.37 0.92 0.60--1.41 0.70 **Notes:** *p*\<0.0001 indicates a significant difference. **Abbreviations:** ECOG-PS, Eastern Cooperative Oncology Group Performance Status; eGFR, estimated glomerular filtration rate; GD, gemcitabine and docetaxel; HR, hazards ratio; RDI, relative dose intensity. ###### Adverse events in 122 patients treated with second-line GD chemotherapy -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Adverse events Group A (n=45) age \< 65 Group B (n=56) 65 ≤ age \<75 Group C (n=21) age ≥ 75 *P*-value ------------------------- -------------------------- ------------------------------ ------------------------- ----------- ----------- ----------- ------------------------------------------------------- ---- **Hematologic** **Decreased WBC** 44 (97.8) 19 (42.2) 52 (92.9) 18 (32.1) 18 (85.7) 9 (42.9) ns **Neutropenia** 43 (95.6) 19 (42.2) 52 (92.9) 20 (35.7) 18 (85.7) 7 (33.3) ns **Decreased Platelets** 43 (95.6) 11 (24.4) 54 (96.4) 17 (30.4) 20 (95.2) 8 (38.1) ns **Anemia** 23 (51.1) 3 (6.7) 33 (58.9) 8 (14.3) 16 (76.2) 4 (19.0) [\*](#tfn17-cmar-10-3669){ref-type="table-fn"} **Non-hematologic** **Elevated amylase** 4 (8.9) 0 (0) 5 (8.9) 0 (0) 2 (9.5) 0 (0) ns **Rash** 8 (17.8) 0 (0) 4 (7.1) 0 (0) 3 (14.3) 0 (0) ns **Fatigue** 13 (28.9) 0 (0) 24 (42.9) 0 (0) 19 (90.5) 0 (0) [\*\*\*\*](#tfn19-cmar-10-3669){ref-type="table-fn"}\ ~\#\#\#\#~ **Elevated AST/ ALT** 3 (6.7) 0 (0) 7 (12.5) 1 (1.8) 1 (4.8) 0 (0) ns **Constipation** 5 (11.1) 0 (0) 12 (21.4) 0 (0) 5 (23.8) 0 (0) ns **Nausea** 11 (24.4) 0 (0) 18 (32.1) 0 (0) 13 (61.9) 0 (0) [\*](#tfn17-cmar-10-3669){ref-type="table-fn"}\ ~\#~ **Gastritis** 8 (17.8) 0 (0) 8 (14.3) 0 (0) 8 (38.1) 0 (0) ns **Alopecia** 28 (62.2) 0 (0) 19 (33.9) 0 (0) 5 (23.8) 0 (0) [\*\*\*](#tfn18-cmar-10-3669){ref-type="table-fn"}\ \$\$\$ -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Notes:** *P*\<0.001, statistically significant comparing Group A vs Group B. *p*\<0.05, *p*\<0.001, and *p*\<0.0001, statistically significant comparing Group A vs Group C. *P*\<0.05 and *P*\<0.0001, statistically significant comparing Group B vs Group C. **Abbreviations:** ALT, alanine aminotransferase; AST, aspartate transaminase; GD, gemcitabine and docetaxel; ns, not significant; pt, patient; WBC, white blood cell.

Introduction {#Sec1} ============ Many studies related to nanoparticles (NPs) show that their release has a negative influence on ecosystems and human health (Nel et al. [@CR20]; Cox et al. [@CR9]). Despite the fears, knowledge on this subject is still insufficient. In the case of plant growth and development, NPs are noted to have a positive and a negative influence (Montes et al. [@CR18]). Rhizoplane microbiome is an important factor in the development of plants but for their part, microorganisms, being particularly sensitive to changes in the environment, in most cases respond negatively to NPs (Hegde et al. [@CR14]). NPs of silver (Beddow et al. [@CR6]) and of titanium oxide (Ge et al. [@CR13]) have been found to be toxic to bacteria in vitro. Other effects may occur in the soil, where physicochemical properties, texture, and organic matter, may change the properties of NPs. Such an interaction may result in increased or decreased bioavailability and toxicity of NPs for plants and microorganisms (Dimkpa [@CR10]). Many studies confirm that exposure to NPs does not change the structure and function of soil microbial communities, while others determine the impact to be negative (Simonin and Richaume [@CR22]). Microorganisms in the rhizosphere influence plants in a number of ways: they activate growth, nutrition, development, susceptibility to diseases, resistance to heavy metals, and degradation of xenobiotics. The microbiota occurring in the roots of plants above all mobilizes and provides nutrients by increasing their bioavailability. A non-nutritional function is greater resistance to biotic and probably abiotic stresses too (Bulgarelli et al. [@CR7]). The results of the interaction of NPs with biological systems vary, depending on the experimental environment. The influence of NPs in combined plant-microbial systems has been assessed in a small number of studies, compared to the relatively large number of separate assessments pertaining to plants and microorganisms (Dimkpa [@CR10]). For this reason, the aim of the research was to preliminary assess a plant**-**rhizoplane microbiome system, which was subjected to selected NPs, found to be particularly useful in many technological applications, i.e.**,** silver nanoparticles (AgNPs) and titanium oxide nanoparticles (TiO~2~NPs). These nanoparticles have in numerous cases been shown to have a negative effect in vitro. The studies intentionally differentiated plants species to Monocot (wheat) and Dicot (flax) growth form using one concentration of NP. Differences in the development of root system and cell wall chemistry between Monocot and Dicot may affect rhizoplane microbiome. This test was carried out in soil conditions. The intentions of evaluating initial growth (germination, seedlings biometry, relative chlorophyll content) and basic parameters of the rhizoplane microbiome (total bacteria, *Pseudomonas*, *Bacillus*, *Clostridium*, toxinogenic *Penicillium*) was to select the hot spot of the NPs activity in systems. This will allow focusing detailed, systematic studies on the impact of NPs on the biological system of the plant**-**rhizoplane microbiome in soil conditions. Materials and methods {#Sec2} ===================== All chemical reagents used in the experiments (silver nitrate, tannic acid, ammonia solution 25%) are commercial products available from Sigma Aldrich (Switzerland). Similarly, the titanium (IV) oxide anatase (TiO~2~NPs1) and titanium (IV) oxide nanopowder (TiO~2~NPs2) were purchased from Sigma Aldrich. These compounds were used without further purification. Aqueous solutions of these compounds were prepared with the use of ultrapure water (conductivity of 0.06 μS·cm^−1^) obtained from a Milli-Q Elix & Simplicity 185 purification system (Millipore SA Molsheim, France). A suspension of silver NPs was prepared according to the procedure described in our previous work (Barbasz et al. [@CR3]). Briefly, the silver nitrate aqueous solution (320 mL, 1 mM) was mixed with the tannic acid solution (10 mL, 0.6 mM) and then the pH of the mixture was adjusted using ammonia solution to the value of 9. After 1 hour of mixing at room temperature, the silver nanoparticle suspension obtained was purified using an ultra-filtration method. The mass concentration of silver in the suspensions was determined using a DMA5000M density meter (Anton Paar, USA), as described previously (Barbasz et al. [@CR3]). In the experiments with the plant culture, the suspensions of TiO~2~NPs and of AgNPs were applied in concentrations of 100 mg·L^−1^. Similarly, the silver nitrate solution with a concentration of silver ions equal to 100 mg·L^−1^ was applied. The electrical conductivity and pH of the colloidal suspension were measured using CPC-505 multifunctional device (Elmetron, Poland) equipped with an EC-60 conductivity sensor and an EPS-1 glass electrode (Elmetron, Poland). The morphology and size distribution of the applied NPs were determined by means of a microscope examination method using a JEOL JSM-7500F microscope (USA). The silver NPs were visualized in transmission mode (TEM). For this purpose, a drop of the silver suspension was placed on a copper grid coated with carbon film. The solvent was evaporated at an elevated temperature (50 °C) and the samples were visualized. In the case of the titanium dioxides, the powders supplied were glued to the special conductive holder using carbon tape and were imaged in the secondary electron mode (SEM). The micrographs were analyzed using MultiScan software supplied by Computer Scanning System. The size of the silver NPs and of both types of the titanium dioxide NPs, in the suspension of a concentration of 100 mg·L^−1^, was determined by the dynamic light scattering (DLS) technique via diffusion coefficient measurements carried out using a Zetasizer Nano ZS instrument (Malvern, UK). The hydrodynamic diameter of the NPs was calculated based on the Stokes--Einstein relationship (Barbasz et al. [@CR3]). The electrokinetic properties of the NPs under controlled conditions of pH, ionic strength, and temperature were determined via electrophoretic mobility measurements carried out using the Zetasizer Nano ZS instrument. The zeta potential of the NPs was calculated using Henry's law (Barbasz et al. [@CR3]). The concentration of silver ions released from the NPs after 14, 30, and 60 days was determined using Seven Compact™ pH/ionometer (Mettler Toledo, Poland) equipped with a perfectION™ silver/sulfide electrode. The dissolved oxygen (DO) concentration in the suspensions of controlled pH, ionic strength, and the silver NPs concentration was measured using a COG-1 t oxygen probe connected with a CPO-505 oxygen meter (Elmetron, Poland). In order to separate the NPs from silver ions, the ultrafiltration method described in our previous work (Barbasz et al. [@CR4]) was applied. The plants used in the experiments were as follows: common spring wheat cv. Bombona and common flax cv. Nike. Sixty seeds of tested plants were put in plastic pots filled with 0.8 kg of homogeneous soil with about 40% maximum water capacity. Next, the soil was evenly soaked with solution (1 mL in 1 L sterile water). Control objects were soaked with sterile water only. Seedlings were allowed to grow up in a phytotron under controlled conditions at 20 °C day and 18 °C night temperature, a 12-h photoperiod with a light intensity of 220 μmol photons m^−2^·s^−1^ PPFD, and 80% RH. The soil was supplemented with water every 3 days in the same amount for each treatment. First, 5 days of growth was allowed to proceed with foil cover and then the germination of seeds was controlled. Germination percentage was measured using the following formula: germination \[%\] = (number of seed germinated/total number of seeds) × 100. After 7 days, the plants were sprayed with the same solutions that were applied the first time to the soil. After 21 days, the plants were analyzed to determine their biometric and relative chlorophyll content (Chlorophyll Meter SPAD-502 Plus Konica Minolta INC, Japan). Total rhizoplane bacteria, *Clostridium* spp. and toxin-producing *Penicillium* were amplified with the Maxima Probe qPCR Master Mix (2X) (Thermo Fisher Scientific, USA) according to the methods used by Yu et al. ([@CR28]), Song et al. ([@CR23]), and Suanthie et al. ([@CR24]), respectively. The phlD gene of PGP (plant growth promoting) *Pseudomonadaceae* was determined according to the method proposed by Hu et al. ([@CR15]). To determine *Bacillus* spp. 16SBACF and 16SBACR primers (Mora et al. [@CR19]) were used. Both *Pseudomonadaceae* and *Bacillus* amplified with the Maxima Sybr Green qPCR Master Mix (2X) (Thermo Fisher Scientific, USA). Efficiency of reaction was 0.91 to 1.00. The resulting data was analyzed using *STATISTICA* software. Duncan's multiple range test (*p* \< 0.01) was used to compare the means. The principal component analysis (PCA) was performed by Pearson (correlation) method. Agglomerating hierarchical clustering was performed by Bray and Curtis dissimilarity statistics with Ward's agglomeration method. Both calculations were made with XLSTAT program (Addinsoft, UK). Results and discussion {#Sec3} ====================== Selected properties of the colloidal suspensions and the NPs were determined and it is shown in Table [1](#Tab1){ref-type="table"}. The response of the plant-rhizoplane microbiome system to NPs is presented in Tables [2](#Tab2){ref-type="table"} and [3](#Tab3){ref-type="table"}. The treatments applied had a limited influence on the germination and early development of the shoots of wheat and flax seedlings compared to the control. This response was different for the two plants tested (Monocot and Dicot growth forms). Enhanced seed germination and seedling growth was only obtained in wheat treated with TiO~2~NPs1. Feizi et al. ([@CR12]) reported significantly faster germination of wheat in response to treatment with TiO~2~NPs and higher shoot and seedling lengths than with untreated wheat, but only at low concentrations, i.e., of 1--2 mg·L^−1^ and no significant effect at concentration of 100 mg·L^−1^. In our experiment, the response of the Dicot growth form, i.e., flax, was weaker and only an enhanced trend was observed. Often, TiO~2~NPs have been noted to increase the germination rate and growth of some plants (Zhang and Karan [@CR29]; Lyu et al. [@CR17]) but under laboratory, in vitro conditions. Positive effects of TiO~2~NPs on germination and early growth of plants cultivated in soil conditions were found in our experiment but only in some cases. The NPs applied did not modify the dry mass of either of the plants in our experiment. Differences were noted in the greenness of leaves which indicates changes in the chlorophyll content of the treated plants. AgNO~3~ decreased the chlorophyll content of both plants treated. The response of flax to NPs was noted as an increase in chlorophyll content. The total bacterial community in wheat rhizoplane was not significantly modified, but there was a declining trend. In turn, there was a modification in the total quantity of bacteria in the flax rhizoplane in the case of the treatments with NPs. The difference in surface charge determined in the characteristics of the compounds used had a decisive influence on the bacterial community in the flax rhizoplane. Positively charged TiO~2~NPs2 significant decreased the quantity of bacteria in contrast to the negatively charged AgNPs and TiO~2~NPs1 which increased it. Abbaszadegan et al. ([@CR1]) have proven that the surface charge of AgNPs is a significant factor affecting bactericidal activity-positively charged NPs showed a high level of effectiveness against five species of pathogenic bacteria. Those authors explain the response based on van der Wal et al. ([@CR26]), citing the fact that the cellular membrane of the bacteria has a negative charge due to the presence of carboxyl, phosphate, and amino groups. The repulsion between the bacteria and negatively charged NPs could result in the formation of an electrostatic barrier by limiting the interaction between the NPs and the bacteria. The same conclusion is presented by Zhu et al. ([@CR30]). The cell membrane consists of an anionic hydrophilic outer surface. In contrast to neutral or anionic NPs, cationic particles attach more readily to the cell surface, from where they may also be taken up more avidly if size permits.Table 1Selected physicochemical properties of NPs suspensionPropertiesType of colloidAgNPsTiO~2~NPs1TiO~2~NPs2Concentration \[mg L^−1^\]100100100Conductivity \[μS cm^−1^\]9.45.48.6pH5.95.45.5Particle size \[nm\]\*16 ± 4207 ± 1768 ± 7Diffusion coefficient \[cm^2^ s^−1^\]\*\*3.1·10^−7^1.1·10^−8^6.1·10^−9^Hydrodynamic diameter \[nm\]\*\*17 ± 3441 ± 40812 ± 30Electrophoretic mobility \[(μmcm) (Vs)^−1^\]\*\*− 3.41 ± 0.09− 3.74 ± 0.05+ 2.13 ± 0.04Zeta potential \[mV\]\*\*− 63 ± 2− 58 ± 131 ± 1Concentration of leached silver ions from suspensions\*\*DO = 7.4 mg L^−1^ after 14/30/60 days7 ± 1.4/11 ± 2.2/10 ± 3.2----*DO* dissolved oxygen\*Determined from size distribution derived from TEM micrographs; \*\*for 100 mg L^−1^, and *T* = 298 KTable 2Effect of NPs on tested parameters of plants seedlingsPlantTreatmentGermination \[%\]Leafs weight \[g\]Leafs length \[cm\]Dry matter \[%\]Greennessof leaves \[SPAD\]Common wheatcontrol68b\*4.80b28.21a12.8a34.04abAgNO~3~64b4.91ab27.10a14.2a30.34cAgNPs72b5.19ab27.32a14.9a32.35bTiO~2~NPs184a5.40a29.51a14.3a33.63abTiO~2~NPs264b5.19ab30.25a13.3a34.42aMean705.1028.2113.932.96Common flaxcontrol75ab2.66b9.65a8.05a21.93bAgNO~3~63b0.78c5.62b8.23a15.01cAgNPs77a3.25a9.72a6.85a26.23aTiO~2~NPs178a2.86b9.63a7.85a25.25abTiO~2~NPs280a2.90b9.31a6.85a26.23aMean752.498.786.5819.72\*Values marked with the same letters in columns are not significantly different (*p* ≤ 0.05) according to the multiple Duncan's testTable 3Changes on load of selected group of bacteria in root zone after NPs treatment \[μg·g^−1^ DM\]PlantTreatmentBacterial DNAPGP *PseudomonasBacillus* spp.*Clostridium* spp.Toxinogenic *Penicillium* spp.Common wheatcontrol6.17a\*2.50ab2.6·10^−4^b0.22a52.2·10^−4^cAgNO~3~5.12a0.19b1.3·10^−4^b0.12b0aAgNPs5.49a1.41ab1.4·10^−4^b0.13b8.6·10^−4^bTiO~2~NPs15.63a2.91a1.9·10^−4^b0.16ab0aTiO~2~NPs25.37a2.99a3.9·10^−4^a0.17ab0.1·10^−4^aMean5.562.002.2·10^−4^0.1612.2·10^−4^Common flaxcontrol10.15c0.08b1.5·10^−5^c0.17aNDAgNO~3~7.48e0.09b2.6·10^−5^bc0.06bNDAgNPs13.32a0.54a3.9·10^−5^ab0.07bNDTiO~2~NPs111.43b0.09b2.6·10^−5^bc0.14aNDTiO~2~NPs29.23d0.20b5.7·10^−5^a0.05bNDMean10.320.202.7·10^−5^0.09--*DM* dry matter, *PGP* plant growth promoting, *ND* no detected (obtained results were below the lowest calibration curve values)\*Values marked with the same letters in columns are not significantly different (*p* ≤ 0.05) according to the multiple Duncan's test Further analysis of the bacterial community conducted in our own study for rhizoplanes of wheat and flax showed specific quantitative modifications under the influence of the applied NPs, most often depending on the plant species. The *Pseudomonas* community rich in wheat rhizoplane was stimulated by TiO~2~NPs independently of the surface charge and did not differ from the control under treatment with AgNPs. In turn, significant growth was observed for the community of *Pseudomonas* that was poor in flax rhizoplane under treatment with AgNPs. There was significant growth in the quantity of *Bacillus* spp. in the wheat rhizoplane only under treatment with TiO~2~NPs2, and in the flax rhizoplane, such growth was caused by all NPs used in the study, whereby TiO~2~NPs2 were found to have the most significant effect. *Clostridium* spp. was significantly limited in the rhizoplane of both species of plant by AgNPs, and to a lesser degree by TiO~2~NPs. The study showed toxinogenic *Penicillium* bacteria in the wheat rhizoplane, and all the treatments used reduced it in quantity. The qualitative study conducted did not confirm the surface charge of NPs to have any influence on an increase/decrease in the quantity of bacteria in the rhizoplanes of both species of plant. However it appears to be significant that the tested NPs were not shown to have any toxic effects on the community of plant growth-promoting bacteria. It was also confirmed that the rhizoplane microbiome is dependent on the species of plant (Lemanceau et al. [@CR16]), and that the bacteria found in the communities are sensitive to NPs to a varying degree. Evidence of the toxic effect of some AgNPs on bacteria in a soil system has been provided (Anjum et al. [@CR2]; Tripathi et al. [@CR25]). Exposure of soil bacteria to 100--1000 mg L^−1^ of various types of AgNPs has been shown to significantly inhibit function. Calder et al. ([@CR8]) evaluated *Pseudomonas bacteria* in sand and soil matrices and found evidence of toxicity of AgNPs at a very low concentration in sand only. Organic matter and the changes in surface charge of NPs involved eliminated their toxicity. Our results showed no toxic effects of negatively charged tannin-reduced AgNPs on both the plants tested and rhizoplane bacteria. A toxic effect was observed for the ionic form of Ag and flax was much more sensitive to this. Silver ions had a significant negative effect on the germination and growth of flax seedlings and significantly increased the quantity of bacteria in flax rhizoplane. The results obtained for treatment with TiO2NPs and ionic silver (AgNO3) are consistent with the analysis provided by Cox et al. ([@CR9]). Studies analyzed according to Cox et al. ([@CR9]) have shown that the toxicity of AgNPs is dependent on the properties of the form that results from the method of reduction applied. Most of the silver NPs examined exhibit phytotoxicity and bacteriostaticity at fairly low concentrations. However, some types of AgNP coating resulting from the reduction method applied have a significantly reduced toxic effect on living organisms, which is important for the protection of the environment. Yang and Watts ([@CR27]) indicated that surface area and surface characteristics play an important role in the phytotoxicity of AuNPs too. Razzaq et al. ([@CR21]), by using soil-applied AgNPs synthesized by a reduction of AgNO~3~ with trisodium citrate at a concentration 100 mg L^−1^ shows higher weights of wheat seedlings and higher chlorophyll content too. But the same NPs caused significant reduction in wheat germination in comparison to the control under in vitro conditions. The effects of AgNPs on the inhibition of germination have been reported for many plants in not-soil conditions (Barrena et al. [@CR5]; El-Temash and Joner [@CR11]). This shows that there is another decisive factor in addition to the surface properties of NPs, i.e., the conditions under which plants are treated. The principal component analysis (PCA) presented in Fig. [1](#Fig1){ref-type="fig"}a confirmed the lack of correlation in the response to NP between wheat (CW) and flax (CF), but average correlation for CW and strong for CF inside group. The main parameter variable for CW and CF was germination, and in less scale SPAD and leaf biometry. These results indicate that NPs have a strong effect on germination and medium on greenness of leaves irrespective of the Monocot and Dicot form plants. The rhizoplane microbiome parameters were a not strong variable, which indicates a poor microbiome response to NP in soil conditions. This allows us to conclude that in the plant-microbiome system in soil conditions, the more effective response induce NP at the level of plant physiology. Results of agglomerative hierarchical clustering (AHC) presented in Fig. [1](#Fig1){ref-type="fig"}b demonstrate differentiation in response of both plant group. In case of wheat, CW-control, CW-TiO~2~NPs1, and CW-TiO~2~NPs2 were not variable, but CW-AgNO~3~ and CW-AgNPs were different. In case of flax, were formed three groups: first CF-control/CF- TiO~2~NPs2, second CF-AgNO~3,~ and third CF-AgNPs/CF-TiO~2~NPs1. Results indicated that AgNO~3~ strong influence on microbiome, nevertheless the plant type was main impact on fate of chemical compounds. Next, grouped in analysis negatively charged AgNPs and TiO~2~NPs1 in CF case indicated important influence of surface charge. However, microbiological AHC method showed some differentiation between those parameters, the PCA indicated smaller weight in these variables with reference to other analysis parameters.Fig. 1**a** Principal component analysis (PCA) shows correlation between objects (red) and discrimination factors (blue) and **b** agglomerative hierarchical clustering (AHC) describe dissimilarity between objects in microbiological parameters contacts. CW common wheat, CF common flax Conclusions {#Sec4} =========== Our experiment above all shows the importance of the test plants used in evaluation of NP influence. First of all, the response of Monocot and Dicot growth form plants to the tested NPs was different. Secondly, rhizoplane microbiome is dependent on the species of plant, and that the bacteria found in the communities are sensitive to NPs to a varying degree. We also found that the surface charge of metal NPs under soil conditions matter for rhizoplane microbiome. Positively charged NPs significantly decreased the quantity of total bacteria in contrast to negatively charged NPs which increased their quantity in Dicot rhizoplane. We are grateful to Dr. Elżbieta Bielańska for her invaluable help in the TEM measurements. The study was supported by University of Agriculture in Krakow Grant of Department of Agricultural Environment Protection No DS 3109/KOSR and by the Research Grant: the Programme Innovative Economy 01.01.02-12-028/09-00. [^1]: Responsible editor: Zhihong Xu

Background {#Sec1} ========== Unintentional injuries among adolescents are a public health concern for several reasons, including the health impacts on an already susceptible, vulnerable subpopulation (twenty-one and younger) and the multifaceted issues leading to these unintentional injuries. One particular area of concern are injuries involving students in supervised, school-sponsored career-technical-vocational education (CTE) programs, and how socioeconomic status (SES) may relate to the occurrence of these injuries. Currently, there is a wealth of information supporting the association between SES and unintentional injuries among adolescents involving motor vehicle-related injuries, recreational-related injuries, falls, and sports-related injuries \[[@CR1]--[@CR4]\]; however, little is known about the association between SES and work-related injuries in adolescents, specifically injuries among students enrolled in CTE programs. Current literature has suggested an inverse association between SES and prevalence of adolescent work-related injuries, where lower SES is associated with a higher prevalence of adolescent work related injuries \[[@CR5]\]. Research also has suggested SES influenced how young adults were treated in the work place by employers (e.g., being assigned dangerous tasks and a lack of overall supervision), which can have a direct impact on work place injuries \[[@CR6], [@CR7]\]. It has been estimated 70--80 % of teens work at some point during their high school career \[[@CR8]\]. Because of their inexperience, and in many instances, their lack of knowledge concerning safety and health (S&H) topics, young workers have about twice the risk of incurring job-related injuries compared to older adults \[[@CR8]--[@CR12]\]. Other factors leading to higher rates of injury among young workers included a lack of physical and/or emotional maturity, lack of proper safety training, being unfamiliar with standard operating procedures for certain tasks, and how young workers maybe not be aware of legal limitations imposed on workers under age 18, including prohibited tasks and prohibited equipment under child labor laws \[[@CR8], [@CR13], [@CR14]\]. The highest rates of work-related injuries have occurred in the 18--24 year old bracket, followed by 15--17 year olds \[[@CR15]\]. It has been estimated in the United States (U.S.) each year about 160,000 young workers experience occupational related injury or illness; however, the National Institute of Occupational Safety and Health acknowledged how these numbers are widely underestimated and approximately two-third of work-related incidents are unreported \[[@CR15]\]. CTE programs offer a great opportunity to prepare students to enter the work force. It has been estimated there are over 20,000 CTE vocational and ready to work programs in the U.S. \[[@CR16]\]. Students in CTE programs are more likely to report having received safety training and having been informed of their legal rights compared to teens working outside of these structured programs \[[@CR17], [@CR18]\]. The U.S. Office of Vocational and Adult Education estimated, on average, every high school student has taken at least one CTE course, and 1-in-4 students have taken three or more courses in a single program area \[[@CR16]\]. Clearly, CTE programs encompass a large percentage of young adults. Understanding injuries related to these programs and their association with SES is of public health significance. Past evidence suggested children of families in the lowest income bracket have the highest rates of unintentional, nonfatal injuries; it is unclear, however, if this association holds true for adolescents enrolled in structured work programs, such as CTE programs \[[@CR19]\]. Reducing the prevalence of work related injuries in this subpopulation would improve their S&H and also decrease medical expenditures, as unintentional injuries are a prominent source of medical spending for youth aged 21 years and younger in the U.S. \[[@CR19]\]. SES is generally acknowledged as an important predictor of health status \[[@CR6], [@CR20]--[@CR22]\]. Proxies of SES utilized throughout the literature usually encompassed some form of measurement of finance, education, and occupation. SES as it relates to chronic disease \[[@CR20], [@CR23]\] mental health, \[[@CR24], [@CR25]\] and infectious disease \[[@CR26], [@CR27]\] is relatively well studied; however, further investigation is warranted to examine potential associations between SES and unintentional injuries among young workers. Moreover, a literature review performed by Cubbin et al. in 2002 reported SES was an important predictor of injury; nevertheless, the direction and strength of this association depended on indicators used to measure SES \[[@CR28]--[@CR30]\]. The New Jersey (NJ) Department of Education, based on NJ Administrative Code 6A:19--6.5, requires by law for accidents/incidents (injury or illness) involving CTE students and/or staff that are treated by a licensed physician, physician's assistant, or advanced practice nurse to be reported to the NJ Commissioner of Education \[[@CR31]--[@CR33]\]. These incidents are directly reported to the NJ Safe Schools Program (NJSS) online surveillance system (via Psychdata) for aggregate analyses. Data from submitted individual injury reports are collected on several key variables including reporting county, school district, and school name. Information is also collected on the gender; whether the injured person was a staff or student; the title of the co-op/structured learning experience program; and where the injured person was treated (doctor/clinic versus hospital/emergency department (ED)). Moreover, parts of body injured, nature of injury, cause of injury, severity of injury, use of personal protective equipment (PPE), and date and time of the injury are recorded \[[@CR31]--[@CR33]\]. Here, we examined potential associations between SES indicators and work-related injuries among adolescents, specifically injuries involving CTE programs in the state of NJ, excluding injuries acquired by school staff members. We hypothesized SES would be associated with several key variables examined including injury cause, injury location on the body, injury type, injury severity, use of PPE, where an injury was treated and gender of the injured individual. Specifically, we predicted individuals in lower SES schools/school districts would have more severe injuries, use PPE less often, or use PPE incorrectly compared to higher SES schools/school districts, and have an increased odds of being treated at a hospital. To test this hypothesis, we utilized data collected through the NJSS incident reporting surveillance system between the years 1998--2013 and District Factor Groups (DFG) scores, a proxy of SES \[[@CR34]\]. Methods {#Sec2} ======= This data analysis represents aggregate injury surveillance data; no personal, identifying information was utilized. The Rutgers University-New Brunswick Institutional Review Board human subjects approved protocol number is 021997W0383. In NJ, DFG scores are a proxy used to estimate a community's relative SES, as calculated using six different variables to estimate SES. The variables are: percent of adults with no high school diploma, percent of adults with some high school education, occupational status, unemployment rate, percent of individuals in poverty, and median family income. Data for these variables were collected from the most recent U.S. Decennial Census. Statistical analyses were conducted within SAS 9.4 (Cary, North Carolina). In order to explore associations between school-district SES indicators and CTE related injuries reported to NJSS, DFG scores were used as a proxy measurement of SES for each reporting school/school district. Schools were given a DFG score ranging from 1 (lowest score/lowest SES) through 8 (highest score/highest SES). From the DFG scores, schools were dichotomized into lower scoring (DFG scores 1--4) and higher scoring (DFG scores 5--8) schools. The NJ Department of Education does not assign DFG Scores to county vocational school districts. Therefore, DFG scores were summed and averaged across the county as a whole, and the mean county score was used to assess SES for each reporting school in that county. Descriptive analyses were then performed to describe the demographics of the overall study population, as well as the study population stratified by DFG score. DFG was stratified in two ways, both as raw scores, ranging from 3--6 (for our study population) and as a dichotomous variable either being low (3 and 4) or high (5 and 6). This dichotomous classification coincides with low SES and high SES, respectively. Chi square tests (Χ^2^) for independence were conducted in order to examine associations between SES (high versus low) and various variables including gender, injury-treatment setting (hospital versus doctor/clinic), injury location on body, injury type, injury cause, severity of injury, and use of PPE, as these variables were hypothesized to be related to SES. Logistic regression was then conducted to further explore associations between SES and injury report variables listed above. Both crude and adjusted models were explored. The adjusted models included several variables hypothesized to potentially confound the association between SES and injury treatment setting. These variables were severity of injury, injury type, body part injured, and injury cause. We further hypothesized how the severity of injury would be the most important predictor of where an injury would be treated, as severity would determine if an injury needed acute attention from the ED, or further care that could be given at a doctor's office; therefore, a preliminary model adjusted solely for severity of injury was examined. To be conservative in the interpretation of results, the final model included adjustments for each of these variables. To assess potential associations between SES and PPE use, data were stratified by the years 2003--2008 and 2008--2013. This stratification of years was chosen because starting in 2008 employers were legally required to pay for properly selected and fitted PPE determined necessary for employees (NJ as of 2/2008, U.S. as of 10/1/2008) \[[@CR35]\]. PPE usage data were also stratified on career cluster, categorized as either being hazardous or nonhazardous. Categorization of career clusters as hazardous or non-hazardous was based on the 17 non-agricultural hazardous occupations orders (HOs) and 11 agricultural HOs of the U.S. Department of Labor (U.S. DOL) \[[@CR36]\]. Based on this classification there were eight career clusters categorized as hazardous: (1) agriculture, food, and natural resources; (2) architecture and construction; (3) transportation, distribution, and logistics; (4) manufacturing; (5) law, public safety, corrections, and security; (6) human services (e.g. cosmetology programs); (7) health sciences; and, (8) science, technology, engineering, and mathematics. Eight career clusters were thus categorized as non-hazardous: (1) marketing; (2) arts, audio/video technology, and communication; (3) business management and administration; (4) education and training; (5) finance; (6) hospitality and tourism; (7) government and public administration; and, (8) information and technology. It should be noted this classification scheme excluded the fact there are hazards present in each career cluster, for example, ergonomic factors which are accounted for in every workplace/workstation environment, including those career clusters classified as non-hazardous. Results and discussion {#Sec3} ====================== The twenty-one counties in NJ had at least one school district submit an injury report between 1998 and 2013. Sixteen counties were classified as higher DFG scoring schools/districts; nine scored a "6", and seven scored a "5". Five counties were classified as lower DFG scoring schools/districts; four scored a "4" and one scored a "3". Demographic results of the study population are described in Table [1](#Tab1){ref-type="table"}. Overall, there was an even distribution of where an injury was treated, with 56 % being treated at hospitals and 44 % at doctors/clinics. The most common body part injured were fingers, making up 38 % of reported injuries. The most common type of injury was cut/laceration (43 %). The most common injury cause was 'struck by' (35 %). Overall, most injuries reported were non-disabling (68 %); 32 % of injuries were temporarily disabling and there was only one reported permanently disabling injury. In this context, the term temporarily disabling meant the student was able to return to their school-sponsored SLE after medical treatment and rest. Overall, the use of PPE was fairly evenly split---47 % of reported injuries stated some kind of PPE was in use at the time of the incident. A majority or 95 % of reporting schools were classified as high SES (scores 5 and 6), and 5 % were classified as low SES (scores 3 and 4).Table 1Summary of descriptive statistics for variables on injury reports to NJ Safe Schools state-law based surveillance, 12/1998-12/2013CharacteristicTotal (*n*)Total (%)DFG 380.4 4874.5 575539.0 6108856.1DFG Score High184795.3 Low914.7Gender Male137371.7 Female54328.3Status Student180996.4 Staff693.7Treatment Hospital74655.6 Doctor59544.4Injury Location Finger72238.1 Hand19910.5 Eye1437.6 Foot683.6 Back502.6 Face452.4 Other28114.9 Multiple1889.9 Arm633.3 Head532.8 Knee331.7 Ankle492.6Injury Type Fracture704.0 Burn1719.7 Bruise/Bump603.4 Sprain905.1 Puncture714.0 Other34419.6 Multiple1508.5 Abrasion452.6 Cut/Laceration75843.1Injury Mode Struck Against29215.9 Caught In/Under/Between1136.2 Extreme Temperature1457.9 Other45124.6 Rubbed/Abraded472.6 Fall (Same Level)864.7 Struck By64135.0 Multiple573.1Severity Non-disabling123068.1 Temporary disabling57531.9 Permanent Disability10.1PPE No23047.2 Yes25852.9 When stratifying by SES (high versus low), the distribution of gender remained similar to the ratio in the overall study population. Males consisted of 72 % of the sample population in high SES schools, and 65 % in low SES schools. When stratified by SES, where injuries were treated differed between high and low SES schools (*p* = 0.029). For high SES schools, location of where injuries were treated was fairly evenly distributed---55 % were treated at hospitals and 45 % were treated by doctors/clinics. On the other hand, in low SES schools, 68 % of injuries were treated at hospitals, while only 32 % were treated by doctors/clinics. Severity of injury did not change substantially when stratified by SES (*p* = 0.87). In high SES schools, 68 % of injuries were non-disabling and 32 % were temporarily disabling. The one reported injury resulting in a permanent disability during the study period, an amputation of fingers, was reported in a high SES school. For low SES schools, 70 % of reported injuries were non-disabling and 30 % were temporarily disabling. For both high and low SES schools, finger was the most common body part injured, 'struck by' the most common cause of injury, and cut/laceration the most common injury type. Distribution of injury reports and use of PPE were analyzed based on the time period in which they were reported (2003--2008 vs. 2008--2013), as well as whether the injury reported was associated with a hazardous or non-hazardous career cluster (Table [2](#Tab2){ref-type="table"}). We were interested in comparing PPE use in high SES and low SES schools. Table [2](#Tab2){ref-type="table"} depicts results from the PPE use analysis. We observed a trend, where, in general, reported incidents with a response of "Yes" to PPE use decreased from the first time block (2003--2008) to the second time block (2008--2013). And, this trend was observed in general, regardless of hazardous and SES classifications. For example, those reporting "Yes" to PPE use in high SES schools in hazardous career clusters decreased from 76 % to 61 % (71 incidents to 62 incidents) between the two time blocks. An exception to this trend included those incidents among students in low SES schools/districts that reported "Yes" to PPE use in non-hazardous career clusters; however, low cell sizes should be considered. We observed no consistent trend for those who reported "No" for PPE use. In general, the number of incidents in which it was reported "No" for PPE use increased from the first time block (2003--2008) to the second time block (2008--2013). For example, those reporting "No" for PPE use in high SES schools in hazardous career clusters increased from 24 % to 39 % (22 incidents to 39 incidents) between the two time blocks. We also observed those schools/districts who reported "No" PPE use in low SES schools in hazardous career clusters remained stable, though cell sizes were low. It should also be noted those who reported "No" PPE use in low SES schools in non-hazardous career clusters decreased by half (from six incidents to three incidents) even if cell sizes were low between years. In summary, as schools/districts which reported using PPE reported fewer injuries than schools/districts which reported not using PPE, these results suggested schools/districts using PPE are selecting properly fitting PPE and using PPE correctly at a higher rate. Further research is needed for more definitive conclusions on the impact of the state/federal laws requiring purchase of PPE by employers for employees and of specific PPE training. Increase reporting of injuries among low SES schools would also strengthen conclusions.Table 2Injury incidents by occupation type and PPE use: New Jersey Safe Schools surveillance data (12/1998 through 12/2013)03/2003-08/20089/2008-10/2013PPE usageOverall injury reports 2003--2013 (N = 393)High SESLow SESHigh SESLow SESHazardousNon-hazardousHazardousNon-hazardousSumHazardousNon-hazardousHazardousNon-hazardousSumYesN214714930*123*62242391(%)54.076.359.860.00.061.425.050.050.0Within Time Period (%)57.739.82.40100.068.126.42.23.3100.0NoN17922332663397223116(%)46.023.740.240.0100.038.675.050.050.0Within Time Period (%)34.952.43.29.5100.033.662.11.72.6100.0Note: Totals include injury reports indicating both use of PPE and a career cluster Descriptive analysis was carried out to better understand associations between SES and injury reports. Chi square test (Χ^2^) for independence (Table [3](#Tab3){ref-type="table"}) revealed statistically significant associations between SES and injury cause \[X^2^ = (7, 14.74), *p* = 0.04\] as well as SES and injury treatment setting \[X^2^ = (1, 4.76), *p* = 0.03\]. A series of logistic regressions were performed to further examine potential associations between SES and the binary outcome injury treatment setting. To understand the potential differences in treatment locations within each SES group (high versus low), logistic regression was performed comparing the location of treatment in the high vs. low SES schools. Unadjusted odds ratios (OR) suggested low SES schools had an increased odds of being treated at a hospital/ED compared to high SES schools (OR = 1.75; 95 % CI = 1.1--2.9; Table [4](#Tab4){ref-type="table"}). Similarly, when adjusting for severity of injury, low SES schools still had an increased odds of being treated at a hospital/ED compared to high SES schools (adjusted odds ratio \[AOR\] = 1.80; 95 % CI = 1.2--1.9; Table [4](#Tab4){ref-type="table"}). Further, when adjusting for severity of injury, injury type, body part injured and injury cause, low SES schools had an increased odds of being treated at a hospital/ED compared to high SES schools (AOR = 2.40; 95 % CI = 1.3--4.3; Table [4](#Tab4){ref-type="table"}).Table 3High versus low SES schools/districts: injury reports to New Jersey Safe Schools state-law based surveillance, 12/1998-12/2013CharacteristicHigh SESn (%)Low SES n (%)p-valueGender Male1,315 (72.0)58 (65.2)0.160 Female512 (28.0)31 (34.8)Treatment Hospital697 (54.9)49 (68.1)0.029 Doctor572 (45.1)23 (31.9)Injured Body Part Finger675 (37.4)47 (53.4)0.140 Hand191 (10.6)8 (9.1) Multiple180 (10.0)8 (9.1) Arm61 (3.4)2 (2.3) Head53 (2.9)0 (0.0) Knee31 (1.7)2 (2.3) Ankle48 (2.7)1 (1.1) Eye136 (7.5)7 (8.0) Foot68 (3.8)0 (0.0) Back48 (2.7)2 (2.3) Face42 (2.3)3 (3.4) Other273 (15.1)8 (9.1)Injury Type Fracture64 (3.8)6 (6.8)0.481 Burn162 (9.7)9 (10.2) Bruise/Bump58 (3.5)2 (2.3) Sprain85 (5.1)5 (5.7) Puncture64 (3.8)7 (8.0) Multiple141 (8.4)9 (10.2) Abrasion43 (2.6)2 (2.3) Cut/Laceration724 (43.4)34 (38.6) Other330 (19.7)14 (15.9)Injury Cause Struck Against281 (16.1)11 (12.6)0.039 Caught In /Under/Between100 (5.7)13 (14.9) Extreme Temperature137 (7.9)8 (9.2) Rubbed/Abraded45 (2.6)2 (2.3) Fall (Same Level)80 (4.6)6 (6.9) Struck By617 (35.4)24 (27.6) Multiple54 (3.1)3 (3.4) Other431 (24.7)20 (23.0)Severity Non-Disabling1,168 (68.0)62 (70.5)0.870 Temporary Disabling549 (32.0)26 (29.5) Permanently Disabling1 (0.10)0 (0.0)PPE Yes246 (53.2)15 (55.6)0.400 No216 (46.8)12 (44.4)Table 4Odds ratio (OR) for being treated at a hospital emergency departmentCharacteristicOR95 % CI*p*-valueSES^a^ HighReferent Low1.751.1--2.90.03SES^b^ HighReferent Low1.81.1--3.10.02SES^c^ HighReferent Low2.41.3--4.3\<0.01^a^Crude OR^b^Adjusted for severity of injury^c^Adjusted for severity of injury, injury type, body part injured, and injury cause Previous literature has suggested childhood injuries varied by SES, especially relating to morbidity and mortality \[[@CR22]\]. The goal of the present analysis was to better understand potential associations between SES indicators and reported CTE-related injuries. Data reported here represents surveillance data captured between 12/1998 through 12/2013 using the NJSS law-based CTE reporting system, aggregated through Psychdata. We hypothesized SES status would be associated with several key variables including injury cause, injury location on the body, and injury type, injury severity, use of PPE, and where an injury was treated. Specifically, we hypothesized the low SES schools/districts would have more severe injuries, use PPE less often and/or use PPE incorrectly, compared to high SES schools/districts, and have an increased odds of being treated at a hospital/ED. Statistically significant differences were observed between SES and injury cause. These results suggest a statistically significant association between SES and CTE related injuries and where injuries are treated. For high SES schools, the distribution of injury treatment setting was fairly evenly distributed, with 55 % being treated at hospitals and 45 % being treated by doctors/clinics. In low SES schools, however, a greater difference was observed, with 68 % of injuries being treated at hospitals/ED and 32 % being treated by doctors. Our logistic regression analyses (both crude and adjusted) supported this observation, suggesting low SES schools/districts have increased odds of being treated at a hospital/ED compared to high SES schools/districts. Thus, an argument can be made for the discrepancy between injury treatment setting and SES being in part an issue of access, both in a physical sense (e.g., lack of private/public transportation given distance to travel) and in regards to health care (e.g., number of providers) and/or health care insurance. It should be noted how these data cannot fully capture the recent changes in health care reform, i.e., the passing of the Patient Protection and Affordable Care Act (ACA) in 2010 and its implementation ongoing in 2013 through the present \[[@CR37]\]. Future research with NJSS surveillance data may better capture changes in health care status in relation to the ACA. Further research should also be conducted to explain why these discrepancies existed. Current literature has suggested poor individuals who are underinsured or uninsured typically seek care in hospital/ED \[[@CR38]\]. It should be noted, however, as these injuries were acquired within school sponsored CTE programs, cost incurred would potentially be covered by the reporting school/district. Although this may suggest how health insurance status does not play as vital role in the injury treatment setting of a reported injury, individuals often times do not know how to navigate the health care system, or lack the knowledge concerning reimbursement by reporting school/district. Irrespective of who is responsible for paying for treatment, the cost to society as a whole is vast, with each trip to an ED expecting to cost around \$200.00 \[[@CR39]\]. Moreover, links have been made between workplace S&H and income inequality by the U.S. DOL, which stated workplace injuries may have a greater effect on low wage workers and those trying to enter the middle class. Similarly, U.S. DOL has noted workplace injuries (examined in the adult population) placed burdens on the workers and their families and have contributed to income inequality \[[@CR40]\]. Further research may examine how outcomes of workplace injuries affect varying SES grades differently. This analysis explores the association of income inequality and workplace S&H as it related to students participating in school-sponsored CTE programs. It should also be considered how effect modification may be at play here, skewing the relationship between workplace injury outcomes and SES \[[@CR41]\]. There are also areas of improvements regarding the data sources used. The NJ Department of Education should strive to provide direct DFG scores for vocational-CTE schools. A limitation of this study was NJ DFG scores were not available for the twenty-one reporting county-based CTE schools/districts. To overcome this limitation, DFG scores were summed and averaged across reporting schools in each county. These county-average DFG scores were used for each reporting school in each respective county. This limitation may have led to non-differential misclassification bias of exposure---whether the school was reported as a low SES school or a high SES school---and a bias towards the null, i.e., underestimating the true association between SES and where incidents were treated (hospital/ED versus doctor). Also limiting current analysis was the lower number incident reports from low SES schools compared to high SES schools due to the small number of counties classified as low SES schools. In total, there were 95 injury reports from low SES schools and 1,843 injury reports from high SES schools. Future efforts should be placed on increasing injury reporting in low SES schools. Another limitation of this study was there were no denominator data---data were reported incidents within CTE programs. There was no information concerning students/staff enrolled in CTE programs that were not injured between the years 1998--2013. Another limitation related to the generalizability of results to the general secondary school/student population. Students enrolled in NJ CTE programs represent a growing yet specific subset of the overall student population. These students may not be generalizable to the general student body throughout NJ, and may not be generalizable to other state CTE students, as CTE programs differ state by state. Another limitation, as is true with most surveillance data, was missing and incomplete data fields. However, as of October 2013, reports are only submitted online to NJSS via Psychdata. This eliminates the ability to leave certain spaces blank as occurred with the past paper based system. Future analyses will compare completeness of reporting factors between the former paper-based and current online reporting system. This study also had several strengths. This study represented surveillance data over fourteen years for the entire state of New Jersey. These findings add to a major gap in the literature by specifically examining injuries relating to secondary school students enrolled in CTE programs by SES. This study also identified areas to be further analyzed in order to (1) reduce the rates of injures for secondary school students enrolled in CTE programs; (2) reduce total medical expenditures resulting from CTE related injuries; and, (3) improve the overall quality of life for secondary school students enrolled in CTE programs. Conclusion {#Sec4} ========== Initial findings from this current analysis suggested there is a statistical difference between high and low SES schools and injury cause, as well as high and low SES schools and injury treatment setting. Results from this analysis suggested injuries occurring in low SES schools have higher odds of being treated at a hospital compared to an injury reported at a high SES school. This association remained true after for controlling for several key variables including injury severity. Future research should explore why this may be the case, and better assess whether this is an issue of medical access, and/or school policy as to where an injury is initially treated. This study's results can guide future development of injury prevention trainings and interventions, which can lead to decreased rates of injuries, decreased medical expenditures, and increased student academic performance and achievements. ACA : Patient Protection and Affordable Care Act AOR : adjusted odds ratio CI : confidence interval CTE : Career and Technical Education DFG : District Factor Group DOL : Department of Labor ED : emergency department/emergency room HOs : hazardous occupations orders NJ : New Jersey NJ SS : NJ Safe Schools Program OR : odds ratio PPE : personal protective equipment S&H : safety and health SES : socioeconomic status SLE : school-sponsored structured learning experiences U.S. : United States **Competing interests** This paper has not been accepted or published previously and is not under simultaneous consideration for publication except with *Environmental Health and Public Health.* We have no financial or other non-financial conflicts of interest to disclose. We will transfer copyright. **Authors' contributions** AAA contributed to the conception and design of this effort and led the data management and analyses, interpretation of results, and the drafting and revising of the manuscript. DGS contributed to the conception and design and assisted in the data analyses, interpretation of results and the drafting and revising of the manuscript. Both authors read and approved the final manuscript. **Author information** AAA is a Research Teaching Specialist V for the Center for School and Community-Based Research and Education (CSCBRE) and the New Jersey Safe Schools Program (NJ SS) at Rutgers School of Public Health (SPH). As of mid-January 2016, AAA is at University of North Carolina-Chapel Hill Gillings School of Public Health. DGS is an Associate Professor of Environmental and Occupational and Environmental Health at Rutgers SPH and the Environmental and Occupational Health Sciences Institute, Co-Director of CSCBRE, and the Director of NJ SS. For other papers on NJ SS injury surveillance see: <http://www.njsafeschools.org/papers.html> For more on DGS and CSCBRE and NJ SS, please see the following websites: CSCBRE and NJ SS: <http://sph.rutgers.edu/centers_institutes/community_based_research/index.html> <http://www.njsafeschools.org> Personal websites including CV content: <http://sph.rutgers.edu/departments/environmental_occupational/faculty_staff.html> \[Go to DGS\] <http://eohsi.rutgers.edu/eohsi-directory/name/derek-shendell/> <http://climatechange.rutgers.edu/people/affiliates/469-shendell-derek> We thank the New Jersey (NJ) Department of Education (NJDOE), Office of Career Readiness (formerly Office of Career and Technical Education) for funding the NJ Safe Schools Program ("Occupational Education and Safety and Health Training and Standards Updating" grant to Rutgers School of Public Health) plus their staff for providing their knowledge and expertise, and other fellow NJ OSHA Alliance for Young Worker Safety and Health members for their support to promote the health and safety of minors. We also graciously acknowledge efforts of other NJ Safe Schools Program staff who have contributed throughout the years. This paper does not constitute an endorsement of authors, or organizations, by NJDOE. Views and opinions expressed are not necessarily those of the NJDOE.

Introduction {#sec1} ============ Due to the improvement of living and the change of lifestyles, the number of people with obesity in China has seen explosive growth in recent years. The accumulation of excess white adipose tissue (WAT) is the main reason of obesity. Brown adipose tissue (BAT), which is different from WAT, can burn fat and promote energy metabolism, thus providing new methods for the prevention of obesity and associated metabolic diseases.[@bib1] WAT from certain depots, in response to appropriate stimuli, can undergo a process known as browning that takes on characteristics of BAT---notably, the induction of uncoupling protein 1 (UCP1) expression and the presence of multilocular lipid droplets and multiple mitochondria.[@bib2] Except for adipocytes, adipose tissue also contains abundant adipose tissue macrophages (ATMs).[@bib3] There are two phenotypes of ATMs. In obesity, a large number of pro-inflammatory M1 macrophages recruited in adipose tissue release amounts of inflammation factors and lead to chronic inflammation and metabolic disorders.[@bib4] There exists a dynamic conversion in the proportion of M1 and M2 macrophages, which is closely related to adipose metabolism and energy expenditure.[@bib5], [@bib6] It has been reported that the transformation of M1 to M2 macrophages promoted the activation of BAT and the browning of WAT.[@bib7], [@bib8] Because PPARγ plays a key role in M2 polarization and in the browning of white adipocytes, and in macrophage polarization in association with its browning capacity, we hypothesize that circNrxn2-miR-103-FGF10 could also affect WAT browning by affecting macrophage polarization. Circular RNAs (circRNAs) are a group of non-coding RNAs (ncRNAs) characterized by the presence of a covalent bond linking 3′ and 5′ ends produced by backsplicing.[@bib9], [@bib10] circRNAs are widely spread with stable structures, conserved sequences, and cell- or tissue-specific expression.[@bib11] Studies have shown that circRNAs can function in multiple ways. They can act as a natural microRNA (miRNA) sponge to regulate miRNA expressions or regulate gene transcription, cell cycle, and other physiological processes through interaction with proteins. ciRS-7/CDR1as has been identified as a miR-7 sponge to inhibit miR-7 activity.[@bib12] Wang and colleagues[@bib13] found that a heart-related circRNA (HRCR), which acted as a miR-223 sponge, was correlated with pathological hypertrophy and heart failure. These studies are strongly supported by the idea that circRNAs can be used as a biomarker for disease diagnosis. However, their functions are still largely unidentified and have broad research prospects, and whether there exist circRNAs acting as a miR-103 sponge has not been studied. miRNAs are a class of small (18--25 nt) single-stranded ncRNAs that function as endogenous regulators by targeting mRNAs.[@bib14] miRNA dysfunction is related to a variety of human diseases, such as obesity and metabolic disorders, suggesting that miRNAs play vital roles in obesity.[@bib15], [@bib16] Aberrant expressions of miRNAs, including miR-221, miR-519d, miR-141, and miR-520e, are associated with human obesity and related metabolic syndrome.[@bib17], [@bib18], [@bib19] Among them, miR-103 belongs to a highly conserved family of miRNA and is abundant in adipose tissue.[@bib20], [@bib21] Dozens of miR-103 target genes, such as Dicer, Cav1, DAPK, KLF4, Fadd, Mef2d, and Wnt3a, have been demonstrated to regulate adipogenesis, insulin sensitivity, cell migration, metastasis, and apoptosis.[@bib21], [@bib22], [@bib23], [@bib24], [@bib25], [@bib26], [@bib27] However, the regulation of miR-103 in ATMs and WAT browning has not been studied. In this study, we present that circNrxn2, miR-103, and FGF10 formed a regulatory circuit in adipose tissue and that they regulated WAT browning by ATM polarization. This study provides a molecular basis for preventing obesity and obesity-associated metabolic disorders. Results {#sec2} ======= circNrxn2 Acted as an Endogenous miR-103 Sponge and Increased FGF10 Expression in HFD Mice Adipose Tissue {#sec2.1} --------------------------------------------------------------------------------------------------------- To search for circRNAs and mRNAs that bind to miR-103, bioinformatics were carried out for predictive analysis. Among the numerous molecules, there is a potential binding site for miR-103 in the circ005661 sequence, which we called circNrxn2, as it was transcribed from the Nrxn2 gene ([Figure 1](#fig1){ref-type="fig"}A), and there are two binding sites for miR-103 in the 3′ UTR of FGF10 mRNA ([Figure 1](#fig1){ref-type="fig"}B). Therefore, circNrxn2 and FGF10 were selected as candidate molecules that interacted with miR-103. A dual luciferase reporter assay was conducted, and we found that overexpressing miR-103 caused a significant decrease in the luciferase activity of the 3′ UTR FGF10 (1,637--1,644), whereas there was no difference between the 3′ UTR FGF10 (702--708) and the control group ([Figure 1](#fig1){ref-type="fig"}C). Moreover, the mutation of 3′ UTR FGF10 (1,637--1,644) displayed no difference from the control group ([Figure 1](#fig1){ref-type="fig"}C). The data suggested that FGF10 (1,637--1,644) may be the potential binding site for miR-103.Figure 1circNrxn2 Acted an Endogenous miR-103 Sponge and Increased FGF10 Expression in Mice Adipose Tissue(A) Scheme of the interaction of circNrxn2 and miR-103. (B) Scheme of the interaction of miR-103 and FGF10. (C) Double luciferase assay was performed. (D) qRT-PCR was used to detected circNrxn2 and GAPDH levels after streptavidin capture. (E) The expression of FGF10 in WAT of HFD mice (8-week-old C57BL/6J mice fed with an HFD for 10 weeks; inguinal WAT (iWAT) was acquired for these experiments). (F) The expression of circNrxn2, miR-103, and FGF10 in iWAT of HFD mice. (G--I) The expressions of circNrxn2 (G), miR-103 (H), and FGF10 (I) in iWAT of HFD mice after circNrxn2, miR-103, or FGF10 treatment. Data represent the mean ± SEM. \*p \< 0.05; \*\*p \< 0.01. n ≥ 3. In addition, circNrxn2 was added to the luciferase activity assay system. It was found that circNrxn2 addition increased the luciferase activity compared to the relative control group ([Figure 1](#fig1){ref-type="fig"}C), meaning that circNrxn2 relieved the inhibitory effect of miR-103 on FGF10. Besides, circNrxn2 was much more enriched in the miR-103-captured fraction compared with the corresponding mutant group ([Figure 1](#fig1){ref-type="fig"}D). Western blot analysis also indicated that circNrxn2 treatment upregulated the FGF10 protein level and that the addition of miR-103 reversed it ([Figure 1](#fig1){ref-type="fig"}E). A high-fat diet (HFD) upregulated the miR-103 level, while circNrxn2 and FGF10 levels were lower in the adipose tissue of HFD mice than that in normal diet (ND) mice ([Figure 1](#fig1){ref-type="fig"}F). These results confirmed that circNrxn2 could sponge miR-103, thus crippling the effect of miR-103 on FGF10. To avoid the effects of expression vectors on other tissues, we used adipose-tissue-specific expression vectors and evaluated their expression in liver, muscle, perirenal fat, WAT, and BAT ([Figures S1](#mmc1){ref-type="supplementary-material"}A--S1C). HFD mice were injected intraperitoneally (i.p.) with the circNrxn2 miR-103 and FGF10 viruses, respectively. It was found that circNrxn2 significantly decreased the miR-103 level and increased the FGF10 mRNA level ([Figure 1](#fig1){ref-type="fig"}G). miR-103 overexpression inhibited circNrxn2 and FGF10 mRNA levels ([Figure 1](#fig1){ref-type="fig"}H). FGF10 treatment enhanced circNrxn2 and inhibited miR-103 levels ([Figure 1](#fig1){ref-type="fig"}I). These results indicated that circNrxn2 acted as an endogenous miR-103 sponge and increased FGF10 expression in mouse adipose tissue. circNrxn2 Promoted WAT Browning in HFD Mice {#sec2.2} ------------------------------------------- To investigate the roles of circNrxn2-miR103-FGF10 in WAT browning, HFD mice were housed at 4°C for 12 h. Cold stimulation led to smaller adipocytes and less crown structure as reported previously ([Figure 2](#fig2){ref-type="fig"}A).[@bib28] Cold stimulation upregulated circNrxn2 and FGF10 levels and downregulated the miR-103 level significantly ([Figure 2](#fig2){ref-type="fig"}B), indicating that circNrxn2-miR-103-FGF10 had a potential role in WAT browning. HFD mice were injected i.p. with circNrxn2 lentivirus, circNrxn2 lentiviruses + miR-103 adenoviruses, circNrxn2 lentiviruses + short interfering (si)FGF10 lentiviruses, or control empty viruses. We found that the size of adipocytes was decreased after circNrxn2 treatment and that the addition of miR-103 or siFGF10 abolished the effect of circNrxn2 ([Figure 2](#fig2){ref-type="fig"}C). Also, the temperature of mice treated with circNrxn2 was significantly increased ([Figure S1](#mmc1){ref-type="supplementary-material"}D). qRT-PCR was used to detect the effect of circNrxn2-miR-103-FGF10 on WAT browning. We found that circNrxn2 facilitated the mRNA levels of UCP1, PPARγ, PRDM16, PGC1α, COX7α, and COX8b ([Figure 2](#fig2){ref-type="fig"}D; [Figures S1](#mmc1){ref-type="supplementary-material"}E, S1F, and [S2](#mmc1){ref-type="supplementary-material"}) and the protein levels of UCP1, PPARγ, PGC1α, and COX4 ([Figure 2](#fig2){ref-type="fig"}E), indicating that circNrxn2 promoted WAT browning, mitochondrial synthesis, and oxidation. In addition, miR-103 or siFGF10 addition reversed the effect of circNrxn2 ([Figures 2](#fig2){ref-type="fig"}D and 2E). Immunofluorescent staining also showed that circNrxn2 increased the expressions of UCP1 and COX4 ([Figures 2](#fig2){ref-type="fig"}F and 2G), and the addition of miR-103 or siFGF10 reversed it. These data suggested that FGF10 promoted WAT browning, mitochondrial synthesis, and oxidation, while miR-103 and siFGF10 inhibited it. circNrxn2-miR-103-FGF10 had a vital role in WAT browning.Figure 2circNrxn2 Promoted WAT Browning in HFD MiceHFD mice were 8-week-old C57BL/6J mice fed with an HFD for 10 weeks. They were injected i.p. with circNrxn2 lentivirus, circNrxn2 lentiviruses + miR-103 adenoviruses, circNrxn2 lentiviruses + siFGF10 lentiviruses, or control empty viruses. (A) H&E staining of adipose tissue section. (B) The changes of circNrxn2, miR-103, and FGF10 after cold stimulation. (C) H&E staining of adipose tissue section after circNrxn2, circNrxn2+miR-103, circNrxn2+siFGF10, or control treatment. (D) The relative mRNA levels of UCP1, PPARγ, PRDM16, PGC1α, COX7α, and COX8b of iWAT after circNrxn2, circNrxn2+miR-103, or circNrxn2+siFGF10 treatment in HFD mice. (E) Western blot analysis of protein levels of UCP1, PPARγ, PGC1α, and COX4 in WAT. (F and G) Immunofluorescent staining of (F) UCP1 and (G) Cox4 in adipocytes under circNrxn2, circNrxn2+miR-103, or circNrxn2+siFGF10 treatment. Data represent the mean ± SEM. Scale bars, 100 μm. \*p \< 0.05; \*\*p \< 0.01. n ≥ 3. To further explore the roles of circNrxn2-miR-103-FGF10 in regulating white adipocytes browning, adipocytes were treated with circNrxn2, circNrxn2 + miR-103, circNrxn2 + siFGF10, or control empty viruses, respectively, and then incubated with CL316, 243 to induce white adipocytes browning. Consistent with the results *in vivo*, circNrxn2 treatment increased the transcriptional and post-transcriptional levels of PRDM16, PPARγ, and PGC1α ([Figure 3](#fig3){ref-type="fig"}A). Besides, circNrxn2 promoted the mRNA levels of Cox7α and Cox8b ([Figure 3](#fig3){ref-type="fig"}A) and the protein levels of Cox4, PPARγ and PGC1α ([Figure 3](#fig3){ref-type="fig"}B), and these effects were reversed after the addition of miR-103 or siFGF10 ([Figures 3](#fig3){ref-type="fig"}A and 3B). Moreover, we found that circNrxn2 increased the mitochondrial number and that the addition of miR-103 or siFGF10 decreased it ([Figures 3](#fig3){ref-type="fig"}C and 3D). Immunofluorescent staining also showed that circNrxn2 promoted UCP1 expression and that miR-103 or siFGF10 addition decreased it ([Figure 3](#fig3){ref-type="fig"}E). These results indicated that FGF10 could promote WAT browning and mitochondrial function, whereas miR-103 and siFGF10 have opposite effects.Figure 3circNrxn2 Promoted White Adipocytes Browning *In Vitro*3T3-L1 adipocytes were treated with circNrxn2, circNrxn2 + miR-103, circNrxn2 + siFGF10, or control empty viruses. (A) The relative mRNA levels of PRDM16, PPARγ, PGC1α, COX7α, and COX8b. (B) The protein levels of PRDM16, PPARγ, PGC1α, and COX4. (C) The mitochondria staining with mitochondrial tracker. (D) The citrate synthase activity was detected in adipocytes under treatment of circNrxn2, circNrxn2 + miR-103, and circNrxn2 + siFGF10. (E) Immunofluorescent staining of UCP1 in adipocytes. Data represent the mean ± SEM. Scale bars, 200 μm. \*p \< 0.05; \*\*p \< 0.01. n ≥ 3. circNrxn2 Increased M2 Macrophage Polarization in Mice {#sec2.3} ------------------------------------------------------ We had found that circNrxn2 decreased crown structure in adipose tissue ([Figure 2](#fig2){ref-type="fig"}C). ATM phenotypic conversion could regulate energy metabolism and BAT activation, so we detected the influence of circNrxn2 on ATM polarization. We found that circNrxn2 treatment promoted the expression of CD206, an M2 macrophage marker, and it was reversed after the addition of miR-103 adenoviruses or siFGF10 ([Figure 4](#fig4){ref-type="fig"}A). In addition, circNrxn2 upregulated the mRNA levels of interleukin (IL)-4 and CD206 and decreased IL-6 and tumor necrosis factor α (TNF-α) mRNA levels ([Figure 4](#fig4){ref-type="fig"}B), which were abolished after the addition of miR-103 or siFGF10 ([Figure 4](#fig4){ref-type="fig"}B). These results suggested that circNrxn2 and FGF10 promoted M2 macrophage polarization and that miR-103 inhibited M2 macrophage polarization *in vivo*.Figure 4circNrxn2 Increased M2 Macrophage Polarization in HFD Mice(A) CD206 immunofluorescent staining of HFD mice adipose tissue section. (B) The mRNA levels of IL-6, TNF-α, IL-4, and CD206 in adipose tissue. (C) The mRNA levels of IL-6, TNF-α, IL-4, and CD206 in macrophages. (D) The protein levels of IL-6, TNF-α, IL-10, and CD206 in macrophages. (E) Immunofluorescent staining of CD206. (F) The morphologic observation of macrophages. (G) The mRNA levels of IL-6, TNF-α, IL-4, and CD206 in plasma were detected to evaluate the systemic effects of circNrxn2, miR-103, and FGF10. Data represented the mean ± SEM. Scale bars, 200 μm. \*p \< 0.05; \*\*p \< 0.01. n ≥ 3. In order to examine the effects of circNrxn2-miR-103-FGF10 on macrophage polarization, we isolated and cultivated macrophages *in vitro* and then treated them with circNrxn2, circNrxn2 + miR-103, circNrxn2 + siFGF10, or control empty viruses. It was found that circNrxn2 promoted the mRNA levels of IL-4 and CD206 ([Figure 4](#fig4){ref-type="fig"}C) and the protein levels of IL-10 and CD206 ([Figure 4](#fig4){ref-type="fig"}D). In addition, the mRNA and protein levels for M1 macrophage markers IL-6 and TNF-α were decreased after circNrxn2 treatment ([Figures 4](#fig4){ref-type="fig"}C and 4D) and were reversed after the addition of miR-103 and siFGF10 ([Figures 4](#fig4){ref-type="fig"}C and 4D). Immunofluorescent staining results also showed that circNrxn2 increased CD206 expression and that the addition of miR-103 or siFGF10 decreased it ([Figure 4](#fig4){ref-type="fig"}E). Besides, the macrophages were mostly circular after circNrxn2 treatment and exhibited spindle type after adding miR-103 or siFGF10 ([Figure 4](#fig4){ref-type="fig"}F). To evaluate the systemic effects of circNrxn2, miR-103, and FGF10, the expression of IL-6, TNF-α, IL-4, and CD206 in plasma of mice was measured ([Figure 4](#fig4){ref-type="fig"}G). These results all indicated that circNrxn2 treatment induced M2 macrophage polarization, which was blocked after the addition of miR-103 or siFGF10. circNrxn2 Promoted WAT Browning and Increased the M2 Macrophage Polarization by Activating PPARγ Signaling Pathway {#sec2.4} ------------------------------------------------------------------------------------------------------------------ It has been reported that PPARγ signaling is essential in regulating WAT browning and ATM polarization.[@bib29], [@bib30] We found that circNrxn2 activated PPARγ expression in mouse adipose tissue, so we tried to explore whether the PPARγ signaling pathway participated in circNrxn2 regulation of WAT browning and ATM polarization. PPARγ agonist and antagonist were used to treat mouse adipose tissue. The results showed that Rosiglitazone, a PPARγ agonist, enhanced circNrxn2-induced PPARγ, UCP1, and CD206 and blocked circNrxn2-inhibited IL-6 ([Figure 5](#fig5){ref-type="fig"}A). Besides, miR-103 and siFGF10 addition suppressed the effects of circNrxn2, and Rosiglitazone treatment also enhanced these effects ([Figure 5](#fig5){ref-type="fig"}A). Furthermore, T0070907, a PPARγ antagonist, was used to verify our results. We found that T0070907 restrained circNrxn2-induced PPARγ, UCP1, and CD206 and increased circNrxn2-inhibited IL-6 ([Figure 5](#fig5){ref-type="fig"}B). T0070907 treatment also enhanced the effects of the addition of miR-103 and siFGF10 ([Figure 5](#fig5){ref-type="fig"}B). These results suggested that circNrxn2 promoted WAT browning and increased M2 ATM polarization by activating the PPARγ signaling pathway.Figure 5circNrxn2 Promoted WAT Browning and Increased M2 Macrophage Polarization by Activating the PPARγ Signaling Pathway(A and B) PPARγ agonist (A) and antagonist (B) were injected in HFD mice adipose tissue combined with circNrxn2, miR-103, and FGF10 after circNrxn2, miR-103, or FGF10 treatment. The mRNA levels of PPARγ, UCP1, CD206, and IL-6 were detected. Data represented the mean ± SEM. ∗p \< 0.05; ∗∗p \< 0.01. n ≥ 3. circNrxn2 Promoted WAT Browning by Increasing M2 Macrophage Polarization in HFD Mice {#sec2.5} ------------------------------------------------------------------------------------ To further explore the relationship between circNrxn2-miR-103-FGF10-regulated ATM polarization and WAT browning, clodronate was used to eliminate macrophages. As a result, it was found out that clodronate treatment efficiently decreased F4/80 expression in WAT of HFD mice ([Figure 6](#fig6){ref-type="fig"}A). Although circNrxn2 promoted the mRNA levels of UCP1, DIO2, PPARγ, COX7α, COX8b, and PGC1α, the addition of miR-103 or siFGF10 caused the downregulation of these factors ([Figures 6](#fig6){ref-type="fig"}B and 6C), and clodronate treatment decreased the degree of WAT browning regulated by circNrxn2-miR-103-FGF10 ([Figures 6](#fig6){ref-type="fig"}B and 6C). Similar results were found in western blot analysis, which also showed that circNrxn2 induced less upregulation of PRDM16, UCP1, PPARγ and COX4 and that the addition of miR-103 or siFGF10 also reduced gene expression after clodronate treatment ([Figure 6](#fig6){ref-type="fig"}D). These results indicated that the circNrxn2-miR-103-FGF10-regulated WAT browning was partially achieved by ATM polarization.Figure 6circNrxn2 Promoted WAT Browning by Increasing M2 Macrophage Polarization in HFD MiceHFD mice were treated with clodronate to eliminate macrophages. (A) Immunofluorescent staining of F4/80 in HFD mouse adipose tissue after clodronate treatment. (B and C) The mRNA levels of (B) UCP1, DIO2, and PPARγ and of (C) PGC1α, COX7α, and COX8b. (D) The protein levels of PRDM16, PPARγ, and COX4. Data represent the mean ± SEM. Scale bars, 200 μm. \*p \< 0.05; \*\*p \< 0.01. n ≥ 3. Furthermore, macrophages were treated with circNrxn2, circNrxn2 + miR-103, circNrxn2 + siFGF10, or control empty viruses, and then the supernatants were collected and added to culture adipocytes. The results showed that circNrxn2 treatment resulted in an increase in the mRNA levels of PRDM16, UCP1, PPARγ, PGC1α, COX7α and COX8b, while the addition of miR-103 or siFGF10 inhibited the expression of these genes ([Figure 7](#fig7){ref-type="fig"}A). Western blot also showed that circNrxn2 treatment increased the protein levels of PRDM16, UCP1, PPARγ, PGC1α and COX4, and that miR-103 or siFGF10 addition reversed the effects ([Figure 7](#fig7){ref-type="fig"}B), which was consistent with UCP1 immunofluorescence results ([Figures 7](#fig7){ref-type="fig"}C). These data indicated that circNrxn2-miR-103-FGF10 promoted WAT browning by increasing M2 macrophage polarization.Figure 7circNrxn2 Promoted White Adipocyte Browning by Increasing M2 Macrophage PolarizationRAW264.7 macrophages were treated with circNrxn2, circNrxn2 + mi-103, or circNrxn2 + siFGF10 and cultured for 24 h in 2% free fatty acid (FFA)-free BSA in 24-well plates, and the supernatant was collected as RAW-conditioned medium. 3T3-L1 adipocytes were incubated with RAW-conditioned medium for 24 h to detect mRNA expression levels and 48 h for protein levels. (A) The mRNA levels of PRDM16, UCP1, PPARγ, PGC1α, COX7α, and COX8b. (B) The protein levels of PRDM16, UCP1, PPARγ, PGC1α, and COX4. (C) Immunofluorescent staining of UCP1 in adipocytes. Data represent the mean ± SEM. Scale bars, 200 μm. \*p \< 0.05; \*\*p \< 0.01. n ≥ 3. Discussion {#sec3} ========== The location of the different types of adipose tissue varies. BAT is mainly distributed in subscapular and interscapular tissue and around the kidneys, and beige adipose tissue and WAT mainly exist in subcutaneous and internal organs of mice.[@bib31] Under appropriate stimulations, such as cold exposure, β-3 adrenergic receptor agonists, and thiazolidinediones, white adipocytes can be transformed into thermogenic adipocytes with higher UCP1 expression like BAT. This process is called browning.[@bib32] Many studies have found that a variety of stimuli and factors, such as miRNA, macrophages, and certain drugs, can regulate WAT browning, which is more likely to occur in the inguinal WAT (iWAT).[@bib33], [@bib34], [@bib35] Besides, PPARγ is widely studied as a classical factor regulating browning. In this paper, the work we did is also on these bases. circRNAs serve as endogenous miRNA sponges to influence the expression of downstream miRNA target genes. For example, ciRS-7/CDR1as binds miR-7 to sequester away miR-7, resulting in high-level expression of target genes.[@bib12], [@bib36] miR-103 has been confirmed to play a critical role in lipid metabolism; it was proved that miR-103 can promote 3T3-L1 cell differentiation and regulate insulin sensitivity.[@bib21], [@bib25] Furthermore, genome-wide approaches also identified that miR-103 could bind with PPARγ.[@bib37] Consistent with these findings, our research displays that circNrxn2 acts as a miR-103 sponge to enhance the FGF10 level, thus promoting WAT browning. Different miRNA types and their effects have also been studied in WAT browning and adipose tissue regulation. Various kinds of miRNAs have been shown to regulate subcutaneous WAT browning and BAT activation in cold exposure. A previous study has shown that miR-32 inhibition reduces FGF21 expression to decrease tolerance to cold; therefore, it plays a role in cold-induced WAT browning in mice.[@bib38] Moreover, miR-455 can also enhance the WAT browning response stimulated by cold and norepinephrine.[@bib39] In contrast, several miRNA types were found to be negative regulators of WAT browning and BAT activity. miR-27 played a negative regulatory role in white fat browning.[@bib40] miR-155 and C/EBPβ constituted a double closed-loop negative-feedback system that regulated the adipogenesis of brown or beige adipocytes.[@bib41] miR-133 restricted the browning of adipocytes by inhibiting the transcription of the PRDM16.[@bib42] Besides, researchers found that inhibition of the miR-327-FGF10-FGFR2 signaling axis can induce WAT browning and increase whole-body metabolic rate. They hypothesized that miRNAs targeting FGF10 would provide an effective new approach for the treatment of obesity and metabolic disorders.[@bib43] What's more, it had been reported that fat-specific knockout of dicer, a target of miR-103, developed a loss of intra-abdominal and subcutaneous white fat and "whitening" of interscapular brown fat,[@bib44] suggesting a potential role of miR-103 in the transformation of WAT and BAT. In this study, with bioinformatics predictive analysis, we found that circNrxn2 and FGF10 can specifically bind to miR-103, and we further explored their promotion of adipose tissue browning. Recent studies have also shown that adipocytes can interact with ATM by transporting miRNAs through vesicles,[@bib45], [@bib46] so we wonder whether the browning of white adipocytes can also be activated by ATM. Under normal circumstances, ATMs disperse in adipose tissue and participate in maintaining adipose tissue homeostasis. ATMs play a key role in the adipose tissue-liver crosstalk in nonalcoholic fatty liver diseases (NAFLDs). Our previous research also found that melatonin, a hormone, could divert adipose-derived exosomes to macrophages, which promoted M2 macrophage polarization to alleviate inflammation.[@bib46] We also found that transcriptional factor Hoxa5 activated the PPARγ signaling pathway to increase M2 macrophage polarization, thereby ameliorating obesity-induced chronic fat inflammation. In obesity, the number of ATMs in WAT increased dramatically and mainly presented in M1 type, a coronal structure.[@bib47] M1 macrophages were involved in the elimination of pathogens through the secretion of proinflammatory cytokines such as IL-6 and TNF-α. M2 macrophages inhibited inflammation and maintained physiological functions of adipose tissue. In addition to secreting the anti-inflammatory factor IL-10, adipose tissue M2 macrophages also played a key role in the browning of white fat. It was found that M2 macrophages secreted a large amount of norepinephrine (about 50% or more of the total) and induced WAT browning, promoted lipolysis, and increased thermogenesis after cold stimulation.[@bib48] Studies have shown that specific knockout of RIP140 in monocytes and macrophages can promote the polarization of adipose tissue M2 macrophages, promote browning of WAT, and improve insulin sensitivity.[@bib49] In addition, it was found that injection of M2 macrophages therapeutically induced WAT browning and improved HFD-induced insulin resistance.[@bib50] The anti-hypertensive drug telmisartan could induce WAT browning by PPARγ-activation-mediated M2 polarization.[@bib51] These studies showed that ATMs play a crucial role in body adipose metabolism. Besides, PPARγ had been reported as a master regulator of WAT browning and insulin sensitizer.[@bib52], [@bib53] Here, we confirmed that circNrxn2 activated the PPARγ signaling pathway to promote WAT browning and increase M2 macrophage polarization. However, there are still some limitations in our research. We find that a new ncRNA regulatory loop, circNrxn2-miR103-FGF10, could regulate WAT browning by affecting ATM polarization. However, it is unclear which specific regulatory factor in macrophages has an effect on WAT browning. Consistent with previous studies, we hypothesized that, due to our treatment, certain molecules in ATM are transported into adipocytes by exosomes and then cause browning. Based on this speculation, we next prepare to extract exosomes of ATM after treatment with circNrxn2, miR-103, and FGF10 and then search target molecules in ATMs that act on browning of adipocytes through proteomics sequencing. These experimental studies will be reflected in our future work. Taken together, the data in our study revealed that circNrxn2 acted as an endogenous miR-103 sponge and relieved the inhibitory effect of FGF10 to promote WAT browning via increasing M2 macrophage polarization, which involved PPARγ activation ([Figure 8](#fig8){ref-type="fig"}). *In vivo*, we show that the delivery of circNrxn2 effectively promotes WAT browning and UCP1 expression in the HFD mice, demonstrating the functional impact of the circRNA on thermogenesis. Transfection of circNrxn2 alone can significantly increase FGF10 levels and promote browning of WAT. circNrxn2, miR-103, and FGF10 interacted with each other to form a regulatory loop in adipose tissue to regulate its browning. Nonetheless, this study provides a good therapeutic strategy for treating obesity and improving obesity-related metabolic disorders.Figure 8circNrxn2 Promoted WAT Browning by Facilitating M2 Macrophage PolarizationcircNrxn2 acted as an endogenous miR-103 sponge and relieved the inhibitory effect of FGF10 to promote WAT browning via increasing M2 ATM polarization, which involved PPARγ activation. Materials and Methods {#sec4} ===================== Animal Experiments {#sec4.1} ------------------ Mice handling protocols were conducted following the guidelines and regulations approved by the Animal Ethics Committee of Northwest A&F University. C57BL/6J mice were purchased from the Laboratory Animal Center of the Fourth Military Medical University (Xi'an, China). Mice were housed at 25°C ± 1°C, 55% ± 5% humidity, in 12-h/12-h light/dark cycles and were provided with water and diet *ad libitum*. 8-week-old male mice were fed with a high fat diet (HFD; fat provides 60% of total energy) for 10 weeks to obtain obese mice. HFD mice were housed at 4°C for 12 h to induce WAT browning. For *in vivo* studies, mice were treated with circNrxn2 lentiviruses, circNrxn2 lentiviruses + miR-103 adenoviruses, miR-103 adenoviruses, miR-103 adenoviruses + FGF10 adenoviruses, or control empty viruses at 1 × 10^9^ plaque-forming units (PFUs) in 0.2 mL PBS through injection (i.p.) for 1 week, respectively. To remove macrophages, half of the mice received a daily injection i.p. of clodronate (Selleck, Shanghai, China), and the other half were injected with PBS. Primary Cell Culture {#sec4.2} -------------------- The proposal of primary preadipocyte culture and differentiation was as described previously.[@bib27], [@bib43] Primary peritoneal macrophages were obtained from 6-week-old C57BL/6J mice at 2--3 days after the injection of 2 mL 4% sterile thioglycolate solution by pelvic washing with PBS containing 3% fetal bovine serum (FBS). Primary cultures of macrophages were maintained in RPMI 1640 medium. All culture media were supplemented by the addition of 10% FBS and penicillin and streptomycin. RNA Extraction and cDNA Synthesis {#sec4.3} --------------------------------- Total RNA, including miRNAs, was extracted using TRIzol Reagent (Invitrogen Life Technologies, Carlsbad, CA, USA). cDNA was synthesized using a PrimeScript II 1st Strand cDNA Synthesis Kit (Takara, Dalian, China). qRT-PCR Analysis {#sec4.4} ---------------- qRT-PCR was used to detect the relative miRNA and mRNA levels with AceQ qRT-PCR SYBR Green Master Mix (Vazyme, Nanjing, China) as described by Cao et al.[@bib44] U6 and GAPDH were used to normalize miRNA and mRNA levels, respectively. The relative gene expression was calculated using the 2^−ΔΔCt^ method. Western Blot {#sec4.5} ------------ Western blot analysis was performed by using standard methods as described by Liu et al.[@bib45] Proteins were separated by SDS-PAGE, transferred to polyvinylidene fluoride (PVDF) nitrocellulose membrane (Millipore, Boston, MA, USA), blocked with 5% fat-free milk for 2 h at room temperature, and then incubated with primary antibodies in 5% milk overnight at 4°C. Then, secondary antibody was added and incubated at room temperature for 2 h. Proteins were visualized using chemiluminescent peroxidase substrate (Millipore, Boston, MA, USA), and then the blots were quantified using the ChemiDoc XRS system (Bio-Rad, Hercules, CA, USA). miRNA Target Genes and circRNA Prediction {#sec4.6} ----------------------------------------- To predict the miRNA target genes, we used three different bioinformatics software tools: TargetScan, miRBase, and PicTar. We chose their intersection to improve the accuracy of the forecast. circRNA prediction was performed in CircNet, circBase, starBase, and Circular RNA Interactome. Biotin-Coupled miRNA Capture {#sec4.7} ---------------------------- Biotin-labeled wild-type (WT) or mutant (mut) miR-103 was transfected into adipocytes. 48 h after transfection, cells were collected and treated as described previously.[@bib31] Dual Luciferase Reporter Assay {#sec4.8} ------------------------------ Dual luciferase reporter assay proposal was performed as described by Liu et al.[@bib45] For luciferase assay, renal luciferase expression plasmid (pRL-TK) vector and pGL3 vector containing a WT fragment of the 3′ UTRs of FGF10 or mutation fragment were transfected into 293T cells with miR-103 or circNrxn2. 48 h later, cells were harvested to analyze the luciferase activity using the Dual Luciferase Reporter Assay Kit (Promega, Madison, WI, USA). Immunofluorescent Staining {#sec4.9} -------------------------- Cells or tissue sections were fixed, permeabilized, and blocked. After incubating overnight with primary antibodies, secondary fluorescent antibodies were added, and DAPI was used for nuclear counterstaining. Data Analysis {#sec4.10} ------------- All experiments were repeated at least three times. Data were expressed as mean ± SEM. The statistical analysis of differences was performed in GraphPad Prism 5.0 using Student's t test. \*p \< 0.05 was considered as statistical significance, and \*\*p \< 0.01 means very significant. Author Contributions {#sec5} ==================== Experimental design: T.Z. and C.S.; Performance of experiments: all authors; Data analysis: Z.Z. and T.X.; Writing the paper: T.Z. All authors are assured that we met the criteria for authorship, and all contributed to and reviewed the manuscript. Conflicts of Interests {#sec6} ====================== The authors declare no competing interests. Supplemental Information {#appsec2} ======================== Document S1. Figures S1 and S2Document S2. Article plus Supplemental Information This study was financially supported by Major National Scientific Research Projects (2015CB943102); the National Natural Science Foundation of China (31572365); the Joint Funds of the National Natural Science Foundation of China (U1804106); the Fundamental Research Funds for the Central Universities (245201971), the Key Sci-tech innovation team of Shaanxi province (2017KCT-24); and the Key Sci-tech innovation team of Northwest A&F University. Supplemental Information can be found online at <https://doi.org/10.1016/j.omtn.2019.06.019>.

Background ========== The major manifestation or clinical consequence of osteoporosis is the appearance of osteoporotic fracture or fragility fracture \[[@B1]\]. It is well known that osteoporotic fractures involve a higher incidence of new fractures and lead to disability \[[@B2]\]. Hip fractures and those of the vertebrae with clinical manifestations are especially important since they carry an increase in mortality \[[@B3],[@B4]\]. There is currently wide consensus regarding the need to develop strategies for the prevention of fractures and in the last years it has been recommended that the decision and the threshold of intervention be based on clinical assessment of risk of fragility fracture \[[@B5]-[@B8]\] and not only on the values of BMD and the relative risk as in the meta-analysis by Marshall D et al. \[[@B9]\]. Multiple epidemiological studies have described different clinical risk factors of osteoporotic fracture (CRFs) and which are been associated with an increased risk of developing osteoporosis and/or fragility fractures. Nonetheless, not all have determined a strong association, and the presence of these CRFs has not been uniform in the different studies and systematic reviews \[[@B10]-[@B14]\]. Most of the most powerful CRFs are concordant in different populations and, in general, similar for different fractures. Fractures related to falls have additional risk factors such as the number of falls, scarce physical activity and others such as the use of a walking stick, the need for help to get up from a chair, etc.). The CRFs associated with lifestyle such as smoking, alcohol intake or caffeine, low calcium consumption and scarce physical exercise have shown greater variability and lesser uniformity among the different studies \[[@B6],[@B7]\]. Finally, the influence of some risk factors on the risk of fragility fracture has been demonstrated in different meta-analyses and systematic reviews \[[@B15]-[@B20]\]. As previously commented, since more than 15 years ago there has been evidence that BMD below the standard values is one of the important risk factors for fragility fracture \[[@B21],[@B22]\]. More recently, however, other CRFs with as great or greater specific weight in the determination of risk of fragility fracture have been reported \[[@B11]-[@B13],[@B22]\]. It is well known that there is an important variation in the relative risk of hip fracture in both men and women at an international level. The WHO itself has performed numerous investigations on this difference. In one of the latest studies this difference was defined as a standardised rate at 10 years, being, in the most extreme cases, 15-fold greater between countries such as Norway and Chile \[[@B23]\]. The studies performed by the Bone and Mineral Research Program, in Garvan Institute of Medical Research show that the combination of BMD and non-invasive clinical risk factors in a nomogram could be useful for identifying high-risk individuals for intervention to reduce the risk of hip fracture \[[@B24]\]. With the objective to make a purpose of when were the better moment and the patient who better benefits of new drugs available for the prevention of osteoporotic, World Health Organization Collaborating Centre for Metabolic Bone Diseases, University of Sheffield, UK developed the FRAX tool. Both are useful tools to estimate absolute risk of fracture for clinical practice but both have limitations: They discriminative ability was only moderate in older women (mean 74 years old) which may limit their clinical utility \[[@B25]\]. Both Garvan and FRAX are widely available tools: <http://www.garvan.org.au/bone-fracture-risk/> and <http://www.shef.ac.uk/FRAX/> but both models still need to be validated in different populations before they can be generalized to other populations and further studies will be needed to validate their contribution in selecting patients who will achieve fracture risk reduction with anti-osteoporosis therapy. With the current available algorithms, a possible clinical application may be to use FRAX as the primary model and to consider using Garvan in patients with recurrent fractures and falls \[[@B25]\]. Since the technical reports of 1994 \[[@B26]\] and their review in 2001 few changes have been made with respect to the WHO recommendations on the management of osteoporosis. In 2007, the WHO published a new tool for the evaluation of absolute risk of fragility fracture: the FRAX tool \[[@B27]-[@B30]\]. This tool was developed by WHO to evaluate fragility fracture risk for a 10 year period in patients for many countries \[[@B31]-[@B33]\]. The extension of a method for calculating the risk (probability) of fractures using the FRAX tool is foreseeable in Spain similar to what is occurring in other countries since its publication \[[@B34],[@B35]\]. But before its clinical use its necessary to validate the calculator in a local cohort \[[@B29],[@B30]\]. Objectives ---------- The objective of this study was to evaluate the discriminative and predictive capacity of the FRAX tool to determine osteoporotic or fragility fracture in Spain at 10 years. This study describes the discriminatory capacity using the AUC-ROC *of the* FRAX tool to determine which Spanish women will have an osteoporotic fracture over the 10 years following the determination of the risk. On the other hand, the global predictive capacity of the FRAX tool has been calculated to detect the osteoporotic fractures on comparing the fractures observed over the 10 years with those expected by the FRAX tool. Methods ======= Methods/design -------------- The protocol, procedures and main characteristics of the study have recently been published \[[@B35]\]. Briefly, the FRIDEX cohort (**F**racture **RI**sk **f**actors and bone **DE**nsitometry type central dual **X**-ray) is constituted of men and women referred by general practitioners and specialists for undergoing central bone densitometry by Dual-energy X-ray absorptiometry (DXA) for the initial study of osteoporosis or treatment follow up, who accept to answer an extensive questionnaire on risk factors (QRF) for osteoporotic fracture (family history of osteoporosis and hip fracture, clinical risk factors and lifestyle habits related to diet and toxic substances) \[[@B35]\]. This cohort was started in 1999 at the Bone Densitometry Unit of the Department of Nuclear Medicine of the University Hospital Vall d'Hebrón, Barcelona, Spain. During the baseline visit at the reference centre informed consent to participate was requested and a QRF for osteoporotic fractures is given during the visit and anthropometric parameters are determined. Ten years after the first QRF and DXA the patients were asked to answer a phone survey to know the evolution of the study variables and outcomes such as new personal or parental fractures, new disease or prescriptions. Study population and enrolment procedures ----------------------------------------- This multicentre study was carried out by family practitioners and other specialists who refer patients to the same reference centre for undertaking BMD. The criteria for referral followed the recommendations of the WHO of not performing a population screening but to select cases among those at greatest risk of having osteoporosis and subsequent osteoporotic fractures or the follow up and control of patients already receiving specific treatment. Participants reside mostly in urban areas and were referred for DXA scan by family doctors, ambulatory specialists and hospital specialists. Randomised sample (simple computerised randomisation stratified by sex) was obtained of women from 40 to 90 years of age in the FRIDEX cohort for 10 years since the baseline DXA and QRF. Eligibility criteria -------------------- ### Patient inclusion criteria The study subjects were Caucasian women, ≥ 40 and ≤ 90 years of age at the time of inclusion in the FRIDEX cohort \[[@B35]\], understood and spoke the Spanish language, were able to respond to the initial questionnaire done at the surgery and a ten-year follow up structured telephone questionnaire (TQ). All accepted to participate in the study providing the corresponding verbal consent. Physically or psychically handicapped patients were included if the relatives or care providers accepted to answer the TQ. Patient exclusion criteria -------------------------- Subjects \< 40 or \> 90 years of age at the time of the first DXA and QRF were excluded since the FRAX tool does not allow the calculation of the adjusted risk outside this age range. Patients with physical or psychological limitations impeding their participation and whose relatives did not accept to respond to the TQ were excluded as were those with Paget's disease, cancer with bone involvement or disease which may simulate osteoporosis (i.e. myeloma). Patients from ethnic groups other than Caucasian were not included since other studies have demonstrated different risk characteristics. Patients not providing consent to the TQ and those without a telephone to contact or did not respond after 3 calls made at different times according to the procedure manual were also excluded from the study. Dead patients were not studied because of the impossibility of obtain all the study variables or to answer the questionnaire by relatives. Data collection --------------- The sample ordering was performed using randomised numbers for each month and the calls were made in this order. The baseline variables of QRF and BMD were collected from January to July 2000. The follow up variables were collected at the same month during 2010 by TQ to complete the 10 years of follow up. The TQ was collected regarding the fragility fractures occurring from the time of inclusion until the date of the TQ as well as other information on known factors of fracture risk and falls. In all cases of fracture the medical records of the patients were reviewed and, when necessary, we requested a medical report for its validation. All cases of fracture that could not be verified or those arising from a motor vehicle accident or major trauma were excluded from analysis. Dead patients were not studied because of the impossibility of obtain all the study variables and to answer the questionnaire by relatives. Baseline variables ------------------ Height, weight, body mass index were obtained during baseline DXA scan. The rest of baseline items were obtained by semi structured questionnaire by interviewer during the same visit. On the other hand, the variables are set according to the instructions of the official website of FRAX \[ <http://www.shef.ac.uk/FRAX/tool.jsp?lang=sp>\]. The variables which are mentioned in the questionnaire were defined as well according to standard units of measurement for each. Regarding the risk of alcohol consumption, the quantification of consumption in standard drinks (UBEs) allows rapid quantification of consumption and its easy conversion into grams of pure alcohol. The value of the UBE in Spain with a slight North--south gap is set to 10 g of alcohol and is equivalent to a consumption of wine (100ml), sparkling wine (100 ml) or beer (200 ml) half and consumption of distilled or combined (25 ml). Weekly risky drinking for women and over 65 years is that is\> 17 UBEs and men\> 28-UBEs. The phone records of alcohol consumption have shown good validity and correlation in Mediterranean countries where alcohol consumption is widespread. Only in case of personal circumstances (deafness, slurred speech, etc.) a part of the information was obtained through regular cohabiting relatives of patients in 15 of 770 cases (1.9%). BMD measurement was determined by central DXA according to the 2007 recommendations of the International Society for Clinical Densitometry (ISCD) (available at: <http://www.iscd.org/Visitors/positions/OfficialPositionsText.cfm>) for the interpretation of the results using a Lunar GE model *Prodigy Advance* densitometer with 11.4 software and with BMD and T-score determination with NHANES III references. The densitometry diagnostic criteria used were the 1994 WHO criteria which classify the results into 3 groups according to the levels of BMD values of the femoral neck: normal (T-score \>−1), osteopenia (T-score between −2.4 and −1 inclusive) and osteoporosis (T-score ≤ −2.5). The estimated absolute risk of fracture during the 10-year period according to the FRAX tool was determined through the official website (version 3.2 accessed on October 2010). The calculations of the probability of fracture with or without the T-score of femoral neck and lumbar spine (L1-L4) were analysed in parallel by two blind investigators (patient entities were kept anonymous and were assigned an alphanumeric code). Two other blinded investigators reviewed the results and recalculated the data on the appearance of any difference. Analysis plan ------------- The hip fractures during the follow up period were taken as the endpoint event. At first, all fractures were collected by TQ (structured interview), but were only included in the analysis if these fractures were verified against patients records. The characteristics of the population were described according to descriptive univariate analysis. We used the Chi-square test to evaluate the association between qualitative variables. The Student's t-test or, if necessary, its non parametric equivalent, the Mann--Whitney U test, was implemented to evaluate the differences in the distribution of a quantitative variable according to the categories defined by a binary exposure. To assess the differences in the distribution of a quantitative variable according to the categories defined by a categorical variable with more than two categories, ANOVA analysis of variance or its corresponding non parametric test (Kruskal-Wallis) were used. The relative risk (RR) was calculated by quotient between prevalence of each risk factor in fractured women and in non-fractured. To know the discriminating ability of the FRAX tool we used AUC-ROC and the Hosmer-Lemeshow goodness-of-fit test. The overall predictive capacity ratio was calculated by comparison of observed fractures (ObsFx) in the cohort and period and the expected fractures (ExpFx) by the FRAX tool \[sum of individual probability of fracture from all women included/100\]. The proportion of fractures expected is calculated by the sum of an individual probability of fracture from all women included/100. Model calibration is done by multiplying the FRAX result by the ratio ObsFx/ExpFx. All the statistical tests were undertaken with a confidence interval of 95% and with the use of the 17th version of the SPSS statistical package. This work follows the guidelines of the STROBE initiative for epidemiological studies \[ <http://www.strobe-statement.org/index.php?id=strobe-publications>\]. Ethics ------ Procedures for human subject protection and the original protocol \[[@B35]\] were approved by the Clinical Research Ethics Committee of the Vall d\'Hebron University Hospital, Barcelona, Spain and by the Ethical Committee of the Institut Universitari d'Investigació en Atenció Primària (IDIAP) Jordi Gol. Barcelona. Spain. Informed consent was obtained before beginning the interviews of all the patients. Results ======= Among the person completing 10 years since their inclusion in the cohort, 1,308 could be contacted for this study (Figure  [1](#F1){ref-type="fig"}). About 69 (5.3%) patients died (43.4%). Thirty nine have been detected by searching the telephone number and detect the death. In the other 30 cases were detected through contact with family and reported only 2 cases of fracture between baseline and the date of death. A total of 770 women fulfilled the inclusion criteria and provided informed consent to participate. {#F1} During the 10 years of study 65 women presented a total of 82 major osteoporotic fractures which could be contrasted: 17 women with 18 hip fractures, 10 with 18 proximal humeral fractures, 25 with 30 forearm fractures, and 14 with 16 vertebral fractures. All the fractures were caused by low intensity impact according to the classical definition of fragility fracture \[[@B26]\]. Table  [1](#T1){ref-type="table"} shows the baseline characteristics of the participants and those selected but did not participate in the study. No significant differences were observed between these two groups except that the participants ware one year younger on average (56.8 vs. 57.8 years) and the participants were taking glucocorticoids (3.7 vs. 5.9%). ###### Risk factors among participants/non participants in the current study of FRIDEX cohort   **Participants** **Non Participants** **p-value** --------------------------------------------------- ------------------ ---------------------- ------------- Age years, mean (SD) 56.8 (8.0) 57.8 (8.5) \<0.001 Weight in Kg., mean (SD) 66.6 (11.5) 65.9 (11.1) ns Height in cm., mean (SD) 155.2 (5.9) 155.6 (6.0) ns BMI in Kg/cm2 , mean SD) 27.7 (4.7) 27.9 (4.2) ns Smoker, n (%) 132 (11.2%) 103 (11.5%) ns Alcohol Risk, n (%) 6 (0.5%) 5 (0.5%) ns Previous Fracture, n (%) 269 (22.8%) 245 (27.4%) ns Parental osteoporosis or fractures, n (%) 185 (15.7%) 138 (15.4%) ns Glucocorticoids, n (%) 44 (3.7%) 53 (5.9%) 0.024 Rheumatoid Arthritis, n (%) 12 (1.0%) 11 (1.2%) ns Calcium/Vit. D Supplements, n (%) 221 (18.7%) 187 (20.9%) ns Active Bone Drugs (antiosteoporotic drugs), n (%) 329 (27.9%) 277 (30.9%) ns Table  [2](#T2){ref-type="table"} describes the main characteristics of the 770 participating women as well as the results of the variables or risk factors included in the FRAX tool and the results of the baseline DXA expressed as the result stratified according to the WHO classification. It also includes the variable of *falls in the previous year* which was assessed at the end of the study. The CRFs showing significant differences between women with fractures and those without fractures are: age, previous fractures, having rheumatoid arthritis and having a diagnosis of osteoporosis on DXA. The relative risks (RR) of the different CRFs are shown separately for major fracture and hip fracture in Tables  [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}. ###### Baseline risk factors and falls in previous year for major fracture   **65 Women with fracture** **705 Women without fracture** **P- value** **CI 95%** **RR** **CI 95% RR** --------------------------------- ---------------------------- -------------------------------- -------------- ------------ ------------- ----------------- Age (SD) 61.2 (9.7) 56.4 (7.7) \<0.001 2.4-7.3 2.62 (\*) (1.63 -- 4.21) Weight (SD) 67.4 (11.5) 66.6 (11.4) 0.559 ns \- \- Height (SD) 155.3 (6.1) 155.1 (5.8) 0.805 ns \- \- BMI (SD) 28.0 (4.4) 27.7 (4.7) 0.653 ns 1.19 (\*\*) (0.31 -- 4.53) Previous Fracture (%) 43.1 18.6 \<0.001 12.1-36.9 2.91 (1.84 -- 4.60) Parental Hip Fracture (%) 15.4 15.6 0.963 ns 0.98 (0.52 -- 1.88) Smoker (%) 9.2 11.3 0.604 ns 0.81 (0.36 -- 1.82) Alcohol Risk (%) 1.5 1 0.508 ns 1.49 (0.23 -- 9.45) Glucocorticoids (%) 7.7 2.8 0.052 ns 1.15 (0.94 -- 1.40) Rheumatoid Arthritis (%) 4.6 0.7 0.024 1.2-9.0 4.61 (1.83 -- 11.63) Falls in previous year (%) 32.3 22.3 0.066 ns 1.59 (0.97 -- 2.60) Osteoporosis (baseline DXA) (%) 50.8 25.8 \<0.001 12.4-37.6 4.96 (‡) (1.98 -- 12.43) Osteopenia (baseline DXA) (%) 41.5 50.9 0.147 ns 2.36 (Φ) (0.93 -- 6.03) Normal (baseline DXA) (%) 7.7 23.3 0.004 8.4-22.8 \- \- RR: Relative Risk. (\*) \< 65 vs. ≥ 65 years. (\*\*) \< 20 vs. ≥ 20. (‡) Osteoporosis vs. normal. (Φ) Osteopenia vs. normal. ###### Baseline risk factors and falls previous year for hip fracture   **17 Women with fracture** **753 Women without fracture** **P- value** **CI 95%** **RR** **CI 95% RR** --------------------------------- ---------------------------- -------------------------------- -------------- ------------ ------------- ----------------- Age (SD) 69.4 (7.1) 56.5 (7.8) \<0.001 9.2-16.7 11.49 (\*) (4.12 -- 32.08) Weight (SD) 64.8 (8.1) 66.7 (11.5) 0.498 ns \- \- Height (SD) 153.1 (7.3) 155.2 (5.8) 0.139 ns \- \- BMI (SD) 27.7 (3.0) 27.7 (4.7) 0.945 ns 1.02 (\*\*) (0.06 -- 16.43) Previous Fracture (%) 47.1 20.1 0.012 3.1- 50.9 3.42 (1.34 -- 8.71) Parental Hip Fracture (%) 17.6 15.5 0.738 ns 1.16 (0.34 -- 3.98) Smoker (%) 0.0 11.4 0.242 ns 0.225 (0.01 -- 3.71) Alcohol Risk (%) 0.0 1.1 1.000 ns 2.42 (0.16 -- 37.26) Glucocorticoids (%) 11.8 3.1 0.102 ns 3.97 (0.96 -- 16.45) Rheumatoid Arthritis (%) 11.8 0.8 0.012 4.3-26.3 12.7 (3.46 - 46.63) Falls in previous year (%) 35.3 22.8 0.246 ns 1.81 (0.68 -- 4.84) Osteoporosis (baseline DXA) (%) 58.8 27.2 0.004 8.0-55.2 7.86 (‡) (1.02 -- 60.80) Osteopenia (baseline DXA) (%) 35.3 50.5 0.215 ns 2.63 (Φ) (0.32 -- 21.65) Normal (baseline DXA) (%) 5.9 22.3 0.106 ns \- \- RR: Relative Risk, ns: no statistical significance. (\*) \< 65 vs. ≥ 65 years. (\*\*) \< 20 vs. ≥ 20. (‡) Osteoporosis vs. normal. (Φ) Osteopenia vs. normal. The values of the different AUC-ROC for major and hip fracture calculated in the cohort of Spanish women are shown in Table  [4](#T4){ref-type="table"}. That is, of BMD by DXA with the T-score of the femoral neck (FN) and with the T-score of spine L1-L4 and the FRAX tool in three ways: without BMD, with the FN T-score and with spine L1-L4 T-score. The best result was for FRAX tool for hip fracture without the T-score (0.888). In all cases the results presented significant differences with the reference (0.50) except for BMD with spine L1-L4 T-score (p=0.067). Figures  [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"} are graphs of the AUC-ROC of the FRAX tool for major fracture and hip fracture. A determination of the AUC-ROC specifically for vertebral fracture was performed, being 0.752 (CI 95%; 0.643-0.861) for the FRAX tool without BMD, 0.815 (CI 95%; 0.725-0.905) with the FN T-score and 0.710 (CI 95%; 0.575-0.844) with L1-L4 T-score, without significant differences among them (p=0.157) (graph not shown). We compare AUC of the ROC curve of FRAX tool for major and hip fracture with a simple model including only age. The AUC in a model that includes only age was 0.668 for major fracture and 0.882 for hip fracture with no significant differences with the results of FRAX tool (p=0.565 and p=0.976 respectively). ###### Area Under Curve (AUC) of Receiver Operating Characteristics (ROC)     **AUC ROC** **CI 95%** **p - value** ---------------------------------------------- ------------------------------------- --------------- --------------- --------------- AUC ROC 10-year prediction of MAJOR FRACTURE BMD with FN T-score 0.661 (0.583-0.739) p\<0.001 BMD with L1-L4 T-score 0.638 (0.565-0.711) p\<0.001 FRAX® tool without BMD 0.693 (0.622-0.763) p\<0.001 FRAX® tool with FN T-score 0.716 (0.646-0.786) p\<0.001 FRAX® tool with spine L1-L4 T-score 0.712 (0.644-0.780) p\<0.001 AUC ROC 10-year prediction of HIP FRACTURE BMD with FN T-score 0.779 (0.631-0.929) p\<0.001 BMD with L1-L4 T-score 0.630 (0.487-0.773) (ns) FRAX® tool without BMD 0.888 (0.824-0.952) p\<0.001 FRAX® tool with FN T-score 0.849 (0.737-0.962) p\<0.001   FRAX® tool with spine L1-L4 T-score 0.767 (0.658-0.876) p\<0.001 AUC ROC: Area Under Curve of Receiver Operating Characteristics; BMD: Bone Mineral Density; FN: Femoral Neck; L1-L4: Lumbar Spine; CI: Confidence Interval; ns: Not significant. {#F2} {#F3} Around 27 major fractures and 6 hip fractures were expected with the FRAX tool without BMD, while around 30 major fractures and 7 hip fractures were expected with the inclusion of the T-score of the femoral neck in the FRAX tool (Table  [5](#T5){ref-type="table"}). ###### Ratio of Observed fractures/Expected fractures by FRAX tool   **MAJOR FRACTURES** **HIP FRACTURES** ---------------------- --------------------- ------------------- ----- ------------- ---- ----- ----- ------------- FRAX without BMD 65 26.7 2.4 (1.9 - 3.1) 17 6.0 2.8 (1.7- 4.6) FRAX with FN T-score 65 29.9 2.2 (1.7 - 2.8) 17 7.3 2.3 (1.4 - 3.8) MAJOR FRACTURES (hip, vertebra, humerus, wrist). Fx: Fracture. ObsFx: Observed fractures; ExpFx: Expected fractures; CI: Confidence Interval; BMD: Bone Mineral Density; FN: Femoral Neck. The ObsFx/ExpFx ratio was 2.4 (CI 95%; 1.9 - 3.1) for major fracture and 2.8 (CI 95%; 1.7 - 4.6) for hip fracture (Table  [5](#T5){ref-type="table"}) with the FRAX tool without BMD and 2.2 (CI 95%; 1.7 - 2.8) and 2.3 (CI 95%; 1.4 - 3.8), respectively with femoral neck T-score. Expressed in percentages, the FRAX tool without BMD predicts 41.1% of the cases of women with major fracture in 10 years and 46% on adding the algorithm of the T-score of the femoral neck, with these values being 35.5% and 42.8% for hip fractures, respectively. With respect to the analysis of the sample of the FRIDEX cohort we performed a goodness-of-fit test which stratifies the results in quintiles of risk associated with quintiles of results of fracture. Figure  [4](#F4){ref-type="fig"} shows the Hosmer-Lemeshow test for major fracture, with the cases of the sample distributed into quintiles and the line of regression for the results of the FRAX tool without BMD and with the FN T-score. The lower part of the figure represents the same results after calibration (simulation) by the number of times that the ObsFx is greater than the ExpFx (Table  [5](#T5){ref-type="table"}). Figure  [5](#F5){ref-type="fig"} shows the results for hip fracture in the same way. {#F4} {#F5} Discussion ========== According to the comparative analysis of the baseline characteristics between the participants and the non participants for any reason we found that the non participants did not differ from the participants except in that they were one year older and more patients were on glucocorticoids. Thus, the participants in the study did not present worst conditions of the cohort. Self-reported generally even structured interview have a significant correlation with those in the medical record. In any case always been found documented as explained. In all cases of fracture the medical records of the patients were reviewed and, when necessary, we requested a medical report for its validation. All cases of fracture that could not be verified or those arising from a motor vehicle accident or major trauma were excluded from analysis, fractures in the history of the subjects under study. A potential limitation of self-reported fractures is in vertebral fractures. In our study the total self-reported fractures were 16% higher than they were registered and so were excluded from the final analysis. It can be an advantage for risk predictions proposed by FRAX. The present study is centered on the discriminatory and predictive capacity of the FRAX. Analysis of the AUC-ROC was used to analyse the discriminatory capacity of this tool. As shown in (Table  [4](#T4){ref-type="table"}) the results of the FRAX without DXA values were greater than the AUC-ROC of BMD with values of the T-score of the femoral neck. Thus, these results demonstrate that the FRAX without the determination of BMD presents a discriminatory capacity not inferior to and even somewhat better than the DXA, according to the AUC-ROC. Analysis of the BMD with the DXA technique for the axial skeleton has traditionally been considered as the best predictive test known to determine fragility fractures \[[@B9],[@B26],[@B36]\] with the strategy of intervention for their prevention in medical practice having been based on this test in Spain \[[@B35]\] and in the remainder of the international scientific community until the appearance of the importance of other risk factors for fracture \[[@B27]-[@B33]\]. On analysing the role of the determination of BMD of L1-L4 in the different tests, it was found that the discriminatory capacity for major fracture using the AUC- ROC was lower than that of the determination of BMD with the T-score of the femoral neck, although statistical significance was maintained (Table  [4](#T4){ref-type="table"}). This inferiority was maintained for hip fracture but with no significant differences since the confidence interval integrates the value 0.50 which is the value of statistical significance for this test. Part of the debate on the possible weaknesses of the FRAX has been centered on the lack of the BMD values of the lumbar spine in its algorithm. This criticism is based on the traditional consideration that the BMD of each area presents the best predictive capacity for fractures in the same area, especially for the vertebrae and the hip \[[@B36]\] and, thus, it has been argued that the prediction of vertebral fractures could be improved. The discriminatory capacity measured with the AUC-ROC worsened with the incorporation of the L1-L4 T-score in the algorithm of the FRAX for major or hip fractures (Table  [4](#T4){ref-type="table"}). This result is congruent, but on introducing the L1-L4 T-score value in the FRAX (as a simulation) to analyse what would happen with vertebral fracture, the result of the AUC-ROC for vertebral fracture worsened slightly with respect to that obtained with the FN T-score, although without significant differences. Thus, on introducing the values of the L1-L4 T-score in the FRAX in this study the result did not provide an improvement in the discrimination of vertebral fractures measured with the AUC-ROC. Although it has described that a correction can adapt the lumbar spine BMD and improve the prediction for major and vertebral fractures of FRAX \[[@B31]\] in our study by incorporating the lumbar spine BMD did not improve the discriminative ability of FRAX measured by AUC with femoral neck BMD neither for major or vertebral fracture (data not showed). The adjusted predictive capacity of the FRAX analysed using the ObsFx/ExpFx ratio was far from the 1 value which would be the desired result in the case of good adjustment of the predictive capacity of the FRAX in our country. In our cohort this ratio was of 2.4 for major fracture and 2.8 for hip fracture. These values improved minimally on the introduction of the T-score of the femoral neck in the algorithm (2.2 and 2.3 respectively). Indeed, the FRAX predicted the risk of major fracture in 41.1% of the women and 35.5% for hip fracture without BMD, with these values improving only slightly with 46% and 42.8%, respectively on performing the BMD with DXA. These data seem to coincide with the analysis recently carried out in two cohorts of French women with a similar overall discriminatory value for fracture and low overall sensitivity (48-50% for FRAX predictions) and better than BMD alone \[[@B33],[@B37]\]. In Spain our group previously demonstrated that the FRAX has good capacity to detect densitometric osteoporosis but also with imbalance in the predictive capacity \[[@B38]-[@B40]\]. Nonetheless, a two recent studies in Spain had shown similar results to ours for major fractures with an ObsFx/ExpFx ratio of 3.1 (CI 95%: 2.8-3.5) and 0.8 (CI 95%: 0.7-1.1) for hip fracture \[[@B41]\]. Although the initial formation of the two cohorts followed very similar schemes, the method of follow up in our study was notably different. In the present study we only analysed fragility fractures reported by the women, which could be contrasted with electronic record or clinical reports. In the second study the results of ratio ObsFx/ExpFx were 0.66 and 1.10 for major and hip fracture respectively \[[@B34]\]. The most important methodological differences were that the study was carry out for a three years period, the authors do not included vertebral fractures \[[@B34]\]. The ROC curve has several problems. For analysis of sensitivity and specificity we have not a gold standard of FRAX for Spanish population. Moreover, ROC needs a gold standard of illness (fracture) and we do not have because of the electronic records are not completely reliable and we needed to make a double check (self-reported validate against records). On the other hand, the area under the ROC curve is important, since it measures the discrimination power of the model. Nevertheless, tests of discrimination alone are not sufficient for model evaluation, since they do not indicate whether calibration is also good \[[@B34],[@B35],[@B42]\]. In our study, on application of the Hosmer-Lemeshow test a good correlation was observed between the different quintiles of risk in all the simulation (Figures  [4](#F4){ref-type="fig"}, [5](#F5){ref-type="fig"}) but with a line which groups the results of the regression deviated from the reference toward the values observed. This circumstance led us to carry out a calibration multiplying each of the values resulting from the prediction made by the FRAX by a constant based on the ObsFx/ExpFx ratio for major fracture and for hip fracture. As shown in the lower part (calibration) of Figures  [4](#F4){ref-type="fig"} and [5](#F5){ref-type="fig"}, on multiplication of the results of the FRAX by the ObsFx/ExpFx ratios, the results with their CI 95% adjust perfectly to the diagonal of reference in the Hosmer-Lemeshow test. The FRAX tool can therefore be considered to present with a poor discriminatory capacity for women to have major osteoporotic fractures within 10 years, with this capacity being good for hip fractures without the need of determining the BMD, although this improves somewhat with its determination. The FRAX tool shows a scarce predictive capacity of the risk of fracture and predicts less than 50% of those which occur. The reason for this underdiagnosis may be because the Spanish cohort introduced as the reference in the FRAX tool is not representative of the current female population since these women present significantly more fractures than those actually predicted by the FRAX tool. We have excluded from the analysis of the cohort of women receiving active treatment for the bone at baseline of the study because of the FRAX has so defined, but we have not been excluded women who received treatment during the 10-year period. This can be a potential confounding factor, however exclude women would mean removing the greatest potential for fracture, but keep going who have received treatment can be reduced the all risk of new fractures observed. Other potential confounding factor can be the Calcium/Vit D supplement intake because we have not excluded at baseline or during the study period. There is important discussion in the literature about the role of these supplements in reducing the risk of fracture, except in a subgroup of patients taking bone active drugs for the potential hypocalcaemia or in patients admitted to nursing homes. These patients are not included in this study. Moreover there is no significant difference between Calcium/Vit D supplement intake between participants and no participants. New epidemiological studies are needed in our country to compare these results on major and other fragility fractures which, although not severe, also affect the quality of life \[[@B43]\]. However, together with other authors in our country \[[@B6],[@B10],[@B34],[@B38]-[@B41]\] we believe that there are sufficient data to promote the habit of investigating the risk factors of fragility fracture among Spanish physicians, especially in primary care, to determine the absolute risk and be able to propose changes in lifestyle in persons with a high risk as well as evaluate which patients should be referred for determination of the BMD by DXA \[[@B38]\]. In our opinion, the current state of the FRAX needs some adjustments such as those proposed in this study. Something similar to this need for adaptation and adjustments happened in Spain with the application of the first Framingham-type cardiovascular risk scales which required adaptations such as the REGICOR scale and others in our country \[[@B44]-[@B46]\]. We know that the promoters of the FRAX are committed to the adaptation of the tool to the different countries with the publication of new studies such as what has been done up to now. We also consider that with improvements this may be a very useful tool especially in the first level of care and this has been demonstrated by the important extension in its use worldwide \[[@B28],[@B35]\]. Conclusions =========== In summary, FRAX without BMD demonstrates a poor discriminative capacity for major fractures and a good discriminative capacity for hip fractures with the AUC ROC for Spanish women but its predictive capacity does not adjust well with the current algorithm leading to underdiagnosis for major fracture and hip fractures. On introducing the values of the L1-L4 T-score in the FRAX tool, the result did not provide an improvement in the discrimination of vertebral fractures measured with the AUC-ROC. Simple models based on age or BMD alone predicted 10-year risk of major and hip osteoporotic fractures, as well as more complex FRAX models. We advise our Spanish colleagues to use the FRAX tool in clinical practice but weighing the resulting value of each individual case of the FRAX without BMD by a calibration value to obtain an absolute risk value of major o hip fracture at 10 years. New studies may allow a single value which is easier to remember in clinical practice. The result obtained will be more adjusted to the reality of the risk of fragility fracture in our country according to the results found in the present and other studies \[[@B34],[@B38],[@B41]\]. Study limitations and strengths ------------------------------- Our study has some strengths and limitations. We assumed that women in the FRIDEX cohort could have a higher risk of osteoporotic fractures than the general population because it is a population that had previously been selected to undergo a DXA scan for some reason. However it is important to know the profile of women who are selected to perform the DXA-scan by general practitioners and other specialists as may higher but close to the general population over 50 years. Fractures occurring in the participants were followed by an ad-hoc TQ taking into account the traditional low response rates by post in previous epidemiological studies conducted in Spanish population \[[@B36]\]. However, all fractures included were verified against patient records. Other potential confounders and biases are that we excluded those who died during the follow-up, the collection of incident fractures is captured in retrospect, the validation records was only for patients with fractures and, as well, usually the electronic registers of fracture tends to be less records than actually occur. To minimize these potential biases we have verified all self-reported fractures and not included in the study which did not fulfill both (self-reported and recorded). Therefore, this type of analysis tends to benefit the predicted fractures in the ratio ObsFx/ExpFx. We are aware that the authors of the FRAX tool apply only the DXA value of the femoral neck because of the absence of improvement in the prediction of major fracture risk with the use of the lumbar spine T-score. This has been one of the main criticisms related to the FRAX tool. As strengths of the study, 4 investigators were involved in the operating systems to verify the calculations of the values of FRAX and all hip fractures included in the analysis were contrasted. The FRIDEX study is a prospective population-based cohort study, being one of the first studies to follow Spanish women over a 10-year period to determine the incidence of fragility fractures. Abbreviations ============= AUC: Area Under Curve; BMD: Bone Mineral Density; CRFs: Clinical risk factors; DXA: Dual-energy X-ray Absorptiometry; FN: Femoral neck; FRAX® and FRAX: Fracture Risk Assessment Tool; FRIDEX: Fracture Risk Factors and Bone Mass Density by DXA cohort; ISCD: International Society for Clinical Densitometry; NAHNES: National Health and Nutrition Examination Survey; NOGG: National Osteoporosis Guideline Group; ObsFx/ExpFx: Observed fractures/Expected fractures ratio; QRF: Questionnaire of Risk Factors; RF: Risk Factor; RR: Relative Risk; ROC: Receiver Operating Characteristics; REGICOR: Registre Gironí del Cor \[ <http://www.regicor.org>\]; TQ: Telephone Questionnaire 10-year follow up; WHO: World Health Organization. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= RA is the principal investigator, project design and direction, preparation and review of the manuscript. GR coordination of field work, preparation and review of the manuscript. GE coordination and management of the cohort, review of the manuscript. AA coordination and analysis of the FRAX values, review of the manuscript. MI, NP, MZ, SG, PS, SS, VB, MR, MCG, FL-E, GL, JJA, GPL, JP-S and AP field work, calculation of the FRAX values and review of the manuscript. JCM statistical analysis and management of the database, review of the manuscript. EG, EC, YF, PT, DP-A and AD-P scientific support and methodological expert, review of the manuscript. All authors have read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2474/13/204/prepub> Acknowledgements ================ This Project was funded by the Spanish Ministry of Health, Instituto de Salud Carlos III (Grant: PI09/90507) and by IDIAP Jordi Gol fellowship. We are grateful to all participants, both professionals (GPs and other specialist) and patients who gave their time and enabled this study to happen.

{#sp1 .7} {#sp2 .8} {#sp3 .9} {#sp4 .10} {#sp5 .11}

Introduction ============ During mammalian fertilization, calcium ions are important for sperm capacitation, motility, acrosome reaction, and penetration into the oocyte ([@B1]). Several ca+^2^-permeable channels have been identified in sperm cells, including high voltage-gated calcium channels, transient receptor potential channels, and cyclic nucleotide-gated channels ([@B2]). CatSpers are unique channels having a six transmembrane- spanning repeat and a single pore resembling that of the voltage-gated Ca^+2-^ permeant channels. These channels are exclusively expressed in the testis and highly conserved in the mouse and human ([@B2], [@B3]). The CatSpers contains four alpha subunits (CatSper 1-4) form a Ca^+2^ selective pore, and two additional auxiliary subunits, CatSper β and CatSper ɣ ([@B2]-[@B5]). CatSper1 and CatSper2 are critical to sperm motility and male fertility; whereas CatSper3 and CatSper3 play an important role in the sperm hyperactivated mobility and acrosome reaction ([@B2], [@B6]-[@B10]). Aging causes many of detrimental alternations in the male reproductive tract, such as testicular damage, decrease in sperm parameters and gene expression. Based on the free radical theory of aging, excessive production of free radicals causes the process of aging ([@B11]). It has been hypothesized that the balance between oxidant and antioxidant spices is one of the key regulators of the aging process. The semen contains a number of antioxidants such as glutathione peroxidase, superoxide dismutase, vitamin E, vitamin C, selenium and carnitine that protect spermatozoa against free radicals ([@B12]). Vitamin E is a lipid-soluble antioxidant that protects sperm cells from lipid peroxidation and oxidative stress ([@B13]). Vitamin E supplementation hasincreased the number and motility of sperm in animals ([@B14], [@B15]). Suleiman *et al* have reported that vitamin E administration improved sperm motility and declined Malondialdehyde (MDA) level in men with asteno or oligoasthenospermia ([@B16]). Conversely, dietary deficiency of vitamin E leads to deleterious changes on the male reproductive tract, such as histological alternations in seminiferous tubules and degenerative spermatogonium ([@B17]). Our previous study indicated that the supplementation with selenium in the aged mice could up-regulate the expression of CatSper genes, and improved sperm quality in the aged mice ([@B18]). Hence, the aim of the present study was to evaluate the effects of vitamin E, synergist of the selenium on the expression of CatSper genes and sperm quality in 11-12 months old aged and 2-3 months old young male mice. Materials and methods ===================== **Chemicals** Vitamin E (ɑ-tocopherol acetate) was manufactured by Sigma Corporation, USA. **Animals** Male BALB/c mice, varying in age (the young group: n=48, 2-3 months old; the aged group: n=48, 11-12 months old) were purchased from the Experimental Animal Center of the Mashhad University of Medical Sciences, Mashhad, Iran. The mice were fed a standard chow and water ad libitum, and exposed to a 12-hour light/dark cycle, at a temperature of 22^o^C. All the experimental protocols were approved by the Ethical Committee of Mashhad University of Medical Science. **Study design** In this experimental study, mice were randomly divided into four groups of twelve animals each (n=12): the Aged control mice (Control 1); the Aged mice receiving vitamin E treatment (Experimental 1); the Young control mice (Control 2) and the Young mice receiving vitamin E treatment (Experimental 2). The control groups received no injection. The experimental groups were administered, intraperitoneally, 106 mg/kg all-rac-a tocopheryl acetate for 35 days ([@B11]). The mice were dissected to collect the left testis and cauda epididymis from each group on the days 21, 28 and 35 after injection. Testis was stored at -80ºC until further analysis and sperm cells from the epididymis were used for sperm parameters. **Sperm quality analysis** Sperm analysis was performed according to WHO protocol given for human sperm examination ([@B19]). **Sperm Motility** The left cauda epididymis was placed in 1 ml of phosphate buffer saline solution. Cauda was minced with scalpels and incubated in a 5% CO~2~ incubator for 15 min. One drop of sperm suspension was placed on a Neubauer chamber, covered by a 22×22 mm cover slip, and the percentage of motile sperm was evaluated under a light microscope at ×400 magnifications. **Sperm Count** Sperms acquired from the epididymis were released into 1 ml of phosphate buffer saline. After 15 min incubation in a 5% CO~2~ incubator, sperm count was determined using a Neubaur hemocytometer under a light microscope (Olympus BH2). The sperm count was expressed as ×10^6^/mL. **Sperm Morphology** One hundred sperm from different fields were counted for each animal to determine the morphological abnormalities. **Sperm Viability** Two volume of Eosin-B was mixed with one volume of sperm sample. Dead sperm cells were stained red while live sperm cells were unstained. One hundred sperm cells were counted for each sperm sample and were expressed as the percent of viable sperm. **RNA isolation** Total RNA was isolated from testis using the RNX plus solution (Cinnagen, Iran) according to the manufacturer\'s instructions. Briefly, the tissue was homogenized in 1 ml of RNX solution and incubated at room temperature for 5 min. Then, 200 µl of chloroform was added to the tube and centrifuged at 12000 g, for 15 min at 4^o^C. The upper phase containing RNA was transferred into a tube and an equal volume of isopropanol was added. The mixture was then centrifuged at 12000 g, at 4^o^C, for 10 min. The pellet was washed with 75% ethanol and resuspended in 50 µl of Diethylpyrocarbonate (DEPC) treated water. The RNA quantity and integrity was evaluated by the absorbance ratio A260/A280 nm and1% agarose gel electrophoresis, respectively. **Reverse transcription** cDNA Synthesis Kit Revert Aid were purchased from Fermentas Corporation (Germany). To synthesize cDNA, 1 µg of total RNA was reverse transcribed with 5X Reaction Buffer, 20 U/μl Ribolock RNase inhibitor, 10 mM dNTP, 200 U/μl MMLV reverse transcriptase, and oligo (dt)18 primer in a 20 μl reaction. The mixture was incubated at 42^o^C for 60 min, and then the enzyme was inactivated at 70^o^C for 5 min. **Real time** Real-time PCR was carried out using SYBER Green fluorescence dye (Fermentas Corporation, Germany) and a M3000P Real-Time instrument (Stratagene, CA, USA). All reactions were run in duplicate for CatSper 1, CatSper 2 and β-Actin. The Primer sets were described elsewhere ([@B20]). Standard curves were constructed by using the logarithmic dilution series of complementary DNA from the high quality sample. Reaction mixtures was included 2 μl cDNA, 12.5 μl SYBER Green PCR Master Mix (Fermentase Corporation, Germany), 0.6 μl reverse primer, 0.6 μl forward primer, and 9.3 μl of water. Reaction conditions were as follows: an initial denaturation step (95^o^C for 10 min) followed by 40 cycles of a three-phase PCR (denaturation at 95^o^C for 25 s; annealing at 60^o^C for 30 s, and extension at 72^o^C for 30 s). β-actin was used as an internal control to normalize the Real time -PCR reaction. Relative gene expressions were calculated by using the Pfaffl method and β-actin as reference gene ([@B21]). **Immunohistochemistry study** Immunohistochemistry was carried out with the sperm cells that were collected from the mouse cauda epididymidis. A 10 μL of sperm sample was pipetted onto a poly-L-lysine-coated slide, air dried and fixed in cold methanol. Antigen retrieval was carried out by heating the slides in Tris/EDTA pH 9.0 at 95^o^C. The slides were blocked in 3% BSA for 2 hours at room temperature. For blocking endogenous peroxides activity, the slides were incubated in 3% H~2~O~2~^-^ methanol for 20 min. After three washes in PBS, slides were incubated in primary antibodies (mouse anti-goat CatSper antibody dilution 1:100) overnight at 4^o^C. The slides were washed three times with PBS, and incubated with secondary antibody (1:100 dilution; HRP-conjugated donkey anti-goat IgG) for 2 hours at room temperature. After 10 min washes in PBS, a 0.03% solution of diaminobenzidine was applied until brown color developed. Following wash, the samples were treated in xylene and dehydrated in a graded ethanol series 70-100%. Samples were mounted with Entellan (Merk, Germany). Slides incubated in PBS without primary antibody were used as negative controls. Microscopic images were obtained on an Olympus BX51 light microscope. After taking photomicrographs, the sperm staining patterns were determined semiquantitatively with the use of a 5-point scale: weak, +; moderate, ++; strong, +++; highly strong, ++++ ([@B22]). **Statistical analysis** All statistical analysis was undertaken sing the statistical package for social sciences (SPSS v.21) software. Data were analyzed using ANOVA followed by Duncan post hoc test. P-value\<0.05 was considered statistically significant. Results ======= **Effects of Vitamin E treatment on sperm parameters in the aged and the young male mice** The results of sperm characteristics for all groups are summarized in [Table I](#T1){ref-type="table"}. The analyses of variance demonstrated a significant difference between the aged control and the young control group (p\<0.001). Treatment of mice with vitamin E caused a remarkable increase in the sperm count on the day 21 and 28 of the vitamin administration. Our findings revealed an increase in sperm motility of mice treated with vitamin E at the end of the experimental period, though this increase was not statistically significant in the experimental group 2 (p\>0.05). In addition, the percentage of normal sperm morphology was increased in mice that received 106 mg/kg vitamin E. The types of sperm abnormalities noted were coiled tail, cooked tail, and vacuolated head. The percentage of live sperm cells was not affected by the vitamin treatment in both aged and young mice (p\>0.05). **Effects of Vitamin E treatment on CatSper genes expression in the aged and the young male mice** The results of relative expression of CatSper genes are shown in [Figure 1](#F1){ref-type="fig"}-[2](#F2){ref-type="fig"}. We observed different pattern in the expression of CatSper genes in the testis of the aged and young mice. The relative expression of CatSper 1 in the experimental group 1 was 2.40-fold greater than in control group 1 (2.40±0.66 vs. control group, p\<0.01). The expression level of CatSper 1 was decreased at 28 days (0.85±0.24 vs. control group), and then the expression increased distinctly at 35 days (3.05-fold). At 21 days, the relative expression of CatSper 1 gene was 2.10-fold greater in experimental group 2 than those in the control group2 (p\<0.001), reached a peak at 28 days, and then declined to its lowest level at 35 days ([Figure 1](#F1){ref-type="fig"}). The level of CatSper expression in experimental group1 decreased at 21 days at 0.70±0.24 fold change (p=0.001), compared to control, and then gradually increased at 28 and 35 days of vitamin treatment. The relative expression of CatSper 2 in the experimental group 2 increased at 21 days. The expression level of CatSper 2 declined significantly at 1.40-fold change at 28 days (1.40±0.41 vs. control group 2, p\<0.01) and continued to decrease at 35 days ([Figure 2](#F2){ref-type="fig"}). [Figure 3](#F3){ref-type="fig"} shows the immunohistochemistry localization of CatSper protein and [Figure 4](#F4){ref-type="fig"} and [Figure 5](#F5){ref-type="fig"} demonstrate the estimate of immunoreactivity of CatSper 1 and CatSper 2 proteins in mouse sperm. Positive staining for CatSper proteins are shown in brown. Immunohistochemistry shows that CatSper 1 and CatSper 2 proteins are localized throughout the head and the flagellum ([Figure 3 C-F](#F3){ref-type="fig"}). The intensity of CatSper staining in the mouse sperm from all groups was quantified. Staining immunoreactivity for CatSper1 was strong in the middle piece and weak in the head and the principal piece of sperm in the experimental group1 on the day 21. As vitamin treatment progressed from days 21 to days 28, CatSper1 reaction dramatically decreased in the middle piece of the sperm, but high level of staining was detected on the 35 day. The staining intensity increased in the middle piece of the sperm tail in the experimental group 2 on day 21 of vitamin administration. The amount of CatSper1 reaction increased gradually until day 28 of vitamin treatment, but low level of staining was recorded afterwards in the experimental group 2 ([Figure 4](#F4){ref-type="fig"}). Moderate CatSper2 immunoreactivity was observed in the head and the tail of sperm on the days 21 and 28 whereas a strong signal was detected in the middle piece on the day 35 in the sperm cells of the experimental group1 ([Figure 5](#F5){ref-type="fig"}). On the 21^st^ day of vitamin treatment, strong labeling was detected in the experimental group 2 in the middle piece of sperm tail but moderate in the head and principal piece of sperm tail. After this period, CatSper 2 reaction did not indicate remarkable changes. These findings were in agreement with their relative expression of CatSper genes. No staining was observed in control slides when PBS was used instead of primary antibody. ###### Effect of vitamin E treatment on sperm parameters in the aged and young mice **Sperm Parameters** **11-12 months old** **2-3 months old** **p-value** ^\*^ ---------------------- -------------------------- -------------------- ------------------ ------------------ ----------------- --------- Day 21 Concentration (10^6^/mL) 3.80 ± 0.11^a^ 4.18 ± 0.24^b^ 4.40 ± 0.18^b^ 4.76 ± 0.37^cd^ \<0.001 Motility (%) 42.00 ± 3.16^a^ 51.25 ± 2.32^b^ 77.75 ± 2.06^c^ 86.50 ± 2.38^d^ \<0.001 Viability (%) 70.25 ± 4.50^a^ 69.75 ± 0.50^a^ 83.75 ± 6.40^a^ 83.75 ± 4.35^a^ \<0.001 Normal morphology (%) 70.50 ± 3.32^a^ 75.25 ± 7.14^ab^ 81.00 ± 2.71^bc^ 85.00 ± 3.74^c^ 0.001 Day 28 Concentration (10^6^/mL) 3.87 ± 0.14^a^ 4.10 ± 0.16^b^ 4.35 ± 0.23^c^ 4.92 ± 0.07^d^ \<0.001 Motility (%) 40.50 ± 4.12^a^ 41.50 ± 4.20^a^ 79.00 ± 2.94^c^ 87.00 ± 1.82^d^ \<0.001 Viability (%) 71.50 ± 5.07^a^ 70.25 ± 4.27^a^ 82.75 ± 5.50^b^ 82.75 ± 5.50^b^ \<0.001 Normal morphology (%) 70.00 ± 4.08^a^ 74.25 ± 6.13^ab^ 79.25 ± 0.96^bc^ 86.00 ± 4.24^c^ \<0.001 Day 35 Concentration (10^6^/mL) 3.75 ± 0.17^a^ 4.11 ± 0.40^ab^ 4.30 ± 0.29^b^ 4.23 ± 0.93^b^ 0.002 Motility (%) 39.50 ± 0.58^a^ 51.00 ± 2.44^b^ 78.25 ± 3.95^c^ 82.00 ± 2.44^c^ \<0.001 Viability (%) 71.75 ± 3.59^a^ 69.50 ± 3.11^a^ 84.00 ± 5.10^b^ 84.00 ± 4.40^b^ \<0.001 Normal morphology (%) 72.25 ± 4.50^ab^ 77.25 ± 1.71^bc^ 78.00 ± 6.48^bc^ 87.00 ± 1.82^d^ \<0.001 Values are expressed as Mean±SD n=12 mice per group. Data were analyzed by ANOVA test (p\>0.05). Different superscripts differed significantly within row at p\<0.05 by Duncan post hoc test. {#F1} {#F2} {#F3} {#F4} {#F5} Discussion ========== The results of the present study indicated that vitamin E treatment increases sperm parameters in the aged mice. Similarly, our previous study showed that administration of 0.2 mg/kg selenium improves sperm quality especially morphology and viability rates in the aged mice ([@B23]). Several mechanisms have been enunciated to explain the effects of aging process on sperm parameters ([@B24]). Epididymis plays an essential role in sperm motility. In epididymis, the site of sperm capacitation and maturation, aging process results in detrimental effects that this may explain how motility of spermatozoa might be affected by age. Sperm morphology is a sensitive index for the status of the germinal epithelium. Age-related changes may affect sperm morphology. Seminal vesicle secretes approximately sixty percents of the semen volume. A decrease in ejaculate volume could be caused by seminal vesicle inadequacy. Besides, smooth muscle atrophy in prostate may affect semen volume ([@B24]). There is a little data in the literature on the CatSper genes regulatory factors and the effects of antioxidant diet on the expression of CatSper genes. Previous studies have shown that gene expression affected by the antioxidants. Our results showed that administration of 106 mg/kg vitamin E could up-regulate CatSper 1 and CatSper 2 genes in the aged as well as young mice. However, administration of vitamin E in the aged mice caused more increase than that in the young mice. At the same time, a remarkable improvement was observed in sperm quality of the aged mice following vitamin E treatment, showing that there was a positive correlation between vitamin E treatment and improvement of sperm parameters. These results were consistent with our previous results that administration of 0.2 mg/kg selenium was significantly increased sperm parameters as well as CatSper genes expression ([@B18]). The relative intensity of CatSper genes expression in the aged mice treated with selenium was highly expressed on the days 21 and 28 but weakly expressed on the days 35 and 42 compared to those of the young mice. In the present study, a peak expression was observed in the aged treated with vitamin E on the day 35, whereas the highest level of CatSper genes expression was detected on the days 21 and 28, which were disagree with the results of our previous study. CatSper1 gene expression changes were more than that of the CatSper 2. In consistent with our results, among all four members of the CatSper genes family, more changes were observed in the expression of CatSper1. It seems that CatSper1 has more powerful regulatory effects. Luo *et al* reported that CatSper1 mRNA was found to be down-regulated in a rat model of cyclophosphamide -induced oligoasthenozoospermia ([@B25]). However, administration of high-dose Yijingfang, a Chinese herbal decoction, increased CatSper1 expression in the epididymal sperm. Besides, sperm count and sperm motility improved following 35 days treatment with high dose of Yijingfang. Similar to this finding, we found that sperm parameters increased by up-regulating of CatSper1 and CatSper 2 expressions following dietary treatment with vitamin E. A study in Canada demonstrated supplementation with vitamin E affects the expression of glutathione S-transferases pi, 8, mu, and superoxide dismutase, all related to oxidative stress in the aged rats. Moreover, vitamin E deficiencies lead to increased expression of oxidative stress-related genes in the aging rat epididymis ([@B26]). In another study, Gan *et al* showed that intra peritoneal injection of 20 µg/kg selenium increased glutathione peroxidase (GSH-Px) expression in the testis and the liver ([@B27]). However, selenium treatment with 40 and 80 µg/kg predominantly decreased GSH-Px mRNA levels in the rat liver and the testis. This result was consistent with our study that an appropriate dose of the vitamin E, as an antioxidant, can increase genes expression in the testis. However, more researches are required to validate these findings. Our immunocytochemistry studies showed the localization of CatSper1 and CatSper2 in the head, middle piece and principal piece of sperm tail. Previous studies have showed that CatSper proteins were found in both the middle piece and acrosomal region of sperm ([@B7], [@B28]). CatSper proteins were also localized to the principal piece of the sperm flagellum ([@B7], [@B8], [@B28]). According to their resembling domain structure, CatSper protein family was suggested to form a heteromeric channel in sperm but heterogeneous expression of CatSper genes family confirmed the hypothesis that CatSper 1 and CatSper 2 are likely responsible for the separate channels of CatSper 3 and CatSper 4 ([@B3], [@B9]). Our results also support this hypothesis. Staining intensity of CatSper1 and CatSper2 proteins was strong in the vitamin E treated mice; in particular, in the middle piece of flagellum. Interestingly, data obtained from CatSper1 and CatSper2 expression were consistent with their relative mRNA level. These findings showed that vitamin E treatment affected in both level of mRNA and protein. Conclusion ========== The present study demonstrated that administration of 106 mg/kg Vitamin E could up-regulate CatSper1 and Catsper2 genes expression, which are two of the responsible genes for sperm mobility. Vitamin E treatment can improve sperm parameters, especially in terms of sperm concentration, motility and morphology rate; therefore, it seems that vitamin E treatment may be useful for treatment of aged subjects as well as infertile men with oligoasthenoteratospermia. This research project was supported by a grant (No. 900714) of Vice Chancellor for Research of Mashhad University of Medical Sciences. The authors are grateful to Fatemeh Motejadded for her excellent technical assistance. Conflict of interest ==================== There is no conflict of interest in this paper.

Background ========== Urbanization has a significant impact on the economy, lifestyles, ecosystems and disease patterns, including malaria \[[@B1],[@B2]\]. An estimated 39% of the population in sub-Saharan Africa (SSA) lived in urban areas in 2003 \[[@B3]\], 198 million Africans lived in urban malaria-endemic areas and 24--103 million clinical attacks occur annually in those areas \[[@B4]\]. An important message addressed in the Pretoria Statement on urban malaria was that the malaria control strategies used in rural areas cannot be directly transferred to the urban context \[[@B5]\]. The epidemiology of urban malaria poses a number of specific challenges: i) the first malaria infection occurs often late in childhood and the acquisition of semi-immunity is delayed \[[@B6]\]; ii) the intensity of the malaria risk is often heterogeneous over small distances, being subjected to the degree of urbanization of particular subdivisions \[[@B7],[@B8]\] and their proximity to possible vector breeding sites \[[@B9],[@B10]\]; iii) rural-urban migration is likely to increase the endemicity of malaria \[[@B11]\]; iv) agricultural and animal husbandry are important economic activities which create a favourable environment for *Anopheles*breeding \[[@B12],[@B13]\]; v) marginalized populations usually lack access to health care, which hampers the effectiveness of case management and the promotion of intermittent antimalarials during pregnancy \[[@B5],[@B14]-[@B16]\]. There is now substantial private sector activity in health care provision in many cities. The private services providers are often untrained or unlicensed, but are seen as a source of inexpensive care by patients. There is not much information about the impact of the private sector on case management. Around 235 papers related to malaria epidemiology in SSA urban settings were published from 1945 to 2004. Entomological profiles and clinical patterns are known to vary between urban, suburban and rural environments \[[@B17]\]. A review of other studies in SSA urban centres showed that transmission patterns vary greatly by city, season and age group. The overall prevalence of parasitaemia was 4.0% in schoolchildren in Brazzaville \[[@B18]\], 2.4--10.3% in Lusaka \[[@B19]\], 2.0% in a Gambian urban area \[[@B20]\] and 3.6--7.5% in Dakar \[[@B21]\]. It was also reported that malaria prevalence in school children varied from 3.0% to 26.4% in different areas of Ouagadougou \[[@B22]\] and varied from 14% in a central urban area to 65% in peri-urban areas in Kinshasa \[[@B23]\]. Evidence showed that the rate of clinical malaria attacks detected in urban health facilities was high and season-dependent. For example, Hendrickse *et al.*found that 36.8% of outpatients were parasitaemic in a hospital in Ibadan \[[@B24]\]. In Niamey, the parasite prevalence was 61.9% during the rainy season but only 5.4% in the dry season in 1989 \[[@B25]\]. In Kinshasa, malaria admissions comprised 29.5% of consultations in 1983, then 38.2% in 1985--86 \[[@B26]\]. In Dakar, malaria fever represented 19.7% of consultations and 34.3% of fever cases were caused by malaria in 1988 \[[@B27]\]; the same authors found that 5.3% (dry season) and 58.8% (rainy season) of febrile outpatients were parasitaemic in 1994 \[[@B28]\]. In Ouagadougou, malaria prevalence accounted for 33% of all outpatients \[[@B29]\], while Dabire reported 22% malaria parasitaemia among children aged 0--14 years in the paediatric ward \[[@B30]\]. Transmission and severity of malaria are influenced by the geographic characteristics of a town and by the socio-economic environment. The heterogeneity and seasonal variation of the entomological inoculation rate, depending on both vector densities and sporozoite rates, have been documented \[[@B31],[@B32]\]. Lindsay *et al.*(1990) showed a difference in the composition of vector species and the vector\'s adaptation in different subdivisions Banjul \[[@B20]\]. To improve interventions, the determinants of the diversity of transmission levels within subdivisions of a city should be understood. Concerns were raised about the association between urban agricultural activities or local irrigation systems and the creation of breeding sites for *Anopheles*sp. \[[@B12],[@B33],[@B34]\]. Peri-urban areas often lack infrastructure, including poor water supply and sanitation, which provides an ideal environment for vector breeding \[[@B35]\]. For example, urban Dakar has \>5,000 market-garden wells which provide permanent sites for mosquito larvae \[[@B13]\]. An identification of vector species, regular larval inspection and larviciding activities should be implemented in the framework of urban malaria control programmes \[[@B36]\]. This article presents the experience of developing a rapid urban malaria appraisal (RUMA) in SSA, carried out with the support of the Roll Back Malaria Partnership. The aims were i) to develop a rapid assessment package that is explicitly evidence-based and can be carried out within a six to ten weeks timeframe; and ii) to assess how rapid malaria appraisal efforts could be best integrated into the municipal health department supervision and to inform control programmes. Methods ======= Study sites ----------- The fieldwork took place in Yopougon municipality/Abidjan (Côte d\'Ivoire), Ouagadougou (Burkina Faso), Cotonou (Benin) and Dar es Salaam (United Republic of Tanzania) (Figure [1](#F1){ref-type="fig"}). {#F1} Abidjan is the economic capital of Côte d\'Ivoire. It is located between latitude 3.7° N--4.0° N and longitude 5.7° E--6.0° E, with a surface area of 454 sq. km. The study was carried out in the large commune of Yopougon (population: 775,000 in 1998) located in the west of Abidjan \[[@B37]\]. The fieldwork in Yopougon municipality (Abidjan) took place from August to September 2002. Ouagadougou, the capital of Burkina Faso, is situated on the Sahelian border between latitude 12.0° N--13.0° N and longitude 1.15° E--1.40° E. The total surface area was estimated to be around 570--655 sq. km in the year 2000 \[[@B38]\]. The population of Ouagadougou was around 1,100,000 inhabitants in 2002. The fieldwork in Ouagadougou took place from November to December 2002. Cotonou is the economic capital of Benin. It is located on a strip of land between Lake Nokou and the Gulf of Guinea (between latitude 6.2° N--6.3° N and longitude 2.2° E--2.3° E). The total population was estimated at 780,000 inhabitants on a territory of 73.8 sq. km in 2002 \[[@B39]\]. The fieldwork in Cotonou took place from February to March, 2003. Dar es Salaam is situated between latitude 6.0° S--7.5° S and longitude 39.0° E--39.6° E on the East African coast. There are 2,500,000 inhabitants on a total surface area of 1,393 sq. km \[[@B40]\]. The fieldwork in Dar es Salaam took place from June to August, 2003. Study design ------------ In July 2002, a generic RUMA protocol was developed based on existing urban malaria research protocols \[[@B41],[@B42]\]. The relevant institutions in each setting were contacted and city-specific proposals were then produced. Parts of health facilities mapping, school and health facility-based survey activities were integrated into the routine surveillance and health system evaluation at the municipal level. All the fieldwork was completed in August 2003. Final reports were completed in June, 2004. The six key components of the RUMA were the following (see also Table [1](#T1){ref-type="table"}): ###### Study design and methodology of RUMA. **Key measures** **Epidemiological measures** **Spatial relationships** **Individual variations** **Institutional factors** ------------------------------------------------- -------------------------------------------- ----------------------------------------- --------------------------- --------------------------- --------------------- -------------------- ------------------------ -------------- -------------------- ---------------------------- --------------------------------- --------------------------- Methodology Age-specific morbidity and mortality rates Fraction of malaria-attributable fevers Overall endemicity Gradient of malaria risk Environmental risks Travelling history Socio-economic factors Bednet usage Treatment strategy Public/private partnership Coverage of treatment providers Degree of drug resistance 1\. Literature review **x** x x x x x x x 2\. Collection of health statistics x x x x x 3\. Risk mapping x x x 4\. School parasitaemia survey x x x x x x x 5\. Health facility-based fever survey x x x x x x x x 6\. Brief description of the health care system x x x 1\. Literature review. A search of the PUBMED bibliographic database was conducted for the time period from 1960 to April 2004, using the terms \"malaria\", \"urban\" and \"sub-Saharan Africa\". The search was limited to the articles published in English, Chinese, French and Spanish. The reference list of all identified papers was screened. Thesis abstracts filed in the medical libraries of universities and national hospitals were collected at each site and local researchers were also contacted. 2\. Collection of routine health statistics. Local experts in ministries of health (MOH) (disease surveillance systems, municipal health departments and national malaria control programmes) and national census and statistics bureaus were contacted to collect demographic data, health system information and statistics, including routine malaria morbidity and mortality reports. 3\. Mapping of health care facilities and major *Anopheles*breeding sites. Three or four trained workers carried out the health facility mapping under the guidance of local health personnel. In order to identify *Anopheles*breeding sites, simple larvae sampling was performed with the assistance of entomological technicians in Dar es Salaam and Ouagadougou. The duration of these tasks varied by site: 12 weeks during the rainy season in Dar es Salaam and around three weeks during the dry season in Ouagadougou. Due to security issues and technical problems, the mapping of breeding sites and health facilities could not be performed in Yopougon municipality (Abidjan) and Cotonou. 4\. School parasitaemia surveys. School surveys were aimed at determining the local endemicity and risk gradient of malaria. In each city, three to four schools with different malaria endemicity (centre/low, intermediate/medium and periphery/high) were investigated. It is a rapid assessment with limited budget; therefore, in each area only one health facility and school were selected for the surveys. The schools were selected near the selected clinics. 200 school children aged 6--10 years were recruited in each school. Additional information on children was collected using a questionnaire with the assistance of teachers (see [Additional file 1](#S1){ref-type="supplementary-material"}). 5\. Health facility-based surveys (See [Additional file 2](#S2){ref-type="supplementary-material"}). The facility-based fever surveys focused on the age-specific fraction of malaria-attributable fevers \[[@B43]\]. Each city was categorised into three to four areas (centre, intermediate, periphery and rural areas) and one clinic from each area was chosen. Health facilities with a high enough volume of outpatients per day were considered for the survey. In urban areas, an estimated 5% to 50% of fever cases among children under 15 years old were due to malaria. A sample size of 200 in each facility gave an estimate of the proportion of cases with parasites with the following approximate lower 95% confidence limits (at 5%, lower 95% CI: 2; at 50%, lower 95% CI: 6). In each clinic, 200 fever cases and 200 non-fever controls were recruited, with half of them being aged \<5 years. Outpatients with a history of fever (past 36 hours) or a measured temperature ≥ 37.5°C were defined as cases. Controls were recruited from another department of the same clinic without current or recent past fever, matched by age and residency. Electronic thermometers were used to measure the armpit temperature. A \"normal\" body temperature is referred to as an oral temperature of 37°C. An armpit temperature reading is usually 0.3°C to 0.6°C lower than an oral temperature reading. Therefore 0.5°C was added to the temperature displayed on the digital readout. Thick and thin blood films were taken to identify malaria infections. Using 100× magnification to read the thick smears, all malaria trophozoites and gametocytes were counted separately. Parasite density was calculated according to parasites per 200 white blood cells in a thick film (assuming 8000 white blood cells per ml of blood). If 200 white blood cells were counted and less than 9 malarial parasites found, the counting continued until 500 white blood cells were identified. 6\. Brief description of the health care system. It focused on i) the municipal malaria control and prevention efforts, ii) the levels and coverage of service delivery, iii) disease surveillance systems, iv) malaria case management and v) trends of parasite resistance to antimalarials. Quality assurance for blood slides ---------------------------------- The diagnostic performance and the quality of blood sample readings were checked twice: first in the field and then at the reference laboratory of the Swiss Tropical Institute (STI) in Basel, Switzerland. The results in Yopougon municipality (Abidjan), Dar es Salaam and Ouagadougou were: sensitivity 87.9%, 83.5% and 98.7%; specificity 89.2%, 99.0% and 98.2%; accuracy rate of slide readings 88.8%, 98.5% and 98.6%. The quality control process was not implemented in Cotonou due to operational problems. Costing ------- The financial cost of the resources required for a RUMA were calculated for each site based on local market prices and salary standards, except for the laboratory material that was purchased in Switzerland. All expenses fell into seven categories: salaries, transportation, communications, stationery, laboratory materials, other cost and administrative fees (Table [2](#T2){ref-type="table"}). A project team was assembled within the existing structure of partner institutions and then the accountants in each site used a setting-specific cost model to identify the cost factors and determine their local value. The preparation and training cost, programme and administrative costs with the partner institution were estimated and an allowance was added for unforeseen circumstances in the finalized budget. The cost for resources like microscopes and drugs for treatment, vehicles and computers were calculated according to the cost structure of the host institution. ###### Budget categories. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ **Type of cost** **Categories** **Valuation** **Information source** -------------------------------------------- ------------------------------------------------ -------------------------------------------------------------- ----------------------------------------------------------- Human resources Project staff\ Gross salary\ Salary slips or personnel records from the project office Health sector staff Per diem Transportation Project vehicles, petrol and maintenance\ Petrol and maintenance of vehicles based on vehicle logbook\ Bills and receipts\ Taxi, motorbike and bus Shipping and packaging Actual expenditure Freight cost Tickets and receipts Invoices Communication Postage and telephone bills Bills or contract documents Stationery Office maintenance cost\ Actual expenditure for items Agreement with site\ Survey materials\ Agreement and receipts\ Photocopies Lap top and printer use Standard local cost\ Agreement with site Laboratory materials & drugs for treatment International trade good price Invoices Other items Bills and receipts Administration Rent of project office, computer and vehicles Agreement with site ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Results ======= One of the principal aims of the present work was to review the feasibility, perceived usefulness and consistency of the collected information. Because RUMA was a cross-sectional assessment the external validity of the findings could not be assessed. However, the internal consistency of the results was assessed. Below, the strengths and weaknesses of each methodology are presented, bearing in mind the constraints imposed by a rapid assessment. Detailed results for each site will be provided in a series of forthcoming publications. Literature review (Tables [1](#T1){ref-type="table"} and [3](#T3){ref-type="table"}) ------------------------------------------------------------------------------------ ###### RUMA methodology strengths and weaknesses. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ **RUMA Methodology** **Strengths** **Weaknesses** ------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------- Literature review • Time-saving, can be done before and afterwards\ • Incomplete information in time and space • Can identify qualified local expertise\ • Comparison of the malaria patterns and trends Collection of health statistics • Good description of malaria burden over a longer time period • Completeness and quality of data Cross-sectional mapping of healthcare facilities & major *Anopheles*breeding sites • Visualization of information for policy makers\ • Time consuming and only limited scale possible\ • Helps to plan urban health programmes and upgrade community infrastructure • Breeding sites may be transient /seasonal School parasitaemia surveys • Good estimates of local endemicity and local risk factors\ • Limited representativeness if only small number of schools were sampled • Good description of fever prevalence in school\ • Malaria risk gradient Health facility-based fever surveys • Estimates malaria-attributable fevers and prevalence of clinical malaria\ • Limited representativeness due to attendance bias • Description of fever management Brief description of the health care system • Understanding of the structure of city health department and of current malaria control activities\ • Only focuses on the available information\ • Limited cost\ • Depends on the efficiency of information dissemination within municipal departments • Review of the efficacy of case management ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ The systematic review of all literature in each city allowed the collection of background information in a time-efficient manner. A substantial body of information was found in each setting, although it was often incomplete in place (for example covering only a part of the city), in time (few time points, only one season) and in content (not all subject areas covered). For the period 1945 to 2004, a total of 109 papers was found (18, 23, 29 and 39 for Abidjan, coastal Benin, Dar es Salaam and Ouagadougou, respectively), relating to malaria epidemiology, socio-economic risk factors of malaria, entomology and drug resistance \[[@B44]\]. Collection of health statistics (Tables [1](#T1){ref-type="table"}, [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}) ------------------------------------------------------------------------------------------------------------------------------ ###### Reported simple malaria cases among total consultations in 4 African cities, all ages. CHU = Centre Hospitalier Universitaire. **a) Abidjan 2001** --------------------------- -------------------- -------- ---------- ----------- --------- ----------------------- ----------------------- ----------- -------------------- Communes Adjamé & Attécoubé Cocody Yopougon Abobo Plateau Treichville & Marcory Port-Bouét & Koumassi Total \% of admission^†^ Health centers 35,714 55,500 \- 71,437 \- \- 62,607 225,258 CHU No CHU 2,525 \- No CHU No CHU 12,375 No CHU 14,900 Total 35,714 58,025 \- 71,437 \- 12,375 62,607 240,158 40.2 **b) Ouagadougou 2001** Sanitary District Kossodo Paul VI Pissy Secteur 30 Total \% of admission Total 16,007 24,527 95,868 67,064 203,466 29.3--41.4 **c) Dar es Salaam 2000** District hospitals^‡^ Ilala Kinondoni Temeke Total \% of admission Total 178,016 498,991 395,566 1,072,573 45.4--53.7^‡^ **d) Cotonou 2002** Sanitary District I II III IV V VI Total \% of admission Total 6,759 9,678 17,339 7,108 29,890 29,483 100,257 32.1--35.9 ^†^Reported number of malaria cases divided by the total number of consultations. ^‡^Both Ilala and Temeke district hospitals have malaria reported weekly and monthly. The raw dataset of malaria reports of district hospitals in Kinondoni was missing in 2001. Total numbers of consultations were estimated. The routine weekly or monthly malaria reports provided a baseline on the burden of malaria in public health facilities, as well as an assessment of the scale of malaria treatment. Overall, case detection in the antenatal clinics and public health services was poor and reporting was not systematic and consistent. In Abidjan, data were collected from the national malaria control programme (Table [4a](#T4){ref-type="table"}). Age-specific monthly data were available. The statistics for 2001 from four out of 10 communes were missing. The malaria cases reported from the main hospitals (Centre Hospitalier Universitaire-CHU) in Yopougon, CHU Cocody and CHU Treichville were separated from the commune data. CHU receive many referral patients and the malaria cases may therefore be over-reported. The data from CHU Yopougon were missing for 2001. In Ouagadougou, the number of malaria-specific cases and the total number of consultations were collected. The raw data were available by season for 1999--2001, but not for 2002. All the data were missing for Paul VI sanitary district from October to December 2001. The reporting of clinical malaria was also inconsistent in Paul VI (Table [4b](#T4){ref-type="table"}). In Dar es Salaam, the weekly malaria reports were collected from the Ilala, Kinondoni and Temeke district health departments. The data were available for 2000-mid 2003, two months before the survey. A discrepancy in records in Kinondoni District was found, as not all health facilities sent their weekly reports to the district municipal office. Moreover, the sums of reported malaria cases in the raw dataset and in the final district reports were not identical. The Kinondoni district health department had lost all of its 2001 weekly reports (Table [4c](#T4){ref-type="table"}). Only Cotonou had complete data sets for 1996--2002, but the raw datasets were unavailable. Hence, it was impossible to review the consistency and accuracy of the data (Table [4d](#T4){ref-type="table"}). Overall, considerable gaps were found in the routine surveillance systems, particularly for remote health services. Often, the data were collected and presented in different formats, making a generalization impossible and this limited their usefulness. Furthermore, the municipal health departments simply summed up the total numbers of reported cases as they lacked the capacity to analyse these data and to extract useful information for management purposes. Mapping activities (Tables [1](#T1){ref-type="table"} and [3](#T3){ref-type="table"}) ------------------------------------------------------------------------------------- As stated above, the mapping activities were only done in Ouagadougou and Dar es Salaam. ### a) Public and private health facilities In Dar es Salaam, the list of existing public and private health facilities was updated and their locations were recorded by a geographic positioning system (GPS). In Ouagadougou, the mapping of health facilities and schools was done in 2002 by the Ecole Inter-Etats d\'Ingénieurs de l\'Equipement Rural (EIER), Burkina Faso. Both digital city maps were updated and available for public use. ### b) Anopheles breeding sites The malaria risks in Dar es Salaam and Ouagadougou were displayed in relation to the location of health facilities and schools. The mapping of *Anopheles*breeding sites in Dar es Salaam was done on a city wide-scale in conjunction with another project \[[@B36],[@B45]\]. In Ouagadougou, in the limited time available, the focus was on permanent and semi-permanent breeding sites instead of searching for the numerous temporary breeding sites. The produced maps of breeding sites indicated mosquito productivity and distribution in the city in a given season. The major drawback of mapping is that ground-truthing is very time-consuming and variable over time. During the rainy season, the city-wide larvae collection, larvae hatching and management of data are difficult tasks. Another disadvantage of this approach is that it tends to be very expensive, unless local Geographic Information Systems (GIS) mapping expertise and/or digital city maps are already available for public use. For future studies, it is recommended focusing on the mapping of health facilities and dropping the breeding sites work as it is difficult to assemble a team with the required expertise within such a short time period. School parasitaemia surveys (Tables [1](#T1){ref-type="table"} and [3](#T3){ref-type="table"}) ---------------------------------------------------------------------------------------------- It was possible to determine the transmission intensity and gradients in different communities. At each site, parasitaemia and fever prevalence rates were obtained for different schools (Figures [2a](#F2){ref-type="fig"}, [2b](#F2){ref-type="fig"}, [2c](#F2){ref-type="fig"}) and by residential areas of children. Around 10 to 70% of children (from city centre to periphery) attended schools with elevated temperature. Malaria prevalence was always higher than the fever prevalence in Ouagadougou since there were many asymptomatic infections. Different communities in Ouagadougou may be exposed to different patterns of malaria transmission and hence the age at first infection and infection patterns may vary. Certainly, the more exposed areas of Ouagadougou experience hyperendemic (if seasonal) malaria. The association between malaria infections and various risk factors were measured and these results are reported elsewhere \[[@B46]-[@B49]\]. {#F2} Health facility-based surveys (Tables [1](#T1){ref-type="table"}, [3](#T3){ref-type="table"} and [5](#T5){ref-type="table"}) ---------------------------------------------------------------------------------------------------------------------------- ###### Age-specific malaria prevalence rates in cases and controls by each site. Health facility-based surveys. Study sites Abidjan Cotonou Dar es Salaam Ouagadougou ---------------------- ---------------- ---------------- --------------- --------------- --------------- -------------- ---------------- ---------------- Infants \<1 year 18/78 (23.1%) 22/169 (13.0%) 0/63 (0%) 2/140 (1.4%) 2/99 (2.0%) 4/116 (3.4%) 7/58 (12.1%) 3/21 (14.3%) Children 1--5 years 61/142 (43.0%) 16/60 (26.7%) 5/68 (6.8%) 4/137 (2.8%) 15/213 (7.0%) 8/178 (4.5%) 45/174 (25.9%) 15/104 (14.4%) Children 6--15 years 39/89 (43.8%) 8/35 (22.9%) 0/35 (0%) 1/78 (1.3%) 7/97 (7.2%) 2/56 (3.6%) 23/62 (37.1%) 20/58 (34.5%) Adults \>15 years 31/120 (25.6%) 17/119 (14.3%) 2/213 (0.9%) 11/529 (2.0%) 13/308 (4.2%) 8/423 (1.9%) 48/266 (18.0%) 72/363 (19.8%) Both the fever and control groups (non-febrile admission) had a medium level of parasitaemia prevalence in the health facilities in Yopougon municipality (Abidjan) and Ouagadougou (Table [5](#T5){ref-type="table"}). Some people in the control groups reported self-medication with paracetamol or traditional herbs before visiting the clinics. This could have led some malaria cases to present without fever at the clinic. The overall prevalence of malaria was surprising low in Cotonou and Dar es Salaam. This might have been due to high Insecticide Treated Nets (ITNs) coverage and/or the dry climate at the time of survey \[[@B46]-[@B49]\]. The detection of malaria parasites in a febrile case does not necessarily indicate clinical malaria. In an effort to improve the case definition and clinical diagnosis, the method of Smith *et al*. \[[@B43]\] was used to estimate the probabilities that individual episodes were really due to a malaria infection. The odds ratio (OR) is the proportion of odds of having parasitaemia in fever cases over controls. The formula for the fraction of fever episodes attributable to malaria parasites is: (1-1/Odds Ratio)\*P. P is the proportion of fever episodes in which the subjects had parasitaemia. These age-specific malaria attributable fractions were very low: 0.12--0.27, 0--0.04, 0--0.02 and 0--0.13 in Yopougon municipality (Abidjan), Benin, Dar es Salaam and Ouagadougou, respectively. These results indicated substantial over-treatment at all sites \[[@B46]-[@B49]\]. The questionnaires (available as a separate file) administered to cases and controls were tailored for local use. They contained four sections: personal information, economic situation of the family, travelling history, clinical signs and malaria history. The information on age, sex, measured axillary temperature, length of febrile illness, types of previous treatment and the reasons for seeking care were obtained. Stay outside the urban area during the previous three months, the type of housing, urban agriculture activities and ITNs usage were also investigated. These data provided indications of disease perception, preventive measures and socio-economic background at community level. The questionnaires administered to cases and controls in health facilities were similar to the ones used in school surveys. In all settings the two sets of data were comparable, which allowed for an internal consistency check. For example, in Dar es Salaam 43.1% and 40.2% of households reported ITN use in both the health facility surveys and the school parasitaemia surveys. In Cotonou, these figures were 36.6% and 28.4%, in Ouagadougou 7.8% and 11.1%. The similarity of both surveys also made possible a combined planning and implementation strategy. Detailed results are presented elsewhere \[[@B46]-[@B49]\], as well as in a series of forthcoming publications. Brief description of the health care system (Tables [1](#T1){ref-type="table"} and [3](#T3){ref-type="table"}) -------------------------------------------------------------------------------------------------------------- The administrative structures of the national and municipal health departments were sketched out and the list of health facilities was updated at each site. The total numbers of registered malaria diagnosis or treatment providers were: 1060 in Abidjan, 365 in Cotonou, 1684 in Dar es Salaam and 315 in Ouagadougou. Non-governmental organizations and religious hospitals play an important role in health care delivery in Cotonou and Ouagadougou. The catchment areas of all public and private health facilities were further calculated \[[@B46]-[@B49]\]. The city malaria control programmes and WHO offices provided information about current malaria control efforts. In order to assess treatment efficacy, the trend of the susceptibility of *P. falciparum*to different antimalarials was reviewed at each site \[[@B46]-[@B49]\]. This component required few resources and brought strong political commitment because it involved representatives of the Ministry of Health and the Directors of the municipal health department. The extra-budgetary resources from RUMA helped the local governments to better monitor the provision of health care services, which facilitated an effective exchange of information. The health information was updated but the quality of health care delivery was not assessed because of restricted scope and time. The disadvantage of this approach is that effective communication and dissemination of official documents depends on the attitude of senior officers. Compared costing of RUMA activities ----------------------------------- The cost for conducting a RUMA in a SSA city with a population of 0.5--3 million is around 8,500--13,000 USD for a six to ten-week period (Table [6](#T6){ref-type="table"}). The cost of human resources in Dar es Salaam and Ouagadougou was highest, mainly because of the additional fieldwork performed there (mapping of breeding sites and health facilities). Indeed, the per diem standard was lower in these cities. The higher savings on transportation, communications and materials in Abidjan and Ouagadougou were made possible by our affiliation with local research institutions. The total expense in Abidjan was much lower because the school survey was not performed (the children did not attend school during a politically troubled time). In Cotonou, the excess of human resource and transportation cost was due to unforeseen supervisory expenses. ###### RUMA expenses by study sites. QA = Quality control, USD = US dollars 1USD = 650 Francs CFA (Communauté Française Africaine), 1 USD = 1,050 Tanzanian Schilling in 2003. Sites by the order of total expenses Human resources Transport Communication Stationery Lab. materials & drugs Others Admin. Total expenses USD Total with QA\* -------------------------------------- ----------------- ----------- --------------- ------------ ------------------------ -------- -------- -------------------- ----------------- Cotonou^§^ 2,375 2,942 500 447 793 0 1,000 8,582 No QA Dar es Salaam 4,321 2,040 93 775 1,030 236 0 8,495 12,435 Ouagadougou 2,493 1,613 198 886 721 98 400 6,970 6,970^¥^ Abidjan 958 411 360 569 707 47 1,000 4,577 7,237 § without GIS mapping and 1st quality control without GIS mapping and school survey \* with second quality control at Swiss Tropical Institute. ^¥^the second quality control was free In general, the difference in the cost of human resources and communications was due to differences in personnel capacity and fluctuations in the amount of work. The costs of stationery and laboratory materials were less variable, because the needs were the same at each site. Discussion and conclusion ========================= This assessment was accomplished in four countries within a period of six to ten weeks in the field and has proven to be a helpful tool in supporting planning of urban malaria control. An ongoing urban malaria control intervention in Dar es Salaam has been initiated on the knowledge basis provided by RUMA. With the incentive of extra-budgetary resources and technical support from STI, local partners were committed to incorporate RUMA into existing activities at the municipal level. Qualified personnel and opportunities for integration, synergy and co-ordination were identified during the meetings with local partners and the collaborations were always very successful. The RUMA methodology is a cross-sectional design and the results are likely to change over time due to seasonality, the dynamics of urbanization and the evolution of malaria transmission. In Dar es Salaam, for example, the surveys were carried out during an exceptionally dry period and results could underestimate the true transmission intensity. Many factors such as the size of the city, the fieldwork logistics, the availability of local expertise and the coordination with local senior officers can influence the schedule and planning, as well as the outcome of such surveys. The study highlighted the need for improved Health Management Information Systems (HMIS) in SSA urban areas. Municipal health departments routinely collect health facility data but information is rarely fed back to the districts and facilities that generate the information. The data are often not available for analysis or accessible due to false registration and under-reporting from health facilities, as well as poor filing and storage of documents at the district or municipal level. In addition, the low number of true malaria cases among fever episodes treated as \"malaria\" raises the issue of the validity of the collected data even further. Hence, much progress needs to be made in order to estimate more accurately the urban malaria burden and plan relevant control measures. GIS provides a platform to display health services and geographic features in relation to population settlements. In this experience policy-makers could readily use the presented information for improved planning, re-allocation of resources and for strengthening the networking between the public and private sectors. While the GIS technology has been shown to be very useful in studying health care delivery and distribution of diseases, its application in an entomological assessment was quite difficult and costly and could only be done in conjunction with other ongoing projects. Hence it should be excluded from the process of RUMA. In contrast, the mapping of health facilities with GIS was feasible and cost-effective. While results from the school surveys gave an indication of the endemicity range and risks in the targeted community, they cannot be considered as being representative without a wider survey. The variations of malaria risk were sometimes related to political divisions or man-made boundaries, but often were due to divergent socio-environmental factors and the degree of urbanization. Because site-specific environmental conditions lead to an aggregated distribution of vectors and different malaria risks, the sampling sites were selected taking into account the population density, the natural environment and urbanization patterns. This should improve the rough categories that previous researchers applied (centre, intermediate and periphery). Despite a potential attendance bias, the health facility surveys allowed the determination of prevalence of parasitaemia among presenting clinical cases, and the calculation of the fraction of malaria-attributable fevers. This allowed to document clearly the high rate of malaria mis-diagnosis in the health facilities. This information is of great importance for urban malaria control. Overall, RUMA is a first step towards understanding malaria endemicity and designing control strategies. It has exemplified the concern for mis-diagnosis of clinical malaria in SSA cities \[[@B25],[@B50]\]. A report by the Tanzania-Japan malaria control programme in Dar es Salaam mentioned that drug administration to diagnosed children was one of the essential interventions that reduced the malaria rates between 1988 and 1996 \[[@B36]\]. An in-depth research is now being implemented in Dar es Salaam to assess the malaria burden with a much larger sample size. The application of RUMA methodology is possible and desirable in other SSA urban areas and it should have a special focus on improved diagnosis. List of abbreviations ===================== CHU Centre Hospitalier Universitaire CNRFP Centre National de Recherche et de Formation sur la Paludisme, Burkina Faso CREC Centre de Recherche Entomologique de Cotonou CSRS Centre Suisse de Recherches Scientifiques, Côte d\'Ivoire EIER Ecole Inter-Etats d\'Ingénieurs et de l\'Equipement Rural, Burkina Faso GIS Geographic Information System HMIS Health Management Information Systems ITNs Insecticide-Treated Nets MOH Ministry of Health RUMA Rapid Urban Malaria Appraisal SSA Sub-Saharan Africa STI Swiss Tropical Institute Authors\' contributions ======================= SW participated in the design of the study, conducted the field work, analysed and interpreted data and drafted the manuscript. CL conceived the study, coordinated the field work and revised the manuscript. TS and PV assisted in the design and the statistic analysis. CG, DD, MA and DM were the key local contacts, facilitated the collaboration and supervised the data collection and laboratory works at each site. AT participated in the design of the study. MT participated in the conception of the work, facilitated the overall coordination and revised it critically at all stages. Supplementary Material ====================== ###### Additional File 1 the questionnaire for health facility-based survey ###### Click here for file ###### Additional File 2 the questionnaire for school parasitaemia survey ###### Click here for file Acknowledgements ================ We would like to acknowledge the support and help of the following institutions and persons. In Benin: Francois Holtz; in Burkina Faso: the Ecole Inter-Etats d\'Ingénieurs de l\'Equipement Rural; in Côte d\'Ivoire: Dr. Joseph Niangue; in Tanzania: Ifakara Health Research and Development Centre. We wish also to express our gratitude to Dr. Andrei Chirokolava for editing and reviewing the city reports. RUMA was supported financially by the Roll Back Malaria Partnership and STI.

Living organisms require a small amount of different metals for their metabolic functions. Among them Cr and Co are important microelements which are found in trace quantities in the human body where they regulate important biological functions. Cr is involved in glucose metabolism, either by facilitating insulin binding to its receptor or by amplifying the effect of insulin on carbohydrate and lipid catabolism[@b1][@b2]. On the other hand Co(III) ions occupy the catalytic site of vitamin B12 and are essential to the vitamin\'s biological activity[@b3]. Under physiological conditions both metals need to be present only in trace quantities because they are required for the biological activity of a very small number of specific cellular proteins[@b4]. Cr and Co toxicity is associated with their oxidation state[@b4][@b5]. Cr can be present as a wide range of possible oxidation states (Cr(I), Cr(III), Cr(IV), Cr(V), Cr(VI)) even though Cr(III) and Cr(VI) are the most energetically stable and thus the most commonly observed[@b6][@b7]. The effects of hexavalent Cr (Cr(VI)) have been previously analyzed since it is classified as group 1 carcinogen (carcinogenic to humans) by the International Agency for Research on Cancer (IARC) mostly related to evidence of excess of malignant lung tumors in exposed workers and is also an environmental pollutant[@b8]. On the other hand, the mechanisms related to the toxicity of Cr(III) are still unknown[@b5][@b6] and therefore it is classified as group 3 carcinogen (unclassifiable due to lack of adequate evidence) by the IARC. Likewise Co can be found in different oxidation states among which Co(II) and Co(III) are the most common[@b9][@b10] and it is classified with its compounds as group 2B carcinogen (possibly carcinogenic to humans) by the IARC. Occupational exposure to Co has been shown to be associated with cardiomyopathy and neurological damage[@b11][@b12][@b13]. Lately, major health concerns have been raised by the toxic levels of Cr and Co released by metal-on-metal (MoM) bearing surface and also metallic junctions of hip implants used worldwide in orthopedic surgery[@b14][@b15]. Whereas in an industrial setting, the toxicity of Cr and Co is contingent upon the duration of exposure, nanoparticle concentration, and entry routes into the body across a barrier membrane (dermal or inhalation absorption), the release of Co and Cr nanoparticles from metallic orthopedic implant devices represents an internal exposure to metals that is not dependent on absorption through barrier membranes and exemplifies a still relatively unexplored field of toxicology[@b8]. It has been determined that these implants release particulate metal debris in the form of corrosion with or without abrasion particles, which are non ionized metal alloys[@b15][@b16]. These non-ionized metal alloys are primarily nanometer-sized particles[@b8][@b17]. Data from the literature indicate a range of 5--100 μg/L of Cr and 5--300 μg/L of Co present in plasma and tissues from patients with MoM implants at revision surgery[@b18]. With regards to metal ions, it is still controversial whether Cr is present as Cr(VI) in patients with MoM bearing surface and/or metal connections, with some reports supporting this idea and others disproving it[@b19][@b20][@b21]. There is a general consensus, however, that Cr(III) is elevated in the biological fluids of all patients with MoM-type implants[@b22][@b23] and also that elevated Co(II), albeit at different levels, is found in serum, blood, and urine of all examined patients[@b24][@b25][@b26]. Histologically, an exuberant inflammatory reaction often associated with tissue necrosis/infarction, macrophagic and lymphocytic infiltrate is observed and originally defined as aseptic lymphocytic vasculitis associated lesions (ALVAL) and later revised as adverse local tissue reaction (ALTR) or adverse reaction to metallic debris (ARMD)[@b27][@b28]. This reaction is likely due to a combination of metal ion toxicity combined with a hypersensitivity reaction which is particularly enhanced in some patients for reasons still not well understood. These types of implant release particulate metal debris in the form of corrosion particles with or without addition of abrasion particles[@b29]. Since it is well established that the major health risk and toxicity of Cr and Co are not related to their elemental composition (Cr^0^ and Co^0^) but to their oxidation state it is of fundamental importance to understand the mechanism underlying metal oxidation in biological samples[@b30][@b31][@b32]. An in depth analysis of the relationship between the concentration of Cr and Co ions[@b30][@b31][@b32] in the biological fluids and the biochemical, molecular and mechanistic analysis leading to biological damage is still missing. The surgical implantation of orthopedic implants represents a "controlled" model for examining the toxicity and mechanism of action of Co and Cr since there are well-defined parameters including duration of exposure and chemical composition of the implant. In this study, peri-implant tissues from revision surgeries were employed to examine the interaction between Cr and Co ions with tissue biomolecules and their relationship with the adverse tissue reaction. Results ======= Histological analysis of metal wear debris in periprosthetic tissues MoM implants --------------------------------------------------------------------------------- Histological examination of periprosthetic tissues recovered from patients with MoM implants demonstrated in some cases the occurrence of the classic layering of a high grade advanced adverse reaction with a thick superficial layer of necrosis/infarction, intermediate layer of macrophagic and lymphocytic infiltrate, and deep layer of perivascular lymphoplasmacytic infiltrate ([Figure 1a](#f1){ref-type="fig"}) associated with presence of large aggregates of corrosion particles of variable size in at least one histological section in most of the examined revision tissues ([Figure 1b](#f1){ref-type="fig"}). Globular and/or irregular small particles of corrosion products were also observed in exfoliated ([Figure 1c](#f1){ref-type="fig"}) and interstitial macrophages ([Figure 1d](#f1){ref-type="fig"}). Ultrastructural analysis of the same case revealed phagosomes filled with predominantly round electron dense particles of variable intensity consistent with corrosion products and a few rod-shaped particles consistent with metallic abrasion particles not identifiable at light microscopy ([Figure 1e](#f1){ref-type="fig"}). Particulate debris was present in the cytoplasm of the macrophages and interstitially as both electron dense rod-shaped and irregular particles and less electron dense, spherical particles that ranged from nanometer to micrometer in size ([Figure 1e, 1f, 1g](#f1){ref-type="fig"}). Increased cobalt and chromium accumulation in periprosthetic tissues surrounding MoM implants --------------------------------------------------------------------------------------------- Periprosthetic tissues were collected from 18 patients undergoing hip revision surgery ([Table 1](#t1){ref-type="table"}). Metal ion concentrations were measured in periprosthetic tissues surrounding MoM implants (9 cases) and in revision tissues from alternative bearings revision samples (9 cases), including metal-on-polyethylene (MoP), ceramic-on-polyethylene (CoP), and ceramic-on-ceramic (CoC). A significant increase in Co and Cr accumulation was detected in revision samples surrounding MoM implants as compared to control (alternative bearings) tissues ([Figures 2a, 2b](#f2){ref-type="fig"}). Inductively coupled plasma mass spectrometry (ICP-MS) measurements revealed a 2.5-fold increase of tissue Co levels surrounding MoM implants as compared to controls ([Figure 2a](#f2){ref-type="fig"}). Periprosthetic tissue Cr levels from MoM implants were found to be 9-fold higher as compared to controls ([Figure 2b](#f2){ref-type="fig"}). To determine the elemental composition of wear particles, backscatter SEM coupled with energy-dispersive X-ray spectroscopy (EDS) was used to examine wear debris powders accumulated in metallic sleeve adapters. EDS spectra demonstrate that both Cr and Co were present in implant wear debris ([Figure 2c](#f2){ref-type="fig"}). Metal ions are produced from phagocytosed Co or Cr nanoparticles in macrophage endosomal compartments ----------------------------------------------------------------------------------------------------- The confirmation of elevated Co and Cr levels in the periprosthetic tissues surrounding MoM implants led to the next series of experiments aimed at addressing the mechanism(s) underlying the formation of metal ions. Histological analysis of the tissues surrounding the tested implants revealed extensive macrophagic infiltrates with phagocytosed pale green corrosion products and in the pictured case, abrasion black metallic particles secondary to implant impingement were also present ([Figure 3a](#f3){ref-type="fig"}). Ultrastructural examination of MoM revision tissue demonstrated that nanometer size electron dense wear debris particles are phagocytosed by macrophages in the periprosthetic tissue ([Figure 3b](#f3){ref-type="fig"}). An accumulation and engorgement of late endosomes and lysosomes, with needle-shaped and spherical wear debris particles, were observed in macrophages present in revision tissues ([Figure 3c](#f3){ref-type="fig"}). The accumulation of phagocytosed wear debris in late endosomal and lysosomal compartments prompted us to examine the effects of the acidic, highly proteolytic micro-environment of these organelles on metal nanoparticles. Wild-type C57Bl/6 bone marrow macrophages were left untreated or exposed to cobalt or chromium nanoparticles at concentrations of 10^9^, 10^10^, 10^11^, and 10^12^ particles per ml for 12 hours. Macrophages treated with cobalt or chromium nanoparticles demonstrated a dose- dependent increase in cobalt and chromium ions concentrations, as measured spectrophotometrically ([Figures 3d, 3e](#f3){ref-type="fig"}). The observed increase in cobalt and chromium ion levels led us to examine the relationship between pH and metal ion generation. To this end we incubated primary macrophages in the presence or absence of NH~4~Cl and leupeptin, prior to treatment with chromium nanoparticles. The pre-treatment of cells with a combination of NH~4~Cl and leupeptin makes the intralysosomal pH more basic and inhibits lysosomal proteolysis, respectively. Macrophage chromium ion levels were determined 12 hours following nanoparticle exposure. Chromium ion levels were significantly elevated in macrophages exposed to nanoparticles alone as compared to cells pre-treated with lysosomal inhibitors and control cells. With increasing concentrations of nanoparticle exposure, NH~4~Cl and leupeptin pre-treated cells did not demonstrate a significant elevation of chromium ion release. Cells not pre-treated with NH~4~Cl and leupeptin, however, revealed a dose dependent increase in chromium ion levels following nanoparticle exposure ([Figure 3f](#f3){ref-type="fig"}). Chromium orthophosphate (CrPO~4~) is among the most commonly observed implant corrosion products in the periprosthetic tissue and it is due to the precipitation of chromium and phosphate ions. The observation of phagocytosed CrPO~4~ within macrophagic dominant infiltrates prompted us to examine the effects of endosomal/lysosomal microenvironment on CrPO~4~, and particularly whether Cr ions could be released from CrPO~4~. Primary macrophages were left untreated or exposed to increasing concentrations of CrPO~4~ ([Figure 3g](#f3){ref-type="fig"}). A dose-dependent increase in chromium ion release was measured in cells exposed to CrPO~4~ ([Figure 3h](#f3){ref-type="fig"}). Histological examination of infiltrates with phagocytosed corrosion products demonstrated a transition from healthy to degenerated macrophages with foamy elements in the periprosthetic tissues surrounding MoM implants ([Figure 3i](#f3){ref-type="fig"}). Together, these data indicate that phagocytosed corrosion products and also abraded metallic wear debris accumulate within late endosomes with the resultant release of metal ions. Metal ion exposure increases oxidative stress and protein carbonylation ----------------------------------------------------------------------- In the next series of experiments, we examined the relationship between metal ion formation and cellular free radical production. Primary macrophages were left untreated or exposed to cobalt or chromium ions at concentrations of 10, 100, and 1000 μM for 12 hours. Using a DiOxyQ probe that fluoresces upon reacting with free radicals, significant increases in ROS levels were measured in primary macrophages exposed to cobalt or chromium at concentrations of 100 μM and 1000 μM when compared to controls ([Figure 4a](#f4){ref-type="fig"}). Significant increases in ROS production were detected in culture supernatants from chromium or cobalt exposed cells at all concentrations when compared to controls ([Figure 4b](#f4){ref-type="fig"}). Next, we investigated whether elevated concentrations of cellular ROS would result in an increase of oxidatively modified proteins in macrophages exposed to metal ions. To address this aim, the cytosol and mitochondria were isolated from untreated and metal ions exposed cells, and subsequently incubated with 2,4-dinitrophenylhydrazine (DNPH). DNPH selectively reacts with or derivatizes carbonyl groups. The DNPH-derivatized samples were separated by a 4--15% gradient SDS PAGE, followed by western blot analysis employing an anti- DNPH primary antibody. Western blot analysis detected carbonylated proteins in the cytosol and mitochondria from untreated and cobalt or chromium exposed macrophages ([Figure 4c](#f4){ref-type="fig"}) A significant increase in the levels of oxidatively modified proteins was detected in the cytosol and mitochondria isolated from cobalt or chromium treated cells when compared to controls ([Figure 4c](#f4){ref-type="fig"}). Western blot analysis of primary macrophages exposed to CrPO~4~ demonstrated a dose dependent increase in the levels of protein carbonylation in CrPO~4~ exposed cells as compared to untreated cells. ([Figure 4d](#f4){ref-type="fig"}). Following the observed increases in cellular ROS levels and protein carbonylation in primary macrophages exposed to metal ions and CrPO~4~, in addition to the significant elevation of metal ions measured in periprosthetic tissues surrounding MoM implants, the next experiments were designed to examine revision tissues surrounding hip implants for protein oxidation. The carbonyl content (carbonylated Arg, Lys, His, and Pro residues) in revision tissues was analyzed spectrophotometrically at an absorbance of 375 nm following the reaction of carbonylated amino acid side-chains of oxidized proteins with DNPH. A significant increase in carbonyl content was detected in revision tissues surrounding MoM implants as compared to control tissues ([Figure 4e](#f4){ref-type="fig"}). The introduction of carbonyl groups into amino acid side chains can result in the loss of catalytic or structural function of the affected protein. Thus, total proteomic analyses were performed on the proteins extracted from revision tissues surrounding MoM implants to identify peptides and proteins with posttranslational modifications, including oxidation sites. The extracted total proteome from MoM tissue homogenates was fractionated on a FPLC gel filtration column subjected to protease digestion with endoproteinase Lys-C and trypsin, and analyzed by tandem mass spectrometry (MS/MS) on a nano LC/Orbitrap system. Nano LC-ESI-MS/MS analysis demonstrated substantial oxidation of amino acids of several proteins in the periprosthetic tissues from MoM implants ([Figure 4f](#f4){ref-type="fig"}). The analysis mapped the side chain posttranslational modifications of several proteins ([Supplement Table 1](#s1){ref-type="supplementary-material"}). Oxidations included the glutamic semi- aldehyde conversions of arginine, lysine, histidine, and tryptophan (Arg- \>GluSA, Lys-\>Allysine, His-\>2-oxo-histidine, and Trp-\>Kynurenine) ([Supplement Table 1](#s1){ref-type="supplementary-material"}). Analysis of the oxidatively damaged proteome indicated that proteins, related to different cellular function and sub-cellular location were oxidatively damaged ([Figure 4g](#f4){ref-type="fig"}, [Supplement Table 1](#s1){ref-type="supplementary-material"}). A directly proportional relationship between Cr levels and carbonyl content was observed in the periprosthetic tissues surrounding hip implants ([Figure 4h](#f4){ref-type="fig"}). Proteins in the MoM revision tissue proteome demonstrate binding affinity for Co and Cr --------------------------------------------------------------------------------------- Immobilized-metal affinity chromatography (IMAC) was performed to separate metal- binding proteins from the total proteome extracted from periprosthetic tissues from MoM implants. Metal-binding proteins were fractionated on the basis of their differential binding affinities of the surface exposed amino acids towards an immobilized metal ion. To determine if some of the known proteins with high affinity for metals could be retained by the IMAC columns charged with cobalt \[Co(II)\] and with chromium \[Cr(III)\], metal depleted tissue homogenates were loaded on an IMAC column/chip saturated with Co(II) or Cr(III). Proteins with affinity for binding Co(II) or Cr(III) were eluted from the immobilized columns with increasing concentrations of imidazole ([Figures 5a, 5b](#f5){ref-type="fig"}). Recovered fractions were pooled into six groups and separated by 1D SDS-PAGE. Silver stain analysis of the SDS-PAGE indicated that several proteins present in the revision tissues surrounding MoM implants were able to bind Co(II) or Cr(III) ([Figure 5c](#f5){ref-type="fig"}, [Supplement Table 2](#s1){ref-type="supplementary-material"}). Following *in situ* tryptic digestion, extracted peptides were analyzed by nano LC--ESI--MS/MS on Orbitrap to identify the metal binding proteins in the revision tissues. Among the proteins eluted from the Cr(III)-loaded column were several enzymes such as enolase (Mg^2+^), catalase (Fe(II)), myeloperoxidase (Fe(III)/Ca^2+^), and arginase (Mn(II)), as well as molecular transporters such as ferritin (Fe(III)), hemoglobin (Fe(III)), ceruloplasmin (Cu(II)) and different Ca^2+^ binding proteins involved in many cellular functions ([Figure 5d](#f5){ref-type="fig"}, [Supplement Table 2](#s1){ref-type="supplementary-material"}). Similarly, among the protein eluted from the Co(II) column, there were several subunits of cytochrome C (Cu(II), Fe(III), and Zn(II) dependent), enzymes that are part of the cellular redox system, and Ca^2+^-binding proteins important for cell motility and cell signaling ([Figure 5d](#f5){ref-type="fig"}, [Supplement Table 2](#s1){ref-type="supplementary-material"}). Altogether, pathway analysis of the cellular proteome binding Co(II) and Cr(III) columns indicate that several cellular pathways and function could be potentially compromised by metal binding. Chromium inhibits the catalytic activity of aldolase and catalase *in vitro* ---------------------------------------------------------------------------- To determine the potential effects of Co(II) or Cr(III) on cellular functions, we selected two enzymes; one, aldolase, which was previously reported to bind Cr(III) at a non-catalytic site[@b33]; and the other catalase, which could replace Fe(III) at its catalytic site with Cr(III). Fructose-bisphosphate aldolase (ALD), catalyzes a key reaction in glycolysis and energy production, i.e. the conversion of D-fructose-1,6-biphosphate into dihydroxyacetone phosphate and D-glyceraldehyde-3-phosphate. Thus, any impairment in this protein\'s catalytic activity would be followed by damage to the energetic balance of any cell[@b34]. The catalytic activity of pure human recombinant aldolase A was assessed to determine any enzymatic differences in the absence or presence of Co(II) or Cr(III) at different concentrations (4 mM, 0.4 mM and 0.04 mM for each Co(II) and Cr(III) solution). In each condition, the enzyme was incubated with the substrate, D-fructose 1,6-biphosphate, and hydrazine sulfate. The resulting hydrazone product, which absorbs at 240 nm, was measured. Spectrophotometric analysis of hydrazone formation demonstrated a dose dependent Cr(III) inhibition of aldolase activity, with complete loss of catalytic activity observed at 4 mM Cr(III) ([Figure 6a](#f6){ref-type="fig"}). Interestingly, spectrophotometric measurements of hydrazone formation in the presence of Co(II) revealed minimal inhibition of aldolase activity at all concentrations used as compared with untreated aldolase ([Figure 6b](#f6){ref-type="fig"}). The second protein examined, catalase, is an enzyme responsible for the conversion of hydrogen peroxide to water and oxygen. Catalase is a tetramer consisting of four identical, tetrahedrally arranged subunits. Each 60 kDa subunit contains a heme group and NADPH in its active center[@b35]. Inhibitors of catalase include metals, which can affect heme structure via the displacement of iron. To examine the effects of metals on catalase function and possible alterations to catalytic activity, H~2~O~2~decomposition was measured spectrophotometrically at 240 nm in the absence or presence of Co(II) or Cr(III). The presence of Cr(III) exerted a dose dependent inhibitory effect on catalase activity, with near abolishment of activity observed at 4 mM Cr(III) ([Figure 6c](#f6){ref-type="fig"}) Spectrophotometric measurements of H~2~O~2~ decomposition in the presence of Co(II) at all concentrations, however, did not reveal any inhibition of catalase activity as compared to untreated catalase ([Figure 6d](#f6){ref-type="fig"}). Together, these results indicate that Cr(III), but not Co(II), affects protein function in a dose dependent manner. A comparison of Co(II) and Cr(III) toxicity from the *in vitro* and *ex vivo* results in this study is presented in [Supplement Table 3](#s1){ref-type="supplementary-material"}. The most notable differences between the metal ions are: a linear correlation between Cr(III) concentration and protein carbonylation and a Cr(III) dose dependent inhibition of enzymatic catalytic activity (*in vitro*). Discussion ========== Chromium and cobalt are minerals, both required as dietary supplements in trace amounts. The adequate daily intake for Cr is 25 to 50 μg/day and 5 to 8 μg/day for Co[@b36]. Both metals display toxic effects when present at high concentrations. Severe and irreversible cardiomyopathy and vision or hearing impairments are reported at serum Co concentrations over 700 μg/L, and reversible hypothyroidism and polycythemia at \~300 μg/L and higher[@b37][@b38][@b39][@b40]. For Cr acute oral toxicity ranges between 1900 and 3300 μg/kg and is associated with damage to multiple tissues and often kidney and liver failure[@b41]. Overexposure to both metals is sporadic and associated with environmental and/or industrial exposure. Recently, groups of patients implanted with hip devices of MoM bearing surface have been shown to develop an accelerated inflammatory reaction frequently associated with tissue necrosis and cellular toxicity. The sense of urgency in determining the molecular association between MoM implants and toxic response relates to the fact that there is a large population worldwide carrying these devices to be monitored long-term. Metal-based implants release micrometer and nanometer sized wear debris, due to the tribocorrosion of the bearing surfaces[@b42]. The particulate metals released from the wear of joint implants are mostly colloidal CrCo alloys and likely some oxides (Cr~2~O~3~and CoO). Herein we determined by TEM and light microscopy of periprosthetic tissues that these metal particles are readily phagocytosed by tissue resident macrophages and transported in endosomal and lysosomal compartments as indicated by the several observed phagocytic cells engulfed with corrosion product particles. In endosomal compartments, due to the presence of an acidic pH, the transition metals release Cr and Co ions with different oxidation states[@b43][@b44][@b45][@b46]. Induction of endosomal alkalinization greatly decreased metal ionization, with a direct correlation between the amount of phagocytosed metal particles, acid pH and released metal ions observed. This determination is important since metal ions are the ones associated with biological damage. Indeed, increased amount of both Cr(III) and Co(II) were found in tissues retrieved from patients with MoM implants as compared with control implants including MoP, CoP, and CoC bearing surfaces which also may contain Cr and Co in their formulation but in much lower amounts. Cr(III) is the most thermodynamically stable species and does not tend to reduce to Cr(II) or oxidize to Cr(VI), which are more unstable oxidation states. Cr(VI), if present could be readily reduced to Cr(III)[@b47]. Cr(III) is the form mostly observed in patients with MoM "metallosis"[@b22][@b48][@b49] and it often forms complexes with the ubiquitously found, negatively charged phosphate group. The resulting Cr orthophosphate (CrPO~4~) is commonly found in the periprosthetic tissue and the synovial fluid of patients with metal-on-metal-related pathology[@b43]. These precipitates are referred to as implant corrosion products. We observed these corrosion products both by TEM, where they present a globular and irregularly shaped appearance, and by histological analysis, as pale green intracellular and extracellular tissue deposits. The equilibrium kinetic between Cr(III) and its combined form (CrPO~4~) and the conditions that skew the reaction toward free Cr(III) ion or its combined form will be worth exploring since these corrosion products serve as a constant source of metal ions that, since they cannot be disposed of, can amplify the tissue damage. Co(II) is the Co form mostly observed in patients with MoM "metallosis". Our analysis on the effects of Cr and Co ions retrieved from MoM patients, indicate that several mechanisms account for the induced damage. First, Cr(III) and Co(II) catalyze the conversion of hydrogen peroxide (H~2~O~2~) into reactive hydroxyl radicals (HO•) in a chemical reaction known as the Fenton reaction[@b5]. Fenton chemistry is regularly known to occur in vivo; however, the level of activity is generally negligible due to only trace amounts of transition metals present in the human body. Additionally, under physiological conditions the cells are equipped with a variety of enzymes, including several isoforms of catalase, glutathione peroxidase and heme oxygenase-1, that dispose of ROS. When higher amounts of metal ions are present, however, the hydroxyl radicals formed by the metal-catalyzed Fenton reaction are greatly increased to levels that cannot be readily inactivated[@b50]. Excess ROS cause deleterious biological effects and irreversible damage by promoting protein, lipid, carbohydrate and DNA oxidation[@b50]. Importantly, we found a strong positive correlation between the amount of Cr and Co ions and tissue oxidative damage. Patients with the highest amount of Cr and Co ions present the highest levels of protein carbonylation, a well-known consequence of oxidative stress[@b50]. It is known that extensive oxidative modifications result in cleavage, sub-unit dissociation, unfolding, and aggregation, with an overall loss of biological function[@b51]. These conditions are effects of oxidative stress and are known to be associated with tissue degeneration and necrotic damage. High levels of oxidative stress can explain the necrotic tissue and formation of masses known as pseudotumors, observed in patients with MoM implants[@b17][@b18][@b27][@b52][@b53][@b54][@b55]. Our *in vivo* proteomic analyses indicate that several proteins associated with different cellular pathways are oxidatively damaged by Cr and Co-induced ROS. Having an overall carbonylated proteome could generally compromise cellular functions. Indeed, our *in vivo* data are further supported by several *in vitro* experiments indicating that Cr(III) induces cellular damage by activating ASK1 kinase, p38, Jun kinase and caspase 3[@b56]. Cobalt has also been shown to induce necrotic and apoptotic cell death[@b57]. As a second mechanism, Cr and Co metal ions can directly bind to proteins and directly induce oxidation and loss of biological function. It has been previously reported that Cr interacts with different serum proteins including hemoglobin, transferrin, and albumin[@b58]. Co has been proposed to bind to several proteins involved in the cellular redox system and ROS clearance[@b59]. Additionally, some of the toxic effects of Co(II) have recently been proposed to be due to inhibition of Ca^2+^ entry and Ca^2+^ signaling and competition with Ca^2+^ for intracellular Ca^2+^- binding proteins[@b60]. Co(II) has also been shown to interact with various receptors, ion channels and biomolecules[@b13][@b61]. Our comprehensive proteomic analysis indicates that both Cr(III) and Co(II) can bind several more cellular proteins including enzymes involved in the cellular redox system (catalase, superoxide dismutase, glutathione peroxidase, metabolism (arginase 1, transglutaminase 3, enolase 1, molecular transport (transferritin, transcobalamin, hemoglobin, cellular motility (annexin A1, annexin A2, lymphocyte cytosolic protein 1, cell signaling (leukotriene A4 hydrolase, ryanodine receptor 2, phosphodiesterase 3A, cGMP-inhibited), and organelles functions. Altogether our data indicate that several cellular pathways and function can be potentially compromised by metal binding. As a proof of principle we determined that incubation of aldolase with Cr(III), and less so with Co(II), decreased the enzyme\'s ability to catalyze the formation of glyceraldehyde 3-phosphate from fructose 1--6 biphosphate, even though Cr(III) does not directly bind to the enzyme\'s catalytic site[@b33]. Finally, as a third mechanism, Cr and Co ions could displace other metal ions present in tissue metalloproteins and affect their activity. Indeed, protein elution from Cr(III) and Co(II) columns retrieved several metalloproteins among which Ca^2+^, Fe(III) and Zn(II) binding proteins were particularly enriched. To determine whether Cr and Co ions can directly replace other ions present in metalloproteins we incubated catalase with Cr(III) ions since catalase contains a heme/Fe group in its active center[@b35]. A significant decrease in the catalytic ability of catalase to decompose H~2~O~2~was observed. In conclusion, our analyses indicate several mechanisms that can substantially explain the exaggerated tissue damage and accumulation of a large amount of necrotic macrophages in the joint fluid. Additional mechanisms, which were not mechanistically explored here, include the adverse tissue reactions associated with Cr and Co hypersensitivity, which are strongly dependent on the particle immunogenicity and patient immunological phenotype and can exponentially increase the inflammatory response. Experimental procedure ---------------------- ### Mice and bone marrow macrophage (BMM) preparation C57BL/6 mice (3 months old) were purchased from Harlan Laboratories. Animal euthanasia and tissue harvesting were conducted according to a protocol approved by the Institutional Animal Care and Use Committee at Albert Einstein College of Medicine. Bone marrow cells were isolated from murine femurs and differentiated in M-CSF (10 ng/ml, RD Systems) for 5 days. ### Primary macrophage cultures with metal nanoparticles and ions In studies examining protein carbonylation and total free radical activity, cells were either left untreated or exposed to cobalt or chromium ions: cobalt (II) sulfate heptahydrate and chromium (III) chloride hexahydrate were prepared at concentrations of 1 μM, 10 μM, 100 μM in culture medium. For the spectroscopic analysis of metal ions, primary macrophages were incubated in the absence or presence of cobalt nanoparticles (purity 99.8%, a mean size of 28 nm) or chromium nanoparticles (purity 99.5%, a mean size of 50 nm, Nanostructured & Amorphous Materials, Inc., Houston, TX) or chromium orthophosphate (chromium (III) salt(1:1); phosphoric acid. Cobalt and chromium nanoparticles were suspended in culture medium at concentrations of 1 × 10^10^, 10^11^, 10^12^ particles per ml. In some experiments, prior to the addition of chromium nanoparticles, macrophages were pre-treated with a final concentration of 20 mM NH~4~Cl and 0.1 mM Leupeptin to inhibit endosome-lysosome system acidification. In all experiments, control and exposed cells were collected 12 hours following treatment. ### Histological analysis of revision tissues Periprosthetic tissues were collected at surgery from a total of 9 cases undergoing revision surgery for adverse local tissue reactions (ALTR). All patients provided informed, written consent regarding participation in this study. Studies and procedures were conducted according to a protocol approved by the Albert Einstein College of Medicine Institutional Review Board. The implant group composition was as follows: MoM CoCr bearing surface, 9 \[5 THA large head (\>36 mm diameter) with metallic CoCr Mo sleeve adapter and 4 resurfacing implants with the same bearing surface\]; Control (CTR) group was comprised of 9 samples: Metal-on-Polyethylene (MoP) bearing surface, 5; Ceramic-on-Polyethylene (CoP) bearing surface, 3; and Ceramic-on-Ceramic (CoC) bearing surface, 1; all with a conventional stem and monoblock modular neck. The tissue was collected fresh from several areas around the implant on ice. Collected tissues were fixed in 10% neutral buffered formalin and processed for hematoxylin-eosin staining. Sections were examined with a Zeiss Axioskop 40 microscope and images captured digitally using a ProgRes® camera (Jenoptik, Germany). ### Backscatter scanning electron microscopy and Energy-dispersive X-ray spectroscopic (EDS) analysis of wear debris powder Wear debris powder was gently scraped or shaken from the implant metallic sleeves with a sterile carbon steel metallic surgical blade and saved in an OPTICLEAR® 15 × 45 mm screw thread borosilicate glass vial w rubber lined cap (Kimble Chase, Rockwood, TN) prior to analysis. Using carbon paste as an adhesive, powders were sprinkled onto a carbon stub and then coated with carbon. Spectra and maps were collected on a Zeiss Supra 40 FESEM with Oxford INCA energy dispersive spectrometer at 10 kV. ### ICP-MS sample preparation and instrumental conditions Each sample was mixed with 0.5 mL of *aqua regia*, and then the sample was diluted to 10 ml with de-ionized water. A series of standard solutions of Co and Cr were prepared under the same acidic conditions. The inductively-coupled plasma (ICP) mass spectrometry (MS) measurements were performed on a PerkinElmer NexION 300X ICP mass spectrometer. ^59^Co and ^52^Cr were measured under the kinetic energy discrimination (KED) mode. ### Spectrophotometric determination of protein carbonyl content The carbonyl content of revision tissues and primary macrophages exposed to metal ions was quantified spectrophotometrically with the Protein Carbonyl Assay kit (Cayman Chemical Company, Ann Arbor, MI). ### Western blot analysis of carbonyl groups Revision tissue homogenates and primary cell lysates were prepared in 150 mM NaCl, 50 mM Tris, 1 mM EDTA, and 1% NP40. Prepared samples were derivatized using the Oxyblot Protein Oxidation Detection Kit (Millipore, USA), separated on 4--15% SDS-PAGE, and transferred membranes were incubated with a rabbit polyclonal anti-DNP antibody followed by a goat anti-rabbit IgG--HRP antibody. Proteins were visualized by chemiluminescence. ### Transmission electron microscopy Periprosthetic tissues, collected fresh on ice, were immediately fixed in 2.5% glutaraldehyde, 2% paraformaldehyde, in sodium cacodylate buffer 0.1 M, pH 7.4 for 3 hours at 4°C, post-fixed in 1.0% aqueous osmium tetroxide, and embedded in LX112-Araldite (Ted Pella Inc, CA, USA). Ultrathin sections were stained with uranyl acetate followed by lead citrate, and viewed with a Jeol JEM-1200EX transmission electron microscope (Jeol Ltd., Akishima, Japan) at 80 kV. ### Fluorometric measurement of total ROS activity The total free radical activity in control and metal ion exposed primary macrophages and culture supernatants was quantified with the OxiSelect *In Vitro* ROS/RNS Assay kit (Cell Biolabs, Inc., San Diego, CA). ### Spectroscopic determination of cobalt (II) ions Cobalt (II) concentrations in total cell lysates prepared from control and cobalt nanoparticle (Co^0^) exposed primary macrophages were measured following the incubation with a solution containing 0.8 ml 6 M HCl, 2.0 ml 50% potassium thiocyanate in 4.8 ml acetone. The absorbance of the complexes of cobalt ions with thiocyanate was measured spectrophotometrically at 620 nm. ### Colorimetric determination of chromium (VI) ions Chromium (VI) concentrations in control and chromium nanoparticle (Cr^0^) exposed primary macrophages and culture supernatants were determined following the incubation with a solution containing 0.5 ml 3 M H~2~SO~4~, 0.5 ml 50% 1.5-Diphenylcarbazide in 9.0 ml deionized water. The absorbance was measured spectrophotometrically at 540 nm. ### Organelle isolation and cytosol preparation Cytosol was prepared was isolated from homogenates following centrifugation for 1 h at 100,000 g at 4°C. Mitochondria were isolated from homogenates following two consecutive centrifugations for 10 minutes each at 7,000 g at 4°C. ### FPLC gel filtration fractionation of the proteome contained in the homogenates from revision tissues Extracted proteins (about 5--6 mg of each homogenate from at least five patient revision tissues) were run through a Superdex S200 gel filtration column (HiLoad 16/60) coupled to a Bio-Rad FPLC system interfaced with Biologic DuoFlow software. One representative run consisted of 240 ml of PBS buffer with 300 mM NaCl, pH 7.4, run through the column at a flow rate of 0.5 ml/min. ### rLys-C/trypsin digestion of the proteome Multiple fractions were lyophilized before mass spectroscopy analysis and subjected to reduction and alkylation. Fractions were digested first with endoproteinase Lys-C followed by trypsin. ### NanoLC-ESI-MS/MS analysis of Lys-C/tryptic peptides Lys-C/tryptic digests were analyzed by nano LC/MS/MS with a Waters NanoAcquity HPLC system interfaced to a ThermoFisher Q Exactive with the Orbitrap operating at 60,000 FWHM and 17,500 FWHM for MS and MS/MS respectively. The fifteen most abundant ions were selected for MS/MS. ### Database searching Raw tandem mass spectra data files were extracted, converted to mgf files using Proteome Discoverer version 1.3, and analyzed using Mascot (Matrix Science, London, UK). Mascot was set up to search the SwissProt-2013-07 database (selected for Homo sapiens, 20265 entries) Mascot was searched with a fragment ion mass tolerance of 0.80 Da and a parent ion tolerance of 10.0 PPM. For the redox proteome analysis we used as variable modifications the amino acids oxidations reported in Mascot. ### Validation of the posttranslational (PTM) modifications and normalizing spectral counts The proteins were considered identified having at least one bold red (BR) significant peptide with an ion score cut-off of 20 or greater (corresponding to p \< 0.05 and a FDR proteins \<1.0). Scaffold (Proteome Software Inc., Portland, OR) was used to validate MS/MS based peptide and protein identifications. ### Immobilized metal affinity chromatography (IMAC) purification of metalloproteins IMAC columns were charged with either Co(II) or Cr(III) (300 mM in PBS, pH 6.5). Columns were equilibrated in 50 mM NaH~2~PO~4~ with 10 mM imidazole and 0.3 M NaCl at 1 ml/min flow rate using the Bio-Rad FPLC system interfaced with Biologic DuoFlow software. 5 mg of protein homogenates were loaded on each IMAC column and 1 ml fractions were collected for the entire run. Eluted fractions at 20--350 mM imidazole were further fractionated on 1D 4%--20% SDS-PAGE and subsequently silver stained. 10 gel bands were cut across each sample lane and subjected to bottom-up proteomics analysis. *In situ* endoproteinase Lys-C and tryptic digestion was performed after reduction and alkylation. Extracted tryptic peptides were analyzed by nano LC--ESI--MS/MS on Orbitrap and the data were processed with Proteome Discoverer, Mascot and Scaffold. ### Aldolase catalytic activity Using the Boyer modification of the hydrazine assay[@b62], in which 3-phosphoglyceraldehyde reacts with hydrazine to form a hydrazone product which absorbs at 240 nm, we designed a microassay in which pure human recombinant aldolase A (1--364 aa) (50 μg/assay mixture) was left untreated or incubated with different concentrations of the two heavy metals (4 mM, 0.4 mM and 0.04 mM for each Co(II) and Cr(III) solution) in the presence of the substrate, D-fructose 1,6-biphosphate (4.0 mM final concentration) and hydrazine sulfate (about 2.5 mM final concentration) in PBS at pH 7.4. The formation of the hydrazone was characterized by an increase in the absorption at 240 nm monitored over 10--15 min incubation time. ### Catalase catalytic activity To test the inhibition of catalase by Co(II) and Cr(III) we developed a microassay adapted from the Worthington\'s assay[@b63] in which the disappearance of peroxide is followed spectrophotometrically at 240 nm. 20--50 μg of pure protein (from human erythrocytes) were incubated in an assay mixture containing 0.05 M potassium phosphate, pH 7.0 and 0.059 M hydrogen peroxide (30%) in 0.05 M potassium phosphate, pH 7.0 as the substrate. The consumption of peroxide was monitored in the presence or absence of Co(II) or Cr(III) using the same concentration range as in the aldolase assay. Author Contributions ==================== B.S., G.P., E.P., S.G., N.C., R.V., L.S. wrote the main manuscript text. B.S., C.C., G.P., L.S. prepared figures 1--6. B.S., C.C., V.Z., G.P., B.Y., G.E., F.M. performed the experiments. All authors reviewed the manuscript. Supplementary Material {#s1} ====================== ###### Supplementary Information Table S1 ###### Supplementary Information Table S2 ###### Supplementary Information Table S3 VZ is supported by the PhD program in Genetics and cell Biology at the University of Tuscia, Department of Ecology and Biology(DEB), Viterbo, Italy. {#f1} {#f2} {#f3} {ref-type="supplementary-material"}. (g) Subcellular distribution of the oxidatively damaged proteome retrieved in the periprosthetic tissues surrounding MoM implants. The list of proteins is reported in [Supplement Table 1](#s1){ref-type="supplementary-material"}. (h) Line graph comparing carbonyl content (nmol/mg) vs Cr (ng/ml) levels in periprosthetic tissues surrounding implants.](srep05729-f4){#f4} {ref-type="supplementary-material"}) were assigned using Protein Knowledgebase (uniProtKB) (<http://www.uniprot.org/>) and MetalMine (<http://metalmine.naist.jp/>) databases.](srep05729-f5){#f5} {#f6} ###### Patient demographics and information regarding prostheses Patient \# Age Sex Implant type Duration of implant Reason for revision ------------ ----- ----- ----------------- --------------------- --------------------- 1 73 F MoP 57 months Dislocation 2 54 F MoP 37 months Dislocation 3 76 M MoP 81 months Osteolysis 4 69 F MoP 101 months Osteolysis 5 79 M MoP 78 months Osteolysis 6 81 F CoP 124 months Osteolysis 7 43 F CoP 24 months Dislocation 8 49 M CoP 38 months Mechanical Failure 9 69 M CoC 80 months Dislocation 10 45 F MoM Resurfacing 19 months ALTR 11 52 F MoM Resurfacing 63 months ALTR 12 44 F MoM Resurfacing 34 months ALTR 13 43 M MoM Resurfacing 72 months ALTR 14 47 M MoM THA 52 months ALTR 15 71 M MoM THA 58 months ALTR 16 56 F MoM THA 36 months ALTR 17 67 M MoM THA 21 months ALTR 18 64 F MoM THA 61 months ALTR MoP: metal on polyethylene; CoP: ceramic on polyethylene; CoC: ceramic on ceramic; MoM: metal on metal. THA: total hip arthroplasty; ALTR: adverse local tissue reaction. Note: All patients did not have occupational exposure or known leisure exposure to products containing significant amounts of cobalt or chromium.

Related literature {#sec1} ================== For information on values of bond lengths, see Allen *et al.* (1987[@bb5]). For related structures of dithio­carbazate derivatives, see, for example: Crouse *et al.* (2004[@bb9]); Fun *et al.* (2008[@bb10]); Shanmuga Sundara Raj *et al.* (2000[@bb11]). For applications and bioactivities of dithio­carbazate derivatives, see, for example: Ali & Tarafder (1977[@bb4]); Ali *et al.* (2001[@bb2], 2002[@bb3], 2008[@bb1]); Chan *et al.* (2008[@bb7]); Chew *et al.* (2004[@bb8]); Crouse *et al.* (2004[@bb9]); Tarafder *et al.* (1978[@bb15], 1981[@bb18], 2001[@bb17], 2008[@bb16]). Experimental {#sec2} ============ {#sec2.1} ### Crystal data {#sec2.1.1} C~17~H~16~N~2~S~2~*M* *~r~* = 312.44Triclinic,*a* = 5.4350 (3) Å*b* = 11.6333 (7) Å*c* = 13.6289 (8) Åα = 66.869 (4)°β = 82.723 (4)°γ = 87.520 (4)°*V* = 786.04 (8) Å^3^*Z* = 2Mo *K*α radiationμ = 0.33 mm^−1^*T* = 100.0 (1) K0.58 × 0.19 × 0.05 mm ### Data collection {#sec2.1.2} Bruker SMART APEX2 CCD area-detector diffractometerAbsorption correction: multi-scan (*SADABS*; Bruker, 2005[@bb6]) *T* ~min~ = 0.829, *T* ~max~ = 0.98216100 measured reflections3570 independent reflections2870 reflections with *I* \> 2σ(*I*)*R* ~int~ = 0.044 ### Refinement {#sec2.1.3} *R*\[*F* ^2^ \> 2σ(*F* ^2^)\] = 0.035*wR*(*F* ^2^) = 0.092*S* = 1.073570 reflections194 parametersH atoms treated by a mixture of independent and constrained refinementΔρ~max~ = 0.28 e Å^−3^Δρ~min~ = −0.27 e Å^−3^ {#d5e542} Data collection: *APEX2* (Bruker, 2005[@bb6]); cell refinement: *APEX2*; data reduction: *SAINT* (Bruker, 2005[@bb6]); program(s) used to solve structure: *SHELXTL* (Sheldrick, 2008[@bb12]); program(s) used to refine structure: *SHELXTL*; molecular graphics: *SHELXTL*; software used to prepare material for publication: *SHELXTL* and *PLATON* (Spek, 2003[@bb13]). Supplementary Material ====================== Crystal structure: contains datablocks global, I. DOI: [10.1107/S1600536808013354/sj2494sup1.cif](http://dx.doi.org/10.1107/S1600536808013354/sj2494sup1.cif) Structure factors: contains datablocks I. DOI: [10.1107/S1600536808013354/sj2494Isup2.hkl](http://dx.doi.org/10.1107/S1600536808013354/sj2494Isup2.hkl) Additional supplementary materials: [crystallographic information](http://scripts.iucr.org/cgi-bin/sendsupfiles?sj2494&file=sj2494sup0.html&mime=text/html); [3D view](http://scripts.iucr.org/cgi-bin/sendcif?sj2494sup1&Qmime=cif); [checkCIF report](http://scripts.iucr.org/cgi-bin/paper?sj2494&checkcif=yes) Supplementary data and figures for this paper are available from the IUCr electronic archives (Reference: [SJ2494](http://scripts.iucr.org/cgi-bin/sendsup?sj2494)). KAC thanks Universiti Putra Malaysia for financial help. MTHT thanks the University of Rajshahi for the provision of laboratory facilities. The authors also thank Universiti Sains Malaysia for the Research University Golden Goose grant No. 1001/PFIZIK/811012. Comment ======= There has been immense interest in nitrogen-sulfur donor ligands since our report on *S*-benzyldithiocarbazate (SBDTC) (Ali & Tarafder, 1977). There have also been a number of reports of Schiff bases derived from SBDTC (Ali *et al.*, 2001, 2002, 2008; Chan *et al.*, 2008; Chew *et al.*, 2004; Tarafder *et al.*, 1978, 1981, 2001; Raj *et al.*, 2000). The intriguing coordination chemistry and increasingly important biomedical properties of ligands derived from SBDTC have also received much attention (Ali *et al.*, 2001, 2002; Crouse *et al.*, 2004; Tarafder *et al.*, 2001, 2008). The synthesis and structure of SBDTC have been reported previously (Ali & Tarafder (1977); Shanmuga Sundara Raj *et al.*, 2000). In continuation of our research, the title compound (I), a ligand with both N and S donor atoms, was synthesized and its crystal structure is reported here. (I) is likely to have biomedical properties similar to other nitrogen-sulfur donor ligands studied by our group. In the title compound (Fig. 1), the 3-phenylprop-2-enylidene amide (N2/C9--C17) and benzyl groups (C1--C7) adopt *trans* and *cis* positions with respect to the terminal thione S2 atom about the C8-N1 and C8-S1 bonds, respectively. The 3-phenylprop-2-enylidene (C9--C17) and the dithiocarbazate (N1/N2/S1/S2/C8) fragments is essentially planar with maximum deviation 0.074 (2) Å for C11, while the dihedral angle between the 3-phenylprop-2-enylidene and the benzyl group is 77.78 (7)°. The bond lengths and angles are in normal ranges (Allen *et al.*, 1987). However the C═S distance of 1.7466 (17) Å is longer than the typical value of dithiocarbazate derivatives (Crouse *et al.*, 2004; Fun *et al.*, 2008; Shanmuga Sundara Raj *et al.*, 2000) but being intermediate between the values of 1.82 Å for a C---S single bond and 1.56 Å for a C═S double bond (Suton, 1965). The C9--N2 distance of 1.285 (2) Å indicates a double bond charactor. The bond angles S1--C8--S2 \[124.67 (10)°\] and N1--C8--S1 \[113.76 (13)°\] also agree with those observed in *trans-cisS*-benzyl dithiocarbazate (Shanmuga Sundara Raj *et al.*, 2004). In the crystal packing (Fig. 2), the molecules are linked by an N1---H1···S2^i^ hydrogen bond (symmetry code: i = -x, 1-y, 1-z) (Table 1) and a weak C9---H9A···S2^i^ interaction involving the terminal thione-S atom forming dimers that are arranged into sheets parallel to the *bc* plane. The crystal is also stabilized by C---H···π interactions (Table 1) involving the C1--C6 phenyl ring (centroid *Cg*1). Experimental {#experimental} ============ The title compound was synthesized by adding cinnamaldehyde (1.34 g, 10 mmol) to a solution of *S*-benzyldithiocarbazate (SBDTC) (1.98 g, 10 mmol) in absolute ethanol (60 ml) and the mixture was refluxed for 40 min. The yellow precipitate, which formed was separated and dried in vacuo over anhydrous CaCl~2~ (Yield: 2.1 g, 63%). Yellow needle shaped single crystals of (I) were obtained after recrystallization from absolute ethanol over 15 days; *M*.p 454 K. Refinement {#refinement} ========== The H1N1 hydrogen atom was located from a difference Fourier map and refined freely with isotropic displacement parameters. The remaining H atoms were positioned geometrically and allowed to ride on their parent atoms, with d(C---H) = 0.93 Å, for CH and aromatic, 0.97 Å, for CH~2~ and *U*~iso~ = 1.2*U*~eq~(C). The highest residual electron density peak is located at 0.96 Å from S1 and the deepest hole is located at 0.72 Å from S1. Figures ======= {#Fap1} {#Fap2} Crystal data {#tablewrapcrystaldatalong} ============ ----------------------- --------------------------------------- C~17~H~16~N~2~S~2~ *Z* = 2 *M~r~* = 312.44 *F*~000~ = 328 Triclinic, *P*1 *D*~x~ = 1.320 Mg m^−3^ Hall symbol: -P 1 Melting point: 454 K *a* = 5.4350 (3) Å Mo *K*α radiation λ = 0.71073 Å *b* = 11.6333 (7) Å Cell parameters from 3570 reflections *c* = 13.6289 (8) Å θ = 1.9--27.5º α = 66.869 (4)º µ = 0.33 mm^−1^ β = 82.723 (4)º *T* = 100.0 (1) K γ = 87.520 (4)º Needle, yellow *V* = 786.04 (8) Å^3^ 0.58 × 0.19 × 0.05 mm ----------------------- --------------------------------------- Data collection {#tablewrapdatacollectionlong} =============== --------------------------------------------------------- -------------------------------------- Bruker SMART APEX2 CCD area-detector diffractometer 3570 independent reflections Radiation source: fine-focus sealed tube 2870 reflections with *I* \> 2σ(*I*) Monochromator: graphite *R*~int~ = 0.044 Detector resolution: 8.33 pixels mm^-1^ θ~max~ = 27.5º *T* = 100.0(1) K θ~min~ = 1.9º ω scans *h* = −7→7 Absorption correction: multi-scan(SADABS; Bruker, 2005) *k* = −14→15 *T*~min~ = 0.829, *T*~max~ = 0.982 *l* = −17→17 16100 measured reflections --------------------------------------------------------- -------------------------------------- Refinement {#tablewraprefinementdatalong} ========== ---------------------------------------------------------------- --------------------------------------------------------------------------------------------------- Refinement on *F*^2^ Secondary atom site location: difference Fourier map Least-squares matrix: full Hydrogen site location: inferred from neighbouring sites *R*\[*F*^2^ \> 2σ(*F*^2^)\] = 0.035 H atoms treated by a mixture of independent and constrained refinement *wR*(*F*^2^) = 0.093   *w* = 1/\[σ^2^(*F*~o~^2^) + (0.0368*P*)^2^ + 0.3179*P*\] where *P* = (*F*~o~^2^ + 2*F*~c~^2^)/3 *S* = 1.07 (Δ/σ)~max~ = 0.001 3570 reflections Δρ~max~ = 0.29 e Å^−3^ 194 parameters Δρ~min~ = −0.27 e Å^−3^ Primary atom site location: structure-invariant direct methods Extinction correction: none ---------------------------------------------------------------- --------------------------------------------------------------------------------------------------- Special details {#specialdetails} =============== ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Experimental. The low-temperature data was collected with the Oxford Cyrosystem Cobra low-temperature attachment. Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes. Refinement. Refinement of F^2^ against ALL reflections. The weighted R-factor wR and goodness of fit S are based on F^2^, conventional R-factors R are based on F, with F set to zero for negative F^2^. The threshold expression of F^2^ \> 2sigma(F^2^) is used only for calculating R-factors(gt) etc. and is not relevant to the choice of reflections for refinement. R-factors based on F^2^ are statistically about twice as large as those based on F, and R- factors based on ALL data will be even larger. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Fractional atomic coordinates and isotropic or equivalent isotropic displacement parameters (Å^2^) {#tablewrapcoords} ================================================================================================== ------ -------------- -------------- -------------- -------------------- -- *x* *y* *z* *U*~iso~\*/*U*~eq~ S1 −0.13630 (8) 0.74106 (4) 0.65560 (3) 0.02137 (13) S2 −0.23454 (8) 0.66804 (4) 0.47403 (4) 0.02329 (13) N1 0.1010 (3) 0.56987 (15) 0.60587 (12) 0.0209 (3) N2 0.2180 (3) 0.55500 (14) 0.69440 (11) 0.0212 (3) C1 −0.2508 (3) 1.02812 (17) 0.63242 (14) 0.0241 (4) H1A −0.1133 1.0404 0.5812 0.029\* C2 −0.2838 (3) 1.10363 (18) 0.68983 (15) 0.0270 (4) H2A −0.1690 1.1664 0.6771 0.032\* C3 −0.4872 (3) 1.08611 (18) 0.76620 (15) 0.0265 (4) H3A −0.5087 1.1364 0.8054 0.032\* C4 −0.6584 (3) 0.99380 (18) 0.78414 (15) 0.0261 (4) H4A −0.7964 0.9824 0.8350 0.031\* C5 −0.6253 (3) 0.91803 (17) 0.72668 (15) 0.0230 (4) H5A −0.7413 0.8558 0.7393 0.028\* C6 −0.4207 (3) 0.93393 (16) 0.65043 (14) 0.0203 (4) C7 −0.3781 (3) 0.84724 (17) 0.59218 (14) 0.0222 (4) H7A −0.3250 0.8938 0.5164 0.027\* H7B −0.5287 0.8014 0.5993 0.027\* C8 −0.0802 (3) 0.65301 (16) 0.57678 (14) 0.0199 (4) C9 0.3953 (3) 0.47537 (16) 0.71089 (14) 0.0205 (4) H9A 0.4371 0.4369 0.6628 0.025\* C10 0.5303 (3) 0.44442 (16) 0.80163 (14) 0.0207 (4) H10A 0.4896 0.4829 0.8497 0.025\* C11 0.7138 (3) 0.36108 (16) 0.81818 (14) 0.0209 (4) H11A 0.7506 0.3277 0.7662 0.025\* C12 0.8636 (3) 0.31572 (16) 0.90742 (14) 0.0202 (4) C13 0.8492 (3) 0.36609 (17) 0.98578 (15) 0.0254 (4) H13A 0.7375 0.4298 0.9830 0.030\* C14 0.9991 (4) 0.32206 (18) 1.06702 (15) 0.0285 (4) H14A 0.9889 0.3570 1.1182 0.034\* C15 1.1645 (4) 0.22635 (19) 1.07320 (15) 0.0298 (4) H15A 1.2654 0.1972 1.1282 0.036\* C16 1.1791 (3) 0.17447 (19) 0.99739 (15) 0.0306 (5) H16A 1.2890 0.1096 1.0016 0.037\* C17 1.0301 (3) 0.21890 (18) 0.91488 (15) 0.0257 (4) H17A 1.0415 0.1836 0.8639 0.031\* H1N1 0.129 (4) 0.519 (2) 0.5733 (18) 0.038 (6)\* ------ -------------- -------------- -------------- -------------------- -- Atomic displacement parameters (Å^2^) {#tablewrapadps} ===================================== ----- ------------- ------------- ------------- -------------- --------------- --------------- *U*^11^ *U*^22^ *U*^33^ *U*^12^ *U*^13^ *U*^23^ S1 0.0237 (2) 0.0250 (2) 0.0198 (2) 0.00424 (18) −0.00880 (17) −0.01190 (19) S2 0.0242 (2) 0.0305 (3) 0.0194 (2) 0.00284 (19) −0.00832 (17) −0.0129 (2) N1 0.0221 (8) 0.0265 (8) 0.0190 (8) 0.0024 (6) −0.0071 (6) −0.0128 (7) N2 0.0218 (8) 0.0265 (8) 0.0165 (7) −0.0001 (6) −0.0052 (6) −0.0090 (6) C1 0.0193 (9) 0.0296 (10) 0.0230 (9) 0.0001 (8) 0.0002 (7) −0.0106 (8) C2 0.0238 (10) 0.0256 (10) 0.0316 (11) −0.0046 (8) −0.0010 (8) −0.0114 (9) C3 0.0255 (10) 0.0279 (10) 0.0311 (10) 0.0026 (8) −0.0030 (8) −0.0172 (9) C4 0.0193 (9) 0.0321 (11) 0.0279 (10) −0.0005 (8) 0.0007 (7) −0.0136 (9) C5 0.0182 (9) 0.0250 (10) 0.0260 (10) −0.0033 (7) −0.0041 (7) −0.0094 (8) C6 0.0189 (9) 0.0223 (9) 0.0195 (9) 0.0031 (7) −0.0076 (7) −0.0067 (7) C7 0.0212 (9) 0.0258 (10) 0.0211 (9) 0.0027 (7) −0.0081 (7) −0.0095 (8) C8 0.0197 (9) 0.0225 (9) 0.0174 (9) −0.0025 (7) −0.0018 (7) −0.0076 (7) C9 0.0199 (9) 0.0229 (9) 0.0212 (9) −0.0018 (7) −0.0027 (7) −0.0109 (8) C10 0.0230 (9) 0.0233 (9) 0.0185 (9) −0.0023 (7) −0.0039 (7) −0.0102 (7) C11 0.0226 (9) 0.0217 (9) 0.0207 (9) −0.0029 (7) −0.0034 (7) −0.0103 (7) C12 0.0179 (9) 0.0206 (9) 0.0209 (9) −0.0033 (7) −0.0032 (7) −0.0063 (7) C13 0.0279 (10) 0.0239 (10) 0.0245 (10) 0.0001 (8) −0.0067 (8) −0.0085 (8) C14 0.0361 (11) 0.0285 (10) 0.0214 (10) −0.0048 (8) −0.0084 (8) −0.0083 (8) C15 0.0236 (10) 0.0373 (11) 0.0217 (10) −0.0033 (8) −0.0074 (8) −0.0025 (9) C16 0.0227 (10) 0.0351 (11) 0.0276 (10) 0.0061 (8) −0.0033 (8) −0.0058 (9) C17 0.0243 (10) 0.0293 (10) 0.0229 (10) 0.0024 (8) −0.0017 (7) −0.0100 (8) ----- ------------- ------------- ------------- -------------- --------------- --------------- Geometric parameters (Å, °) {#tablewrapgeomlong} =========================== ------------------- -------------- ----------------------- -------------- S1---C8 1.7466 (17) C7---H7A 0.9700 S1---C7 1.8187 (17) C7---H7B 0.9700 S2---C8 1.6696 (18) C9---C10 1.433 (2) N1---C8 1.334 (2) C9---H9A 0.9300 N1---N2 1.382 (2) C10---C11 1.337 (2) N1---H1N1 0.87 (2) C10---H10A 0.9300 N2---C9 1.285 (2) C11---C12 1.460 (2) C1---C2 1.381 (3) C11---H11A 0.9300 C1---C6 1.390 (3) C12---C17 1.394 (2) C1---H1A 0.9300 C12---C13 1.399 (3) C2---C3 1.381 (3) C13---C14 1.378 (3) C2---H2A 0.9300 C13---H13A 0.9300 C3---C4 1.379 (3) C14---C15 1.384 (3) C3---H3A 0.9300 C14---H14A 0.9300 C4---C5 1.384 (3) C15---C16 1.380 (3) C4---H4A 0.9300 C15---H15A 0.9300 C5---C6 1.387 (2) C16---C17 1.387 (3) C5---H5A 0.9300 C16---H16A 0.9300 C6---C7 1.504 (2) C17---H17A 0.9300 C8---S1---C7 102.56 (8) N1---C8---S1 113.76 (13) C8---N1---N2 120.49 (15) S2---C8---S1 124.67 (10) C8---N1---H1N1 118.0 (15) N2---C9---C10 121.56 (16) N2---N1---H1N1 120.9 (15) N2---C9---H9A 119.2 C9---N2---N1 114.00 (14) C10---C9---H9A 119.2 C2---C1---C6 120.70 (17) C11---C10---C9 121.02 (16) C2---C1---H1A 119.6 C11---C10---H10A 119.5 C6---C1---H1A 119.6 C9---C10---H10A 119.5 C1---C2---C3 120.11 (18) C10---C11---C12 128.25 (16) C1---C2---H2A 119.9 C10---C11---H11A 115.9 C3---C2---H2A 119.9 C12---C11---H11A 115.9 C4---C3---C2 119.76 (17) C17---C12---C13 118.27 (16) C4---C3---H3A 120.1 C17---C12---C11 118.96 (16) C2---C3---H3A 120.1 C13---C12---C11 122.77 (16) C3---C4---C5 120.16 (17) C14---C13---C12 120.56 (17) C3---C4---H4A 119.9 C14---C13---H13A 119.7 C5---C4---H4A 119.9 C12---C13---H13A 119.7 C4---C5---C6 120.65 (17) C13---C14---C15 120.60 (18) C4---C5---H5A 119.7 C13---C14---H14A 119.7 C6---C5---H5A 119.7 C15---C14---H14A 119.7 C5---C6---C1 118.61 (16) C16---C15---C14 119.64 (17) C5---C6---C7 120.57 (17) C16---C15---H15A 120.2 C1---C6---C7 120.76 (16) C14---C15---H15A 120.2 C6---C7---S1 105.49 (12) C15---C16---C17 120.11 (18) C6---C7---H7A 110.6 C15---C16---H16A 119.9 S1---C7---H7A 110.6 C17---C16---H16A 119.9 C6---C7---H7B 110.6 C16---C17---C12 120.81 (18) S1---C7---H7B 110.6 C16---C17---H17A 119.6 H7A---C7---H7B 108.8 C12---C17---H17A 119.6 N1---C8---S2 121.57 (13) C8---N1---N2---C9 −177.45 (16) C7---S1---C8---S2 −2.52 (14) C6---C1---C2---C3 0.1 (3) N1---N2---C9---C10 −177.41 (16) C1---C2---C3---C4 −0.6 (3) N2---C9---C10---C11 179.74 (17) C2---C3---C4---C5 0.7 (3) C9---C10---C11---C12 −177.74 (17) C3---C4---C5---C6 −0.1 (3) C10---C11---C12---C17 173.65 (18) C4---C5---C6---C1 −0.5 (3) C10---C11---C12---C13 −6.9 (3) C4---C5---C6---C7 176.79 (16) C17---C12---C13---C14 1.0 (3) C2---C1---C6---C5 0.5 (3) C11---C12---C13---C14 −178.40 (17) C2---C1---C6---C7 −176.74 (17) C12---C13---C14---C15 −0.7 (3) C5---C6---C7---S1 −102.64 (16) C13---C14---C15---C16 −0.1 (3) C1---C6---C7---S1 74.54 (18) C14---C15---C16---C17 0.6 (3) C8---S1---C7---C6 −175.27 (12) C15---C16---C17---C12 −0.2 (3) N2---N1---C8---S2 −176.88 (13) C13---C12---C17---C16 −0.6 (3) N2---N1---C8---S1 3.0 (2) C11---C12---C17---C16 178.89 (17) C7---S1---C8---N1 177.58 (14) ------------------- -------------- ----------------------- -------------- Hydrogen-bond geometry (Å, °) {#tablewraphbondslong} ============================= ---------------------- ---------- ---------- ------------- --------------- *D*---H···*A* *D*---H H···*A* *D*···*A* *D*---H···*A* N1---H1N1···S2^i^ 0.87 (2) 2.53 (2) 3.3714 (19) 165 (2) C9---H9A···S2^i^ 0.93 2.93 3.7264 (18) 144 C15---H15A···Cg1^ii^ 0.93 2.83 3.649 (2) 148 ---------------------- ---------- ---------- ------------- --------------- Symmetry codes: (i) −*x*, −*y*+1, −*z*+1; (ii) −*x*+1, −*y*+1, −*z*+2. ###### Hydrogen-bond geometry (Å, °) *D*---H⋯*A* *D*---H H⋯*A* *D*⋯*A* *D*---H⋯*A* ------------------------ ---------- ---------- ------------- ------------- N1---H1*N*1⋯S2^i^ 0.87 (2) 2.53 (2) 3.3714 (19) 165 (2) C9---H9*A*⋯S2^i^ 0.93 2.93 3.7264 (18) 144 C15---H15*A*⋯*Cg*1^ii^ 0.93 2.83 3.649 (2) 148 Symmetry codes: (i) ; (ii) . *Cg*1 is the centroid of the C1--C6 phenyl ring. [^1]: Additional correspondence author, e-mail: suchada.c\@psu.ac.th. [^2]: Additional correspondence author, e-mail: hkfun\@usm.my.

1. Introduction {#sec1-materials-09-00461} =============== Geopolymers are a family of inorganic materials obtained by reaction between an aqueous alkaline silicate solution and an aluminosilicate source \[[@B1-materials-09-00461],[@B2-materials-09-00461]\]. This reaction yields an amorphous three-dimensional structure in which SiO~4~ and AlO~4~ tetrahedra are linked by corner-shared O atoms. Geopolymers are characterized by interesting mechanical properties, low shrinkage, thermal stability, freeze-thaw, chemical and fire resistance, long term durability and recyclability. For these reasons, they have the potential for utilization as Ordinary Portland Cement (OPC) replacement in a wide range of applications, such as fireproof barriers, materials for high temperatures, matrices for hazardous waste stabilization, toolings and moldings \[[@B3-materials-09-00461],[@B4-materials-09-00461]\]. Moreover, with respect to the manufacturing of OPC that consumes a significant amount of natural materials and energy release of a large quantity of carbon dioxide in the atmosphere \[[@B5-materials-09-00461],[@B6-materials-09-00461],[@B7-materials-09-00461],[@B8-materials-09-00461]\], the use of geopolymer-based materials in concrete applications could significantly reduce the CO~2~ emissions \[[@B9-materials-09-00461]\] thanks to the "low carbon" footprint of several raw materials with a high concentration of aluminosilicates from which they can be prepared, *i.e.*, dehydroxylated kaolinite (metakaolin, MK) or industrial waste such as fly ash. Fly ash (FA) is a fine powder by-product transported by flue gas after the combustion of coal in coal-fired power stations typically made up of small glass spheres, consisting primarily of silicon, aluminium, iron, and calcium oxides \[[@B10-materials-09-00461],[@B11-materials-09-00461],[@B12-materials-09-00461],[@B13-materials-09-00461],[@B14-materials-09-00461]\]. The development of FA-based geopolymer concretes could contribute to recycling waste into construction material, thus reducing, at the same time, CO~2~ emissions \[[@B15-materials-09-00461],[@B16-materials-09-00461],[@B17-materials-09-00461]\]. However, FA-based geopolymers typically exhibit brittle behavior with low tensile strength, ductility, and fracture toughness, thus limiting up to now the actual possibility to use these very promising materials for extensive and practical applications in constructions. This limit could be, in principle, overcome by developing geopolymer composites, but, to the best of our knowledge, very little is reported on this topic. Only recently, Li *et al.* \[[@B18-materials-09-00461]\] described the utilization of chitosan biopolymers for the implementation of composite systems based on geopolymers obtained from fly ash, in which the formation of a three-dimensional cross-linked composite is achieved by means of a network of hydrogen bonds between geopolymer and chitosan macromolecules ([Figure 1](#materials-09-00461-f001){ref-type="fig"}). In that case, the interaction between the inorganic matrix and the macromolecule is effective up to 0.1% by weight of *N*-carboxymethyl chitosan content, while a further increase of the chitosan amount produces a decrease of the geopolymerization degree, probably due to an encapsulation effect of the fly ash particles that become un-reactive \[[@B18-materials-09-00461]\]. It is worth noting that, in a bid to obtain advanced geopolymer-based composite materials, a wide range of organic polymers have been studied in literature as organic fillers, such as polyvinyl acetate \[[@B19-materials-09-00461]\], polypropylene \[[@B20-materials-09-00461]\], polyvinyl alcohol \[[@B21-materials-09-00461]\], or water-soluble organic polymers \[[@B22-materials-09-00461]\]. These kind of composites are usually obtained by blending the polymer with geopolymers, sometimes in the presence of compatibilizers \[[@B23-materials-09-00461],[@B24-materials-09-00461],[@B25-materials-09-00461]\]. Nevertheless, a worsening effect of the mechanical properties of the final product \[[@B26-materials-09-00461]\] for very low concentrations (up to \~1 wt %), similar to that previously described in the case of chitosan, has been observed. In these last cases, this detrimental effect is probably due to the use of organic materials poorly compatible with the inorganic matrix that, at high amounts of the organic components, causes a phase separation between the organic filler and the inorganic geopolymer matrix. Very recently, we have developed an innovative, easy and cost-effective synthetic strategy to realize hybrid composite materials by using metakaolin-based geopolymers as inorganic components and different organic resins up to 25% in weight \[[@B27-materials-09-00461],[@B28-materials-09-00461],[@B29-materials-09-00461],[@B30-materials-09-00461],[@B31-materials-09-00461],[@B32-materials-09-00461]\]. These materials show significantly improved physical and mechanical properties and remarkably reduced brittleness with respect to the neat geopolymers, still preserving good thermal and fire resistance \[[@B30-materials-09-00461],[@B32-materials-09-00461]\]. This approach consists of the concurrent co-reticulation of both phases that are mixed together when each polymerization reaction is already started but is far from being completed, thus allowing for realization of a chemical interaction between the organic component and the geopolymeric mixture based on the formation of a wide network of hydrogen bonding ([Figure 2](#materials-09-00461-f002){ref-type="fig"}) \[[@B29-materials-09-00461]\]. This approach ensures high compatibility between the phases and a very good dispersion even at the nanometric level of the organic phase that can be included in the mixture up to about 25% by weight, without addition of external additives or compatibilizers \[[@B30-materials-09-00461]\]. In the present paper, we report on the preparation of novel FA-based geopolymer composites, obtained by extending the synthetic method developed by us in the case of more expensive metakaolin-based geopolymers to raw waste materials. In such a way, we have succeeded in combining an easy and sustainable synthetic approach with the use of industrial by-products for the realization of novel "eco-friendly" aluminosilicate binders in order to preserve the environment and limit the cost of construction materials. These new promising materials have been characterized by means of several techniques, such as scanning electron microscopy (SEM), thermal analysis (TGA), X-ray diffraction (XRD), and compressive strength tests. Their properties have been compared with neat geopolymer samples and analogous samples obtained from metakaolin \[[@B28-materials-09-00461]\]. 2. Experimental Section {#sec2-materials-09-00461} ======================= 2.1. Materials and Methods {#sec2dot1-materials-09-00461} -------------------------- Class F coal fly ash employed in this work was supplied by the ENEL S.p.A. power plant located in Brindisi (Southern Italy) and was used as received, without drying treatment. The water content was found to be 4% after drying in an oven at 105 °C until reaching a constant mass. Its chemical composition was obtained by means of a Perkin-Elmer Optima 2100 DV ICP-OES apparatus (Waltham, MA, USA) \[[@B14-materials-09-00461],[@B33-materials-09-00461]\] and is reported in [Table 1](#materials-09-00461-t001){ref-type="table"}. The water content of fly ash was taken into account in the mix design. Metakaolin was kindly provided by Neuchem S.r.l. (Milan, Italy), and its composition is reported in [Table 1](#materials-09-00461-t001){ref-type="table"}. Epojet^®^ epoxy resin \[[@B34-materials-09-00461]\] was purchased by Mapei S.p.A. (Milan, Italy). Sodium hydroxide was purchased from Sigma Aldrich (St. Louis, MO, USA) and used without further purification. The sodium silicate solution was supplied by Prochin Italia S.r.l. (Naples, Italy), and its composition is reported in [Table 1](#materials-09-00461-t001){ref-type="table"}. The fly ash particle size distribution was determined by means of a Malvern Mastersizer 3000 laser particle analyser (Malvern, UK). Thermogravimetric analyses (TGA) were performed by a Mettler-Toledo TGA/DSC2 STAR^e^ SYSTEM (Columbus, OH, USA). The thermographs were obtained with a heating rate of 10 °C/min using ≈10 mg of the powdered sample under air flow. X-ray diffraction patterns were obtained at room temperature with an automatic Rigaku powder diffractometer mod. Miniflex 600 (Tokyo, Japan), operating in the θ/2θ Bragg-Brentano geometry. The phase recognition was carried out by using the PDF-4+ 2014 (International Centre for Diffraction Data^®^, Tokyo, Japan) database and the Rigaku PDXL2 software (Rigaky, Tokyo, Japan). SEM analysis was carried out by means of a Nova NanoSem 450 FEI Microscope (Hillsboro, OR, USA). The compressive strength was evaluated according to EN 196-1 and measured by testing cubic paste specimens (30 × 30 × 30 mm^3^) in a Controls MCC8 multipurpose testing machine (CONTROLS s.r.l., Liscate, Milan, Italy) with a capacity of 100 kN. The tests were performed after 28 days of curing at room temperature, and the values reported are the averages of the five compression strength values. 2.2. Specimen Preparation {#sec2dot2-materials-09-00461} ------------------------- ### 2.2.1. Preparation of Metakaolin-Based Geopolymer (G-MK) {#sec2dot2dot1-materials-09-00461} The alkaline activating solution was prepared by dissolving solid sodium hydroxide into the sodium silicate solution. The solution was then allowed to equilibrate and cool for 24 h. The composition of the obtained solution can be expressed as Na~2~O·1.34SiO~2~ 10.5H~2~O. Then, the metakaolin was incorporated to the activating solution with a liquid to solid ratio of 1.4:1 by weight and mixed by a mechanical mixer for 10 min at 800 rpm. The composition of the whole geopolymeric system can be expressed as Al~2~O~3~ 3.5SiO~2~ 1.0Na~2~O·10.5H~2~O, assuming that geopolymerization occurred at 100%. Such composition is in good agreement with that actually determined by EDS analyses on the prepared samples. ### 2.2.2. Preparation of Fly Ash-Based Geopolymer (G-FA) {#sec2dot2dot2-materials-09-00461} The alkaline activating solution was prepared by mixing the sodium silicate solution with an aqueous solution of sodium hydroxide 10 M. The solution was then allowed to equilibrate and cool for 24 h. The composition of the obtained solution can be expressed as Na~2~O 0.9SiO~2~ 14.7H~2~O. Then, fly ash was incorporated into the activating solution with a liquid to solid ratio of 0.66:1 by weight and mixed by a mechanical mixer for 20 min at 800 rpm. The composition of the whole geopolymeric system can be expressed as Al~2~O~3~ 3.79SiO~2~ 0.66Na~2~O 8.9H~2~O, assuming that geopolymerization occurred at 100% (see the related discussion in [Section 3.1.4](#sec3dot1dot4-materials-09-00461){ref-type="sec"}). ### 2.2.3. Preparation of the EPOXY-Geopolymer Composites {#sec2dot2dot3-materials-09-00461} Geopolymer-based composites have been obtained by adding Epojet^®^ resin to the freshly-prepared geopolymeric suspension, and were quickly incorporated by controlled mixing (5 min at 1350 rpm). Epojet^®^ is a commercial two-component epoxy adhesive for injection, which, after mixing, takes on the qualities of a low viscosity liquid. It is obtained by mixing its components in 4:1 ratio in weight, as specified in the technical data sheet, supplied by the manufacturer \[[@B34-materials-09-00461]\], and it is usable for 40 min at room temperature. Before being added to the geopolymeric mixture, Epojet^®^ was cured at room temperature for 10 min. The resin was added when it was still easily workable and long before its complete crosslinking and hardening (that takes place in about 5--7 h at 23 °C). It is worth noting that the addition of the unreacted components of the resins to the inorganic suspension produces phase segregation and, on the contrary, a late mixing of the two components (the cured organic resin and the geopolymer) results in a strongly reduced homogeneity of the final material. Different specimens containing up to 20% *w*/*w* of resin were prepared: in particular, G-FA-Ep10 and G-FA-Ep20 specimens were obtained by mixing Epojet^®^ epoxy resin (10% and 20% by weight, respectively) with fly ash-based geopolymer (G-FA); G-MK-Ep10 and G-MK-Ep20 were obtained by mixing Epojet^®^ epoxy resin (10% and 20% by weight, respectively) with metakaolin-based geopolymer (G-MK). All the composites started solidifying in few minutes \[[@B28-materials-09-00461]\]. The composition of the studied samples is reported in [Table 2](#materials-09-00461-t002){ref-type="table"}. 2.3. Curing Treatments {#sec2dot3-materials-09-00461} ---------------------- As soon as prepared, MK-based specimens were casted in cubic molds and cured at room temperature (=25 °C) in \>95% relative humidity conditions for seven days. The evaporation of water was prevented by sealing the top of the molds with a thin plastic film during the curing stage. The specimens were left for a further 21 days in air at room temperature before being characterized. FA-based specimens, as soon as prepared, were casted into cubic molds and cured at 60 °C for 48 h in \>95% relative humidity conditions (the evaporation of water was prevented by sealing the top of the molds with a thin plastic film during the curing stage) and further five days in \>95% relative humidity conditions at room temperature. Finally, the specimens were left for a further 21 days in air at room temperature before being characterized. 3. Results and Discussion {#sec3-materials-09-00461} ========================= 3.1. Characterization {#sec3dot1-materials-09-00461} --------------------- ### 3.1.1. X-ray Diffraction Characterization {#sec3dot1dot1-materials-09-00461} [Figure 3](#materials-09-00461-f003){ref-type="fig"} shows the XRD patterns of the used fly ash (FA), the cured neat FA-based geopolymer (G-FA), and the composite sample (G-FA-Ep20) containing 20 wt % of organic resin. The diffraction pattern of the fly ash is characterized by a wide and diffused hump in the interval range 15°--35° 2θ with a maximum at 2θ°--25°. Minor crystalline phases such as quartz (JCPDS 01-070-2517), mullite (JCPDS 01-076-2579) and hematite (JCPDS 00-013-0534) may also be identified. This amorphous halo is shifted towards slightly higher angular values (maximum at 2θ°--30°) in the G-FA sample, indicating the formation of an alkaline aluminosilicate hydrate gel (N-A-S-H) with a 3D amorphous structure \[[@B35-materials-09-00461]\]. The crystalline phases detected in the initial material (quartz, mullite and hematite) remain substantially unaltered. In this respect, it is known that, in the fly ash only, the amorphous aluminosilicate component is reactive in the geopolymerization reaction \[[@B36-materials-09-00461]\]. Finally, as far as the composite specimen is concerned, the X-ray diffraction pattern is very similar to that of the FA-based geopolymer, even if it is characterized by a more pronounced amorphous halo due to the presence of the organic resin. ### 3.1.2. Thermal Analysis {#sec3dot1dot2-materials-09-00461} Thermogravimetric analysis was performed on the neat G-FA geopolymer, on the organic resin, and on the G-FA-Ep20 composite ([Figure 4](#materials-09-00461-f004){ref-type="fig"}) after the curing. In the case of the neat G-FA geopolymer specimen, weight loss starts at =30 °C and is completed at 750°. The overall weight loss is 15% and can be attributed to the removal of water molecules absorbed (up to ≈100 °C) or differently linked (up to ≈200 °C, free water in the pores; at higher temperatures, structural water and bound water in the nanopores) to the silicate molecules \[[@B37-materials-09-00461],[@B38-materials-09-00461],[@B39-materials-09-00461]\]. Epojet^®^ resin shows a degradation mechanism involving two main steps. The resin is thermally stable up to about 250 °C. Above this temperature, a first degradation step that finishes at ≈480 °C is observed, resulting in a weight loss of 51%. The second degradation process is completed at about 650 °C and a combustion residual of about 5% remains. As far as the G-FA-Ep20 composite specimen, the weight loss shows a complex mechanism involving different steps: in particular, a first step, corresponding to a weight loss of ≈4%, is recorded up to ≈300 °C, while a second step, characterized by a complex path, is observed from ≈300 °C up to ≈700 °C and corresponds to a further weight loss equal to ≈22%. It was found \[[@B28-materials-09-00461]\] that the first degradation step is associated mainly with the loss of water of the geopolymeric phase while the second one corresponds to the degradation of the dispersed organic phase. The combustion residual at 800 °C is about 78%. Moreover, it is worth pointing out that, from the DTA of the curves reported in [Figure 4](#materials-09-00461-f004){ref-type="fig"}, the peak temperature of water loss for the composite (130 °C) is higher than that of the pure geopolymer (95 °C): probably, the polar groups of the resin interact with the water molecules, delaying their evaporation \[[@B28-materials-09-00461]\] (see column numbers 4 and 5 of [Table 3](#materials-09-00461-t003){ref-type="table"}). Degradation temperatures and weight losses for the studied systems are summarized in [Table 3](#materials-09-00461-t003){ref-type="table"}. ### 3.1.3. Microstructural Analysis of Fly Ash {#sec3dot1dot3-materials-09-00461} [Figure 5](#materials-09-00461-f005){ref-type="fig"} shows SEM micrographs of the used fly ash and their particle distribution. It is apparent that the fly ash consists mostly of glassy cenospheres. Such microstructure is in good agreement with that reported in the literature \[[@B2-materials-09-00461],[@B12-materials-09-00461],[@B13-materials-09-00461],[@B14-materials-09-00461],[@B15-materials-09-00461]\]. Moreover, the dimensions of the particles and their aggregates range between a few microns to ≈30 microns have been detected. In particular, the D~50~ is 30.1 μm ([Figure 5](#materials-09-00461-f005){ref-type="fig"}B). ### 3.1.4. Microstructural Analysis of Geopolymers and Geopolymer Based Composites {#sec3dot1dot4-materials-09-00461} The SEM micrographs of freshly obtained fracture surfaces of the MK-based geopolymer and FA-based geopolymer are reported in [Figure 6](#materials-09-00461-f006){ref-type="fig"}. MK based-geopolymer ([Figure 6](#materials-09-00461-f006){ref-type="fig"}A--C) is characterized by a compact morphology that only at very high magnification ([Figure 6](#materials-09-00461-f006){ref-type="fig"}C) reveals some unreacted kaolinite crystals and the presence of small spheroidal domains, probably reminiscent of the gelation process of the geopolymer. The morphology of FA-based geopolymer ([Figure 6](#materials-09-00461-f006){ref-type="fig"}A'--C') instead is dominated by the presence of the unreacted FA particles that are well dispersed in the geopolymer matrix. This disaggregated morphology is typical of fly ash-based geopolymers \[[@B1-materials-09-00461],[@B33-materials-09-00461],[@B40-materials-09-00461]\] Moreover, in this sample, the globular morphology of the geopolymeric matrix is rather evident already at 10,000 magnification ([Figure 6](#materials-09-00461-f006){ref-type="fig"}B') and, at variance with the MK-based geopolymer for which the particle dimensions are in the range 20--50 nm ([Figure 6](#materials-09-00461-f006){ref-type="fig"}C), in this last case, matrix globules have bigger average diameters, in the range 50--100 nm ([Figure 6](#materials-09-00461-f006){ref-type="fig"}C'). The limited reactivity of the fly ash particles causing the non-completeness of the geopolymerization reaction also influences the composition of the geopolymer matrix that, as determined by EDS analysis performed on a homogeneous moiety of the surface, is characterized by Si:Al and Na:Al ratios equal to 1.11 and 0.16, respectively. In [Figure 7](#materials-09-00461-f007){ref-type="fig"}, the SEM micrographs of the FA-based geopolymer composite containing the epoxy resin (10% by weight) are reported (A'--C') and compared to the images of the MK-based samples having the same resin content (A--C). As far as the MK-based sample, resin particles are homogeneously dispersed in the geopolymer matrix as well-defined microspheres with diameters in the range 1--10 μm. No segregation phenomena are observed ([Figure 7](#materials-09-00461-f007){ref-type="fig"}A,B) and a good adhesion between the organic phase (resin) and the inorganic one (geopolymer matrix) is apparent ([Figure 7](#materials-09-00461-f007){ref-type="fig"}C) \[[@B28-materials-09-00461]\]. In the case of the FA geopolymer composite, particles of organic resin strongly interacting with the matrix are still observed ([Figure 7](#materials-09-00461-f007){ref-type="fig"}C') but, in respect to the MK-based composite sample, an increase of the dimensions and a less homogeneous distribution of the particles is observed, probably due to the lower reactivity of FA in respect to MK, that allows the coalescence of the drops of resin during the geopolymerization reaction. In addition, several microspheres of unreacted fly ash particles are still present. These particles could act as a reinforcing agent of the matrix, improving the mechanical properties with respect to the neat geopolymer (see [Section 3.1.5](#sec3dot1dot5-materials-09-00461){ref-type="sec"}). In order to help distinguish between these two types of particles (*i.e.*, the organic resin particles containing carbon and the unreacted fly ash particles), [Figure 7](#materials-09-00461-f007){ref-type="fig"}D,D' shows the EDS maps of the elements carbon (in red) and silicon (in blue) for the G-MK-Ep10 and G-FA-Ep10 composites. It is worth noting that, in the case of G-MK-Ep10, it is evident that the organic particles are scratched when the samples are broken to prepare the SEM specimens ([Figure 7](#materials-09-00461-f007){ref-type="fig"}B,C), while the unreacted fly ash spheroidal particles preserve their very smooth surface ([Figure 7](#materials-09-00461-f007){ref-type="fig"}B'). These experimental pieces of evidence reflect the different interactions of the resin and FA particles with the geopolymeric matrix, as also shown in [Figure 8](#materials-09-00461-f008){ref-type="fig"}. In particular, the interaction of the resin particles with the inorganic matrix is very strong due to the network of hydrogen bonds between the phases ([Figure 7](#materials-09-00461-f007){ref-type="fig"}C' and [Figure 8](#materials-09-00461-f008){ref-type="fig"}C) \[[@B28-materials-09-00461]\]. On the contrary, there is a clear gap between unreacted FA particles and geopolymer matrix ([Figure 8](#materials-09-00461-f008){ref-type="fig"}D). [Figure 8](#materials-09-00461-f008){ref-type="fig"}B reports the EDS analysis taken along a line starting on a resin particle, passing through the geopolymer matrix and coming to an FA particle (the arrow is shown in [Figure 8](#materials-09-00461-f008){ref-type="fig"}A in yellow). By recording the variation of the chemical composition of carbon, silicon and aluminium along this line, it is apparent that the resin particle is characterized as expected by a high content of carbon and by the presence of minor Si and Al due to the adhesion of geopolymer matrix on its surface. Both geopolymeric matrix and fly ash particle instead show a negligible content of C while Si and Al are present. In particular, in agreement with the chemical composition (see Paragraph 2.1), FA particles are characterized by a similar content of Si and Al, while, in the geopolymer matrix, the silicon content is higher than that of aluminium. ### 3.1.5. Compressive Strength Test {#sec3dot1dot5-materials-09-00461} The compressive strengths of FA-based geopolymers (G-FA) and of the corresponding composites containing, respectively, 10% and 20% by weight of epoxy resin (G-FA-Ep10, G-FA-Ep20) are reported in [Figure 9](#materials-09-00461-f009){ref-type="fig"} and compared with the mechanical performances of the composite specimens obtained starting from metakaolin and containing the same resin content (G-MK, G-MK-Ep10, G-MK-Ep20). As expected on the basis of the limited reactivity of the weathered fly-ash with respect to MK-based geopolymers \[[@B14-materials-09-00461],[@B33-materials-09-00461]\], G-FA samples show a lower compressive strength than G-MK. In particular, by comparing the values of mechanical strength ([Figure 9](#materials-09-00461-f009){ref-type="fig"}), it is possible to observe that, in the case of neat geopolymer samples, G-MK mixture is characterized by a noticeable increase of ≈40% of compressive strength in respect to G-FA mixture. As far as composite samples, for both the matrices, the incorporation of the organic resin in the neat geopolymeric material significantly influences their mechanical properties, as the compressive strength increases with the organic content (compare G-FA-Ep10 and G-FA-Ep20 *vs.* G-FA and G-MK-Ep10 and G-MK-Ep20 *vs.* G-MK) and, in both cases, the best mechanical performance was obtained for the specimen containing 20% by weight of organic resin. As already demonstrated for the metakaolin composites, this evident improvement of mechanical properties is probably due to the presence of the organic resin that acts as reinforcement, thanks to a crack deviation mechanism and absorbing part of the load by plastic deformation \[[@B28-materials-09-00461]\]. Moreover, it is worth noting that the differences in the mechanical performances of FA-based composites and MK-based ones strongly decrease as the organic resin content increases. In particular, the difference in compressive strength of composite samples is ≈30% for the samples containing 10% by weight of resin (G-MK-Ep10 and G-FA-Ep10) and is ≈5% only in the case of the samples containing 20% by weight of resin (G-MK-Ep20 and G-FA-Ep20). These data allow for concluding that the presence of the resin is more effective when it is used with the fly ash-based geopolymer samples, rather than the metakaolin ones. This observation supports the idea of a valid use of fly ash in place of more expensive raw materials, such as metakaolin, in particular for those applications for which it is important to save materials and limit the costs. Finally, preliminary data (not reported) indicate that, if compared to neat G-FA, G-FA composite mixtures show an improved adhesion to common construction supports, minimizing pouring phenomena and avoiding aggregate segregation. 4. Conclusions {#sec4-materials-09-00461} ============== Through a co-reticulation reaction of commercial epoxy-based organic resins and a fly ash-based geopolymer, new organic--inorganic composite materials with improved mechanical properties were prepared. This strategy allows the organic phase to chemically interact with the geopolymeric mixture during the geopolymerization process through the formation of a wide network of hydrogen bonding due to the presence of several hydroxyl groups. In such a way, a high compatibility between the organic and inorganic phases, even at appreciable concentration of resin (20% *w*/*w*), was realized up to micrometric level and a good and homogeneous dispersion (without the formation of agglomerates) of the organic particles was achieved. These new materials show enhanced mechanical properties in respect to the neat geopolymer. In particular, compressive strength of the new composite material is significantly higher than that of the neat fly-ash based geopolymer, being comparable or even superior to that obtained starting from the more expensive metakaolin. It is worth pointing out that, despite the high concentration of organic resin, similarly to the analogous composites containing melamine based resins \[[@B30-materials-09-00461]\], preliminary data show that these new materials are not flammable and do not produce smoke in significant amounts. Considering the increasing demand for materials with low environmental impact in today's construction and housing industry, this paper tries to add new results in the field of sustainable building materials with reduced environmental footprint. In fact, the following conclusions can be drawn: the compressive strength of the composite materials is significantly better than that of the neat geopolymer;fly ash-based composite materials represent a valid alternative in place of more expensive raw materials, such as metakaolin, in particular for those applications for which it is important to save materials and limit the costs;the procedure is inexpensive and uses easily available reagents. Finally, having successfully replaced the metakaolin with fly ash, we can suggest that the novel composites may have all the conditions to be an Environmentally Friendly Material. In order to confirm this hypothesis, a complete LCA (Life Cycle Assessment) study is in progress. The authors thank Neuvendis S.p.A. (San Vittore Olona, Milan, Italy) for the metakaolin supply and Prochin Italia S.r.l. (Naples, Italy) for the silicate solution supply. Thanks for the technical assistance are due to Giovanni Morieri, Ornella D'Andria and Luciana Cimino. G.R. and O.T. conceived and designed the experiments; G.R., L.R., V.R. and F.C. performed the experiments; G.R., C.F., F.C. and R.C. analyzed and discussed the data; G.R., L.R., and O.T. and C.F. wrote the paper. The authors declare no conflict of interest. ![Schematic representation of the mechanism of interaction between geopolymer (on the left and on the right, in the hypothesis of Na/Al = 1:1) and *N*-carboxymethyl chitosan macromolecules (in the middle). Dashed lines represent hydrogen bonds \[[@B18-materials-09-00461]\].](materials-09-00461-g001){#materials-09-00461-f001} ![Schematic representation of some possible interactions between the epoxy resin Epojet^®^ \[[@B31-materials-09-00461]\] (in the middle) and the geopolymeric matrix (on the left and the right, in the hypothesis of Na/Al = 1:1). Dashed lines represent hydrogen bonds.](materials-09-00461-g002){#materials-09-00461-f002} {#materials-09-00461-f003} {#materials-09-00461-f004} ###### (**A**) scanning electron microscope (SEM) micrographs and (**B**) particle size distribution (volume density *vs.* size) of the used fly ash.   ###### Scanning electron microscope (SEM) micrographs of MK-based geopolymer (**A**--**C**) and FA-based geopolymer (**A'**--**C'**).   {#materials-09-00461-f007} {ref-type="fig"}B' that shows, following the direction of the yellow arrow: (i) the resin particle; (ii) the matrix and (iii) the not-reacted ash particle; (**B**) % distribution of C (**red**), Si (**blue**), Al (**green**) along the arrow obtained by EDS analyses; (**C**,**D**) 40,000× magnification image of the interface zone between the geopolymer matrix and a resin particle, and between the matrix and an ash particle, respectively.](materials-09-00461-g008){#materials-09-00461-f008} {#materials-09-00461-f009} materials-09-00461-t001_Table 1 ###### Chemical composition (weight%) of weathered fly ash used in this paper after thermal treatment at 950 °C, metakaolin and the sodium silicate solution. ------------------------------ -------- ------- ----------- -------- ------ ------ -------- **Fly Ash** Al~2~O~3~ SiO~2~ K~2~O Fe~2~O~3~ Na~2~O MgO CaO others 28.12 53.75 1.89 6.99 0.87 1.59 4.32 2.47 **Metakaolin** Al~2~O~3~ SiO~2~ K~2~O Fe~2~O~3~ TiO~2~ MgO CaO others 41.90 52.90 0.77 1.60 1.80 0.19 0.17 0.67 **Sodium Silicate Solution** SiO~2~ Na~2~O H~2~O 27.40 8.15 64.45 ------------------------------ -------- ------- ----------- -------- ------ ------ -------- materials-09-00461-t002_Table 2 ###### Composition (wt %) of the samples used in this study. Mix ID MK FA SS NaOH NaOH ~soln~ Resin ----------- ------ ------ ------ ------ ------------- ------- G-MK 41.6 \- 50.0 8.4 \- \- G-MK-Ep10 37.4 \- 45.0 7.6 \- 10 G-MK-Ep20 33.3 \- 40.0 6.7 \- 20 G-FA \- 60.2 19.9 \- 19.9 \- G-FA-Ep10 54.2 17.9 \- 17.9 10 G-FA-Ep20 48.2 15.9 \- 15.9 20 MK = metakaolin; FA = fly ash; SS = sodium silicate solution; NaOH ~soln~ = aqueous sodium hydroxide solution 10 M; Resin = Epojet^®^ Mapei S.p.A. \[[@B34-materials-09-00461]\]. materials-09-00461-t003_Table 3 ###### Thermal properties of the neat geopolymer (G-FA), pure epoxy resins (Epojet), and the composite specimen (G-FA-Ep20). Mix ID Weight Loss Starting Temperature (°C) Weight Loss Ending Temperature (°C) Weight Loss at 200 °C (wt %) Weight Loss at 400°C (wt %) Residual at 800°C (wt %) ----------- --------------------------------------- ------------------------------------- ------------------------------ ----------------------------- -------------------------- G-FA 30 750 7.2 8.7 85 Epojet 250 650 2.1 39.4 5 G-FA-Ep20 30 700 2.6 9.2 78

This article has been accepted for publication and undergone full peer review but has not been through the copyediting, typesetting, pagination and proofreading process, which may lead to differences between this version and the Version of Record. Please cite this article as doi:10.1111/jth.14988

Pulmonary embolism (PE) is a serious cardiovascular and pulmonary condition with mortality ranging from 2% to 7% worldwide, even after thromboprophylaxis.\[[@ref1]\] It is the third most common cause of cardiovascular death, after myocardial infarction and cerebrovascular stroke.\[[@ref2]\] Regardless of the advancement in thromboprophylaxis, diagnosis, and treatment, PE remains an important determinant of in-hospital complications and outcome.\[[@ref3]\] In some cases with acute massive PE, severe hemodynamic compromise and systemic hypotension causes sudden death secondary to circulatory collapse and heart failure.\[[@ref4][@ref5]\] This occurs due to increased pulmonary vascular resistance with high pulmonary arterial and right ventricle (RV) pressure which overwhelms the RV failure.\[[@ref6]\] Risk stratification of patients with acute PE is crucial for early diagnosis and appropriate selection of the treatment.\[[@ref7]\] Echocardiography is a useful diagnostic and prognostic tool for RV dysfunction which is considered as an independent predictor of short-term outcome of PE.\[[@ref8]\] Moreover, computed tomography pulmonary angiography (CTPA) has become the modality of choice for the diagnosis of PE as it offers a multiplanar view for the assessment of pulmonary vessels to the subsegmental levels.\[[@ref9][@ref10]\] RV dysfunction and extent of pulmonary artery (PA) obstruction at computed tomography (CT) are useful prognostic parameters.\[[@ref11]\] Patients with RV dysfunction and PE have a high mortality rate, and hence, RV dysfunction is considered as a poor prognostic marker.\[[@ref8]\] Furthermore, the RV/left ventricular (LV) diameter ratio and pulmonary vascular obstruction score are important parameters to evaluate RV dilatation which is associated with significant mortality in PE.\[[@ref12][@ref13]\] In addition, the presence of RV dysfunction indicates a high risk of recurrent or fatal PE. To date, four scoring systems have been proposed by different investigators to determine the presence, location, and degree of obstruction of arterial clots in patients with PE.\[[@ref14]\] Among them, Qanadli PE index\[[@ref13]\] is easy to calculate, identify complete or partial obstruction due to proximal clot, and thus provides information about the residual pulmonary perfusion. The clot burden score has significant prognostic and therapeutic potential as it facilitates direct visualization of the clot within the PA with a standard reproducible score to quantitate the improvement after therapeutic intervention.\[[@ref13][@ref14]\] There are few clinical studies from the Middle East to evaluate the role of CTPA for diagnosis and prognosis of PE.\[[@ref14][@ref15][@ref16]\] The aim of this study is to assess the diagnostic implications of different CT parameters, and to correlate the clot burden, cardiovascular measurements, and clinical presentation of patients with acute PE. Methods {#sec1-1} ======= We conducted a retrospective observational study of fifty consecutive patients presented with a clinically suspected PE and underwent CTPA. Patients were identified from the radiology database, and the relevant clinical information was retrieved from the medical records at Hamad General Hospital between May 2011 and February 2015. Clinically suspected patients with PE who were not confirmed by CT imaging were excluded from the study. Images from 16- or 64-slice multidetector CT angiography were reviewed in 45 patients, after excluding patients in whom images were not available on the database or in whom poor image quality precluded accurate assessment. Clinical presentation, preexisting comorbidities, radiologic imaging, detailed notes for in-hospital course, and discharge summaries were reviewed for each patient. Data collection included patient demographics (age, sex, and nationality), clinical presentation and predisposing factors, laboratory investigations (D-dimer testing, coagulation profile), Doppler ultrasound, finding of echocardiography (ejection fraction \[%\], RV wall hypokinesis and pulmonary arteries dilation) and CTPA, management, hospital length of stay, complications, and outcome. Patients were followed up for routine clinical care as per the standard practice. The plasma D-dimer concentrations were measured at the clinical laboratory, Hamad General Hospital, using a well-validated commercial assay used for routine D-dimer testing. The normal range of plasma D-dimer concentration in our hospital laboratory is 0--0.55 mg/L. Coagulation profile includes protein S deficiency, protein C deficiency, hyperhomocysteine, antithrombin III deficiency, and antiphospholipid syndrome. PE was defined radiologically as the presence of an endoluminal central filling defect partially or completely occluding the pulmonary arteries.\[[@ref17]\] The RV dysfunction refers to the presence of RV hypokinesis, explained by a qualitative evaluation of the RV wall motion.\[[@ref18]\] Computed tomography pulmonary angiography {#sec2-1} ----------------------------------------- The occurrence of PE was confirmed by CTPA examination. In brief, all patients were examined using Siemens SOMATOM Sensation (Siemens AG, Munich, Germany) 16- or 64-slice machine. For all examinations, kV was set as 100, collimation of 0.6, rotation time of 0.5 s, slice thickness of 5 mm, and pitch of 1.0. Nonionic intravenous contrast (100 ml of Omnipaque 350 mg/ml) was injected at a rate of 4.0--5.0 ml/s. Pulmonary arterial phase bolus tracking through main pulmonary trunk was carried out followed by aortic phase immediately following pulmonary arterial phase. (In pregnant patients, only pulmonary arterial phase was performed.) Moreover, three-dimensional reconstruction was performed, if indicated. Imaging analysis {#sec2-2} ---------------- ### Clot burden {#sec3-1} Qanadli *et al*.\'s\[[@ref13]\] scoring system was used to quantify the vascular obstruction index using CTPA which is based on the percentage of vascular obstruction of the pulmonary arterial tree developed secondary to PE. In brief, the Qanadli index determines the number of blocked segmental arterial branches and then to be adjusted by a factor of one for partial blockage or a factor of two for completely obstructive PE. The maximum attainable score in this system is 40 (pulmonary trunk completely obstructed by thrombus), which represents complete obstruction index (100%). In this study, we used the same scoring system based on the site and degree of occlusion of pulmonary arteries. Computed tomography signs of right heart dysfunction {#sec2-3} ---------------------------------------------------- CT findings considered for the functional cardiovascular measurements include the ratio of RV to LV diameter (RV/LV ratio), ratio of main PA diameter to ascending aorta (AO) diameter (PA/AO ratio), superior vena cava (SVC) and azygos vein diameter, bowing of interventricular septum, clot burden (Qanadli score \[QS\]). The other observations such as infarction, pulmonary effusion (right, left, and bilateral), pneumothorax and rib fracture, and pulmonary metastasis were also recorded. RV and LV dimensions were identified based on the maximal distance between the ventricular endocardium and the interventricular septum, perpendicular to the long axis. RV dysfunction was diagnosed if the RV to LV diameter ratio was \>1.2.\[[@ref14]\] Inferior vena cava reflux to determine signs of right ventricular dysfunction {#sec2-4} ----------------------------------------------------------------------------- Aviram *et al*.\[[@ref19]\] classified reflux of contrast medium into inferior vena cava (IVC) or hepatic veins based on severity into six different categories such as (a) no reflux into IVC, (b) trace of reflux into IVC only, (c) reflux into IVC but not hepatic veins, (d) reflux into IVC with opacification of proximal hepatic veins, (e) reflux into IVC with opacification of hepatic veins down to the mid-portion of the liver, and (f) reflux into IVC with opacification of distal hepatic veins. The Institutional Review Board (IRB\# 15139/15) of the Hamad Medical Corporation has approved and granted exempt status for this retrospective study. Statistical analysis {#sec2-5} -------------------- Data were reported as percentage, mean (±standard deviation), median, and range, where applicable. Baseline characteristics, echocardiography, and right ventricular dysfunction (RVD) parameters were analyzed according to clot burden (QS ≤18 vs. QS \>18) and outcome (survivors vs. nonsurvivors). Student\'s *t*-test was used to compare continuous variables, and Pearson\'s Chi-square test was used for categorical variables. Pearson\'s correlation coefficient (r) has been calculated between QS and cardiovascular measurements. Receiver operating characteristic (ROC) curve was constructed to establish the best cutoff for QS and RV dysfunction. To look for predictors of mortality, multivariate regression analysis was performed after adjustment for the significant variables on univariate analysis, and data were expressed as odd ratio and 95% confidence interval (CI). Two-tailed *P* \< 0.05 was considered statistically significant. Data analysis was carried out using the Statistical Package for Social Sciences version 18 (SPSS Inc., Chicago, IL, USA). Results {#sec1-2} ======= A total of 45 consecutive patients with radiologically confirmed diagnosis of PE were eligible for enrollment in the present study. [Table 1](#T1){ref-type="table"} shows the demographics, comorbidities, and outcome of PE patients. The mean age of the patients was 49 ± 11.4 years, and 29 (64.4%) were males. Thirty (67%) patients developed PE during hospitalization whereas 15 (33%) patients presented to the emergency department with PE. Deep vein thrombosis (DVT) (60.0%), hypertension (37.8%), obesity (average BMI 32.8 ± 8.6; 28.9%), diabetes mellitus (26.7%), and abnormal coagulation (24.4%) were the frequent preexisting comorbidities. Recurrent PE was observed in 6 (14.6%) patients. Overall mortality was 18% during the hospital course. ###### Demographics, clinical presentation, comorbidities, and outcome of patients with pulmonary embolism (*n*=45)  [Table 2](#T2){ref-type="table"} shows the coagulation profile and echocardiography findings. Thirty-three patients (73.3%) underwent echocardiography, of which 13 (39.4%) cases had RV dysfunction in terms of RV dilatation and hypokinesia. Pulmonary arteries dilation was found in 8 (24.2%) patients. Protein S deficiency was found in one-third (30%) of the patients, and the median D-dimer value was 5.7 (1.2--72.2) mg/L FEU. ###### Echocardiography and coagulation profile  Parameters of the right heart dysfunction as identified by the CTPA are shown in [Table 3](#T3){ref-type="table"}. The mean clot burden (QS) was 19.5 ± 11.3 points and 53% of patients had QS \>18 points. One-third (33.3%) of the patients had PA/AO ratio \>1.0. The number of patients with RV dysfunction (RV/LV ratio \>1.2) was 27 (60.0%). IVC reflux was found to be mild-to-moderate (Grades I--III) in 73.3% of patients and severe (Grades IV--VI) in 26.7% of patients. Interventricular septal abnormality was identified in 29 (64.4%) patients. ###### Computed tomography pulmonary angiography findings of right ventricular dysfunction  [Table 4](#T4){ref-type="table"} demonstrates univariate association between cardiovascular measurements and mortality. The RV/LV ratio, PA/AO ratio, bowing of interventricular septum, IVC reflux, SVC, and azygos vein diameters were comparable in survivor and nonsurvivor groups. In our study, cancer patients were more likely to die due to paraneoplastic embolism than those without malignancy (71.4% vs. 16.7%; *P* = 0.002). ###### Univariate association between cardiovascular measurements and mortality  Clinicoradiological characteristics of PE patients according to clot burden (QS ≤18 vs. \>18) are shown in [Table 5](#T5){ref-type="table"}. The two groups were comparable for age, gender, and preexisting comorbidities/risk factors and echocardiography findings. However, obesity (52.4% vs. 12.5%; *P* = 0.01), hypertension (54.4% vs. 23.8%; *P* = 0.03), and median D-dimer levels (7.8 vs. 3.4; *P* = 0.03) were significantly higher in patients with QS \>18. Among RV dysfunction parameters, only RV/LV ratio (1.61 ± 0.48 vs. 1.04 ± 0.29; *P* = 0.001) and bowing of interventricular septum (91.7% vs. 33.3%; *P* = 0.001) were associated with higher QS. The mortality rate was comparable in patients with higher clot scores (QS \>18) (16.7% vs. 19.0%; *P* = 0.83) as compared to low scores (QS ≤18). A significant positive correlation was found between RV short axis (*r* = 0.499, *P* = 0.001), RV/LV ratio (*r* = 0.592, *P* = 0.001), and PA diameter (*r* = 0.301, *P* = 0.04) with the PA clot burden. Moreover, LV short axis showed a significant negative correlation with clot burden (*r* = −0.515, *P* = 0.001) \[[Table 6](#T6){ref-type="table"}\]. Figures [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"} demonstrate CTPA findings of two of our PE patients. ###### Clinicoradiological characteristics of pulmonary embolism patients according to clot burden (Qanadli score)  ###### Correlation between Qanadli score and cardiovascular measurements  {#F1} {#F2} ROC curve showed an area under the curve (AUC) for the prediction of RV dysfunction based on PA clot burden to be 0.822 (95% CI: 0.698--0.947, *P* = 0.001). In our study, the optimal cutoff value of QS for the identification of RV dysfunction was considered based on highest pair of sensitivity (77.8%) and specificity (72.2%). In addition, the AUC was 0.822; 95% CI: 0.698--0.947, *P* = 0.001, which is considered a good diagnostic value. Hence, the clot burden cutoff level of 17.5 points was found to predict RV dysfunction which is also consistent with Qanadli *et al*.\[[@ref13]\] scoring system. Multivariate regression analysis showed that after adjustment for age and cancer, neither RV/LV ratio (odds ratio \[OR\] 1.6; 95% CI: 0.13--20.15, *P* = 0.71) nor clot scores (OR 0.99; 95% CI: 0.87--1.14, *P* = 0.92) were independent predictor for mortality. Discussion {#sec1-3} ========== PE is a serious medical condition that could ultimately lead to death within few hours due to RV failure and circulatory collapse. Therefore, RV burden should be detected rapidly to identify patients who could benefit from early aggressive therapy. The noninvasive CTPA can safely and quickly identify the presence and extent of PE. The present study highlights the diagnostic implications of CTPA parameters and assesses the correlation between clinicoradiological characteristics of 45 patients with PE. In our series, echocardiographic evaluation revealed RV wall hypokinesis and PA dilation in 39.4% and 24.2% of cases, respectively. Earlier studies have suggested that echocardiography might be effective in predicting RV dysfunction and clinical outcome.\[[@ref20]\] However, due to its lower sensitivity to diagnose PE, echocardiography is recommended mainly for hemodynamically unstable patients.\[[@ref15]\] At present, CTPA is considered as the modality of choice for accurate diagnosis of PE.\[[@ref6]\] In comparison to echocardiography, CTPA enables a thorough assessment of embolization in the pulmonary arteries and detects associated underlying pulmonary disorders and other causes of acute chest pain.\[[@ref21]\] Moreover, it facilitates risk stratification of patients based on the degree of vascular obstruction (i.e., clot burden) which could be used as a marker for appropriate selection of treatment.\[[@ref22]\] Apart from the size of embolus, the clinical outcome of patients with PE also depends on various cardiopulmonary measurements (RV, LV, RV/LV ratio, PA, AO diameter, and PA/AO ratio) identified by CTPA. In our study, patients with PE presented at a young age, more likely to be males, and two-thirds of them were identified during hospitalization. Our findings are consistent with an earlier study from Saudi Arabia with a similar age at presentation, male predominance, and in-hospital diagnosis of PE.\[[@ref15]\] Prior hospitalization could not represent a real limit in our study as only one patient had prolonged immobilization (bedridden) which could associate with venous thromboembolism (VTE). We do not believe that our findings are influenced by other acute diseases as we anticipated all major risk factors in our study. A previous study from our center reported mean age of 50 years in patients with confirmed diagnosis of thromboembolism.\[[@ref23]\] In that study, the proportion of PE was 12.2% among 662 thromboembolism cases who were screened over 5 years. However, other investigators reported the diagnosis of PE with an advanced mean age of 63 years.\[[@ref24]\] We observed a higher rate of protein S deficiency (30%) in patients who developed PE at relatively younger age. Similarly, an earlier study showed that low free protein S carried a high risk of developing or recurring of VTE in young patients with a mean age 39 years.\[[@ref25]\] In addition, preexisting comorbidities were frequent in our study cohort. Al Otair *et al*.\[[@ref15]\] reported recent surgery, obesity, immobilization, and recurrent DVT to be the frequent risk factors associated with PE. PA obstructive index is important as it facilitates direct visualization of the clot and allows accurate diagnosis of PE.\[[@ref26]\] There are several scoring systems reported to evaluate the pulmonary vascular tree and assessment of clot burden score.\[[@ref16]\] The present study quantified clot burden scoring according to QS as it is more objective, easy to calculate, distinguish between partial and complete obstruction, and has less interobserver variability.\[[@ref13][@ref19]\] Serial imaging studies could be used to assess the degree of vascular obstruction which helps a clinician determine real-time risk stratification and monitoring of treatment in an objective way. Previous studies have identified the clot burden score to be a predictor of RV dysfunction and poor outcomes. Qanadli *et al*.\[[@ref13]\] found that an obstruction score \>40% correlates well with RV dilatation. It has been observed that PE patients with 40% or higher obstruction index had 11-fold increased risk of mortality.\[[@ref11]\] Another study reported significant correlation between RV/LV ratio, IVC reflux, QS, and the short-term PE-related mortality.\[[@ref14]\] Bazeed *et al*.\[[@ref26]\] observed a significant difference in clot burden scoring among PE patients who died than those who survived. On the other hand, other investigators reported no significant association of obstruction score with PE-related mortality.\[[@ref14]\] Consistent with these reports, our findings revealed no correlation of clot burden with in-hospital mortality. This variability could be explained by the fact that pulmonary obstruction could be related to other factors such as mechanical obstruction, vasoactive agents, reflex vasoconstriction, and systemic hypoxemia occurred during PE.\[[@ref14]\] However, obstructive index can be accurately used as an indicator of the severity of PE and for treatment response.\[[@ref27]\] The assessment of cardiorespiratory status secondary to an acute PE should not only rely on the degree of pulmonary obstruction but also consider the signs of RV strain (RV/LV diameter ratio \>1) as a potential indicator for the RV dysfunctions, which occurs due to sudden increase in the afterload caused by mechanical obstruction and vasoconstriction.\[[@ref28]\] It has been suggested that the severity of PE could be assessed more accurately by RV failure rather than the degree of obstruction.\[[@ref29]\] Earlier studies have suggested threshold values for RV/LV ratio ranged from 1 to 1.5 as indicative of RV strain\[[@ref12][@ref30]\] and can be used as a predictor of PE mortality.\[[@ref14]\] In our series, we used a threshold of RV/LV ratio of 1.2 for the diagnosis of RV dysfunction. However, we did not observe a significant correlation between RV dysfunction and PE-related mortality. Our findings are in accordance with previous studies which suggest that RV/LV ratio and clot burden are not associated with PE-related mortality.\[[@ref31][@ref32]\] The reason for these controversies could be partially explained by the variability in measurements of RV/LV ratio. In our study, obesity, hypertension, and D-dimer were significantly associated with severity of obstruction (QS \>18). Similar to our findings, Attia *et al*.\[[@ref16]\] reported a significant association between obesity and higher mean obstructive index in patients with acute PE. In our study, patients with right heart dilatation (RV/LV ratio \>1.2) were more likely to have larger clot volume as compared to those without RV dilatation. Similarly, the larger clot volume (QS \>18) was observed in patients with interventricular septal defect in comparison to patients with normal interventricular septal morphology. The higher clot volume is likely to obstruct the pulmonary circulation, causing RV pressure overload and dilatation. Therefore, estimation of the degree of vascular obstruction (clot burden) would help in the PE risk stratification in clinical practice and also enable treatment monitoring. Moreover, we observed a significant positive correlation between RV/LV ratio and PA diameter with clot burden (QS). Our findings coincide with earlier studies which also observed a good correlation between RV dysfunction parameters (RV/LV ratio and PA diameter) and the severity of obstruction.\[[@ref16][@ref22]\] In the present study, ROC curve analysis determined the cutoff point for the obstructive index (QS) to be 17.5 points (clot burden of 43.7%) which could predict RV dysfunction with reasonable sensitivity (77.8%) and specificity (72.2%). In line with findings of our study, other investigators observed the best cutoff point for clot burden which ranges between 40% and 49% with high sensitivity and specificity, and values ≥40% could potentially identify more than 90% of patients with RV dysfunction.\[[@ref16]\] On the other hand, PE patients with a CT burden of \<40% are unlikely to develop acute RV dysfunction.\[[@ref13][@ref22][@ref33]\] Therefore, the findings of the present study highlighted the clinicoradiological implications of CTPA for the accurate diagnosis and risk stratification of patients with PE. We observed no significant difference between mortality and clot burden which is consistent with an earlier study by Rodrigues *et al*.\[[@ref22]\] Another important factor was reflux of the dye in the IVC which is an indirect sign of tricuspid valve insufficiency.\[[@ref29]\] Kang *et al*.\[[@ref34]\] found that contrast reflux is a bad prognostic factor. Our study did not observe any prognostic role of IVC reflux, and similar results were shown by Collomb *et al*.\[[@ref35]\] The interventricular septum may shift toward the LV due to increased right heart pressure with severe pulmonary obstruction.\[[@ref36]\] The prognostic significance of interventricular septum abnormality is debatable. Araoz *et al*.\[[@ref31]\] reported that patients with ventricular septal defects are more likely to be admitted to the ICU. On the other hand, Van der Meer *et al*.\[[@ref11]\] did not find any relation to mortality from acute PE. It has been revealed that the mean diameter of the azygos vein and SVC was higher in nonsurvivors.\[[@ref21]\] A recent study by Furlan *et al*.\[[@ref17]\] showed no significant association between the diameters of the SVC, the azygos vein diameters, and short-term mortality which corroborates our findings. Recently, Meinel *et al*.\[[@ref37]\] concluded in a meta-analysis that increased RV/LV diameter ratio measured on transverse CT images was the strongest risk with a 2.5-fold higher risk for all-cause mortality and 5-fold higher risk for PE-related mortality. Furthermore, the degree of thrombus load and central thrombus location were not predictive for all-cause mortality although both were associated with adverse clinical outcome. In the present study, cancer patients were more likely to die due to paraneoplastic embolism than those without malignancy. Consistent with our findings, a large prospective study based on RIETE registry revealed a higher 3-month mortality rate in the patients with cancer-related VTE as compared to those without cancer (4%).\[[@ref38]\] After adjustment for age and history of cancer, the present study showed that RV/LV ratio and clot scores were not independent predictors for mortality. One of the limitations of our study is the retrospective design and underpowered due to small number of cases which needs further validation based on larger sample size. We did not consider the preexisting structural heart defects or pulmonary disease which could affect the cardiopulmonary measurements on CT angiography. Finally, the clot burden score does not consider clots situated within the small pulmonary arteries at the periphery as well as unresolved previous episodes of PE in cases of recurrence. Although multivariate logistic regression analysis could not identify the independent predictors of outcomes, this could be limited by the sample size which is insufficient to draw findings of statistical significance to be used for the risk stratification. The total number of PE cases in this study underestimates and does not reflect the real number in our institution as we retrospectively analyzed only cases with detailed CTPA data. Conclusion {#sec1-4} ========== Clot burden \>18 is associated with RV dysfunction in patients with acute PE. Echocardiography and RVD parameters showed no correlation with in-hospital deaths. CTPA has clinicoradiological implications for risk stratification in PE patients. As the sample size is small, our findings warrant further larger prospective studies. Financial support and sponsorship {#sec2-6} --------------------------------- Nil. Conflicts of interest {#sec2-7} --------------------- There are no conflicts of interest. We would like to thank all the staff of Radiology Department at Hamad General Hospital, for their contribution and support. This study was approved by the Medical Research Center at HMC, Doha, Qatar (IRB\# 15139/15).

A salivary mucocele is defined as an accumulation of leaking salivary secretion in single or multiloculated cavities in the connective tissue of the mouth or neck contiguous to a salivary gland or duct[@r1]. In general, it can be observed in the form of a cyst lined with granulation tissue and the absence of any epithelial lining. A salivary mucocele can occur in several kinds of animals including dogs of any breed[@r1]. The cause is generally not identified; however, blunt trauma, foreign body and sialolith have been suspected as major causes of salivary mucocele. It arises most commonly from the sublingual salivary gland, either from individual units of the polystomatic portion or from the duct of the monostomatic portion in dogs[@r1]. Saliva usually leaks from the torn portion, and accumulates in the adjacent tissue. Consequently, the accumulated saliva induces an inflammatory response. A wall of granulation tissue is gradually developed in response to the inflammation. The diagnosis is usually made based on the history, physical examination by palpation and cytological examination after aspiration of the cyst and can be confirmed by histopathological examination[@r4]. In a toxicological study, we encountered a laboratory beagle with a salivary mucocele characterized by lining epithelial-like cells. It was considered to be a spontaneous lesion because there were no similar lesions in the other animals. Unexpectedly, there are only a few reports focused on the histological features of salivary mucoceles in animals. Therefore, we describe the histopathological characteristics of this lesion. {#fig_001} The animal was a male beagle (Kitayama Labes Co., Ltd., Yamaguchi, Japan) treated with a compound in a 2-week repeated-dose toxicity study with a 4-week recovery period. The experiment procedures were approved by the Animal Ethics Committee of Takeda Pharmaceutical Company Limited. No abnormality was observed in the appearance of the neck and mandible or in the clinical pathology. The animal was sacrificed at 14 months old by exsanguination under thiopental sodium anesthesia and subjected to a complete necropsy. At necropsy, a pale yellowish cyst, 10 × 40 × 12 mm in size, containing frothy mucus was observed under the mandibular skin. The proximal edge of the cyst was not clearly identified because it was buried between the digastric and mylohyoid muscles. The other organs had no abnormal gross findings. The cyst was fixed in 10% neutral buffered formalin solution. Routine paraffin embedded sections were stained with hematoxylin and eosin (HE). Alcian blue (pH 2.5)-periodic acid-Schiff (AB-PAS) double staining and immunohistochemical staining for human cytokeratin (hCK), which identifies cytokeratin 5, 6, 8, 17 and 19, and macrophage scavenger receptor A (MSR-A) were performed. The antibodies used for this study were as follows: anti-human cytokeratin mouse monoclonal antibody (diluted 1:500, clone: MNF116, Dako Cytomation, Tokyo, Japan) and anti-human macrophage scavenger receptor A mouse monoclonal antibody (diluted 1:100, clone: SRA-E5, Transgenic Inc., Kobe, Hyogo, Japan). For electron microscopy, small pieces of the cyst wall, originally fixed in 10% neutral buffered formalin solution, were postfixed in osmium tetroxide, dehydrated and embedded in epoxy resin. Ultrathin sections were cut and stained with lead citrate and uranyl acetate and examined using a transmission electron microscope. {#fig_002} Light microscopically, the cyst was encapsulated by dense connective tissue and numerous villous projections arose from the internal surface of the cyst ([Fig. 1a](#fig_001){ref-type="fig"}). Villous projections had fibrovascular stalks with lymphoplasmacytic cells and pigmented macrophages and were lined by stratified epithelial-like cells ([Fig. 1b](#fig_001){ref-type="fig"}). The epithelial-like lining cells had round to oval nuclei and slightly eosinophilic and foamy cytoplasm. The lumen of the cyst was filled with eosinophilic amorphous material with a few desquamated cells ([Fig. 1b](#fig_001){ref-type="fig"}). Some multinucleated giant cells were also observed on the surface of the lining cells ([Fig. 1c](#fig_001){ref-type="fig"}). The sublingual gland ([Fig. 1d](#fig_001){ref-type="fig"}) and a ruptured sublingual interlobar duct ([Fig. 1e](#fig_001){ref-type="fig"}) connected to the cyst were observed in the surrounding connective tissue. The amorphous material within the cyst showed a positive reaction for AB-PAS with a bluish or reddish violet coloration ([Fig. 1f](#fig_001){ref-type="fig"}). The epithelial-like cells also had AB-PAS-positive reddish violet colored granules in their cytoplasm ([Fig. 1g](#fig_001){ref-type="fig"}). As expected, secretory granules of the sublingual gland and saliva in the ducts showed a similar positive reaction for AB-PAS ([Fig. 1h](#fig_001){ref-type="fig"}). {#fig_003} Immunohistochemically, the epithelial-like cells were positive for MSR-A, a macrophage marker in dogs[@r6] ([Fig. 2a-b](#fig_002){ref-type="fig"}), which strongly suggests that these cells were macrophages. On the other hand, the multinucleated giant cells were negative for MSR-A ([Fig. 2b](#fig_002){ref-type="fig"}: Inset). As described in previous papers, the expression of MSR-A increases during differentiation and maturation of macrophages[@r7]. Therefore, expression of MSR-A might be changed during the process of multinucleated giant cell transformation. Predictably, the epithelial-like cells were negative for hCK ([Fig. 2c](#fig_002){ref-type="fig"}), while the epithelial cells of the interlobar duct were strongly positive for hCK ([Fig. 2d](#fig_002){ref-type="fig"}). Electron microscopically, the epithelial-like cells had numerous vacuoles containing electron-lucent material, which was presumed to be lysosomal in origin ([Fig. 3a](#fig_003){ref-type="fig"}). This finding probably corresponded to the reddish violet colored granules stained by AB-PAS. In addition, the epithelial-like cells had pseudopods on their cell surfaces interdigitating with those on the adjacent cells ([Fig. 3b](#fig_003){ref-type="fig"}). Structures of the basal lamina or the cell junctions were not observed. Based on the results described above, we concluded that the epithelial-like cells lining the cyst were macrophages because they had typical macrophage characteristics, such as positive staining for MSR-A, and contained many lysosomes without a basement membrane. As far as we know, cystic structures characterized by epithelial-like macrophages have not been reported except for salivary mucocele of the soft palate with fibrocytes giving the appearance of an epithelial lining[@r8]. In addition, the macro-phages had no nuclear atypia, few mitoses and no monotonous proliferation. Therefore, this lesion was not considered to be a neoplastic lesion but an inflammatory response to sublingual mucoid material. In animals, cystic structures in the salivary gland are classified into two subtype, the extravasation type, a salivary mucocele, also called a sialocele or extravasated pseudocyst, and the retention type, a true salivary cyst, also called a dilatation of the duct[@r1]. Salivary mucoceles lack an inner luminal epithelial lining. Conversely, true salivary cysts have an epithelial lining, such as a nonkeratinized stratified squamous or interlobular duct epithelium[@r1]. In the present case, the cyst was lined by epithelial-like macrophages instead of epithelial tissues. Therefore, we diagnosed it as a salivary mucocele. In humans, a salivary cyst on the floor of the mouth is clinically called a ranula with or without an epithelial lining[@r9]. On the other hand, a textbook for animals[@r1] states that the term "ranula" should be only applied to a cystic distension of the duct, although some of the literature for animals has also applied the term "ranula" to a salivary gland cyst occurring from the sublingual gland with or without an epithelial lining[@r5]. It seems that the nomenclature of these lesions is controversial in animals. Therefore, we did not apply the term "ranula" in the diagnosis. In the present case, fortunately, we found a ruptured sublingual interlobar duct that could be a cause of this lesion. However, it is not adequate to allege that the ruptured duct was the primary site of leakage. It is reported that salivary mucoceles sometimes have more than one communication between the sublingual gland and the cyst.[@r14] In view of the orientation of the cyst, it was speculated that the primary leakage probably occurred in the individual units of the polystomatic portion or the duct of the monostomatic portion. The histological appearance of salivary mucoceles varies greatly depending on the stage of development[@r1]. This case was recognized as a relatively long-standing lesion because the cyst was circumscribed by mature dense connective tissue with abundant vessels. On the other hand, the lumen of the cyst was filled with eosinophilic amorphous material and desquamated cells, suggesting that sublingual secretion could still be leaking out consistently and that this lesion would be in the process of formation. These prolonged irritations might induce accumulation of a large number of macrophages and influence its morphological changes, leading to an epithelial-like appearance. We believe this case report helps to understand a diversity of the background findings in beagles used in toxicity studies. The authors would like to thank Mr. Ryo Fukuda, Mr. Ryotaro Hori and Mr. Yoshiyuki Furukawa for their support during this work.

Introduction ============ The kinetoplastid *Trypanosoma brucei* is the causative agent of human African trypanosomiasis, also known as 'sleeping sickness', and the cattle disease Nagana. Human African trypanosomiasis is always fatal if untreated and constitutes a major public health problem in sub-Saharan Africa ([@b22]). The disease is characterized by an asymptomatic period of several weeks or months that progresses through an early stage, characterized by malaria-like symptoms, to an advanced stage, where severe neurological and mental disorders appear. There is a clear need for new therapeutics to control the disease. *Trypanosoma brucei* and the related trypanosomatid parasites *Trypanosoma cruzi* and *Leishmania* sp. synthesize complex cell-surface glycoconjugates, several of which are essential to parasite survival and/or infectivity. Examples for the infective bloodstream form of *T. brucei* include the variant surface glycoprotein (VSG), the transferrin receptor ([@b69]; [@b16]; [@b42]; [@b50]; [@b41]), the p67 lysosomal glycoprotein ([@b51]) and the membrane-bound histidine acid phosphatase *Tb*MBAP1 ([@b21]). This has led to the investigation of the enzymes of sugar nucleotide biosynthesis as potential therapeutic targets ([@b17]; [@b76], [@b78]; [@b74]; [@b75]; [@b70]; [@b18]; [@b37]; [@b40]; [@b39]). Sugar nucleotides are activated forms of sugars used as donors in glycosylation reactions. They are synthesized either by a *de novo* pathway, requiring the conversion of a precursor sugar/sugar nucleotide, and/or by a salvage pathway, in which the sugar is activated using a kinase and a pyrophosphorylase. In most eukaryotes, sugar nucleotides are formed and used in the cytosol and/or transported in the lumen of the Golgi apparatus and/or the endoplasmic reticulum where they are used by glycosyltransferases ([@b25]). In the last few years, our knowledge of *T. brucei* sugar nucleotide biosynthesis ([Fig. 1](#fig01){ref-type="fig"}) has widened to demonstrate that several steps in the biosynthesis of GDP-fucose (GDP-Fuc) ([@b75]), UDP-galactose (UDP-Gal) ([@b60]; [@b61]; [@b78]), UDP-*N*-acetylglucosamine (UDP-GlcNAc) ([@b70]; [@b39]) and GDP-mannose (GDP-Man) ([@b18]) are essential for parasite growth. Furthermore, most of these, and other ([@b40]), reports show that many of the enzymes of sugar nucleotide biosynthesis are located in the kinetoplastid peroxisome-like organelle, the glycosome. {#fig01} Despite these advances, and its importance as a metabolite, relatively little is known about the synthesis of UDP-glucose (UDP-Glc) in *T. brucei*. In eukaryotic cells, UDP-Glc is required for the synthesis of diverse glucose-containing glycoconjugates and secondary metabolites ([@b24]). It also plays a crucial role in the 'quality control' of newly synthesized glycoproteins taking place in the endoplasmic reticulum ([@b73]; [@b62]), as unfolded glycoprotein glucosyltransferase (UGGT), part of the calnexin- and/or calreticulin-mediated glycoprotein quality control system, requires UDP-Glc as donor substrate ([@b68]; [@b12]). In *T. brucei* UGGT has been shown to be essential for parasite growth and survival at 40°C ([@b32]). UDP-Glc is also the presumed donor for the synthesis of base J (β-D-glucosylhydroxymethyluracil), a rare deoxynucleotide of unknown function found in the DNA of trypanosomatids ([@b8] and references within). Finally, via the action of UDP-Glc 4′-epimerase, UDP-Glc is the only source of the essential sugar nucleotide UDP-Gal in *T. brucei* and *T. cruzi* ([@b60]; [@b61]; [@b38]) because their hexose transporters do not take up D-Gal ([@b5]; [@b71]). Generally, UDP-Glc is synthesized in a three-step process: (i) phosphorylation of glucose to glucose-6-P via hexokinase, (ii) transfer of the phosphate group from C6 to C1 to produce glucose-1-phosphate (Glc-1-P) via phosphoglucomutase (PGM, E.C. 5.4.2.2) and (iii) coupling of glucose-1-P to UTP via UDP-Glc pyrophosphorylase (UGP). In *T. brucei*, the enzymes responsible for the first and last step have been characterized ([@b46]; [@b40]) and *T. cruzi* and *Leishmania* have easily identifiable PGM genes ([@b52]; [@b53]). The absence of a putative *T. brucei* PGM gene is therefore perplexing. The PGMs belong to the α-D-phosphohexomutase (PHM) superfamily of enzymes together with eukaryotic phospho-*N*-acetylglucosamine mutases (PAGM, E.C. 5.4.2.3) and bacterial phosphomannomutase/phosphoglucomutases (PMM/PGM, E.C. 5.4.2.2) and phosphoglucosamine mutases (E.C. 5.4.2.10) ([@b65]). Although eukaryotic phosphomannomutases (PMM, E.C. 5.4.2.8) reversibly catalyse the transfer of the phosphate group of a mannosyl phosphate between the C6 and C1 positions, these enzymes belong to the haloacid dehydrogenase superfamily of phosphotransferases ([@b15]). The apparent lack of a gene encoding for *T. brucei* PGM could be explained if *Tb*PGM bears little or no resemblance to conventional PGMs, or if the conversion of glucose-phosphates can be catalysed by *Tb*PMM and/or *Tb*PAGM. The latter seems the more likely as, although eukaryotic PMMs and PAGMs are generally highly specific for their sugar phosphate substrates, the ability of some PMMs and PAGMs to interconvert glucose-phosphates has been shown ([@b23]; [@b7]; [@b31]; [@b47]; [@b55]; [@b54]; [@b34]; [@b56]). In this paper, we identified the genes encoding functional *T. brucei* phosphomannomutase (*Tb*PMM) and phospho-*N*-acetylglucosamine mutase (*Tb*PAGM) and studied their subcellular location and metabolic activities in bloodstream form *T. brucei*. As well as interconverting mannose- and *N*-acetylglucosamine-phosphates respectively, both were shown to be capable of producing Glc-1-P from glucose-6-phosphate (Glc-6-P), albeit with different efficiencies. *Tb*PMM was also crystallized and its structure solved at high resolution. The dual specificities of these enzymes suggested an explanation for the absence of a *Tb*PGM gene in this parasite and knockdown of *Tb*PMM and *Tb*PAGM by RNA interference (RNAi) showed that both enzymes are mutually redundant for their PGM activity. Results ======= Identification of *T. brucei* PMM and PAGM encoding genes --------------------------------------------------------- *Saccharomyces cerevisiae* PMM (NCBI accession no.: NP116609) was previously used as a template for a BLASTp search of *T. brucei* predicted proteins and **Tb10.70.0370** was identified as a putative *TbPMM*. Putative PMM genes could be also found in the *Leishmania major* (GeneDB ID: **LmjF36.1960**) and *T. cruzi* (GeneDB ID: **TcCLB.510187.480**) genomes ([@b74]) and the *Leishmania mexicana* PMM has been structurally characterized (**LmxM.36.1960**; [@b35]). As shown in [Fig. 2A](#fig02){ref-type="fig"}, *Tb*PMM belongs to the eukaryotic PMM family. Eukaryotic PMMs are members of the haloacid dehydrogenase superfamily of phosphotransferases. *Tb*PMM, like the other eukaryotic PMMs, belongs to the type II class of this superfamily ([@b35]; [@b67]). This can be deduced by the location of the cap domain (aa 85--183), which in this class is found between motif II and III ([Fig. 2A](#fig02){ref-type="fig"}). Eukaryotic PMMs are characterized by a conserved DVDGT motif (motif I, aa 9--13), localized at the protein N-terminus. The first Asp of this motif (D9) is phosphorylated during the catalytic cycle. *Tb*PMM shows 57% sequence identity to its *S. cerevisiae* homologue. {#fig02} A BLASTp search of the *T. brucei* predicted protein database with *Homo sapiens* and *S. cerevisiae* PAGM amino acid sequences (NCBI accession nos. NP_056414.1 and NP_010856.1 respectively) revealed a single putative *TbPAGM* gene (GeneDB ID: **Tb927.8.980**) ([@b74]). Two putative PAGMs could also be identified in *T. cruzi* (GeneDB ID: **TcCLB.503733.70**, **TcCLB.508569.80**). Both genes are more than 97% identical, probably accounting for the two haplotypes (Esmeraldo and non-Esmeraldo-like) present in the CL-Brener strain that was sequenced ([@b19]). A single PAGM gene was also detected in *L. major* (GeneDB ID: **LmjF07.0805**). For *Tb*PAGM, the predicted amino acid sequence contains the conserved motifs for the α-D-PHM family, such as the phosphate-binding site (RPSGTE, aa 564--569), the Mg^2+^ binding site (DGDADR, aa 338--343), and also several invariant residues such as R564, S566, G418, D338, D340, D342, R343 (marked by asterisks, [Fig. 2B](#fig02){ref-type="fig"}) ([@b65]). However, in *Tb*PAGM, the sugar-binding site (aa 435--442) contains an unusual variation, whereby the normally highly conserved A437 residue is replaced by a serine. As observed in other PAGMs, the phosphoserine loop (TASHNP, aa 80--85, black box, [Fig. 2B](#fig02){ref-type="fig"}), essential for catalytic activity, is translocated to an earlier position compared to other α-PHM family members, in order to accommodate the bulky *N*-acetyl group present in the phosphorylated GlcNAc substrate ([@b65]). *Tb*PAGM shares 28% and 35% identity to its human and yeast orthologues respectively. According to a recent quantitative proteomic analysis, both *Tb*PAGM and *Tb*PMM are present at the protein level in both life cycle stages and the relative concentrations of the proteins (procyclic levels divided by bloodstream form levels) are 1.06 ± 0.30 and 1.17 ± 0.30 respectively ([@b77]). The activities of recombinant *Tb*PMM and *Tb*PAGM on mannose- and N-acetylglucosamine-phosphates ------------------------------------------------------------------------------------------------- The *TbPMM* and *TbPAGM* open reading frames (ORFs) were amplified by PCR from genomic DNA obtained from *T. brucei* strain 427, sequenced (EMBL nucleotide sequence database accession number FR851306 and FR851307) and cloned into an expression vector containing *N*-terminal His~6~-tags and expressed in *E. coli*. After Nickel column purification, *Tb*PMM protein was further purified by size exclusion chromatography ([Fig. S1](#SD1){ref-type="supplementary-material"}) and the His-tag was removed by incubation with TEV protease resulting in a band of 27.9 kDa by SDS-PAGE ([Fig. S1](#SD1){ref-type="supplementary-material"}). For *Tb*PAGM, the His-tag was cleaved with PreScission protease™ and the untagged protein was further purified by size exclusion chromatography ([Fig. S2](#SD1){ref-type="supplementary-material"}). The molecular masses and identities of the tagged and untagged recombinant proteins were verified by MALDI-TOF and by LC-MS/MS peptide mass fingerprinting respectively (data not shown). Although both recombinant enzymes were active with and without the His-tag, the untagged recombinant enzymes were used for activity assays and, in the case of *Tb*PMM, for crystallization trials. A high-pH anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) system was used to assay the activity of the recombinant *Tb*PMM, as described in *Experimental procedures*. It was clear that *Tb*PMM can convert Man-1-P to Man-6-P in the presence of glucose-1,6-biphosphate (Glc-1,6-biP) as a cofactor ([Fig. 3B](#fig03){ref-type="fig"}), whereas no Man-6-P could be detected when *Tb*PMM was absent ([Fig. 3A](#fig03){ref-type="fig"}). We used the same system to see whether *Tb*PMM could convert GlcNAc-1-P to GlcNAc-6-P, but this was not the case ([Fig. S3A](#SD1){ref-type="supplementary-material"}). The same procedure was also used to determine the kinetic properties of the enzyme for Man-1-P ([Fig. 3C](#fig03){ref-type="fig"}). Different concentrations of Glc-1,6-biP were tested before selecting an optimal concentration of 1 µM. The apparent *K~m~, V~max~* and catalytic efficiency of recombinant *Tb*PMM using Man-1-P as substrate are shown in [Table 1](#tbl1){ref-type="table"}. It was not possible to saturate the reaction, as the *Tb*PMM activity showed inhibition at the highest Man-1-P concentrations ([Fig. 3C](#fig03){ref-type="fig"}). Consequently, an equation for high-substrate inhibition was used to calculate the kinetic parameters of the reaction. The pH dependence of *Tb*PMM PMM activity was analysed over the pH range 4.0 to 10.0. The enzyme showed activity between pH 5.0 and 8.0 with an optimum at pH 6.5 (data not shown). {ref-type="table"}).](mmi0085-0513-f3){#fig03} ###### Comparison of the kinetic properties of recombinant *Tb*PMM and *Tb*PAGM with different sugar phosphates as substrates *Tb*PMM *Tb*PAGM ------------ ------------------------------------------ --------------- ------------- ------------------------------------------ --------------- ------------- ----------- ----------- Man-1-P 327 ± 66[b](#tf1-2){ref-type="table-fn"} 2.6 ± 0.4 3.7 × 10^5^ Glc-6-P 96 ± 8[c](#tf1-3){ref-type="table-fn"} 0.331 ± 0.008 5.7 × 10^4^ 107 ± 17[b](#tf1-2){ref-type="table-fn"} 0.23 ± 0.03 9.4 × 10^4^ 8.3 ± 0.4 6.3 ± 2.6 GlcNAc-6-P 14 ± 6[c](#tf1-3){ref-type="table-fn"} 0.051 ± 0.006 1.6 × 10^5^ IC~50~ for GlcNAc phosphates as competitive inhibitors in the formation of Glc-1-P. The kinetic constants were calculated using an equation for high-substrate inhibition based on the best-fit non-linear lines. The kinetic constants were calculated from the Michaelis--Menten equation. The HPAEC-PAD system was also used to show the activity of recombinant *Tb*PAGM on GlcNAc-1-P and GlcNAc-6-P, with Glc-1,6-biP as cofactor ([Fig. 4B](#fig04){ref-type="fig"}). *Tb*PAGM was also able to convert Man-1-P to Man-6-P *in vitro* ([Fig. S3B](#SD1){ref-type="supplementary-material"}); however, as discussed later, this activity was not able to compensate for the knockdown of *Tb*PMM *in vivo*. The kinetic parameters for the conversion of GlcNAc-6-P to GlcNAc-1-P by *Tb*PAGM were determined using a *Tb*UAP and pyrophosphorylase coupled assay with colorimetric detection ([@b70]) in a 96-well plate format, as described under *Experimental procedures*. The *Tb*PAGM enzyme followed classic Michaelis--Menten kinetics ([Fig. 4C](#fig04){ref-type="fig"}) and the apparent *K~m~, V~max~* and catalytic efficiency of recombinant *Tb*PMM with using GlcNAc-6-P as the substrate are shown in [Table 1](#tbl1){ref-type="table"}. The pH dependence of *Tb*PAGM was studied over a 4.0 to 10.0 pH range. The enzyme showed a broad optimum between pH 7.2 and 9.1 (data not shown). {ref-type="table"}).](mmi0085-0513-f4){#fig04} Both *Tb*PMM and *Tb*PAGM can convert Glc-6-P to Glc-1-P -------------------------------------------------------- The ability of *Tb*PMM and *Tb*PAGM to catalyse this reaction was assessed by incubating each recombinant enzyme in the presence of Glc-1-P. In both cases the formation of Glc-6-P could be detected by HPAEC-PAD ([Fig. S4](#SD1){ref-type="supplementary-material"}). As Glc-1-P is the substrate for the UGP, we were able to set up an assay for the conversion of Glc-6-P to Glc-1-P by coupling *Tb*PMM and *Tb*PAGM with *Tb*UGP to produce UDP-Glc in the presence of UTP ([@b40]). Thus, a single peak that co-eluted with UDP-Glc could be detected on a HPAEC-UV system at A~260~, when either *Tb*PMM or *Tb*PAGM was coupled with recombinant *Tb*UGP ([Fig. 5B and D](#fig05){ref-type="fig"}). Removal of either PHM from the reaction resulted in the loss of the UDP-Glc peak ([Fig. 5A and C](#fig05){ref-type="fig"}). {ref-type="table"}.](mmi0085-0513-f5){#fig05} The kinetic parameters for the conversion of Glc-6-P to Glc-1-P by *Tb*PMM and *Tb*PAGM were determined in a discontinuous coupled assay with *Tb*UGP with colorimetric detection. This assay measures the pyrophosphate (PPi) product of the Glc-1-P + UTP → UDP-Glc + PPi coupling reaction via conversion of the PPi to inorganic phosphate and detection with a molybdate reagent (see *Experimental procedures*). The kinetic values of *Tb*PMM and *Tb*PAGM are listed in [Table 1](#tbl1){ref-type="table"}. Interestingly, *Tb*PAGM activity was inhibited at the highest concentrations of Glc-6-P so that no saturation could be observed in the kinetic analysis ([Fig. 5F](#fig05){ref-type="fig"}). No such inhibition could be detected in the case of *Tb*PMM ([Fig. 5E](#fig05){ref-type="fig"}). As the affinity of *Tb*PAGM for GlcNAc-6-P is almost 10-fold higher than for Glc-6-P, we investigated whether GlcNAc-6-P could act as a competitive inhibitor in the formation of Glc-1-P. This was confirmed by running the assay at 100 µM Glc-6-P (i.e. at the apparent K~m~ of *Tb*PAGM for Glc-6-P) in the presence of increasing concentrations of GlcNAc-6-P, which had an IC~50~ of 8.3 ± 0.4 µM ([Fig. 6A](#fig06){ref-type="fig"}). GlcNAc-1-P also inhibited the conversion of Glc-6-P to Glc-1-P with a comparable IC~50~ of 6.3 ± 2.6 µM ([Fig. 6B](#fig06){ref-type="fig"}). {ref-type="table"}).](mmi0085-0513-f6){#fig06} Determination and analysis of the *Tb*PMM structure --------------------------------------------------- The crystal structure of the full-length *Tb*PMM was solved (PDB accession number **3F9R**) by molecular replacement using a modified form of the *L. mexicana* PMM (*Lm*PMM; PDB accession number **2i54**). The crystallographic and model refinement data for the *Tb*PMM structure are reported in [Table 2](#tbl2){ref-type="table"}. Two polypeptide units could be seen in the crystals asymmetric unit, chains A and B. When symmetry mates were extended, each monomer formed a dimer with its own symmetry mates, i.e. A : A and B : B. These dimers mimicked the dimerization mode seen in both human and *Lm*PMM structures ([Fig. S6](#SD1){ref-type="supplementary-material"}). ###### Data collection and crystallographic refinement ------------------------ ---------------- Data collection Space group P21212 Cell dimensions  a (Å) 105.93  b (Å) 46.99  c (Å) 94.49 Wavelength 1.5418 Resolution 50--1.8 (1.86) Unique reflections 44 877 R~merge~ 0.122 (0.862) Mean I/σ(I) 17.83 (1.40) Completeness (%) 99.9 (99.3) Redundancy 6.2 (4.4) PDB code 3F9R Refinement Resolution 50--1.85 Number of reflections 41 076 Test set 2074 R~work~/R~free~ 0.202/0.266 RMS deviations  Bond lengths (Å) 0.016  Bond angles (°) 1.471 Average B value (Å^2^) 31.63 ------------------------ ---------------- The oligomeric state of *Tb*PMM was studied by analytical ultracentrifugation (AUC) and size exclusion chromatography. Recombinant *Tb*PMM was analysed by AUC in two different buffers: one containing sodium chloride at physiological concentrations (10 mM HEPES pH 7.5, 150 mM NaCl) and one without any salt (10 mM Tris-HCl pH 7.5, 1 mM DTT). This last buffer is the same used in the crystallization of *Lm*PMM ([@b35]). In both cases, *Tb*PMM behaved as a monomer and no oligomers could be observed ([Fig. S5](#SD1){ref-type="supplementary-material"}). It should be noted that when *Tb*PMM was analysed by size exclusion chromatography in high salt during the purification process, the retention time was consistent with the apparent molecular weight (MW) of a homodimer (data not shown). Because the dimeric state of *Tb*PMM could be observed only at high concentrations of salt, but not under physiological conditions, we concluded the biological assembly of *Tb*PMM to be a monomer. Comparison of the *Tb*PMM crystal structure with of the monomers of the *L. mexicana* and human enzymes showed that the overall structure is very well conserved ([Fig. 7A](#fig07){ref-type="fig"}), as are the residues forming the active site ([Fig. 7B](#fig07){ref-type="fig"}). As previously described, the residues involved in the catalysis are located in the core domain (aa 5--82 and 187--246): D9 (Motif I) corresponds to the Asp that is phosphorylated in the reaction, S45 (motif II) and K187 (Motif III) are involved in the binding of the phosphate, and D9, D11 (Motif I), D206 and D214 (Motif IV) co-ordinate the Mg^2+^ ion ([Fig. 7B and C](#fig07){ref-type="fig"}) ([@b1]; [@b57]). The cap domain contributes the residues involved in substrate recognition (E119, R121, M124, N126, R132, R139, S177), so that the active site is located in the groove at the interface of the core and cap domains ([Fig. 7A](#fig07){ref-type="fig"}) ([@b35]; [@b67]). The electron density map for the solved structure revealed a sulphate ion in each monomer\'s active site, in approximately the same position in which the 1-phosphate moiety of the Glc-1,6-biP cofactor would be found ([@b35]). This ion co-ordinates directly with R139 and R132, and through linking waters with S177 and R121 ([Fig. 7C](#fig07){ref-type="fig"}). Overlaying the structure with **2i55**, the *Lm*PMM structure with Glc-1,6-biP bound, the phospho-sugar fits quite well within the closed pocket in *Tb*PMM. In [Fig. 7B](#fig07){ref-type="fig"}, we can see the strict conservation within this pocket. In the *Lm*PMM structure, D10, D207 and D215 are engaged in co-ordinating a Mg^2+^ ion and D10 is supposed to form the phosphoprotein intermediate and in the *Tb*PMM structure, all four Asp residues are conserved (D9, D11, D206 and D214), even though no Mg^2+^ ion was found in the electron density. No additional magnesium was added to the protein prior to crystallization, although density for a magnesium ion was found in a distal part of the enzyme at another magnesium-binding site found in other PMMs structures (**2i55**, **2fue**). {#fig07} As the quaternary structure of *Tb*PMM is the main difference between this enzyme and other described eukaryotic PMMs, which are all homodimers ([@b36]; [@b35]; [@b67]), we analysed the residues at the hypothetical dimer interface. Analysis using PISA-EBI identified 12 residues potentially involved in H-bonding or salt bridge interactions (N91, K99, R103, A106, D107, D109, P111, R114, T116, V118, Y120 and N133) and 14 residues involved in hydrophobic or other interactions in the *Tb*PMM structure. These residues are well conserved in the *Lm*PMM interface, with the exception of three amino acids. Two of these (I94 and Y120) were partially involved in hydrophobic interactions and correspond to the *Lm*PMM residues V95 and F121. The third substitution occurs at R103, which in *Lm*PMM is H104 ([Fig. 7D](#fig07){ref-type="fig"}). Interestingly, the PISA-EBI macromolecular analysis tool finds a low complexation significance score (CSS) for *Tb*PMM (0.017) further suggesting a monomeric quaternary structure. As a comparison, the CSS for *Lm*PMM is 0.249, consistent with the interface residues having a role in multimer formation. Subcellular localization of *Tb*PMM and *Tb*PAGM in bloodstream form *T. brucei* -------------------------------------------------------------------------------- Subcellular localization for both enzymes was studied by immunofluorescence microscopy. Paraformaldehyde-fixed wild type bloodstream form *T. brucei* cells were incubated with polyclonal mouse anti-*Tb*PMM ([Fig. 8A](#fig08){ref-type="fig"}) or anti-*Tb*PAGM ([Fig. 8B](#fig08){ref-type="fig"}) combined either with rabbit polyclonal anti-GAPDH, as glycosomal marker (images a to d) or with rabbit anti-enolase, as cytosolic marker (images e to h). Cells were washed and incubated with a mix of Alexafluor 488 anti-mouse IgG (green) and Alexafluor 594 anti-rabbit IgG (red). Anti-*Tb*PMM showed a punctate staining pattern and substantial colocalization with anti-GAPDH ([Fig. 8A](#fig08){ref-type="fig"}, images a to d), and an absence of colocalization with anti-enolase ([Fig. 8A](#fig08){ref-type="fig"}, images e to h), indicating that *Tb*PMM is a glycosomal enzyme. On the other hand, *Tb*PAGM showed both punctate and diffuse staining, in agreement with partial colocalization with GAPDH ([Fig. 8B](#fig08){ref-type="fig"}, images a to d) and with enolase ([Fig. 8B](#fig08){ref-type="fig"}, images e to h), respectively, indicating that *Tb*PAGM is located in both the glycosomes and the cytosol. {#fig08} The subcellular localization of *Tb*PMM and *Tb*PAGM was also studied by digitonin latency, as described previously ([@b27]), using Western blotting of the proteins released by escalating digitonin concentrations ([Fig. 8C](#fig08){ref-type="fig"}). *Tb*PMM was very poorly released at lower concentrations of digitonin but increased gradually with increasing digitonin concentration, reaching a maximum at 0.1 mg digitonin/mg protein. This pattern is very similar to that for aldolase (aldo), a lumenal glycosome resident enzyme, indicating that *Tb*PMM is also a glycosomal enzyme. Lower MW bands were observed at higher concentrations of digitonin and at 0.1% Triton X-100 ([Fig. 8C](#fig08){ref-type="fig"}, lanes 7 and 8 respectively) are probably due to proteolytic degradation of *Tb*PMM, and not due to non-specificity of the affinity-purified antibody that specifically immunoprecipitated a single band of the correct apparent MW from 2 × 10^8^ cell equivalents of a trypanosome detergent lysate ([Fig. 8D](#fig08){ref-type="fig"}, lane 2). Digitonin latency was also used to study the cellular location of *Tb*PAGM. Unlike *Tb*PMM, *Tb*PAGM was substantially released at low concentrations of digitonin, a pattern similar to enolase (eno), a cytosolic protein. However, the release of *Tb*PAGM increased gradually with the increase in digitonin concentration ([Fig. 8A](#fig08){ref-type="fig"}, compare lanes 2 and 7) indicating that PAGM has a dual localization in the cell, glycosomal and cytosolic. RNAi knockdown of *Tb*PMM and *Tb*PAGM and the effects on sugar nucleotide metabolism in bloodstream form *T. brucei* --------------------------------------------------------------------------------------------------------------------- The contributions of *Tb*PMM and *Tb*PAGM to the biosynthesis of UDP-Glc and UDP-Gal*p* were studied by RNAi. Fragments of 464 and 497 bp of *TbPMM* and *TbPAGM* ORFs, respectively, were cloned into p2T7TA^Blue^ and the resulting constructs were used to generate two cell lines expressing tetracycline inducible double-stranded RNA (dsRNA) targeting *TbPMM* and *TbPAGM* respectively. Induction of dsRNA targeting *TbPMM* resulted in a 68 ± 18% knockdown of *Tb*PMM mRNA after 48 h (as judged by qRT-PCR with four replicates) and a reduction of growth rate that led to cell death after 72 h, demonstrating the essentiality of *Tb*PMM ([Fig. 9A](#fig09){ref-type="fig"}). Phenotypic analyses of this knockdown cell line were carried out at 24 h and 48 h post dsRNA induction. The expression of *Tb*PAGM protein was evaluated by Western blot in case this enzyme, which can interconvert Man-1-P and Man-6-P, might be upregulated in the induced *Tb*PMM RNAi cells to compensate for the loss of *Tb*PMM, but this was not the case ([Fig. 9B](#fig09){ref-type="fig"}). Sugar nucleotides were extracted from induced and un-induced *Tb*PAGM and *Tb*PMM RNAi cell lines and from wild type cells, separated by HPLC and quantified by electrospray ionization tandem mass spectrometry, as previously described ([@b74]). In the *Tb*PMM RNAi cell line the GDP-Man levels were reduced upon induction, as expected. Thus, GDP-Man was reduced to 55% and 50% (*P* \< 0.005) of wild type levels after 24 and 48 h of dsRNA induction respectively, and the downstream GDP-Man metabolite, GDP-Fuc, was reduced to 40% (*P* = 0.001) of wild type levels after 48 h ([Fig. 9C](#fig09){ref-type="fig"}). However, under these latter conditions all of the UDP-sugar levels were similar to those of wild type or un-induced cells ([Fig. 9C](#fig09){ref-type="fig"}). These data show that, unlike GDP-Man and GDP-Fuc, the synthesis of UDP-Glc and UDP-Gal is not critically dependent on the expression of *Tb*PMM. Further evidence of mannose starvation in the induced *Tb*PMM RNAi cell line was obtained by Western blotting with anti-VSG antibodies ([Fig. 9D](#fig09){ref-type="fig"}). Thus, as already evident after 24 h but very clear after 48 h, *Tb*PMM knockdown leads to under *N*-glycosylation of sVSG221, as previously described for the conditional null mutant of *T. brucei* GDP-Man pyrophosphorylase ([@b18]). {#fig09} The induction of dsRNA targeting *TbPAGM* caused only a slight reduction in cell growth ([Fig. 10A](#fig10){ref-type="fig"}), even though the level of *Tb*PAGM protein after 2 and 3 days of induction was reduced to 15% of wild type levels or 30% of un-induced levels respectively ([Fig. 10B](#fig10){ref-type="fig"}). No compensatory difference in the RNA levels for *Tb*PMM was observed in the *Tb*PAGM RNAi cell line, either un-induced or induced for 2 or 3 days (data not shown). After 48 h of induction, the sugar nucleotide levels in the *Tb*PAGM RNAi cell line agreed reasonably well with those of un-induced and wild type cells ([Fig. 10C](#fig10){ref-type="fig"}). However, after 72 h of induction, the UDP-GlcNAc levels were 25% and 30% (*P* \< 0.05) of un-induced and wild type levels respectively, while all other sugar nucleotide levels remained comparable to those in un-induced cells ([Fig. 10C](#fig10){ref-type="fig"}). Despite the significant knockdown in *Tb*PAGM protein levels, evidence for GlcNAc starvation, i.e. under *N*-glycosylation of sVSG221 as previously described for the conditional null mutants of *T. brucei* UDP-GlcNAc pyrophosphorylase and glucosamine-6-P *N*-acetyltransferase ([@b70]; [@b39]), was barely detectable ([Fig. 10D](#fig10){ref-type="fig"}). It was noted previously that growth defects do not occur in bloodstream form *T. brucei* until UDP-GlcNAc levels fall below 20% of wild type levels ([@b70]) and it would appear that the RNAi knockdown of *Tb*PAGM achieved here is not deep enough to significantly affect protein glycosylation and cell growth. {#fig10} In conclusion, the analysis of the sugar nucleotide levels in *Tb*PMM and *Tb*PAGM RNAi cell lines suggests that these two enzymes are mutually redundant for the conversion of Glc-6-P to Glc-1-P. Thus, although a reduction in GDP-sugars levels could be observed upon knockdown of *Tb*PMM expression, and likewise reduction in UDP-GlcNAc levels was observed in the case of *Tb*PAGM knockdown, no significant reduction in UDP-Glc or UDP-Gal was observed in either cell line. Discussion ========== Although genes encoding putative PMM and PAGM enzymes were found in the *T. brucei* genome, and biochemically confirmed here, no gene encoding a putative PGM, the enzyme responsible for the synthesis of Glc-1-P from Glc-6-P, could be found ([@b6]; [@b74]). The absence of a *T. brucei* PGM gene was particularly perplexing. First, because *T. cruzi* and *Leishmania* have easily identifiable *PGM* genes and the *T. cruzi* PGM enzyme has been biochemically characterized ([@b52]; [@b53]) and, second, because the only known route to the identified *T. brucei* metabolite UDP-Glc ([@b74]) is via UGP ([@b40]). As UGP uses Glc-1-P and UTP as substrates, and as the only known route to Glc-1-P is from Glc-6-P via PGM activity, it follows that *T. brucei* must possess an enzyme with PGM activity. Furthermore, the flux through UDP-Glc, and therefore also through Glc-1-P, is known to be very considerable in bloodstream form and procyclic form *T. brucei* ([@b74]). This is because UDP-Glc is the obligate precursor of UDP-Gal ([@b60]; [@b61]; [@b66]; [@b78]) that, in turn, is used to synthesize the many galactose-containing glycoproteins of the parasite, including the highly abundant VSG coat glycoprotein and the procyclins in the bloodstream and procyclic forms of the parasite respectively. The biochemical characterization of *Tb*PMM and *Tb*PAGM described in this paper resolves this metabolic paradox, as both enzymes are capable of converting Glc-6-P to Glc-1-P, as well as interconverting their definitive mannose- and N-acetylglucosamine-phosphate substrates. Certain other PMMs and PAGMs have been shown to be similarly promiscuous in other organisms ([@b23]; [@b7]; [@b31]; [@b47]; [@b54]; [@b34]; [@b56]), but this is the first known example of PGM activity being completely replaced by PMM and/or PAGM. Analysis of the synteny between trypanosomatid genomes ([Fig. S7](#SD1){ref-type="supplementary-material"}; [@b20]) suggests that *T. brucei* has deleted its PGM gene at some point in evolution, presumably after one or both of PMM and PAGM had attained sufficient catalytic flexibility to support Glc-6-P to Glc-1-P conversion as well. The opposite situation was recently reported for another protozoan parasite, *Giardia lamblia*. This organism has no obvious PMM gene but instead has evolved two PGMs: a canonical glucose phosphate-specific PGM and a divergent one able to convert mannose-phosphates ([@b44]). The PGM activities of *Tb*PMM and *Tb*PAGM appear to be mutually redundant, such that RNAi knockdown of each in turn has no significant effect on UDP-Glc and UDP-Gal levels. This is despite the fact that Western blot and unpublished proteomic data strongly suggest that *Tb*PAGM is much more abundant than *Tb*PMM. Indeed, it is interesting that *Tb*PAGM, which can also interconvert mannose-phosphates, cannot substitute for *Tb*PMM and that, therefore, RNAi knockdown of *Tb*PMM affects protein glycosylation and suppresses parasite growth. This is consistent with previous studies using conditional null mutants that show that downstream enzymes on the pathway to GDP-Man and its further metabolite GDP-Fuc are essential ([@b75]; [@b18]). In contrast, the RNAi knockdown of *Tb*PAGM described here was insufficient to significantly affect protein glycosylation or suppress growth. This is despite the fact that the conditional null mutants of glucosamine-6-P *N*-acetyltransferase and UDP-GlcNAc pyrophosphorylase show that the downstream enzymes leading to UDP-GlcNAc are clearly essential ([@b70]; [@b39]). The notion that any other phospho-sugar mutase(s) (in addition to *Tb*PMM and *Tb*PAGM) might exist in *T. brucei* that could perform PGM function was excluded by searches of the updated trypanosomatid genomes at TriTrypDB ([@b4]) for all of the protein family (PFAM) domains relevant to PMM, PAGM and PGM activities (PF03332, PF02878, PF02879, PF02880 and PF00408). These searches returned only PMM and PAGM genes for the African trypanosomes, whereas they returned PMM, PAGM and PGM genes for *T. cruzi* and the Leishmania. The *Tb*PMM enzyme has some unusual features when compared to other eukaryotic PMMs. First, its apparent *K~m~* for Man-1-P is relatively high (327 ± 66 µM) in comparison to the values reported for *Arabidopsis thaliana* (29.7 µM) ([@b56]), *Galderia sulphuraria* (50 µM) ([@b47]), and human PMM1 (3.2 µM) and PMM2 (18 µM) ([@b55]; [@b54]). Second, unlike the aforementioned enzymes, *Tb*PMM is inhibited by high levels of Man-1-P. Third, whereas PMMs, including that of *L. mexicana* ([@b35]), are typically homodimers, *Tb*PMM appears to be a monomer under physiological conditions. The properties of both *T. brucei* GDP-Man pyrophosphorylase, which shows significant product inhibition ([@b18]), and *Tb*PMM, which shows product inhibition in the Man-6-P to Man-1-P direction, may assist the coupling of GDP-sugar synthesis to glycoprotein synthesis. Thus, a reduction in glycoprotein precursor synthesis would reduce the demand for GDP-Man and lead to an accumulation of Man-6-P, which, via the reversible phosphomannose isomerase reaction, would lead to fructose-6-phosphate that would enter the glycolytic pathway. The ability to divert a significant portion of carbon metabolism into GDP-sugar nucleotides to fuel glycoprotein synthesis in rapidly dividing trypanosomes, and to automatically shunt this back into glycolysis when glycoprotein synthesis is repressed in non-dividing cells, suggests a mechanism that might help trypanosomes manage transformations between non-dividing metacyclic trypomastigotes and rapidly dividing slender trypomastigotes and from slender to non-dividing stumpy trypomastigotes. In most organisms, PMM and PAGM activities are located in the cytosol. However, in bloodstream form *T. brucei, Tb*PMM localizes to the glycosomes and *Tb*PAGM localizes to both the glycosomes and the cytosol. The glycosomes of *T. brucei* and related trypanosomatids belong, together with the glyoxysomes of plants and peroxisomes of other eukaryotes, to the microbody family of organelles. When compared to the other microbodies, trypanosomatid glycosomes are unique in their essential role in carbohydrate metabolism, i.e. the first seven steps of glycolysis, gluconeogenesis, glycerol metabolism and the pentose-phosphate pathway ([@b43]; [@b29]). Glycosomal proteins are typically targeted to these organelles via Peroxisomal Targeting Sequences (PTS), which can be located in the C-terminus (PTS-1) or N-terminus (PTS-2) of a protein. Internal targeting sequences have also been proposed as another targeting mechanism, as in *T. cruzi* PGM and *T. brucei* triosephosphate isomerase ([@b53]; [@b26]) or proteins may reach these organelles by 'piggybacking' on other glycosome-targeted proteins ([@b72]). As canonical trypanosome PTS sequences are not apparent for *Tb*PAGM or *Tb*PMM \[although the *T. cruzi* and *L. major* PMM orthologues do contain PTS-1 sequences ([@b48])\], it seems likely that both either use the piggybacking mechanism or some unidentified internal targeting sequence. An inefficient import of *Tb*PAGM in the glycosomes could explain the partial localization of this enzyme in the cytosol of bloodstream form *T. brucei*. In any case, both *Tb*PMM and *Tb*PAGM join the growing list of enzymes of *de novo* sugar nucleotide biosynthesis that have been localized to the glycosomes. These include hexokinase and phosphoglucose isomerase ([@b79]), glucosamine-6-P *N*-acetyltransferase ([@b39]), *Tb*UAP ([@b70]), *Tb*UGP ([@b40]), GalE ([@b61]), phosphomannose isomerase ([@b14]) and GDP-mannose 4,6-dehydratase ([@b75]). It is possible that all or most of the sugar nucleotide biosynthesis machinery is localized to the glycosome in *T. brucei* and that the rate of synthesis of these metabolites is controlled, to some degree, by 'compartmentalization'. Such a mechanism has already been proposed as an alternative to allosteric regulation for glycolysis in *T. brucei* ([@b29]). In summary, we have expressed and characterized the PMM and PAGM enzymes of *T. brucei* and determined their subcellular locations and the high-resolution crystal structure of *Tb*PMM. Further, by analysing their substrate specificities, we have provided a solution for the apparently paradoxical absence of a PGM gene in *T. brucei* by demonstrating that both *Tb*PMM and *Tb*PAGM can interconvert Glc-6-P and Glc-1-P. Selective knockdown of *Tb*PMM and *Tb*PAGM by inducible RNAi and analysis of sugar nucleotide levels suggests that they are mutually redundant for this particular function. It is also clear that the routes to three key sugar-1-phosphates (Man-1-P, GlcNAc-1-P and Glc-1-P) and thence to five key sugar nucleotides (GDP-Man, GDP-Fuc, UDP-GlcNAc, UDP-Glc and UDP-Gal) are dependent on just two enzymes: *Tb*PMM and *Tb*PAGM, each of which could have therapeutic potential. Experimental procedures ======================= DNA isolation and manipulation ------------------------------ Plasmid DNA was purified from *Escherichia coli* (DH5α) using the Qiagen Miniprep or Maxiprep kit as appropriate. Gel extraction was performed using QIAquick kits. Custom oligonucleotides were obtained from the University of Dundee oligonucleotide facility. *T. brucei* genomic DNA was isolated from 1 × 10^8^ bloodstream form cells using DNAzol (Helena Biosciences). All *T. brucei* cell cultures are mycoplasma-free. All plasmids were verified by sequencing (DNA Sequencing Service, College of Life Sciences, University of Dundee; <http://www.dnaseq.co.uk>). Recombinant proteins were identified in the Proteomics and Mass Spectrometry Facility, College of Life Sciences, University of Dundee. Protein sequence multiple alignments were assembled using ClustalW ([@b13]) and Jalview ([@b80]). *Tb*PMM expression and purification for antibody production ----------------------------------------------------------- The *TbPMM* (**Tb927.10.6440**) ORF was amplified by PCR from *T. brucei* genomic DNA with Platinum Taq DNA polymerase High Fidelity (Invitrogen) using forward and reverse primers containing *Nde*I and *Xho*I restriction sites (underlined) respectively: 5′-GGAATTC[CATATG]{.ul}AAAAGAGTTCTTTTACTCTTTGAC-3′ and 5′-CCG[CTCGAG]{.ul}TTACTTCATGGCAATTATTTTTTCC-3′. The *Nde*I and *Xho*I restriction sites were used to clone the *TbPMM* ORF into a pET15b (Novagen) modified with a PreScission™ Protease (PP) cleavage site, generating the following expression construct: pET15b-His~6~-PP-*TbPMM*. The recombinant *Tb*PMM was expressed in *E. coli* BL21(DE3). The cells were grown overnight at 16°C after induction with 0.75 mM isopropyl β-D-1-thiogalactopyranoside (IPTG). After harvesting, the cells were resuspended in 50 mM Tris-HCl pH 7.3, 150 mM NaCl, 100 mM imidazole, 1 mM DTT, 1 mg ml^−1^ lysozyme and Complete Protease Inhibitor Cocktail Tablets (Roche) and then lysed using a French press. The cell lysate was cleared by centrifugation (17 000 *g*, 4°C, 30 min), filtered through a 0.2 µm cellulose acetate membrane (Whatman) and loaded onto a 5 ml Ni^2+^ HiTrap™ Chelating HP column (GE Healthcare). Fractions were collected and checked by SDS-PAGE. The *Tb*PMM containing fractions were pooled and digested with 0.1 mg of PreScission™ protease GST-tagged (a kind gift of Daan van Aalten, University of Dundee) overnight at 4°C. The PreScission™ protease was removed by incubating the sample for 2 h at 4°C with 50 µl of Glutathione Sepharose™ 4 Fast Flow (GE Healthcare) slurry and then by separating the beads by centrifugation. A fraction of the untagged *Tb*PMM, present in the supernatant, was used for mouse and rabbit immunization. The remaining recombinant protein was used for coupling to CNBr Sepharose beads. *Tb*PMM protein expression and purification for activity assays and crystallization trials ------------------------------------------------------------------------------------------ *Tb*PMM ORF was amplified as described above and cloned into the pET15-MHL vector (GenBank ID EF456738) and transformed into *E. coli* BL21-(DE3)-V2R-pRARE2. A single colony was inoculated into 100 ml of LB medium containing ampicillin/chloramphenicol (100 µg ml^−1^ and 34 µg ml^−1^ respectively) inside a 250 ml baffled flask and incubated with shaking at 250 r.p.m. overnight at 37°C. The culture was transferred into 1.0 l of TB with the same antibiotic formulation pre-added. The culture was allowed to grow in the LEX system (Harbinger Biotechnology and Engineering) to an OD~600~ of 5--6, cooled to 15°C and induced with 0.5 mM IPTG overnight at that temperature. The culture was harvested by centrifugation. The resulting pellets were resuspended to approximately 40 ml l^−1^ cell culture in a Binding Buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 5 mM imidazole and 5% glycerol) with protease inhibitors (1 mM benzamidine and 1 mM phenylmethyl sulphonyl fluoride) added, and stored at −80°C. On the day before purification, the pellets were thawed overnight at 4°C. Each pellet from 1 l of culture was pretreated with 0.5% CHAPS and 500 units of benzonase for 40 min at room temperature, and subsequently sonicated. After 6 min sonication, the cell lysate was centrifuged using a Beckman JA-25 rotor at 24 000 r.p.m. for 20 min at 4°C. The cleared lysate was loaded onto a 1.0--2.5 ml Ni-NTA (Qiagen) open column (pre-equilibrated with Binding Buffer) at approximately 1.5--2.0 ml min^−1^. The Ni-NTA column was then washed with 150 ml of Wash Buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 30 mM imidazole and 5% glycerol) at 2--2.5 ml min^−1^. The protein was then eluted with Elution Buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 250 mM imidazole and 5% glycerol). The eluted sample was applied to a Sephadex S200 16/60 gel filtration column (GE Healthcare) pre-equilibrated with Gel filtration Buffer (10 mM HEPES, pH 7.5, 500 mM NaCl) on an AKTA explorer system (GE Healthcare). The fractions corresponding to the eluted protein peak were pooled and further treated with TEV protease overnight to cut the His tag. The mixture was loaded onto another 1.0 ml Ni-NTA open column and the cut protein was collected from the flow through. Its identity and purity were evaluated by mass spectroscopy and SDS-PAGE gel. The sample was then concentrated using a 15 ml Amicon Ultra centrifugal filter device (Millipore) to 10 mg ml^−1^ and stored at 4°C. *Tb*PAGM protein expression and purification for activity assays and antibody production ---------------------------------------------------------------------------------------- *TbPAGM* (**Tb927.8.980**) was amplified by PCR from *T. brucei* genomic DNA with Platinum Taq DNA polymerase High Fidelity (Invitrogen) using forward and reverse primers containing *Nde*I and *Xho*I restriction sites (underlined) respectively: 5′-GGAATTC[CATATG]{.ul}GTGCTGCAGGCT-3′ and 5′-CGC[CTCGAG]{.ul}CTACGCTCCACCGCAGA-3′. The *Nde*I and *Xho*I restriction sites were used to clone the *TbPAGM* ORF into a pET15b (Novagen) modified with a PreScission™ Protease (PP) cleavage site, generating the following expression construct: pET15b-His~6~-PP-*TbPAGM*. The recombinant *Tb*PAGM was expressed in *E. coli* BL21(DE3). Cultures were grown at 37°C, in LB medium containing ampicillin, until OD~600~ was approximately 0.5. IPTG was added to a final concentration of 0.5 mM and the cultures were grown overnight at 16°C. The cell pellets were obtained by centrifugation and resuspended in 50 mM Tris-HCl pH 7.7, 200 mM NaCl, 50 mM imidazole, in the presence of DNase I, EDTA-free Complete Protease Inhibitor Cocktail Tablets (Roche) and 1 mg ml^−1^ lysozyme. The cells were lysed using a French press and the cell lysate cleared by centrifugation (20 000 *g*, 4°C, 30 min) filtered through a 0.2 µm cellulose acetate membrane (Whatman) and loaded onto a 5 ml Ni^2+^ HiTrap™ Chelating HP column (GE Healthcare). The *Tb*PAGM-containing fractions were pooled and digested with 0.1 mg of PreScission™ protease GST-tagged as described above. The cleaved protein was further purified by gel filtration using a Superdex G-200 10/30 column (Amersham) and eluted in 50 mM Tris-HCl pH 7.7, 200 mM NaCl. Analytical ultracentrifugation ------------------------------ Recombinant and untagged *Tb*PMM was analysed by sedimentation velocity on a Beckman Optima XL-1 Analytical Ultracentrifuge with an AN50-Ti rotor at 4°C, 32 000 r.p.m. The quaternary structure of the protein was studied at 0.25, 0.5 and 0.75 mg ml^−1^ in two different buffers: (i) 10 mM HEPES pH 7.5, 150 mM NaCl, (ii) 10 mM Tris-HCl pH 7.5, 1 mM DTT. Absorbance data were collected and analysed using the SEDFIT software ([@b64]). *Tb*PMM was assumed to be a globular protein and its amino acid composition was used to determine its density. Activity assay for the conversion of Man-1-P to Man-6-P ------------------------------------------------------- To assay activity, 10 ng of recombinant untagged *Tb*PMM was incubated in 100 µl reaction volume for 20 min at room temperature in buffer D (2 mM Bis-Tris propane pH 7.3, 5 mM MgCl~2~, 1 µM glucose-1,6-biphosphate, 50 ng of BSA and 250 µM Man-1-P). For *Tb*PAGM, 100 ng of enzyme was incubated in buffer D for 1 h at 37°C. The reaction was then stopped by addition of an equal volume of 0.2 M NaOH, raising the pH above 12. The samples were analysed by High Pressure Anion Exchange Chromatography coupled to a Pulse Amperometric Detector (HPAEC-PAD, Dionex) using a CarboPac PA1 column and conditions adapted from [@b82]. For the kinetic analysis, the concentration of Man-1-P in the reaction was varied between 10 and 1000 µM. Each concentration was analysed in triplicates. A high-substrate inhibition [equation (1)](#m1){ref-type="disp-formula"} based on a non-linear fit was used to calculate the kinetic parameters of the reaction. All kinetic data were fitted using GraFit5. Activity assays for the conversion of Glc-6-P and Glc-1-P --------------------------------------------------------- The conversion of Glc-1-P to Glc-6-P by recombinant *Tb*PMM and *Tb*PAGM was first followed by HPAEC-PAD. The conditions and reaction buffer are the same as described above for the Man-1-P to Man-6-P conversion, except that 25 ng of enzyme was used and the substrate was changed to Glc-1-P. The conversion of Glc-6-P to Glc-1-P by *Tb*PMM and *Tb*PAGM was also analysed using a coupled assay with *Tb*UGP. The formation of the final product (UDP-glucose) was followed on a HPAEC system (Dionex) using a CarboPac PA1 column and a UV detector set at 260 nm. Different amounts of recombinant proteins (50--400 ng) were incubated 15 min at room temperature in 100 µl reaction volume containing 2 mM Bis-Tris propane pH 7.3, 10 mM MgCl~2~, 1 µM glucose-1,6-biphosphate, 500 µM Glc-6-P, 1 mM DTT, 0.1 mg ml^−1^ BSA, 500 µM UTP and 1 µg rx^−1^*Tb*UGP (Buffer CA). The reaction was stopped by addition of an equal volume of 0.2 M NaOH. Kinetic parameters were determined using a colorimetric assay on a 96-well plate format (Cellstar) as described in [@b70]. Briefly, 25 ng of recombinant *Tb*PMM or *Tb*PAGM was incubated in 100 µl reaction volume for 15 min at room temperature in buffer CA, containing also 0.08 U rx^−1^ pyrophosphatase (Sigma). The concentration of Glc-6-P was varied between 10 and 1000 µM. After addition of 100 µl of Biomol Green (Biomol International) and 20 min incubation, the A~620~ was measured on a SpectraMax 340PC (Molecular Devices) plate reader. Triplicates were run for each point. The same high-substrate inhibition equation as above was used to fit the kinetic data for *Tb*PAGM. IC~50~ for GlcNAc-1-P and GlcNAc-6-P was measured using the colorimetric coupled assay. Briefly, 25 ng of *Tb*PAGM was incubated in the same conditions described above. A fixed concentration for Glc-6-P of 100 µM was used, while the GlcNAc-1-P and GlcNAc-6-P concentration was varied between 1 and 250 µM. The detection was carried out as above. Activity assays for the conversion of GlcNAc-6-P and GlcNAc-1-P --------------------------------------------------------------- The conversion of GlcNAc-1-P to GlcNAc-6-P by recombinant *Tb*PAGM was first followed by HPAEC-PAD. The conditions and reaction buffer (Buffer D) are the same as described above for the Man-1-P to Man-6-P conversion, expect that 25 ng of enzyme was used and the substrate was changed to GlcNAc-1-P. For *Tb*PMM, 250 ng of enzyme was incubated in Buffer D containing 800 µM GlcNAc-1-P for 1 h at 37°C before analysis by HPAEC-PAD. The conversion of GlcNAc-6-P to GlcNAc-1-P by *Tb*PAGM was analysed using a coupled assay with *Tb*UAP with colorimetric detection on a 96-well plate format (Cellstar) as described above. Briefly, 25 ng of recombinant *Tb*PAGM was incubated in buffer CA (2 mM Bis-Tris propane pH 7.3, 10 mM MgCl~2~, 1 µM glucose-1,6-biphosphate, 1 mM DTT, 0.1 mg ml^−1^ BSA, 500 µM UTP and 1.5 µg rx^−1^*Tb*UAP). Kinetic parameters were determined by varying the concentration of GlcNAc-6-P between 5 and 100 µM. Triplicates were run for each point. Parasite culture ---------------- Bloodstream form *T. brucei* parasites (strain 427, variant MITat1.2) expressing T7 polymerase and tetracycline repressor protein under G418 selection ([@b81]) were grown in HMI-9t medium at 37°C with 5% CO~2~. HMI-9T is a variant of the HMI-9 medium described in [@b30] where thioglycerol is used instead of β-mercaptoethanol ([@b28]). Generation of RNAi constructs and transformation of bloodstream form *T. brucei* -------------------------------------------------------------------------------- RNAit (<http://trypanofan.path.cam.ac.uk/software/RNAit.html>; [@b59]) was used to identify suitable RNAi internal coding sequences for *TbPMM* (**Tb927.10.6440**, nt. 204--667) and *TbPAGM* (**Tb927.8.980**, nt. 676--1172). The 464 nt (*TbPMM*) and 497 bp (*TbPAGM*) fragments, containing *Xho*I and *BamH*I restriction sites at their 5′ and 3′ respectively, were synthesized by Dundee Cell Products and were then cloned into p2T7TA^Blue^ ([@b2]) using the *Xho*I and *BamH*I sites to generate p2T7-*TbPMM* and p2T7-*TbPAGM*. These constructs were digested with NotI, precipitated, washed with 70% ethanol and redissolved in sterile water, before being used for *T. brucei* transformation by Amaxa nucleofection as previously described ([@b11]). The final constructs were verified by sequencing (DNA Sequencing Service, College of Life Sciences, University of Dundee) before transformation. Sugar nucleotide analysis ------------------------- Sugar nucleotide extraction and analysis were performed as previously described ([@b74]). Briefly, cells were pelleted by centrifugation, washed in ice-cold phosphate buffer saline and lysed in 70% ethanol in the presence of 20 pmol of the GDP-glucose internal standard (Sigma). The lysate was centrifuged to remove insoluble material and the supernatant extracted with butan-1-ol to remove lipids. Sugar nucleotides were extracted from the resulting aqueous phase using EnviCarb graphitized carbon columns (Supelco) as previously described ([@b58]). The eluted sugar nucleotides were analysed by multiple reaction monitoring LC-MS/MS ([@b74]). The *P*-values were determined using Student\'s *t*-test. Crystallization and data collection ----------------------------------- Crystals for *Tb*PMM were obtained by setting up the protein sample obtained as described above in a hanging drop vapour diffusion experiment at 20°C. The successful crystallization conditions leading to the structure were: 2 M (NH~4~)~2~SO~4~, 0.2 M NaCl, 0.1 M sodium cacodylate, pH 6.0. Crystals for data collection were transferred to paratone-N as a cryoprotectant, flash frozen in liquid nitrogen and data collected on a Rigaku FRE Superbright anode with a RAXIS-IV plate imager. A complete data set was collected, then indexed and scaled with the HKL2000 program ([@b49]). Structural determination and model refinement --------------------------------------------- The structure for *Tb*PMM was solved by molecular replacement using the program PHASER and the *Lmex*PMM structure (**2i54**) modified to have its side-chains replaced with the Tb sequence by FFAS03 ([@b33]). The model was refined against 1.85A data using an initial round of ARP/WARP, followed by iterative manual rebuilding in COOT and refinement with REFMAC 5.5.0109 (ccp4 Suite, [@b45]). The final model was refined with good geometry and statistics, and checked with MOLPROBITY with no outliers in the Ramachandran plot. Final data information can be found in [Table 2](#tbl2){ref-type="table"}. *Tb*PMM residues in this paper are numbered from the start methionine. Production of antibodies and cell localization of *Tb*PMM and *Tb*PAGM by immunofluorescence -------------------------------------------------------------------------------------------- Two adult Balb/c mice were used to raise polyclonal antibodies against *E. coli* overexpressed and untagged *Tb*PMM protein with Freund\'s complete adjuvant. Each animal received two further immunizations with Freund\'s incomplete adjuvant over 2 months. A pool of these mice sera was used in Western blots and immunofluorescence. Similar procedures were performed for *Tb*PAGM. For the immunofluorescence experiments, wild type bloodstream form *T. brucei* cells were grown in HMI-9T medium ([@b30]; [@b28]) to a density of 1 to 2 × 10^6^ cells ml^−1^, harvested by centrifugation at 800 *g* for 10 min at 4°C and resuspended in trypanosome dilution buffer (20 mM Na~2~HPO~4~, 2 mM NaH~2~PO~4~, 5 mM KCl, 80 mM NaCl, 1 mM MgSO~4~, 20 mM glucose pH 7.8) to a density of 2 × 10^7^ cells ml^−1^. Aliquots (10 µl) were placed on 13 mm pre-cooled coverslips inside 12-well plate, covered, incubated for 15 min at 4°C and subsequently fixed in 1 ml of freshly prepared 4% paraformaldehyde in phosphate-buffered saline (PBS) for 30 min at 4°C followed by three washes with 1 ml of PBS. Cells were permeabilized with 0.1% Triton X-100 in PBS for 10 min at room temperature. Samples were then blocked in 5% fish skin gelatin (FSG) in PBS containing 10% normal goat serum. The coverslips were incubated in a humid chamber with 1:1000 mouse anti-*Tb*PMM mixed with either 1:2000 rabbit anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH) antiserum or 1:4000 rabbit anti-enolase antiserum diluted in 1% FSG in PBS, 0.05% Triton X-100. Both anti-GAPDH and anti-enolase were kindly provided by Paul Michels (Catholic University of Louvain, Belgium). Samples were subsequently washed with 1% FSG in PBS, 0.05% Triton X-100 and incubated with 50 µl of 1:500 diluted Alexa 488-conjugated goat anti-mouse IgG and 1:500 diluted Alexa 594-conjugated goat anti-rabbit IgG for 1 h. Coverslips were washed and mounted on glass slides over a drop of Hydromount containing 2.5% DABCO and left to set in the dark for 30 min. Microscopy was performed on a Zeiss LSM 510 META confocal microscope. Digitonin latency and Western blotting -------------------------------------- Digitonin latency was performed as previously reported ([@b27]). Briefly, *T. brucei* bloodstream form cells (2 × 10^9^ cells) were washed twice with 10 ml of STE buffer (250 mM sucrose, 25 mM Tris-HCl pH 7.4 and 1 mM EDTA) and resuspended in 1.5 ml of STEN (STE containing 0.15 M NaCl). Aliquots (0.15 ml) were treated with equal volume of digitonin at various concentrations in the presence of 0.1 mM TLCK, 1 µg ml^−1^ leupeptin, 1 µg ml^−1^ aprotinin and 1 mM PMSF. Digitonin stock was prepared at 10 mg ml^−1^ in DMSO and diluted in STEN to the required concentrations. Complete extraction was obtained in parallel by treating an equivalent number of cells with 0.1% Triton X-100 in STEN. The lysates were incubated for 5 min at room temperature and the insoluble materials removed by centrifugation at 16 000 *g* for 2 min. The pellets were discarded; aliquots of the supernatants were run on a reducing 4--12% gradient NuPage gel (Invitrogen) and transferred to nitrocellulose for Western blotting. Note: Different amounts of same samples were loaded onto separate gels for each antibody Western blot, in order to keep within the detection by ECL and film exposure linear during the Western blot development; for *Tb*PMM 1 × 10^7^ cell equivalents per lane, for *Tb*PAGM 5 × 10^6^ cell equivalents per lane, for aldolase (aldo) 1 × 10^5^ cell equivalents per lane, for PEX13 5 × 10^6^ per lane and for enolase (eno) 1 × 10^5^ cell equivalents per lane. The membranes were blocked with 0.25% bovine serum albumin, 0.05% Tween-20 (Sigma), 0.15 M NaCl in 50 mM Tris-HCl pH 7.4. The membranes were probed in parallel with the following antibodies 1:1000 diluted mouse anti-*Tb*PMM, 1:1000 mouse anti-*Tb*PAGM, 1:4000 rabbit anti-aldolase, 1:5000 rabbit anti-Pex13 or 1:5000 rabbit anti-enolase for 1 h at 37°C. The membranes were washed and incubated for 1 h at room temperature with 1:100 000 goat anti-mouse or goat anti-rabbit horseradish peroxidase conjugate. After washing, the membranes were developed by chemiluminescent detection (Super Signal WestPico, Thermo Scientific) and film (Hyperbond ECL, GE Healthcare). For the immunoprecipitation of native *Tb*PMM (to check its mono-specificity), 2 × 10^8^*T*. *brucei* bloodstream form cells were lysed in 1% (w/v) SDS in 20 mM Tris-HCl pH 6.8 containing 0.1 M DTT and heated at 50°C for 15 min. SDS was diluted to 0.03% with 1% (w/v) Triton X-100 in 20 mM Tris-HCl pH 6.8 with 0.15 M NaCl in the presence of 0.1 mM TLCK, 1 µg ml^−1^ leupeptin, 1 µg ml^−1^ aprotinin and 1 mM PMSF. The insoluble material was removed by centrifugation, and 5 µl of pooled mice anti-*Tb*PMM sera was added to the supernatant and incubated for 1 h at 4°C. Subsequently, 50 µl of protein G Dynabeads (Invitrogen) was added and incubated for 1 h at 4°C. The beads were recovered by placing on the magnet for 2 min and washed three times with cold 20 mM Tris-HCl pH 7.2, 0.15 M NaCl, 0.03% SDS and 1% Triton X-100. The beads were boiled in SDS sample buffer with 0.1 M DTT, run on a 4--12% Nupage gel and Western blotted as described above. Western blot anti-*Tb*PAGM and anti-sVSG ---------------------------------------- *Tb*PAGM RNAi cells were induced or not with tetracycline for 2 and 3 days, lysed in SDS sample buffer, loaded (5 × 10^6^ cell equivalents per lane) on SDS-PAGE gel and blotted to nitrocellulose. The membrane was blocked with 2% (w/v) FSG in blocking buffer described above and developed with 1:1000 mouse anti-*TbP*AGM mixed with 1:4000 rabbit anti-aldolase. The membrane was washed three times with 0.1% Tween-20 in PBS, incubated with 1:15 000 donkey anti-mouse green infrared conjugate (IRDye 800 CW, Licor) mixed with 1:20 000 donkey anti-rabbit red conjugate (IRDye 680, Licor) and scanned in a Licor Odyssey Infrared Imaging System. The Licor imager software was used to quantify the signal for both infrared channels and the gel loading was normalized against the aldolase signal. For the sVSG analysis, *Tb*PAGM RNAi cells were induced or not with tetracycline for 2 and 3 days while *Tb*PMM RNAi cells for 2 days, then they were lysed in SDS sample buffer, loaded (5 × 10^6^ cell equivalents per lane) on SDS-PAGE gel and Western blotted to nitrocellulose. The Licor development was as described above, but instead rabbit anti-sVSG221 was used at 1:2000 dilution. Quantitative reverse transcriptase PCR (qRT-PCR) ------------------------------------------------ RNA was extracted using the RNeasy RNA extraction kit (Quiagen). cDNA was synthesized from 1 µg of RNA using oligo(dT) and random primers with the qScript cDNA synthesis system (Quanta Biosciences), and then diluted 1 in 100 with nuclease-free water. qRT-PCR reactions included 3 µl of diluted cDNA, SYBR Green Master Mix (Quanta Biosciences), and 0.6 µM forward (atgagggctttgataaagcgagc) and reverse (cgtccgagttgctcaacctgc) gene-specific primers. Amplification was carried out in an iCycler iQ5 PCR thermal cycler (Bio-Rad). Mean and standard deviation were determined by combining four replicate analyses. The authors would like to thank Mark Agacan and the University of Dundee Proteomics Facility (Wellcome Strategic Award 083481) for performing the AUC and proteomics experiments and Alan R. Prescott (Division of Cell Signalling and Immunology, University of Dundee) for his generous help with the microscopy. We would also like to thank Michael Urbaniak for the SILAC data and Alan Fairlamb for the high substrate inhibition equation. We are grateful to Daan van Aalten for providing reagents and Paul Michels for providing the anti-GAPDH and anti-enolase antibodies. Supporting information ====================== Additional supporting information may be found in the online version of this article. Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article. [^1]: Present address: Department of Molecular and Cell Biology, Boston University, Henry Goldman School of Dental Medicine, Boston, MA 02118, USA; [^2]: Present address: National Institute for Bioprocessing Research and Training, Dublin-Oxford Glycobiology Laboratory, Conway Institute, University College Dublin, Belfield, Dublin 4, Ireland; [^3]: Present address: Division of Gene Regulation and Expression, College of Life Sciences, University of Dundee, Dundee DD1 5EH, UK. [^4]: Re-use of this article is permitted in accordance with the Terms and Conditions set out at <http://wileyonlinelibrary.com/onlineopen#OnlineOpen_Terms>

Introduction {#S1} ============ Subjective tinnitus refers to the perception of a sound in one or both ears, or from inside the head, in the absence of an external acoustic source ([@B1], [@B2]). The phantom sound of tinnitus is a serious condition affecting 10--15% of the general population, with \~1% of the population experiencing a debilitating form of chronic tinnitus that interferes with daily life ([@B3]). Hyperacusis, defined as a hypersensitivity to moderate-intensity sounds ([@B4]--[@B7]), is a condition affecting \~6% of the general population and often co-occurs with tinnitus ([@B4]). The prevalence of hyperacusis in the tinnitus population has been estimated to be as high as 80% ([@B8]), suggesting a common mechanism of dysfunction for these two perceptual disorders. At present, the neural basis of tinnitus and hyperacusis remains elusive, and there are no widely accepted treatments or cures for individuals suffering from these conditions. However, studies in both humans and animals have led to a number of proposed neurophysiological models thought to underlie these conditions including tonotopic map reorganization, changes in spontaneous activity, or altered neural synchrony along the auditory pathway \[for review, see Ref. ([@B9])\]. Rigorous testing of these hypotheses, as well as screening for potential therapeutic treatments, requires a reliable animal behavioral paradigm that not only identifies animals with tinnitus and/or hyperacusis but also allows for the use of invasive techniques, such as electrophysiological recordings from the brain and neuroanatomy, which are inappropriate for use in human patients. In order to identify and investigate potential underlying mechanisms of tinnitus and hyperacusis, a number of animal models have been developed ([@B10]--[@B15]). Since an animal cannot directly communicate its subjective experiences, behavioral paradigms that extrapolate an animal's perception based on changes in behavioral performance have been devised to indicate whether an animal is experiencing tinnitus and/or hyperacusis. Such paradigms have utilized a number of behavioral training techniques including lick or lever pressing suppression ([@B10]--[@B14]), two-choice operant conditioning ([@B16]--[@B18]), and reflex modification ([@B15]). Ultimately, any behavioral model of hyperacusis or tinnitus should closely mirror what we know about these disorders in the human population. When evaluating animal models of tinnitus and hyperacusis, a number of important factors should be considered, including whether the method, time-course, and variability of tinnitus or hyperacusis induction, as well as any measures of pitch or loudness, are consistent with evidence from human tinnitus/hyperacusis patients. Furthermore, behavioral paradigms should be resistant to confounding influences, such as hearing loss, that often accompany noise or drug-induced tinnitus/hyperacusis. Thus, the goal of this review will be to evaluate current animal behavioral models of tinnitus and hyperacusis with a focus on the most widely used and newest paradigms in the literature. For each paradigm, a brief summary will be provided as well as a discussion of the paradigm's major advantages and disadvantages. Important factors such as consistency with the human condition and resilience to the secondary effects of drug or noise exposure will also be discussed. Although previous reviews have thoroughly evaluated many of the proposed animal models of tinnitus ([@B19]--[@B25]), to our knowledge, this review will be the first to incorporate evaluations of animal models of hyperacusis, as well as models of tinnitus, which is necessary given the frequent co-occurrence of these two disorders. Human Studies of Tinnitus {#S2} ========================= In order to evaluate animal models of tinnitus, it is important to have an understanding of the key characteristics of tinnitus in the human population. Much of what we know about the features of tinnitus comes from subjective descriptions by tinnitus patients. Studies in which tinnitus is induced in individuals following exposure to loud sound or ototoxic agents, as well as studies of individuals with long-standing tinnitus, provide us with information regarding tinnitus pitch, intensity, time course for onset following exposure, and variability of induction. Measurements of tinnitus pitch are commonly conducted using pitch matching techniques in which individuals are presented with tones of varying frequency and asked to select the tone that is closest in pitch to their tinnitus \[for a methodological review, see Ref. ([@B26])\]. A number of studies have demonstrated a link between the pitch of tinnitus and the configuration of an individual's hearing loss. Tinnitus pitch measurements have been made in individuals immediately following exposure to loud sounds ([@B27]--[@B29]) and in individuals with long-standing tinnitus ([@B30]--[@B33]). Following acute exposure to loud sound, the pitch of tinnitus was found to occur above the frequency of the noise exposure, either in the high-frequency edge of a sharply localized hearing loss ([@B27]) or close to the region of maximum hearing loss produced by the noise exposure ([@B28], [@B29]) (Figure [1](#F1){ref-type="fig"}A). Similarly, in individuals with long-standing hearing loss and tinnitus, the tinnitus pitch was matched either to the frequencies at the edge of the hearing loss ([@B31]), or to the frequency region of maximal hearing loss ([@B32]) (Figure [1](#F1){ref-type="fig"}B). Tinnitus pitch matching measures have also been completed in individuals exposed to sodium salicylate, a drug known to reliably induce temporary hearing loss and tinnitus in humans and animals when taken at high doses \[for review, see Ref. ([@B34], [@B35])\]. Although some studies have matched the pitch of salicylate-induced tinnitus and hearing loss across a broad range of frequencies, salicylate-induced tinnitus and hearing loss are both most commonly reported to occur at the high frequencies ([@B34]). In general, the pitch of tinnitus resulting from acute noise exposure or long-standing hearing loss commonly occurs within the region of hearing loss and above the frequency of the noise exposure used to induce hearing loss ([@B36]). However, measurements of tinnitus pitch are complicated by findings that tinnitus pitch also depends on the etiology of the tinnitus ([@B37]) and that tinnitus pitch tends to be lower in frequency for individuals with normal audiometric profiles, i.e., normal thresholds at 250--8000 Hz ([@B30]). ![**Measurements of tinnitus pitch in human subjects with acute noise-induced tinnitus or long-standing tinnitus**. **(A)** Following acute noise exposure, tinnitus pitch has been found to occur above the frequency of the noise exposure close to the region of maximal hearing loss generated by the noise exposure \[from Ref. ([@B29]) with permission; NIST, noise-induced short duration tinnitus; MTS, maximum threshold shift\]. **(B)** Similarly, tinnitus pitch has been matched to frequencies in the region of maximum hearing loss for individuals with long-standing tinnitus \[from Ref. ([@B32])\].](fneur-05-00179-g001){#F1} Loudness matching has also been conducted on individuals experiencing tinnitus following exposure to loud sounds or sodium salicylate, as well as in individuals with long-standing hearing loss and tinnitus. The median intensity of tinnitus resulting from acute exposure to loud noise was found to be 9 dB SL, whereas the tinnitus resulting from sodium salicylate was matched at 5--15 dB SL ([@B29], [@B34]). In individuals with long-standing tinnitus, the intensity reported by individuals ranges from 7.1 to 18.9 dB SL ([@B31]). However, loudness recruitment, steeper than normal loudness growth functions found in patients with hearing loss ([@B5]), can contaminate tinnitus-loudness matches by underestimating the perceived loudness of tinnitus ([@B26]). Additional factors to consider regarding tinnitus in the human population are the time course for onset following exposure to noise or sodium salicylate and the variability of its induction. For individuals exposed briefly to loud noise, tinnitus onset is reported to occur immediately following the exposure ([@B28], [@B29]). Given the short duration of noise exposure in some of these studies (5 min), tinnitus was reported to last for nearly 15 min following the exposure ([@B29]). However, following exposure to a high dose of sodium salicylate, tinnitus onset occurs between 1 and 3 h post exposure and typically dissipates within 1--2 days following the exposure ([@B34]). Despite the differences in tinnitus onset for acute noise exposure and salicylate, both methods result in variability in the induction of tinnitus. Following acute noise or salicylate exposure, not all individuals develop tinnitus ([@B28], [@B29], [@B34]). Interestingly, some individuals with a preexisting hearing loss exposed to high concentrations of salicylate failed to experience tinnitus, suggesting an individual variability for susceptibility to salicylate-induced tinnitus in this population ([@B34]). Thus, tinnitus pitch, intensity, onset, and variability in the human population are important factors to consider when evaluating animal models of tinnitus. Animal Models of Tinnitus {#S3} ========================= Jastreboff {#S3-1} ---------- The first behavioral model of tinnitus in animals was developed by Jastreboff et al. ([@B10]). In this *conditioned lick-suppression paradigm*, rats were trained to lick for water during periods in which a steady background noise was present, and to suppress their licking during brief periods of silence (conditioned stimulus), which were followed by a foot-shock (unconditioned stimulus). In the initial study, tinnitus was induced using an injection of sodium salicylate following training. During the testing phase, the foot-shock was turned off resulting in the eventual extinction of the lick-suppression behavior. The rate of extinction was assessed over multiple test days and was used as an indicator for the presence of tinnitus; specifically, animals given the tinnitus inducer salicylate began licking during the silent intervals earlier than animals given saline presumably because they heard their tinnitus during the quiet intervals. The rapid extinction of the lick-suppression behavior in the salicylate-treated rats was interpreted as the presence of tinnitus because animals with tinnitus do not experience silence and were expected to behave as if a sound is being presented. A series of important controls were performed using this paradigm to ensure that the observed behavior was representative of tinnitus and not another confounding factor associated with salicylate administration. First, to demonstrate that the observed results were auditory-specific, a light stimulus was used in place of the background sound and animals were trained to suppress licking when the light was turned off. Salicylate administration following training with the light stimulus had no effect on the behavior, indicating that the effects of salicylate are auditory-specific. The effects of salicylate on thirst and motivation were also controlled by administering salicylate during training as well as during testing. Rats given salicylate during training associated their tinnitus perceived during the silent intervals with the foot-shock and suppressed their licking during the silent (tinnitus) intervals during testing. Salicylate administration did not result in a general effect on drinking behavior since rats administered salicylate during training *decreased* their licking during silent intervals whereas rats administered salicylate after training *increased* their licking during silent intervals. Furthermore, hearing loss as a confounding factor was also investigated by decreasing the intensity of the background sound, which did not affect the behavior. The conditioned lick-suppression paradigm offers a number of advantages including its relatively short training time and the observation that it is not affected by confounding factors related to tinnitus induction such as hearing loss and non-auditory effects of salicylate. However, this lick-suppression paradigm is not useful for long-term studies of tinnitus because the behavior extinguishes. Since the animals are tested in extinction over a period of several days, they no longer remain under stimulus control when the shock is turned off. Additionally, the paradigm requires comparison of groups of animals (tinnitus versus control) and has not been used for assessing the presence of tinnitus in individual animals. Bauer and Brozoski {#S3-2} ------------------ As mentioned above, one limitation of the Jastreboff lick-suppression paradigm was its inability to test for long-term tinnitus in rodents. Motivated by the need to test pharmaceuticals for treating tinnitus, Bauer and Brozoski developed an aversive conditioning behavioral paradigm derived from the Jastreboff model in order to provide long-term quantitative and qualitative assessment of tinnitus perception in rats ([@B11]). Similar to the Jastreboff model, Bauer and Brozoski developed a shock-avoidance paradigm in which rats were trained to discriminate sound (white noise or tones of various frequencies and intensities) from silence (0 dB SPL). Initially, the rat's behavior was shaped to frequently press a lever while a white noise or a tone was presented. This behavior was reinforced by a variable interval of reinforcement with a food pellet. One minute silent periods interrupted the white noise and were followed by a brief foot-shock if the lever was pressed. This procedure quickly trained the rat to avoid pressing the lever only during the silent periods. Following initial training, the foot-shock was turned on infrequently, occurring approximately once per week per rat. A lever suppression ratio, $R = \frac{B}{\left( {A + B} \right)},$ was used to quantify whether or not the number of lever presses during the current 1 min period, *A*, differed from the immediately preceding period, *B*. A value of *R* = 0.0 indicated complete suppression of lever pressing (i.e., rats reported no sound was present), whereas a value of *R * = 0.5 indicated no suppression of lever pressing compared to the previous white noise stimulus (i.e., a sound was present). Behavioral performance on this task has been observed following chronic exposure to either sodium salicylate in drinking water ([@B11]) or acute unilateral noise trauma ([@B38]). In the initial study using sodium salicylate, blood-serum levels of salicylate were similar to those measured in humans with salicylate-induced tinnitus. Rats treated with salicylate demonstrated no difference in lever pressing during silent intervals compared to control animals, but did demonstrate higher *R* values (more lever pressing) to tone stimuli, with the maximum increase in *R* value occurring with the 15-kHz tonal stimulus. This was interpreted to represent the presence of tinnitus in the salicylate-treated subjects, as a maximal interaction was expected to occur at tonal frequencies that most closely resembled the tinnitus frequency. The explanation for the behavioral shift to tonal stimuli was that the salicylate-treated subjects heard the tones differently than control subjects and perceived the tones as more noise-like due to their tinnitus. In other words, the tinnitus percept was expected to interact with the perception of the externally presented tonal stimuli to produce a "noisy" percept if the tones were similar to the tinnitus pitch. Since the animals were trained to press during noise stimuli, the animals were expected to press more often when the test tones interacted with their tinnitus to create this noise-like percept. The tinnitus-like behavior was reversed in the experimental group soon after treatment with salicylate was stopped. In a subsequent study ([@B38]), a 1- or 2-h unilateral traumatizing noise exposure centered at 16 kHz was used to induce tinnitus in rats trained on this behavioral paradigm. Following the 1-h noise exposure, the maximum shift in *R* value occurred during testing sessions where the 20-kHz test tone stimulus was presented. However, unlike the previous study in which salicylate-induced *increases* in *R* value were interpreted as the evidence of tinnitus, unilateral noise exposure resulted in a significant *decrease* in *R* value for tonal stimuli which was interpreted as the presence of tinnitus. The reduction in *R* value in the noise-exposed animals was reported for up to 17 months following noise exposure, i.e., persistent tinnitus. Importantly, in the same study, the possible confounding effects of hearing loss were controlled by including an additional experimental group, which were not noise-exposed but were outfitted with foam earplugs fixed in the ear canal with cyanoacrylate. The suppression ratio, *R*, of this group was unaffected by the \~40 dB conductive hearing loss due to the earplug. This is strong evidence that the behavioral paradigm is robust to a moderate unilateral conductive hearing loss; however, this does not exclude the possibility that unilateral sensori-neural hearing loss, and the subsequent loudness recruitment, may be a confounding factor. There are a few notable strengths of this behavioral paradigm for the assessment of tinnitus in rats including the ability to test subjects over long periods of time, its resilience to unilateral conductive hearing loss, ability to determine tinnitus pitch, and the rigor through which the paradigm has been tested by its creators. One limitation of the paradigm is that the results are presented as a mean of all animals performing within a group. While this approach may be effective and appropriate for testing the viability of various pharmaceuticals using group statistics, it is less effective in assessing tinnitus in an individual subject, and the time course of tinnitus onset. Another disadvantage of the paradigm is that since only one tone frequency is presented during each session, many testing sessions are required to determine the pitch of tinnitus. Additionally, using this paradigm, tinnitus-like behavior does not appear until weeks following exposure to unilateral noise trauma, a result at odds with the human literature in which tinnitus onset typically occurs immediately following exposure to intense noise ([@B27]--[@B29]). Furthermore, there is little evidence in the human literature indicating that tinnitus interferes with the perception of external tones as suggested by the authors. Indeed, external sounds \~5--15 dB above thresholds may be quite effective in suppressing tinnitus ([@B26]). Heffner {#S3-3} ------- In an attempt to improve upon the Jastreboff lick-suppression paradigm, Heffner and Harrington ([@B12]) trained hamsters in a *conditioned suppression/avoidance procedure* to drink in the presence of a broad-band noise or various tones, and to stop drinking in the absence of these sounds (silence) to avoid a shock. Although the lick-suppression methods were similar, two crucial differences exist between the Jastreboff and Heffner paradigms. For one, the animals in the Heffner conditioned suppression paradigm underwent extensive training in the hopes of testing individual animals for tinnitus. Jastreboff, on the other hand, took less time to train each animal but could only assess groups of animals for tinnitus. Another key difference between the two paradigms is that the shock was avoidable in the Heffner conditioned suppression paradigm whereas the electric shock was unavoidable in the Jastreboff paradigm. Despite these differences, the hypothesis for both procedures remained the same. Namely, when the shock was turned off during testing, animals with noise-induced ([@B12]) or salicylate-induced ([@B10]) tinnitus were hypothesized to extinguish faster than animals without tinnitus, because animals with tinnitus no longer experience silence, and therefore, should not know when to suppress their licking. Although tinnitus could be assessed in individual animals, the major drawback of the Heffner conditioned suppression paradigm was that a significant overlap in performance existed between the control and noise-exposed animals ([@B25]). Moreover, since this paradigm used extinction as its behavioral measure of tinnitus, this paradigm cannot be used to detect chronic tinnitus ([@B21]). For these reasons, Heffner and colleagues developed another tinnitus paradigm using a *two-choice sound localization procedure*. During the sound localization procedure, animals were trained to lateralize sounds by responding to the right side of a test box for sounds coming from a speaker on the right side and to respond to the left side of a test box for sounds coming from the left side ([@B16], [@B39]). Animals were given water reward for correct responses and were shocked for incorrect responses. Importantly, silent trials, or probes, were interspersed on \~24% of trials, which were neither reinforced nor punished but the animals were forced to choose a side. The side preference during silent trials was determined for each animal prior to noise exposure. Tinnitus was induced by sound-exposing the ear opposite each animal's side preference during silent trials ([@B16]). For instance, an animal with a right-side bias for silent trials during training would be given a left ear sound exposure. Accordingly, animals with tinnitus should switch their side preference during silent trials from the side preferred prior to noise exposure (the right side in the above example) to the opposite, noise-exposed side (the left side in the above example) because the animals now hear a phantom sound on that exposed side. In accordance with their hypothesis, the researchers found that hamsters ([@B16]) and rats ([@B39]) will shift their side preference on silent trials to the noise-exposed, previously non-preferred ear, suggesting that they perceive a phantom sound in that ear. As an important control condition, simply plugging one ear and producing a conductive hearing loss does not result in a shift in behavior on silent trials. However, the key assumption of this paradigm is that exposing one ear to loud sound will always induce tinnitus lateralized to that ear and never produce bilateral tinnitus or tinnitus in the opposite ear. Therefore, a major drawback of this paradigm is that it cannot be used to test drug-induced tinnitus or binaural noise exposures that would likely induce bilateral tinnitus. Yet, the advantage of using a two-choice paradigm to detect tinnitus is that animals with tinnitus make a *qualitatively* different response than animals without tinnitus, whereas animals with tinnitus in lick-suppression paradigms lick more or less than animals without tinnitus, resulting in *quantitative* differences between the two groups that may be the result of other factors, such as hearing loss, that accompany drug or noise-induced tinnitus ([@B12]). In other words, animals perceiving tinnitus in a two-choice paradigm go to a different side, or press a different lever, than animals without tinnitus, whereas animals perceiving tinnitus in suppression paradigms lick more or less than animals without tinnitus but still perform the same licking behavior in both cases. Since tinnitus animals in two-choice tasks make a qualitatively different response than non-tinnitus animals, the behavior in two-choice experiments is more resistant to changes in motivation, stress, hearing loss, or hyperacusis that frequently co-occur with drug or noise-induced tinnitus ([@B12], [@B17]). Rüttiger {#S3-4} -------- Rüttiger and colleagues have developed a water-reinforced conditioned avoidance paradigm for rats with the goal of limiting the need for long periods of water deprivation as well as unavoidable shock ([@B13]). Animals are trained to shuttle between two water spouts during the presentation of a 70-dB SPL white noise background sound in order to receive a reward of 3% sugar water. During silent periods, however, the animals receive a mild foot-shock if they access the water spouts. The animals are trained over a period of weeks until their responses to the water spouts during silent periods are sufficiently suppressed compared to the responses during white noise background sound presentation. Importantly, a variable reinforcement rate is introduced during the final stages of training in order to reduce extinction of the responses, given that both the reward (sugar water) and foot-shock are turned off during the testing phase. Animals with tinnitus are expected to increase their responses at the water spouts during silent periods indicating that they are experiencing a phantom sound. This paradigm has been used to determine the presence of both salicylate and noise-induced tinnitus ([@B13], [@B40]). It has also been used to determine the intensity of tinnitus by comparing the response rate of rats treated with salicylate to that of animals in which the background sound was presented at varying intensities, and was estimated to be \~30 dB SPL. This paradigm has a number of advantages. As mentioned previously, it does not require long-term deprivation of water or presentation of an unavoidable foot-shock. It can be used to test individual animals for tinnitus by injecting saline, as a control, or salicylate in the same animals on different days and comparing behavioral performance. However, although trained animals are reported to go 6--8 months without training and still perform the task to criterion, it cannot be used to test animals repeatedly for tinnitus over long durations due to extinction of the response when reward/punishment is turned off during testing and when persistent tinnitus is present following noise exposure. Furthermore, although the paradigm has been used to determine the intensity of tinnitus, it has not been used to determine the frequency of tinnitus. Lobarinas {#S3-5} --------- Lobarinas and colleagues introduced a schedule-induced polydipsia avoidance conditioning (SIP-AC) paradigm to assess rats for salicylate-induced tinnitus ([@B14]). This paradigm differed from the previously mentioned lick-suppression paradigms in two ways. For one, the animals were not water restricted, but were food restricted and trained to lick a water spout while waiting for a food pellet to drop into a trough. In the SIP-AC paradigm, animals would receive one food pellet per minute and would consume water, even though they were not water deprived, while waiting for another pellet to drop. This behavior is referred to as polydipsia because the animals drank the water even though they were not water deprived ([@B14]). Second, unlike the other lick-suppression paradigms where the animals were trained to stop licking in the absence of sound (quiet), animals in the SIP-AC paradigm were trained to stop licking in the presence of sounds. In other words, animals could lick for water in quiet but were shocked for licking during sound trials. Therefore, if animals experience tinnitus in the SIP-AC paradigm, they should cease licking during quiet trials because they now hear a sound. The benefit of this procedure is that the shock never has to be turned off because an animal with tinnitus should not lick during sound or quiet trials, and therefore, would not get incorrectly shocked. Since the shock is never turned off, extinction is not a problem with this paradigm; therefore, it can potentially be used to measure chronic tinnitus. However, like all of the previous lick-suppression paradigms, SIP-AC is not robust to changes in motivation or hearing loss that accompany salicylate and/or noise-induced tinnitus ([@B23]). For instance, an animal given a large dose of salicylate might become sick and less motivated to drink overall regardless of whether it has tinnitus or not. Similarly, an animal with a substantial drug or noise-induced hearing loss might mistakenly lick during a sound trial, perceiving it as a quiet trial, and receive a shock. Since the animal is not water deprived, the animal might stop licking altogether to avoid the shock. Therefore, an animal with hearing loss could conceivably test positive for tinnitus ([@B23]). Sederholm and Swedberg {#S3-6} ---------------------- Recently, Sederholm and Swedberg ([@B17]) trained rats in a *two-choice operant conditioning procedure* to identify rats with salicylate and noise-induced tinnitus. For this procedure, each rat was trained to press a "tone" lever when it heard a tone and to press a "0 Hz" (silence) lever when no sound was presented. Correct responses were rewarded with food and incorrect responses resulted in a reset of the fixed ratio requirement (20 additional lever presses) on the appropriate lever. After the animals were trained to criterion, the animals were tested following salicylate administration or intense noise exposure in a quiet chamber where their lever presses to the "tone" lever were counted ([@B17]). Animals with tinnitus should have a greater number of "tone" lever presses than "0 Hz" lever presses during testing even though no sound is presented. The benefits of this procedure are that no shock is required to train the animals; and, given that this is a two-choice paradigm, it is more resistant to confounding factors in tinnitus induction, such as hearing loss, hyperacusis, motor impairment, and loss of motivation, since the animals make a qualitatively different response when they experience tinnitus ([@B12], [@B17]). However, this procedure requires extensive training (2--3 months) and the authors could only train their animals to criterion using high stimulus levels (55--65 dB SL) that are likely much higher than the perceived tinnitus intensity. Furthermore, since the rats were always reinforced for responding during testing, it is not clear how long the animals remain under stimulus control. For instance, after several testing sessions, it is possible that the animals would randomly press either lever since both levers result in food reward, making this paradigm problematic for studies of chronic tinnitus ([@B17]). In addition, lever pressing during testing appears highly variable across animals making tinnitus assessment difficult (Figures [2](#F2){ref-type="fig"}A,B). However, despite these criticisms, this is a new tinnitus paradigm that, with more testing, might prove useful in future tinnitus studies. ![**Rats trained in a two-choice operant conditioning paradigm to press a "tone" lever when tones were presented or a "0 Hz" (silence) lever during periods in which no sound was presented in order to receive a food reward**. Changes in tone lever pressing during silent intervals were observed following exposure to salicylate **(A)** or unilateral acoustic trauma **(B)**. Thin lines represent individual animal data while bold lines depict group data. An increase in tone lever pressing during silence was used to indicate the presence of tinnitus following exposure \[from Ref. ([@B17]), with permission\].](fneur-05-00179-g002){#F2} Stolzberg {#S3-7} --------- A central goal of tinnitus research is to identify a putative neurophysiological correlate of tinnitus perception. Major advances have been made toward this objective in electrophysiological studies in human tinnitus patients ([@B41]). While many reports exist on neurophysiological changes following induction of tinnitus in animal models, only a small subset have done so with behavioral confirmation of tinnitus in alert animals without the influence of anesthesia \[e.g., Ref. ([@B42])\]. In an attempt to overcome this issue, Stolzberg and colleagues developed a novel appetitive *two-alternative forced choice* assay, which is better suited to investigate possible neurophysiological correlates of tinnitus in an animal model by allowing neural activity to be recorded while the animal is actively exhibiting tinnitus-like behavior ([@B18]). Stolzberg and colleagues trained rats to access a left feeder port in the presence of a steady, unmodulated narrow-band noise (NBN; 1/8th octave band-pass noise with a frequency center selected randomly across trials), and to access a right feeder port in the presence of a sinusoidally amplitude modulated noise (AM; broad-band noise with amplitude modulated 100% at 5 Hz) or silence (Quiet) (Figure [3](#F3){ref-type="fig"}B). One of the three acoustic conditions (NBN, AM, Quiet) was continuously present in the testing chamber at the start of each trial. Rats initiated a trial by nose-poking into a center port (Figure [3](#F3){ref-type="fig"}A) and maintained this position for a randomized period of 4--8 s until a light cued them ("go cue") to respond to a feeder port based on the acoustic condition. Correct responses were reinforced with a food pellet and incorrect responses resulted in a "time-out" in which the rat was unable to initiate a new trial. During training, the reinforcement rate was reduced from 100 to 70% in order to minimize extinction of the learned behavior, and the percentage of trial types was divided evenly between the two feeders (NBN at 50%; AM at 30%; Quiet at 20%). ![**Illustration of behavioral tinnitus assay reported in Ref. ([@B18])**. **(A)** One of the three conditions (AM, NBN, Quiet) is present. The rat self-initiates a trial by nose-poking in the center port. **(B)** After a variable delay, a light cues the rat to respond to a feeder trough located to the left or right of the center port. The NBN condition is paired with the left feeder trough. AM and Quiet conditions are paired with the right feeder trough. **(C)** Following induction of tinnitus, if a rat hears a steady phantom sound it should respond to the left feeder during the Quiet condition while still correctly identifying AM and NBN conditions. **(D)** Comparison of performance of rats (*n* = 7; each circle--square pair represents one rat) between saline and salicylate treatments. Following saline treatment AM and Quiet conditions were infrequently misidentified as NBN, whereas NBN was identified correctly. Following salicylate treatment Quiet conditions were significantly more likely to be misidentified as NBN, indicating that salicylate induced a phantom sound perception (ns = not significant, *p* \< 0.001) \[from Ref. ([@B18]), with permission\].](fneur-05-00179-g003){#F3} On testing days, in which rats received either an injection of saline or a high dose of salicylate to induce tinnitus, Quiet trials were neither reinforced nor punished. Tinnitus-like behavior was indicated by the rat shifting its response during Quiet trials from the right feeder associated with AM and Quiet during training, to the left feeder associated with the steady NBN (Figure [3](#F3){ref-type="fig"}C). Following treatment with salicylate, rats incorrectly identified the Quiet condition as a NBN significantly more than baseline or saline, indicating the presence of a steady NBN-like phantom sound (i.e., tinnitus) during Quiet trials (Figure [3](#F3){ref-type="fig"}D). Importantly, this change in behavior during Quiet trials was only observed following injection of salicylate and not following injection with saline. Furthermore, the rats still correctly identified AM and NBN stimuli suggesting that they were not performing randomly (Figure [3](#F3){ref-type="fig"}D). This novel tinnitus behavioral assay has some distinct advantages over other tinnitus paradigms, including the ability to identify tinnitus-like behavior in individual animals and the specific design for simultaneous acquisition of electrophysiological data. During the 4--8 s period in which the animals hold their head in the center port to initiate a trial, neural activity from chronically implanted electrodes can be recorded with minimal artifact when the animal is largely immobile with its head in a fixed position in the sound field. In addition, this paradigm is very robust to the secondary effects of salicylate-induced tinnitus, such as hearing loss, hyperacusis, and hyper-reactivity, because the rats still maintain correct performance on AM and NBN stimuli so it is clear that the animals are under stimulus control. Furthermore, as in the other two-choice paradigms ([@B16], [@B17]), animals with tinnitus make a *qualitatively* different response than animals without tinnitus, unlike in the suppression paradigms where it is difficult to differentiate tinnitus from hearing loss, stress, or other factors associated with drug or noise exposure ([@B12]). However, it is uncertain if the paradigm in its present form is appropriate for assessing chronic tinnitus. Additionally, this paradigm does not provide information regarding tinnitus pitch or loudness. Turner (gap pre-pulse inhibition) {#S3-8} --------------------------------- In 2006, Turner and colleagues introduced a novel tinnitus behavioral paradigm, referred to as gap pre-pulse inhibition of the acoustic startle reflex (GPIAS), which utilizes an animal's motoric response (startle reflex) to a sudden loud sound (startle stimulus) that is recorded by a motion sensitive transducer ([@B15]). Presentation of the acoustic startle stimulus evokes a robust acoustic startle reflex (ASR); however, this reflex can be suppressed by insertion of a short duration silent gap in a continuous background sound just prior to the startle-eliciting stimulus ([@B15], [@B43]). In most studies, the ratio between the startle amplitude during trials in which the startle stimulus is presented alone (no-gap trials) and trials in which a gap is presented prior to the startle-eliciting stimulus (gap trials) is calculated as the GPIAS ratio. This ratio is used as an indicator of the effectiveness of the silent gap to inhibit the startle reflex. For an animal with tinnitus, it is expected that if the background sound in which the gap is embedded is qualitatively similar to the animal's tinnitus, then tinnitus will 'fill in' the gap resulting in an impaired ability of the silent gap to inhibit the startle reflex. By comparing the ability of silent gaps in continuous background sounds of varying frequency and bandwidth to inhibit the startle reflex, the paradigm has been used to determine the pitch of an animal's tinnitus. Using this paradigm, tinnitus has been assessed following exposure to salicylate as well as noise in a variety of species including rats, mice, guinea pigs, and hamsters ([@B15], [@B44]--[@B47]). GPIAS has quickly become the most widely used tinnitus behavioral paradigm because it carries a number of advantages over the other previously reported tinnitus paradigms. It requires no behavioral training, no food or water deprivation, can assess tinnitus pitch, and allows for high-throughput screening for tinnitus. Because, there is no training involved, it can also be used to monitor animals for tinnitus repeatedly over long durations. In Turner's original publication, group data was presented since baseline GPIAS measures were not completed prior to tinnitus induction via unilateral noise exposure ([@B15]). However, by collecting baseline and post-tinnitus induction GPIAS measures, the paradigm can be used to identify individual animals with tinnitus allowing for animals to be separated into tinnitus-positive and tinnitus-negative groups ([@B48], [@B49]). However, despite these advantages, a number of concerns have recently been raised regarding the GPIAS paradigm and its use for screening animals for tinnitus. One concern is the discrepancy in tinnitus pitch reported in animals using the GPIAS paradigm following induction of tinnitus via exposure to high-frequency noise. Some studies, including the original Turner study, have reported the tinnitus pitch to fall below the noise exposure frequency ([@B15], [@B50]), while others report the tinnitus pitch to fall above the noise exposure frequency ([@B44], [@B48], [@B51]--[@B53]). Equally interesting is the finding that immediately following noise exposure, tinnitus has been found to occur across a wide range of frequencies but then becomes specific to a limited frequency band over the following weeks ([@B54]). In contrast, the tinnitus pitch in human subjects exposed to loud sound most frequently occurs at or above the noise exposure frequency ([@B28], [@B29]). Another issue is the effect of hearing loss following exposure to noise or ototoxic drugs on the startle reflex amplitude used to assess pre-pulse inhibition. Hearing loss can potentially affect the outcome of GPIAS screening in a number ways: by interfering with audibility of the background sound in which the silent gaps are imbedded or by altering the amplitude of the startle reflex to the startle stimulus alone (no-gap condition). Previous studies in both rodents ([@B55]) and humans ([@B56]) have demonstrated that hearing loss alone, induced by sodium salicylate exposure, can interfere in detection of gaps in low-level continuous noise. This issue has been addressed in the GPIAS paradigm by using intensities of background sounds (60 dB SPL) shown to be resilient to the effects of hearing loss, and by carrying out noise-burst pre-pulse detection measures. During noise-burst detection measures, a short duration noise-burst of the same intensity as the background sound used in GPIAS testing is presented prior to the startle-eliciting stimulus to serve as the pre-pulse cue. It is assumed that if the noise-burst reliably inhibits the startle reflex, then audibility of the background sound in which the silent gaps are embedded during GPIAS testing should not be an issue. In addition to the potential confounding effects of hearing loss on audibility of the background sound, another issue is that unilateral noise exposure can reduce the amplitude of the startle reflex during startle-alone (no-gap) trials ([@B2], [@B44], [@B47], [@B50]). In one study using rats, unilateral noise exposure resulted in a 57% reduction in the startle amplitude during startle-alone trials ([@B2]), while in another study using mice, unilateral noise exposure resulted in a 52% reduction of the acoustic startle reflex even after hearing thresholds recovered to pre-noise exposure levels ([@B44]). Alterations in startle reactivity pose a number of issues. First, as the dependent measure in the GPIAS paradigm, a robust startle reflex is needed in order to observe its inhibition. If animals fail to startle following manipulations to induce tinnitus, they will need to be excluded from further analysis, a practice reported in some previous studies using this paradigm ([@B2], [@B50]). Exclusion of animals from analysis not only reduces the high-throughput nature of the paradigm but may also result in the exclusion of animals that actually have tinnitus, but cannot be tested due to the absence of a robust startle reflex. Second, alterations in baseline startle magnitude (i.e., no-gap trials) can potentially confound the interpretation of pre-pulse inhibition measures for both rodents and human subjects ([@B2], [@B57]). Given that GPIAS is calculated as a ratio between the startle amplitude in no-gap versus gap trials, a change in either parameter can result in a change in the GPIAS ratio. Traditionally, it was assumed that a change in the GPIAS ratio indicative of tinnitus was the result of an *increase* in the startle amplitude during gap trials if animals failed to detect the silent gap due to tinnitus "filling-in" the gap (Figures [4](#F4){ref-type="fig"}A,B; Scenario A). However, a change in the GPAIS ratio indicative of tinnitus can also occur if the no-gap-startle amplitude *decreases*, similar to what is seen following unilateral noise exposure (Figures [4](#F4){ref-type="fig"}A,B; Scenario B). Moreover, unilateral conductive hearing loss via an earplug has been shown to result in a false-positive screening for tinnitus in rats as a result of a reduction in startle magnitude during no-gap trials (Figure [4](#F4){ref-type="fig"}C) ([@B2]). Importantly, the false-positive screening for tinnitus could be eliminated by replacing the acoustic stimulus with a multi-modal airpuff stimulus (acoustic and somatosensory stimulation), which was more resilient to the effects of hearing loss (Figure [4](#F4){ref-type="fig"}C). Any changes in the no-gap-startle magnitude post-tinnitus induction need to be accounted for when using the GPIAS ratio as an indicator for tinnitus in order to ensure that GPIAS ratio changes are truly reflective of impaired detection of silent gaps, and not simply due to hearing loss ([@B2]). Close inspection of raw startle amplitudes before and after tinnitus induction, as well as controlling for methodological issues such as stimulus parameters and animal handling is strongly recommended when using and interpreting behavioral measures using the GPIAS paradigm ([@B58]). ![**Effects of alterations in startle magnitude on gap pre-pulse inhibition of the acoustic startle reflex (GPIAS) interpretation**. **(A,B)** A change in startle amplitude for either gap trials (Scenario A, gap-startle amplitude increases) or no-gap trials (Scenario B, no-gap-startle amplitude decreases) can result in changes in the GPIAS-Startle ratio indicative of tinnitus. **(C)** A decrease in no-gap trial startle amplitude, similar to the schematic of Scenario B, following temporary unilateral conductive hearing loss via an earplug has been shown to result in a false-positive screening for tinnitus in rats using the GPIAS paradigm. The false-positive screening for tinnitus could be eliminated by replacing the acoustic startle stimulus with a multi-modal airpuff startle stimulus which was more resilient to the effects of unilateral hearing loss on startle reflex amplitude (from Ref. ([@B2]), with permission; \*indicates significant difference between startle amplitudes in gap versus no-gap trials; ^†^indicates significant differences between startle ratio values between baseline and post earplug measures; n.s. indicates no significant difference).](fneur-05-00179-g004){#F4} In addition to the potential effects of hearing loss on GPIAS measures, another issue is whether the hypothesis that tinnitus "fills in the gap" is accurate. Recent studies in both humans ([@B59], [@B60]) and rodents ([@B61], [@B62]) have addressed this issue. In one GPIAS study, human patients with high-frequency tinnitus were found to have impaired gap detection for gaps presented in both low and high-frequency background stimuli compared to control subjects ([@B60]) (Figure [5](#F5){ref-type="fig"}A). Since these patients with high-frequency hearing loss also had high-frequency tinnitus, GPIAS should only be impaired at high frequencies, not at low frequencies. Because their tinnitus patients were found to have significantly less inhibition of the startle response irrespective of frequency, the authors suggest that the impaired gap detection may be reflective of a more general cortical processing disorder rather than tinnitus "filling in the gaps." ![**Recent studies addressing the "tinnitus gap filling" hypothesis**. **(A)** Gap pre-pulse inhibition of the acoustic startle reflex measures in human tinnitus and control subjects. Tinnitus subjects demonstrate impaired gap detection (lower inhibition) for both low and high-frequency background sounds compared to control subjects \[from Ref. ([@B60]), with permission\]. **(B)** Subjective gap detection ability assessed in human tinnitus and control subjects. Both control and tinnitus subjects had no difficulty detecting 50 ms gaps presented in noise at 15 dB SPL. Narrow-band noises were presented 1 octave above, below or at the matched tinnitus pitch for tinnitus subjects and at 1.2, 8, and 12 kHz for control subjects without tinnitus \[data from Ref. ([@B59])\]. **(C)** Gap detection assessed in rats trained on a go/no-go operant gap detection task to identify silent gaps embedded in continuous broad-band noise (BBN) and 10--20 or 15--17 kHz narrow-band noise (NBN) presented at 60 dB SPL. Salicylate had no significant effect on gap detection, as gap duration thresholds remained below 6 ms \[data from Ref. ([@B62])\].](fneur-05-00179-g005){#F5} In another study, gap detection ability was assessed in human tinnitus patients by asking whether they could perceive 50 ms gaps in 15 dB SL background sounds presented either at their tinnitus pitch, or one octave above and below their tinnitus pitch ([@B59]) (Figure [5](#F5){ref-type="fig"}B). Both control and tinnitus subjects had no difficulty detecting the silent gaps irrespective of background frequency (i.e., tinnitus did not "fill in" the silent gaps at frequencies above, below, or at the tinnitus pitch). It is important to note, however, that a direct comparison in gap detection cannot be made between the tinnitus and no-tinnitus subjects in this study because the two groups were tested with different background stimuli (control subjects did not have tinnitus and therefore could not have background stimuli matched to their tinnitus pitch). Instead, by comparing within the group of tinnitus subjects across frequency, the data demonstrated that tinnitus did not "fill in" or interfere with detection of silent gaps in background sounds at the tinnitus frequency or at frequencies one octave above and below the matched tinnitus frequency (Figure [5](#F5){ref-type="fig"}B, compare gray bars for each frequency). In a similar study, Boyen et al. ([@B63]) recently found that human tinnitus patients had similar gap detection thresholds compared to a matched non-tinnitus control group even when the test frequency matched the patient's tinnitus frequency ([@B63]). In addition to human studies, a number of animal studies have also investigated the "tinnitus filling the gap" hypothesis. Hickox and Liberman ([@B61]) demonstrated that gap detection deficits in noise-exposed rodents tested with the GPIAS paradigm are dependent on the interval between the silent gap and the startle-eliciting stimulus ([@B43]). Noise-exposed animals demonstrated GPIAS deficits only when the silent gap was placed immediately before the startle stimulus, but not when it was placed 80 ms before the startle stimulus. The authors concluded that these results are inconsistent with the "tinnitus filling the gap" hypothesis, as gap detection deficits from tinnitus filling in the gap should be seen irrespective of where the silent gap is placed (i.e., at any placement where GPIAS would normally be observed) ([@B61]). Lastly, gap detection has also been assessed in rodents trained on a go/no-go operant gap detection task to determine the threshold for silent gaps embedded in continuous background sounds ([@B62]) (Figure [5](#F5){ref-type="fig"}C). In this study, rats were treated with a dose of sodium salicylate known to reliably induce tinnitus ([@B18]). Following salicylate administration, gap detection thresholds were unchanged for gaps embedded in broad-band noise or narrow-band noises presented at 60 dB SPL (the same intensity background noise commonly used in the GPIAS paradigm). These results indicate that salicylate-induced tinnitus does not "fill in the silent gaps." Taken together, the results suggest that tinnitus assessment with the GPIAS paradigm should be interpreted with considerable caution. Ultimately, the rationale for using GPIAS as a behavioral test for tinnitus in animals should be based on the ability of the paradigm to accurately assess tinnitus in human patients. Human Studies of Hyperacusis {#S4} ============================ Hyperacusis, defined as a hypersensitivity to moderate-intensity sounds or abnormal loudness perception ([@B4]--[@B7]), often co-occurs with tinnitus ([@B6], [@B64]--[@B66]). The prevalence of hyperacusis in the tinnitus population was estimated to be \~80% ([@B8]). The frequent co-occurrence of these two perceptual disorders suggests a common mechanism(s) of dysfunction ([@B7], [@B64], [@B66]), such as an increase in central gain following hearing loss ([@B67]--[@B69]). Given the high rate of overlap between these two disorders, it is important to discuss hyperacusis when assessing models of tinnitus ([@B24], [@B66], [@B70]). To clarify, hyperacusis is distinct from loudness recruitment, the abnormally rapid growth in perceived loudness with increasing intensity, in that it does not necessarily coincide with threshold elevation and hair cell damage, but does feature reduced loudness discomfort levels ([@B6], [@B61], [@B71]). In addition, hyperacusis is not sound-specific and anxiety can aggravate symptoms ([@B5], [@B71]). Generally, hyperacusis is measured using loudness rating scales because the primary feature of this auditory perceptual disorder is a reduced tolerance for moderate-level and intense sounds ([@B6], [@B7], [@B47]). To assess a listener's sensitivity to sounds, participants are typically instructed to rate sounds according to pre-determined categories of loudness (such as 1 for quiet and 10 for painfully loud) ([@B72], [@B73]). Although these subjective rating scales are useful measures in humans, these methods are impossible to use in animals because they require a listener's ability to follow instructions and adjust, or rate, stimuli accordingly ([@B74]). Therefore, researchers have turned to objective behavioral measures of loudness perception, such as the amplitude of the ASR and operant conditioning techniques using reaction time (RT) measures, as methods for estimating perceived loudness in both animals and humans. Animal Models of Hyperacusis {#S5} ============================ Acoustic startle reflex paradigm {#S5-9} -------------------------------- The acoustic startle reflex (ASR) paradigm has been used by a number of researchers to assess animals for age-related ([@B75]), drug-induced ([@B76], [@B77]), and noise-induced ([@B47], [@B61], [@B78], [@B79]) hyperacusis. According to these studies, an animal is thought to have hyperacusis if the amplitude of its startle reflex, a short-latency, robust motoric response ([@B80], [@B81]), increases after some manipulation, such as an injection of sodium salicylate or a noise exposure ([@B47], [@B77]) (Figures [6](#F6){ref-type="fig"}A,B). Like the GPIAS reflex paradigm for assessing tinnitus, the ASR paradigm is an efficient, high-throughput behavioral method because it requires no training or learning. In addition, the ASR paradigm is attractive because it does not require any food or water restriction or the use of electric shock. ![**Animal models of hyperacusis using the acoustic startle reflex (ASR) paradigm**. **(A)** Mean ASR amplitudes in rats pre-sodium salicylate injection (open circles) and 1 h post-sodium salicylate injection (250 mg/kg i.p.) (open triangles). Startle amplitudes increased significantly at high sound intensities 1 h after salicylate injection \[from Ref. ([@B77]), with permission\]. **(B)** Mean ASR amplitudes in hamsters following noise exposure (10 kHz, 115 dB SPL, 4 h) (black circles) or in unexposed control hamsters (open circles). Startle amplitudes were significantly higher at high sound intensities in noise-exposed hamsters than in unexposed hamsters, suggesting increased loudness sensitivity in noise-exposed animals \[from Ref. ([@B47]), with permission\].](fneur-05-00179-g006){#F6} However, the ASR paradigm can be problematic for several reasons. For one, it is difficult to discriminate hyperacusis from generalized, non-auditory-specific hyperactivity with ASR alone ([@B23], [@B82]). Secondly, enhanced ASRs have been reported to be more predictive of tinnitus rather than hyperacusis in humans, questioning ASR's usefulness in assessing hyperacusis in animals ([@B83]). In addition, it is unclear how the ASR paradigm can be used to differentiate hyperacusis from loudness recruitment. Lastly, changes in the ASR tend to occur only for high-intensity sounds (90 dB SPL or higher), which suggests that this method may not be sensitive to changes in loudness perception at moderate sound levels, one of the defining features of hyperacusis ([@B78], [@B84]). Although these potential limitations need to be addressed, the ASR paradigm appears, for now, to be an effective tool for estimating loudness perception, hyperacusis, or sound-evoked hyper-reactivity in animals at intensities greater than 80 dB SPL. Operant conditioning methods with reaction time measures {#S5-10} -------------------------------------------------------- Since RT is a reliable surrogate of perceived loudness ([@B72], [@B85]--[@B89]), and loudness growth functions appear to be the perceptual correlate of hyperacusis in humans ([@B7]), researchers have obtained RT measures using operant conditioning techniques to assess loudness perception and hyperacusis in both animals and humans ([@B74]). The first study to examine hyperacusis in animals and humans using RT measures was Lauer and Dooling ([@B74]). They measured RTs in both normal-hearing canaries and an inbred strain of canary with a permanent hereditary high-frequency hearing loss, the Belgian Waterslager (BWS). They hypothesized that animals with hyperacusis should show faster than normal RTs because sound stimuli are perceived as being louder than in normal-hearing animals. In accordance with that hypothesis, they found that BWS canaries had near-normal RTs at lower sound levels but much faster RTs than normal canaries at higher sound levels ([@B74]). Importantly, Lauer and Dooling ([@B74]) verified these methods by testing two humans, one with hyperacusis and one without, using the same behavioral methods as in the canary experiment (Figures [7](#F7){ref-type="fig"}A,B). ![**Median reaction times (RT) for a normal-hearing human listener (Subject 1, black circles) and a listener with reduced loudness tolerance (Subject 2, black squares)**. Median RTs for both listeners for **(A)** 1000 Hz tones and **(B)** 4000 Hz tones presented at various intensities. Subject 2 had significantly faster RTs to moderate and high-intensity sounds than Subject 1, suggesting that the participant with hyperacusis perceived these sounds as louder than the normal-hearing participant, and that RT can potentially measure hyperacusis in humans \[modified from Ref. ([@B74])\].](fneur-05-00179-g007){#F7} Recently, Chen et al. ([@B84]) measured RTs in rats both before and after a large dose of sodium salicylate that is known to induce hearing loss and tinnitus in animals ([@B18]). Briefly, the rats were trained to detect broad-band noise bursts in an otherwise quiet chamber using a *Go/no-go operant conditioning paradigm*. RT measures were taken from the onset of the noise burst to the time the rat made a response, and only RTs for "hits" (when the animal correctly detected the stimulus) were included in the analysis ([@B84]). Chen et al. ([@B84]) found that RTs post-salicylate were faster than pre-salicylate for noise bursts 70 dB SPL or greater but were the same or even slower to respond to noise bursts at 50 dB SPL or less ([@B84]) (Figure [8](#F8){ref-type="fig"}A). Importantly, this go/no-go paradigm can differentiate hyperacusis from loudness recruitment. Hyperacusis is evidenced by faster RTs than normal for moderate to intense sounds, while loudness recruitment associated with hearing loss is characterized by slower RTs for low-level sounds with a rapid loudness growth function and normal RTs to moderate to intense sounds (Figure [8](#F8){ref-type="fig"}B). ![**Animal models of hyperacusis using operant conditioning procedures**. **(A)** Mean reaction times (RT) in rats to broad-band noise bursts for baseline (open circles), saline (open triangles), and salicylate (200 mg/kg i.p.) (filled squares) conditions. RTs decreased significantly for 70, 80, and 90 dB SPL noise bursts following the injection of salicylate, suggesting an increased sensitivity to loud sounds \[modified from Ref. ([@B84])\]. **(B)** Hypothetical data indicating hyperacusis (red dashes) and loudness recruitment (green dots) using the go/no-go reaction time model. Top arrow indicates steep loudness growth function seen in listeners with loudness recruitment. Bottom arrow shows faster reaction times for moderate to intense sounds, indicative of increased loudness perception in listeners with hyperacusis. **(C,D)** Mean number of responses to the 100-dB SPL feeder pre-salicylate (open circles) and 1 h post-salicylate (250 mg/kg i.p.) (red triangles) for individual rats. **(C)** The number of responses to the high-intensity (100 dB SPL) feeder increased significantly following salicylate administration (arrow indicates shift to the left), suggesting increased loudness sensitivity and hyperacusis. **(D)** The number of responses to the high-intensity (100 dB SPL) feeder decreased significantly following salicylate administration (arrow shows shift to the right), indicative of loudness recruitment \[from Ref. ([@B90]), with permission\].](fneur-05-00179-g008){#F8} In another operant conditioning paradigm, Sun et al. ([@B91]) tested adult rats with tympanic membrane (TM) damage as pups, and Zhang et al. ([@B90]) tested rats before and after a large dose of sodium salicylate for hyperacusis in a *two-alternative forced choice (2AFC) operant conditioning task*. In both experiments, animals were trained to make a response to the right nose-poke hole when they heard a 100-dB SPL sound and to the left side when they heard a 60-dB SPL sound. After the rats were correctly identifying these two stimuli, probe stimuli were included from 40 to 110 dB SPL that were always reinforced with a food pellet regardless of which side the animals chose ([@B90], [@B91]). They found that rats with TM damage as pups, and 40% of rats given salicylate, labeled mid-intensity stimuli (70 and 80 dB SPL) as 100 dB SPL more often than control rats, suggesting that rats with prior TM damage and rats given salicylate perceived these stimuli as louder than normal (Figure [8](#F8){ref-type="fig"}C). Interestingly, several rats given salicylate appeared to experience loudness recruitment following an injection of salicylate. Instead of responding more frequently to the 100-dB SPL feeder, these rats responded more frequently to the 60-dB SPL feeder, indicating a loss in sensitivity at lower sound levels and an abnormally steep loudness growth function (Figure [8](#F8){ref-type="fig"}D). Although the above-mentioned go/no-go and 2AFC operant conditioning tasks deliver reliable and stable measures in animals and humans, these tasks require weeks of animal behavioral training and the animals need to be food or water-restricted in order to perform in the task. Nevertheless, the RT paradigm has been extensively validated as a measure of loudness ([@B72], [@B85]--[@B89]) and it may provide the most reliable and sensitive estimates of loudness perception in *both* animals and humans. Conclusion {#S6} ========== A reliable behavioral paradigm is vital for identifying the underlying neural mechanisms and potential therapeutic treatments for tinnitus and hyperacusis. Ultimately, any behavioral model of hyperacusis or tinnitus should closely mirror what we know about these disorders in the human population. Given the range of behavioral training techniques and methodologies used to assess tinnitus and hyperacusis, there are a number of important characteristics to consider when evaluating their utility and consistency with human data. An ideal behavioral paradigm should be able to identify the presence of the condition in individual animals, allow for long-term testing of animals for studies of chronic tinnitus and hyperacusis, should be resilient to any secondary effects of the induction method including hearing loss, and would have a relatively short training and testing time. Additionally, an ideal model of tinnitus would also allow for measurements of tinnitus pitch and loudness, while an ideal model of hyperacusis should allow for the differentiation between the presence of hyperacusis and loudness recruitment. Clearly, the advancements in animal models for tinnitus and hyperacusis have come a long way and will continue to play an important role in revealing the underlying mechanisms and treatments for tinnitus and hyperacusis. Author Contributions {#S7} ==================== Sarah H. Hayes, Kelly E. Radziwon, Daniel J. Stolzberg, and Richard J. Salvi wrote the manuscript and approved the final version. Conflict of Interest Statement {#S8} ============================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. Supported in part by NIH National Institute on Deafness and Other Communication Disorders (R01DC011808 and R01DC009219) and the Office of Naval Research (N000141210731). Sarah H. Hayes was supported by the National Defense Science and Engineering Graduate Research Fellowship. [^1]: Edited by: Arnaud Norena, Université de Provence, France [^2]: Reviewed by: Martin Pienkowski, Salus University, USA; Alexander Galazyuk, Northest Ohio Medical University, USA [^3]: This article was submitted to Neuro-otology, a section of the journal Frontiers in Neurology.

Background {#Sec1} ========== Concurrent use of long-acting muscarinic antagonist (LAMA) and long-acting β~2~-agonist (LABA) therapy has been shown to maximize the bronchodilator response \[[@CR1]\], and dual LAMA/LABA combination therapy is now recommended as an alternative option for patients whose chronic obstructive pulmonary disease (COPD) is not well managed with bronchodilator monotherapy \[[@CR2], [@CR3]\]. In addition to providing improvements in airflow limitation and symptom control, dual bronchodilator therapy reduces the risk of adverse effects that may be associated with increased dosages of a single bronchodilator \[[@CR1]\]. The use of a fixed-dose combination (FDC), in which the two agents are combined in a single device, may improve adherence, leading to improved outcomes and reduced costs \[[@CR4], [@CR5]\]. There are currently options for both once-daily (QD) and twice-daily (BID) dosing of LAMA/LABA FDCs \[[@CR6]\], with some evidence suggesting that BID dosing may be preferable for patients who experience night-time symptoms \[[@CR7]\]. However, choice of delivery device has been limited to LAMA/LABA FDCs delivered by dry powder inhalers (DPIs) or Soft Mist™ devices. Thus there is an opportunity to expand patient choice by developing a metered dose inhaler (MDI) formulation-based LAMA/LABA FDC. A Co-Suspension™ Delivery Technology has been developed to overcome the variability and instability associated with drug delivery via traditional hydrofluoroalkane propellant-based MDI devices. This technology uses a novel formulation technique in which active-agent particles (typically in the form of micronized drug crystals) form strong and non-specific associations within the propellant with specially engineered porous microparticles, made from distearoylphosphatidylcholine and calcium chloride. These formulations possess excellent stability and dose uniformity, and allow simultaneous delivery of multiple drugs from one MDI without one drug affecting the delivery of the others, and maintain consistency of dose and aerosol properties between monotherapies and their combinations \[[@CR8]\]. This Phase IIb study investigated the efficacy and safety of a LAMA/LABA FDC MDI, GFF MDI, containing glycopyrrolate (GP; equivalent to the bromide salt, glycopyrronium bromide) and formoterol fumarate (FF) formulated using the Co-Suspension delivery technology, using the dose of glycopyrrolate identified in a dose-response study in patients with moderate-to-severe COPD \[[@CR9]\]. Earlier Phase I/IIa studies evaluated the efficacy and safety of the individual components, using a previous formulation of phospholipid porous particles \[[@CR10], [@CR11]\]. The primary objective of this two-part study was to assess lung function, specifically the improvement in forced expiratory volume in 1 s (FEV~1~) area under curve (AUC) from 0 to 12 h post-dose (FEV~1~ AUC~0--12~). Firstly to assess improvements in FEV~1~ AUC~0--12~ with GFF MDI 72/9.6 μg or 36/9.6 μg BID compared with individual component MDIs, placebo MDI (suspension of porous microparticles only), or the open-label comparators FF DPI and tiotropium bromide DPI in patients with COPD. Secondly, to assess improvements in FEV~1~ AUC~0--12~ reported with FF 9.6 μg MDI versus placebo MDI in patients with COPD. The safety profiles of GFF MDI and FF MDI, as well as the pharmacokinetic (PK) profiles of both glycopyrrolate and formoterol after chronic administration of GFF MDI, were also investigated. Methods {#Sec2} ======= Patients {#Sec3} -------- Patients were 40--80 years of age with a diagnosis of COPD and were current or former smokers, with a smoking history of at least 10 pack-years. Key lung function criteria were: pre- and post-bronchodilator FEV~1~/forced vital capacity (FVC) ratio \<0.7; post-bronchodilator FEV~1~ ≥750 mL, and ≥30% and \<80% of the predicted value at screening; and pre-bronchodilator FEV~1~ ≤ 80% at baseline. Key exclusion criteria were: pregnancy or lactation; respiratory disease other than COPD; poorly managed COPD that had required hospitalization within 3 months of screening, or treatment with corticosteroids or antibiotics within 6 weeks of screening. In addition, patients who did not meet American Thoracic Society criteria for acceptable spirometry were excluded. Patients provided informed consent before undergoing any screening assessments. Study design {#Sec4} ------------ This was a Phase IIb randomized, multicenter, placebo-controlled, double-blind, chronic dosing (7 days), four-period, eight-treatment, incomplete-block, crossover study, conducted in two parts in the USA, Australia, and New Zealand ([NCT01085045](https://clinicaltrials.gov/ct2/show/NCT01085045)). Patients recruited to Part A were not eligible for Part B. The design of the study is depicted in Additional file [1](#MOESM1){ref-type="media"}: Figure S1. Part A was a four-period, eight-treatment, incomplete-block crossover study, designed to evaluate eight treatments: (i) GFF MDI 72/9.6 μg BID; (ii) GFF MDI 36/9.6 μg BID; (iii) GP MDI 36 μg BID; (iv) FF MDI 9.6 μg BID; (v) FF MDI 7.2 μg BID; (vi) placebo MDI BID; (vii) FF DPI 12 μg BID; and (viii) tiotropium DPI 18 μg QD. In this report, GP was expressed as glycopyrrolate (also known as glycopyrronium bromide) for which ex-actuator doses of 36 μg and 72 μg are equivalent to glycopyrronium (active moiety) 28.8 μg and 57.6 μg, respectively. Similarly, FF was expressed as formoterol fumarate, for which the dose of 9.6 μg (ex-actuator) is equivalent to formoterol fumarate dihydrate 10 μg. Each patient received four of eight possible treatments. A given treatment sequence included a GP MDI or an FF MDI component in no more than two treatment periods, whether administered as an FDC or as a single agent. Six combinations of four treatments were chosen for the study and then 48 treatment sequences created. Patients were randomized to one of the 48 treatment sequences which were generated centrally using an Interactive Web-based Response System based on Williams Square layouts. Part B was a four-period, four-treatment, full-crossover study designed to evaluate: (i) FF MDI 9.6 μg BID; (ii) FF MDI 7.2 μg BID; (iii) placebo MDI BID; and (iv) FF DPI 12 μg BID. Patients were randomized to one of 24 possible treatment sequences, which were generated in the same way as Part A. Patients administered each of their four assigned treatments for 1 week, followed by a 7- to 21-day washout period between treatments. All inhalers were dispensed in a blinded manner, with the exception of FF DPI and tiotropium DPI, which were provided open-label. The first dose of study drug was administered at the clinic under the supervision of a study coordinator (patients had been assessed previously for correct use of the MDI by study staff when using an albuterol MDI for the bronchodilator reversibility assessment). Self-administration continued at home. Each dose comprised two MDI actuations. Patients used their study drug BID for 1 week. FF DPI and open-label tiotropium DPI were administered for 7 days, according to the manufacturer's instructions. This study was conducted in accordance with International Conference on Harmonization guidelines, the Declaration of Helsinki and the US Code of Federal Regulations. The protocol, its amendments and patient informed consent form were approved by an Independent Ethics Committee or Institutional Review Board. Efficacy endpoints {#Sec5} ------------------ In both parts of the study, the primary endpoint was FEV~1~ AUC~0--12~ on Day 7 relative to baseline FEV~1~. In Part A, there were two primary comparisons: (i) GFF MDI 72/9.6 μg BID versus GP MDI 36 μg BID and (ii) GFF MDI 72/9.6 μg BID versus FF MDI 9.6 μg BID. To demonstrate efficacy for GFF MDI, superiority to both monocomponent MDIs was required. In Part B, the primary endpoint was based on the comparison of FF MDI 9.6 μg BID with placebo MDI BID. The secondary efficacy endpoints of the study included measurements on both Day 1 and Day 7: peak change from baseline in FEV~1~, time to onset of action (≥10% improvement in FEV~1~ relative to baseline), proportion of patients achieving ≥12% improvement in FEV~1~ relative to baseline, peak change from baseline in inspiratory capacity (IC) and change in morning pre-dose trough FEV~1~ and IC. An exploratory endpoint included change from baseline in FVC. These endpoints provided additional information on the dose-response of bronchodilator effects on lung function for GFF MDI and FF MDI by exploring two doses for each versus active comparators and placebo MDI: GFF MDI 72/9.6 μg and 36/9.6 μg, and FF MDI 7.2 μg and 9.6 μg. Pharmacokinetics {#Sec6} ---------------- PK parameters were derived from the plasma concentrations of glycopyrrolate and formoterol fumarate obtained on approximately Day 7 (Day 7 ± 2) of each treatment regimen during Study Parts A and B. In Part A, the PK profiles of glycopyrrolate and formoterol after chronic administration of GFF MDI were compared with those after chronic administration of the monocomponent MDIs. In Part B, the PK profile of formoterol after chronic administration of two dose levels of FF MDI was compared with those after chronic administration of FF DPI. PK samples were collected at pre-dose, at 2, 6 and 20 min, and at 1, 2, 4, 8, 10 and 12 h post-dose. PK analyses were performed by Pharsight Inc using a validated version of WinNonlin^®^ Enterprise (Version 5.2). Safety evaluations {#Sec7} ------------------ In addition to monitoring adverse events (AEs) and serious AEs (SAEs), the following safety evaluations were performed: 12-lead electrocardiogram (ECG), vital signs, physical examination, clinical laboratory tests, and evaluation for symptoms of AEs of interest including dry mouth, tremor and paradoxical bronchospasm. Statistical analysis {#Sec8} -------------------- Data processing, data screening, descriptive reporting and analysis of the efficacy and safety data were performed using SAS Version 9.2 (SAS Institute, Inc., Cary, NC). The PK data were analyzed using WinNonLin Version 5.2 (Pharsight Corp., USA). PK graphs were prepared using SigmaPlot for Windows Version 9.01 (Systatsoftware, Inc., San Jose, CA). Power calculations were performed using R software. Further details of statistical methods and analysis are detailed in the Additional file [1](#MOESM1){ref-type="media"}. The primary efficacy analysis was based on a modified intent-to treat (mITT) population, defined as patients who completed at least two treatment periods up to at least 2 h post-dose on Day 7 (with no more than one missing data point from the 15-min to the 2-h post-dose timepoint, inclusive); patients whose baseline FEV~1~ at Visits 4, 6 and 8 was not within 15% of baseline FEV~1~ at Visit 2 (reproducibility criteria) were also excluded from the mITT population. A separate population, PK-mITT, was defined for use in the PK analyses. Data from Part A and Part B were combined for analysis using a linear mixed-effects model of the primary endpoint FEV~1~ AUC~0--12~ (baseline FEV~1~ was included as a covariate). The AUC was calculated using trapezoidal integration on the available timepoints. Superiority testing was performed using a two-sided 0.05 level of significance; non-inferiority testing was performed using a 0.025 level of significance based on a one-sided confidence interval (CI). The pre-defined non-inferiority margin for continuous spirometry variables was 100 mL, selected on the basis that it is the minimally clinically significant difference, defined as the change in FEV~1~ that can be perceived by the patient \[[@CR12]\]. As such, non-inferiority was only confirmed for a treatment group if the relevant bound of the two-sided 95% CI for the difference was above −100 mL or below 100 mL. Mean changes from baseline in FEV~1~ were provided with 95% CI to support any conclusions of non-inferiority. Sample size calculations {#Sec9} ------------------------ Calculations to determine adequate sample size were based on the primary endpoint, FEV~1~ AUC~0--12~. For superiority testing of spirometry parameters, a difference of 100 mL was the pre-defined minimally clinically significant difference to be observed in FEV~1~ AUC~0--12~. Combined data from Part A and Part B gave the study the power of approximately 82--95% to detect the minimally clinically significant difference in FEV~1~ AUC~0--12~ between the treatment comparisons of interest. Role of the funding source {#Sec10} -------------------------- The funder of the study was involved in study design, data collection, data analysis, data interpretation and writing of the report. All authors had full access to all the data in the study and the corresponding author had the final responsibility for the decision to submit for publication. No restrictions were placed on authors regarding the statements made in the manuscript. Results {#Sec11} ======= Patient disposition {#Sec12} ------------------- A total of 169 patients were screened and 122 were randomized between 24 March 2010 and 28 October 2010 to receive treatment at sites in Australia, New Zealand and the USA. Following review of data from four sentinel patients, 118 patients were randomized: 68 patients into Part A and 50 patients into Part B (Fig. [1](#Fig1){ref-type="fig"}). All 118 patients received at least one dose of study drug and were included in the ITT population and 104 patients (88.1%) were included in the mITT population. Major reasons for exclusion from the mITT population were: (i) patient did not complete at least two treatment periods up to at least 2 h post-dose on Day 7 (Part A, 10.3% of patients and Part B, 14.0% of patients), and (ii) patient failed reproducibility criteria or had missing data on Day 7 (maximum of 3.4% patients in any one period of the study). There were 82 patients (69.5%) included in the per-protocol population (patients who completed all four treatment periods). The majority of patients (80.5 to 96.2% across treatment groups) received 80 to 100% of their assigned treatment regimen.Fig. 1Patient disposition. In Part A, patients were randomized to receive any of the eight treatments in each of the four periods of the study in an incomplete block crossover design. In Part B, patients were randomized to receive all three formoterol doses and placebo in each of the four periods of the study in a full crossover design. ^a^Five patients met multiple criteria for exclusion from randomization (not meeting inclusion criteria and/or meeting exclusion criteria). ^b^Patients randomized to treatment, who received at least one dose of study drug. ^c^Patients who completed at least two treatment periods with at least 2 h of post-dose data on Day 7, with no more than one missing data-point from 15 min to 2 h post-dose, inclusive. COPD, chronic obstructive pulmonary disease; CT, computed tomography; ITT, intent-to-treat; LRTI, lower respiratory tract infection; mITT, modified ITT Baseline characteristics {#Sec13} ------------------------ Patients' baseline and demographic characteristics are shown in Table [1](#Tab1){ref-type="table"} (mITT population). Briefly, the mean (± standard deviation \[SD\]) duration of patients' history of COPD was 7.7 (±5.9) years; the mean post-bronchodilator FEV~1~ was 50.8 (±12.7) % of predicted; and the mean FEV~1~ bronchodilator reversibility was 16.9 (±14.9) %. Overall, 52.9% (55/104) of patients had moderate COPD, 44.2% (46/104) had severe COPD and 2.9% (3/104) had very severe COPD.Table 1Baseline demographics (mITT population)ParameterGFF MDIGP MDI 36 μg (*N* = 41)Open-label tiotropium 18 μg (*N* = 58)FF MDIPlacebo MDI (*N* = 52)Open-label FF^a^ DPI 12 μg (*N* = 55)72/9.6 μg (*N* = 41)36/9.6 μg (*N* = 43)9.6 μg (*N* = 64)7.2 μg (*N* = 64)Age, years Mean (SD)62.4 (9.4)63.3 (8.3)66.3 (6.1)64.1 (7.9)63.4 (8.9)63.6 (8.9)62.8 (9.6)60.6 (9.0)Gender, *n* (%) Male25 (61.0)24 (55.8)23 (56.1)34 (58.6)34 (53.1)36 (56.3)29 (55.8)34 (61.8)Race, *n* (%) Black/African00003 (4.7)3 (4.7)3 (5.8)3 (5.5) White39 (95.1)42 (97.7)41 (100)57 (98.3)61 (95.3)61 (95.3)48 (92.3)52 (94.5) Australia/New Zealand (indigenous)2 (4.9)1 (2.3)01 (1.7)001 (1.9)0Smoking status, *n* (%) Current16 (39.0)18 (41.9)15 (36.6)24 (41.4)29 (45.3)28 (43.8)24 (46.2)25 (45.5) Former25 (61.0)25 (58.1)26 (63.4)34 (58.6)35 (54.7)36 (56.3)28 (53.8)30 (54.5)Duration of COPD, years Mean (SD)7.6 (7.3)^b^6.2 (5.4)^c^7.8 (6.2)^b^7.4 (6.7)^d^8.6 (6.1)^e^7.7 (4.4)^f^8.3 (5.2)^g^7.3 (4.3)^h^Mean % predicted FEV~1~ (SD) Pre-bronchodilator44.1 (13.9)^b^46.8 (14.1)^c^45.8 (13.5)^b^44.9 (13.9)^d^44.7 (12.6)^e^43.9 (12.0)^f^43.7 (11.6)^g^44.0 (13.3)^h^ Post-bronchodilator50.6 (13.0)^b^53.0 (13.1)^c^51.5 (13.3)^b^51.3 (13.4)^d^51.4 (12.5)^e^50.2 (12.6)^f^51.1 (12.4)^g^50.9 (12.9)^h^Mean FEV~1~, L (SD) Pre-bronchodilator1.33 (0.48)^b^1.38 (0.47)^c^1.30 (0.41)^b^1.33 (0.47)^d^1.29 (0.43)^e^1.28 (0.40)^f^1.30 (0.41)^g^1.35 (0.46)^h^ Post-bronchodilator1.52 (0.47)^b^1.56 (0.47)^c^1.46 (0.40)^b^1.51 (0.46)^d^1.49 (0.46)^e^1.47 (0.43)^f^1.52 (0.47)^g^1.56 (0.48)^h^FEV~1~ bronchodilator reversibility, L (SD)^i^ Mean (SD)17.8 (16.3)^b^16.3 (17.2)^c^14.2 (14.5)^b^17.1 (16.2)^d^17.5 (14.7)^e^15.9 (12.7)^f^18.6 (12.9)^g^18.5 (15.5)^ha^Foradil^®^ Aerolizer^®^; ^b^ *n* = 38; ^c^ *n* = 39; ^d^ *n* = 56; ^e^ *n* = 58; ^f^ *n* = 63; ^g^ *n* = 45; ^h^ *n* = 54; ^i^percentage change from pre-albuterol at 30 min post-albuterol for FEV~1~% = 100 × n/N, where *n* = number of patients in category and *N* = number of patients in the groupDuration of COPD = (date of first dose of study treatment in the study -- COPD onset date)/365.25Data from four sentinel patients were included in the mITT population in the analyses of demographic and baseline characteristics only*COPD* chronic obstructive pulmonary disease, *DPI* dry powder inhaler, *FEV* ~1~ forced expiratory volume in 1 s, *FF* formoterol fumarate, *GFF* glycopyrrolate/formoterol fumarate, *GP* glycopyrrolate, *MDI* metered dose inhaler, *mITT* modified intent-to-treat, *SD* standard deviation FEV~1~ AUC~0--12~ on Day 7 {#Sec14} -------------------------- Figure [2a](#Fig2){ref-type="fig"} shows the least squares mean (LSM) change from baseline in FEV~1~ over 12 h on Day 7. All active treatments were superior to placebo MDI for FEV~1~ AUC~0--12~ on Day 7 (*p* \< 0.0001) (Figure [2b](#Fig2){ref-type="fig"}). GFF MDI 36/9.6 μg was non-inferior to GFF MDI 72/9.6 μg in FEV~1~ AUC~0--12~ on Day 7 since the upper bound of the CI was \<100 mL (LSM difference between treatments = 0.008 L; 95% CI = −0.039, 0.054 L). GFF MDI 72/9.6 μg and GFF MDI 36/9.6 μg each demonstrated superior bronchodilation of 101 to 124 mL compared with their individual component MDIs, GP MDI 36 μg, FF MDI 9.6 μg and FF MDI 7.2 μg, as well as superior bronchodilation compared with the open-label comparators FF DPI and tiotropium DPI (*p* ≤ 0.0002) for FEV~1~ AUC~0--12~ on Day 7 (Table [2](#Tab2){ref-type="table"}). GP MDI 36 μg, demonstrated non-inferiority to the LAMA comparator open-label tiotropium DPI for FEV~1~ AUC~0--12~ on Day 7 (LSM difference between treatments = −0.006 L; 95% CI = −0.049, 0.038 L). Both doses of FF MDI (7.2 and 9.6 μg) demonstrated non-inferiority to the open-label comparator FF DPI (Table [2](#Tab2){ref-type="table"}).Fig. 2FEV~1~ AUC~0--12~ on Day 7 efficacy endpoint. **a** LSM change (95% CI) in FEV~1~ over 0--12 h on Day 7 by treatment; **b** LSM (95% CI) FEV~1~ AUC~0--12~ difference from placebo on Day 7 by treatment (mITT population). ^a^Foradil^®^ Aerolizer^®^ ^.\ b^LSM allows for any imbalances in baseline covariates that relate to responses to be adjusted for in order to avoid bias in treatment effect estimates. AUC~0--12~, area under the curve from 0 to 12 h post-dose; DPI, dry powder inhaler; FEV~1~, forced expiratory volume in 1 s; FF, formoterol fumarate; GFF, glycopyrrolate/formoterol fumarate; GP, glycopyrrolate; LSM, least squares mean; MDI, metered dose inhaler; mITT, modified intent-to-treat Table 2FEV~1~ AUC~0--12~ at Day 7: GFF MDI 72/9.6 μg and 36/9.6 μg comparisons (mITT population)LSM treatment differences for GFF MDI in FEV~1~ AUC~0--12~ at Day 7GFF MDIGP MDI 36 μgOpen-label tiotropium 18 μgFF MDIPlacebo MDIOpen-label FF^a^ DPI 12 μgComparator72/9.6 μg36/9.6 μg9.6 μg7.2 μgGFF MDI 72/9.6 μg LSM^b^ difference (SE), LNA0.008 (0.0236)0.109 (0.0250)^†^0.103 (0.0216)^†^0.116 (0.0245)^†^0.124 (0.0237)^†^0.298 (0.0261)^†^0.101 (0.0241)^†^ 95% CI−0.039, 0.0540.059, 0.158^c^0.060, 0.1450.068, 0.1650.078, 0.1710.247, 0.3490.053, 0.148GFF MDI 36/9.6 μg LSM^b^ difference (SE), LSee aboveNA0.101 (0.0245)^†^0.095 (0.0213)^†^0.109 (0.0242)^†^0.116 (0.0236)^†^0.290 (0.0261)^†^0.093 (0.0241)^\*\*\*^ 95% CI0.053, 0.1490.053, 0.1370.061, 0.1560.070, 0.1630.239, 0.3420.045, 0.140^\*\*\*^ *p* \< 0.001; ^†^ *p* \< 0.0001^a^Foradil^®^ Aerolizer^®^; ^b^LSM allows for any imbalances in baseline covariates that relate to responses to be adjusted for in order to avoid bias in treatment effect estimates; ^c^non-inferiority comparisonCI, confidence interval; DPI, dry powder inhaler; FEV~1~ AUC~0--12~, forced expiratory volume in 1 s area under the curve from 0 to 12 h post-dose; FF, formoterol fumarate; GFF, glycopyrrolate/formoterol fumarate; GP, glycopyrrolate; LSM, least squares mean; MDI, metered dose inhaler; mITT, modified intent-to-treat; NA, not available; SE, standard error Secondary endpoints {#Sec15} ------------------- All active treatments were superior to placebo MDI for the lung function secondary endpoints (peak change from baseline in FEV~1~; time to onset of action on Day 1 \[≥10% improvement in FEV~1~ relative to baseline\]; peak change from baseline FEV~1~ and change from baseline in morning pre-dose trough FEV~1~ and 12-h post-dose trough FEV~1~; peak change from baseline in IC, change from baseline in morning pre-dose trough IC and 12-h post-dose trough IC; mean daily peak flow readings) on Days 1 and 7 (*p* ≤ 0.0056). The percentage of patients achieving ≥12% improvement in FEV~1~ was 86.8% (GFF MDI 72/9.6 μg), 87.2% (GFF MDI 36/9.6 μg), 73.7% (GP MDI 36 μg) 66.1% (open-label tiotropium DPI 18 μg), 84.5% (FF MDI 9.6 μg), 82.5% (FF MDI 7.2 μg), 40.0% (placebo MDI) and 85.2% (FF DPI 12 μg). Inferential comparisons of the percentage of patients achieving ≥12% improvement in FEV~1~ were not possible for several comparisons due to the limited number of patients (≤5) receiving each pair of treatments. However, numerical benefits were observed for all active treatments compared to placebo. GFF MDI 36/9.6 μg demonstrated non-inferiority to GFF MDI 72/9.6 μg (Table [3](#Tab3){ref-type="table"}; Additional file [1](#MOESM1){ref-type="media"}). At Day 7, GFF MDI 72/9.6 μg and GFF MDI 36/9.6 μg demonstrated superiority to each monocomponent MDI and to open-label FF DPI and tiotropium for morning peak FEV~1~, peak change in FEV~1~, and morning pre-dose trough IC (Table [3](#Tab3){ref-type="table"}; Additional file [1](#MOESM1){ref-type="media"}). FF MDI 7.2 μg and FF MDI 9.6 μg were both non-inferior to FF DPI, and FF MDI 7.2 μg was non-inferior to FF MDI 9.6 μg in secondary endpoints at Day 7 (Additional file [1](#MOESM1){ref-type="media"}).Table 3Secondary efficacy endpoints: Days 1 and 7 -- GFF MDI 72/9.6 μg and GFF MDI 36/9.6 μg comparisons (mITT population)ComparatorTreatment differences for GFF MDI comparisonsGP MDI 36 μgOpen-label tiotropium 18 μgFF MDIPlacebo MDIOpen-label FF^a^ DPI 12 μg9.6 μg7.2 μgDAY 7Change from baseline in morning pre-dose trough FEV~1~, LGFF MDI 72/9.6 μg LSM^b^ difference (SE)0.0960 (0.0280)^\*\*\*^0.096 (0.0247)^†^0.129 (0.0278)^†^0.120 (0.0271)^†^0.234 (0.0302)^†^0.091 (0.0277)^\*\*^GFF MDI 36/9.6 μgLSM^b^ difference (SE)0.073 (0.0273)^\*\*^0.073 (0.0245)^\*\*^0.106 (0.0274)^†^0.097 (0.0270)^\*\*\*^0.211 (0.0300)^†^0.068 (0.0275)^\*^Peak change from baseline in FEV~1~, LGFF MDI 72/9.6 μg LSM^b^ difference (SE)0.125 (0.0282)^†^0.140 (0.0248)^†^0.101 (0.0279)^\*\*\*^0.108 (0.0271)^†^0.342 (0.0300)^†^0.082 (0.0278)^\*\*^GFF MDI 36/9.6 μg LSM^b^ difference (SE)0.127 (0.0273)^†^0.141 (0.0245)^†^0.103 (0.0273)^\*\*\*^0.110 (0.0268)^†^0.344 (0.0298)^†^0.083 (0.0276)^\*\*^Change from baseline in morning pre-dose trough IC, LGFF MDI 72/9.6 μg LSM^b^ difference (SE)0.083 (0.0445)0.090 (0.0399)^\*^0.156 (0.0452)^\*\*\*^0.110 (0.0436)^\*^0.255 (0.0483)^†^0.096 (0.0447)^\*^GFF MDI 36/9.6 μg LSM^b^ difference (SE)0.098 (0.0445)^\*^0.105 (0.0387)^\*\*^0.172 (0.0433)^†^0.126 (0.0428)^\*\*^0.271 (0.0471)^†^0.111 (0.0434)^\*^Peak change from baseline in IC, LGFF MDI 72/9.6 μg LSM^b^ difference (SE)0.078 (0.0532)0.095 (0.0470)^\*^0.050 (0.0529)0.033 (0.0513)0.265 (0.0572)^†^0.016 (0.0527)GFF MDI 36/9.6 μg LSM^b^ difference (SE)0.107 (0.0513)^\*^0.124 (0.0461)^\*\*^0.078 (0.0513)0.062 (0.0503)0.293 (0.0559)^†^0.045 (0.0518)DAY 1Peak change from baseline in FEV~1~, LGFF MDI 72/9.6 μg LSM^b^ difference (SE)0.081 (0.0309)^\*\*^0.104 (0.0268)^†^0.062 (0.0307)^\*^0.060 (0.0297)^\*^0.265 (0.0328)^†^0.072 (0.0306)^\*^GFF MDI 36/9.6 μg LSM^b^ difference (SE)0.068 (0.300)^\*^0.090 (0.0266)^\*\*\*^0.048 (0.0300)0.046 (0.0293)0.251 (0.0326)^†^0.058 (0.0303)Peak change from baseline in IC, LGFF MDI 72/9.6 μg LSM^b^ difference (SE)0.065 (0.0567)0.149 (0.0493)^\*\*^0.134 (0.0564)^\*^0.144 (0.0547)^\*\*^0.412 (0.0607)^†^0.121 (0.0561)^\*^GFF MDI 36/9.6 μg LSM^b^ difference (SE)−0.019 (0.0555)0.065 (0.0491)0.050 (0.0554)0.060 (0.0542)0.328 (0.0602)^†^0.037 (0.0557)Time to onset of action, hazard ratio^c^GFF MDI 72/9.6 μg HR1.399^\*^1.754^\*\*\*^0.9801.1503.475^†^0.971 95% CI1.038, 1.8841.300, 2.3670.746, 1.2890.904, 1.4652.095, 5.7650.713, 1.321GFF MDI 36/9.6 μg HR1.3231.695^\*\*\*^0.8881.0623.358^†^0.878 95% CI0.936, 1.8701.275, 2.2530.671, 1.1750.0806, 1.4002.091, 5.3910.660, 1.169^\*^ *p* \< 0.05; ^\*\*^ *p* \< 0.01; ^\*\*\*^ *p* \< 0.001; ^†^ *p* ≤ 0.0001^a^Foradil^®^ Aerolizer^®^; ^b^LSM allows for any imbalances in baseline covariates that relate to responses to be adjusted for in order to avoid bias in treatment effect estimates; ^c^a hazard ratio of 1.399 signifies a 39.9% higher probability of onset of action at any time point post-dose*CI* confidence interval, *DPI* dry powder inhaler, *FEV* ~1~ forced expiratory volume in 1 s, *FF* formoterol fumarate, *GFF* glycopyrrolate/formoterol fumarate, *GP* glycopyrrolate, *HR* hazard ratio, *IC* inspiratory capacity, *LSM* least squares mean, *MDI* metered dose inhaler, *mITT* modified intent-to-treat, *SE* standard error Exploratory endpoint {#Sec16} -------------------- All active treatments were superior to placebo (*p* \< 0.0001) for change from baseline FVC (calculated as AUC~0--12~) on Day 7 (Additional file [1](#MOESM1){ref-type="media"}: Figure S2). Treatment comparisons are shown in the Additional file [1](#MOESM1){ref-type="media"}. Safety and tolerability {#Sec17} ----------------------- Most AEs were of mild (32.0%) or moderate (29.5%) intensity. Treatment-emergent AEs (TEAEs) reported in more than two patients receiving treatment are displayed in Table [4](#Tab4){ref-type="table"}. The most commonly reported TEAEs were: dry mouth, headache, tremor, cough and dysphonia (Table [4](#Tab4){ref-type="table"}). Dry mouth was reported more frequently by patients receiving GP MDI, GFF MDI, and open-label tiotropium compared to the other groups, while headache and tremor were reported more frequently by patients receiving GFF MDI. No patient in any treatment period reported paradoxical bronchospasm. The incidence of TEAEs was similar for the two doses of GFF MDI (31.7% vs 27.9%).Table 4Summary of adverse events (safety population)GFF MDIGP MDI 36 μg (*N* = 41)Open-label tiotropium 18 μg (*N* = 58)FF MDIPlacebo MDI (*N* = 52)Open-label FF^a^ DPI 12 μg (*N* = 55)72/9.6 μg (*N* = 41)36/9.6 μg (*N* = 43)9.6 μg (*N* = 64)7.2 μg (*N* = 64)Patients with at least one AE, *n* (%)17 (41.5)18 (41.9)11 (26.8)22 (37.9)24 (37.5)16 (25.0)9 (17.3)17 (30.9)Patients with AE related to study treatment, *n* (%)13 (31.7)12 (27.9)7 (17.1)7 (12.1)7 (10.9)4 (6.3)2 (3.8)7 (12.7)Patients with SAE, *n* (%)01 (2.3)02 (3.4)1 (1.6)2 (3.1)00Patients with SAE related to study treatment, *n* (%)00000000Patients with AE leading to early withdrawal, *n* (%)1 (2.4)01 (2.4)1 (1.7)4 (6.3)3 (4.7)1 (1.9)0Patients with SAE leading to early withdrawal, *n* (%)000002 (3.1)00TEAEs reported in ≥2 patients in any treatment group Dry mouth8 (19.5)3 (7.0)5 (12.2)4 (6.9)3 (4.7)2 (3.1)1 (1.9)2 (3.6) Headache3 (7.3)4 (9.3)1 (2.4)1 (1.7)1 (1.6)01 (1.9)2 (3.6) Tremor1 (2.4)5 (11.6)000000 Cough02 (4.7)01 (1.7)0000 Dysphonia1 (2.4)2 (4.7)000000% = 100 × n/N: *n* = no. of patients in the preferred term category for treatment group^a^Foradil^®^ Aerolizer^®^*AE* adverse event, *DPI* dry powder inhaler, *FF* formoterol fumarate, *GFF* glycopyrrolate/formoterol fumarate, *GP* glycopyrrolate, *MDI* metered dose inhaler, *SAE* serious adverse event, *TEAE* treatment-emergent adverse event Six SAEs were reported in five patients, none of which was related to study drug (one patient in the GFF MDI 36/9.6 μg group \[ruptured appendix\]; two patients in the open-label tiotropium group \[inhaled foreign body; abdominal aortic aneurysm\]; one patient in the FF MDI 9.6 μg group \[gastritis\]; and two patients in the FF MDI 7.2 μg group \[COPD exacerbation; atypical chest pain leading to early withdrawal\]). No deaths were reported in the study. There were no notable changes in hematology or chemistry laboratory values and no clinically significant abnormalities in vital signs, ECG, or physical examination. Pharmacokinetics {#Sec18} ---------------- Following chronic administration of GFF MDI 36/9.6 μg, the geometric LSM of glycopyrrolate was approximately 9% (AUC~0--12~) and 14% (maximum observed plasma concentration \[C~max~\]) lower than those observed following GP MDI 36 μg (Fig. [3a](#Fig3){ref-type="fig"}). In addition, the geometric LSM for formoterol was approximately 7% (AUC~0--12~) and 14% (C~max~) lower than those observed when FF MDI 9.6 μg was administered alone (Fig. [3b](#Fig3){ref-type="fig"}).Fig. 3Ratio of geometric LSMs and 90% CIs. **a** GFF MDI 36/9.6 μg versus GP MDI 36 μg (**b**) GFF MDI 36/9.6 μg versus FF MDI 9.6 μg (**c**) FF MDI 9.6 μg versus FF DPI (PK-mITT population). ^a^LSM allows for any imbalances in baseline covariates that relate to responses to be adjusted for in order to avoid bias in treatment effect estimates. ^b^Foradil^®^ Aerolizer^®^ ^.^ AUC~0--inf~, area under the curve from time 0 to infinity; AUC~0--12~, area under the curve from 0 to 12 h post-dose; CI, confidence interval; C~max~, maximum observed plasma concentration; DPI, dry powder inhaler; FF, formoterol fumarate; GFF, glycopyrrolate/formoterol fumarate; GP, glycopyrrolate; LSM, least squares mean; MDI, metered dose inhaler; PK-mITT, pharmacokinetic modified intent-to-treat Results from an analysis of variance (ANOVA) of the dose-normalized exposure parameters of formoterol between the FF MDI 9.6 μg and the FF DPI showed that 90% CIs for the ratios of LSM for the exposure parameters AUC~0--12~ and C~max~ were within the 80--125% interval, demonstrating that monocomponent FF MDI 9.6 μg was bioequivalent to the FF DPI formulation (equivalent to an FF 10 μg dose), with dose normalization. Furthermore, ANOVA results based on non-dose-normalized PK parameters also demonstrated equivalence between FF MDI and FF DPI (Fig. [3c](#Fig3){ref-type="fig"}). Discussion {#Sec19} ========== This 7-day Phase IIb study of GFF MDI 72/9.6 μg and 36/9.6 μg BID is the first study to investigate the FDC GFF MDI formulated using Co-Suspension delivery technology in patients with COPD. Previous studies have investigated the dose-response of the individual components delivered via MDI using Co-Suspension delivery technology \[[@CR9], [@CR13]\]. GFF MDI 72/9.6 μg and 36/9.6 μg BID led to statistically significant and clinically relevant improvement in the primary endpoint FEV~1~ AUC~0--12~ at Day 7 compared with the monocomponent MDIs and placebo MDI in patients with moderate-to-very severe COPD and were well tolerated (*p* ≤ 0.0001). It has become established that the combination of a LAMA and LABA provides benefits in lung function in patients with COPD over LAMA or LABA monotherapy in those patients who are not adequately controlled by a single long-acting bronchodilator. Within this therapeutic approach, LAMA/LABA FDCs glycopyrrolate/indacaterol, tiotropium/olodaterol, aclidinium/formoterol, and umeclidinium/vilanterol have previously demonstrated lung function benefits over monocomponents \[[@CR14]--[@CR19]\]. This study was part of a Phase IIb program including dose-ranging studies of GFF MDI and monocomponents to define the optimal doses of the GFF MDI FDC to take forward to Phase III studies, and utilized non-inferiority testing, based on the targeted 100 mL minimally clinically significant difference, as a well-defined and reproducible treatment effect for trough FEV~1~ to define therapeutic effect \[[@CR12]\]. In this study, non-inferiority was confirmed between GFF MDI 72/9.6 μg and 36/9.6 μg, incorporating GP doses at the higher end of the dose range, and both doses showed statistically significantly greater FEV~1~ AUC~0--12~ at Day 7 versus open-label tiotropium DPI and FF DPI. Notably, GFF MDI demonstrated superiority to placebo and statistically significant (GFF MDI 72/9.6 μg) and numerically greater (GFF MDI 36/9.6 μg) improvements in IC compared with open-label tiotropium DPI. In patients with COPD, changes in IC reflect changes in hyperinflation and have shown a higher correlation to patient-focused outcomes, such as dyspnea with exercise, than other standard spirometric measurements \[[@CR20]\]. Additional information was gained concerning the monocomponents, whereby the doses of the two monocomponents, GP and FF, demonstrated non-inferiority to the open-label active comparators such that GP MDI 36 μg BID was non-inferior to open-label tiotropium DPI and both doses of FF MDI demonstrated non-inferiority to open-label FF delivered via DPI. The PK component of the study characterized the systemic exposure of glycopyrrolate and formoterol delivered as an FDC compared with individual components delivered using Co-Suspension delivery technology. The findings support the absence of a significant drug-drug interaction for formoterol and glycopyrrolate following administration of GFF MDI relative to the individual MDI formulations. It was also shown that formoterol exposure (AUC~0--12~) increases in a dose-proportional manner when delivered via the MDI and that FF MDI 9.6 μg delivered by MDI using Co-Suspension delivery technology was bioequivalent to FF delivered using DPI, which taken together endorse the dose of FF MDI 9.6 μg in the FDC. All treatments were generally well tolerated in this study. The tolerability and safety profiles observed for GFF MDI 36/9.6 μg and 72/9.6 μg BID were consistent with the patient population and drug classes \[[@CR21]\]. The most commonly reported TEAEs (more than two patients in any treatment group) were dry mouth, headache, tremor, cough and dysphonia in descending order of incidence. Whilst the sample size of this first study of GFF delivered by MDI using novel Co-Suspension delivery technology was calculated to provide reasonable information to characterize the response, we recognize that the study was limited by a relatively small population and enrolled a lower than anticipated number of patients with very severe COPD. In addition, the study was only conducted over 7 days. However, information from this study guided study design and sample size for subsequent studies. GFF MDI will potentially widen treatment options for patients with COPD by providing LAMA/LABA therapy in an MDI when other LAMA/LABA FDCs are available as DPI and soft-mist inhaler devices. Unlike other devices, MDIs are not breath-actuated, consequently, MDIs may be able to offer patients with quite severe airflow limitation, who may be unable to breathe deeply enough to release the medication from a DPI, a more practical and reliable method of delivery of their medication. The PKs of GFF MDI support twice-daily administration and the improvements in lung function demonstrated here after BID dosing with GFF MDI as maintenance dual-bronchodilator therapy may provide benefits over QD dosing, preventing excessive worsening of symptoms during the night or towards the morning in patients with this pattern of symptoms \[[@CR7]\]. The results of a patient evaluation survey conducted by Partridge et al., and similarly in the ASSESS study, show that patients with COPD begin to experience worsening symptoms in the evening through the night-time, with the worst symptoms, affecting activity and productivity, occurring in the morning \[[@CR22], [@CR23]\]. This is also an area for future investigation of the effects of GFF MDI on COPD symptom burden. Conclusions {#Sec20} =========== Co-Suspension delivery technology allows formulation of FDCs of different drug classes, at different concentrations, in a single MDI. GFF MDI 72/9.6 μg and 36/9.6 μg BID were associated with a similarly greater magnitude of effect on FEV~1~ AUC~0--12~ at Day 7 compared with the monocomponent MDIs and placebo MDI in patients with moderate-to-very severe COPD, which is likely to be the maximal therapeutic effect. Additional studies are required to establish the optimal doses of GP and FF for combination in the FDC GFF MDI. Additional file {#Sec21} =============== Additional file 1:Study design; Additional Fig. 1. Study design schematic; Additional Fig. 2. Mean change from baseline in PEFR over time by treatment on Day 7 (mITT population); Additional Fig. 3. Mean change from baseline FVC over time by treatment on Day 7 (mITT population); Additional Table 1. Secondary efficacy endpoints: Days 1 and 7 -- FF 9.6 μg and FF 7.2 μg comparisons (mITT population). (DOCX 466 kb) AEs : Adverse events AUC : Area under curve BID : Twice daily CI : Confidence interval COPD : Chronic obstructive pulmonary disease CT : Computed tomography DPI : Dry powder inhaler ECG : Electrocardiogram FDC : Fixed-dose combination FEV~1~ AUC~0--12~ : Forced expiratory volume in 1 s area under the curve from 0 to 12 h FEV~1~ : Forced expiratory volume in 1 s FF : Formoterol fumarate FVC : Forced vital capacity GFF MDI : Glycopyrrolate/formoterol fumarate MDI GP : Glycopyrrolate HFA : Hydrofluoroalkane HR : Hazard ratio IC : Inspiratory capacity ITT : Intent-to-treat LABA : Long-acting beta agonist LAMA : Long-acting muscarinic antagonist LRTI : Lower respiratory tract infection LSM : Least squares mean MDI : Metered dose inhaler mITT : Modified intent-to treat NA : Not available PK : Pharmacokinetic QD : Once-daily SAEs : Serious adverse events SD : Standard deviation SE : Standard error TEAEs : Treatment-emergent AEs The authors would like to thank all of the patients and their families, the team of investigators, research nurses and operations staff involved in this study. The authors would also like to thank Catherine Stanton, of Complete Medical Communications, who provided medical writing support under the direction of the authors, funded by AstraZeneca. The authors would also like to thank Mervyn Thomas for his work in the statistical design and planning of this study and Everest Clinical Research, who conducted the statistical analyses of the study data. Funding {#FPar1} ======= Pearl Therapeutics Inc., a member of the AstraZeneca Group. Availability of data and material {#FPar2} ================================= All relevant data generated or analyzed during this study are included in this published article (and its additional information files). Authors' contributions {#FPar3} ====================== KFR, FJM, JFD, LMF, GTF, SD, CR, CO, PD, ES, TF, MG, SSt, SSp made substantial contributions to the conception or design of the work reported. EMK, DQ, CF, SSp participated in the acquisition of reported data. SD, CR, CO, PD participated in the analysis of reported data. KFR, FJM, JFD, LMF, GTF, EMK, DQ, CF, SSt, SSp, SD, CR, CO, PD participated in the interpretation of reported data. All authors contributed to the writing of the report and participated in the review and interpretation of the data. All authors read and approved the final report before submission. Authors' information {#FPar4} ==================== \[as per title page\]. Co-Suspension™ is a trademark of the AstraZeneca group of companies. **List of principal investigators** JoAnne Marjason, Robyn Huttenmeister, Maria Quartararo, Peter Clyne, Paul Seale, Philip Thompson, Philip Bardin, Dean Quinn, Andrew Veale, Margaret Wilsher, Graham Mills, Michael Chia, Charles M. Fogarty, Selwyn Spangenthal, Edward M. Kerwin, Leonard Dunn. Competing interests {#FPar5} =================== CR is an employee of Pearl Therapeutics, Inc., a member of the AstraZeneca Group. LMF has received grants, personal fees and non-financial support from Almirall, AstraZeneca, Boehringer Ingelheim, Chiesi, GlaxoSmithKline, Laboratori Guidotti, Merck Sharp & Dohme, Menarini, Novartis and Takeda. He has received personal fees and non-financial support from Boston Scientific, Mundipharma and Pearl Therapeutics Inc. He has received personal fees from Bayer, Kyorin and Zambon. He has received grants from Biofutura Italia, Dompè, Malesci, Pfizer and Vree Health Italia. EMK has served on advisory boards, speaker panels, or received travel reimbursement from Amphastar, Boehringer Ingelheim, Forest, Mylan, Novartis, Pearl Therapeutics Inc. (a member of the AstraZeneca Group), Sunovion, Teva and Theravance. He has conducted multicenter clinical trials for \~40 pharmaceutical companies. CF is a consultant and investigator for Pearl Therapeutics, Inc. SSp has no potential conflicts of interest to disclose. KFR has received grants from Boehringer Ingelheim, the German Federal Ministry of Education and Research (BMBF) and Novartis. He has received personal fees from AstraZeneca, Boehringer Ingelheim, Chiesi, Intermune, Novartis and Takeda. GTF has received grants and personal fees from AstraZeneca, Boehringer Ingelheim Novartis, Pearl Therapeutics Inc., Sunovian and Theravance. He has received grants from Forest. He has received personal fees GlaxoSmithKline, Meda, Mylan and Verona. FJM has received personal fees from Adept, Afferent, Amgen, AstraZeneca, Axon, Axon Communication, Boehringer Ingelheim, Clarion, ConCert, Forest, Genentech, GlaxoSmithKline, Ikaria/Bellerophon, Informa, Janssen, Kadmon, Lucid, Methodist Hospital, Novartis, Nycomed/Takeda, Pearl Therapeutics Inc., Pfizer, Prime, Roche, Sunovion, Theravance, Unity Biotechnology, Veracyte and WebMD. He has received non-financial support from Biogen/Stromedix, Boehringer Ingelheim, Centocor and Gilead and grants from the National Institutes of Health. He has received personal fees for delivering CME programs for Academic CME, American Thoracic Society, Annenberg, California Society for Allergy and Immunology, CME Incite, Falco, Haymarket Communications, Integritas, InThought, Miller Medical, National Association for Continuing Education, Paradigm, Peer Voice, Potomac, UpToDate and Western Society of Allergy and Immunology. He has received royalty fees from Informa. He has spoken on behalf of AstraZeneca and Nycomed/Takeda. He is currently a member of the GOLD Scientific Committee. JFD has received personal fees from AstraZeneca, GlaxoSmithKline, Novartis, Sunovion and Teva. PD, ES, CO, SSt, TF, MG and SD are employees of Pearl Therapeutics, Inc., a member of the AstraZeneca Group. Consent for publication {#FPar6} ======================= Not applicable. Ethics approval and consent to participate {#FPar7} ========================================== This study was conducted in accordance with International Conference on Harmonization guidelines, the Declaration of Helsinki, and the US Code of Federal Regulations. The protocol, its amendments and patient informed consent form were approved by an Independent Ethics Committee or Institutional Review Board.

Introduction {#s1} ============ Primary ovarian insufficiency (POI), also known as premature ovarian failure (POF) or premature menopause, is defined as cessation of menstruation before the expected age of menopause. This age is traditionally defined to be prior to 40 years and diagnosis is confirmed by elevated serum FSH levels (\>40IU/l). Although frequently stated that ∼1% of the population is affected with POI before the age of 40 years and 0.1% before age 30 years, the prevalence is actually less certain ([@DMV036C45]). The disorder is clearly heterogeneous, with a wide spectrum of causes, namely cytogenetic, genetic, infectious or iatrogenic. Autoimmune and metabolic etiologies may or may not be genetic. Irrespective, etiology remains to be elucidated in most cases and until a decade ago few specific causes were known beyond X-chromosomal abnormalities, Fragile X mental retardation 1 (*FMR1*) premutation and FSH receptor (*FSHR*) in the Finnish population ([@DMV036C196]; [@DMV036C4]; [@DMV036C236]). Most cases of isolated POI still appear sporadically, but ∼10--15% has an affected first-degree relative, indicating significant genetic etiology ([@DMV036C218]). Pedigrees with multiple affected relatives are not rare (recessive and dominant). Presence of POI as one component of a pleiotropic genetic disorder is also well recognized. Yet identifying precise causative genes has been challenging. Here, we enumerate known genetic causes of POI, most elucidated within the last 5--10 years. Confusion exists concerning nomenclature, namely the use of POF or POI. It is the view of the authors that POI can be taken to encompass occult, biochemical and overt stages, whereas POF is best considered as only the final stage of POI. The designation POI is thus best reserved as alluding to the entire gamut of disorders having diminished ovarian reserve---occult, subclinical, iatrogenic. Although many authors espouse POI in lieu of POF, the canonical genetic reference---Online Mendelian inheritance in man (OMIM)---has long used, and continues to use POF, to nominate causative genes. These designations now apply to POF1--POF9 ([Supplementary Table SI](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)), and the list growing. Methods {#s2} ======= A search for original articles published up to May 2015 was performed using PubMed and Google Scholar to identify studies on genetic variants associated with the human disease. The key word combinations include 'premature ovarian failure', 'primary ovarian insufficiency', 'early menopause', 'genetic', 'gene mutation', 'variant' and \'genome wide study'. For a study to be included in our review, it had to focus on chromosomal analysis, candidate gene screening, or a genome-wide study in different POI cohorts. In addition, studies on mitochondrial genes causing POI and multiple malformation syndromes characterized by POI were included. Reports on the role of candidate genes in animal models were not included. Where appropriate, reference lists of identified articles were also searched for further relevant papers. However, articles identified were restricted to English language full-text papers. Results {#s3} ======= Chromosomal abnormalities in POI {#s3a} -------------------------------- Chromosomal abnormalities have long been recognized as a cause of POI, but percentages vary widely among reported series. This clearly reflects biases of ascertainment, for example reflecting whether a cohort was derived from a referral cytogenetic lab, a gynecologic practice, or a pediatric practice. Numerous different karyotypic anomalies have been found, ranging from numerical defects (monosomy X; X chromosomal mosaicism), X-deletions, X-autosome translocations, and X-isochromosomes and other rearrangements. Aggregate frequency of chromosomal abnormalities in reported studies is summarized in Table [I](#DMV036TB1){ref-type="table"}. Small sample sizes as well as selection biases and differing ages of ascertainment probably account for different prevalence in different populations. However, each of the five largest studies with respect to sample size reported frequencies between 10.0 and 12.9% ([@DMV036C242]; [@DMV036C118]; [@DMV036C12]; [@DMV036C99]; [@DMV036C101]); thus, a prevalence of 10--13% seems reasonable. Table IFrequency of chromosomal abnormalities (CA) in different population studies.ReferenceFrequency of CA (%)No. of CASample sizeClinical characteristicsPopulation[@DMV036C8]18.018100PA, SATunisian[@DMV036C101]10.0518179PA, SAIranian[@DMV036C99]12.164531PA, SAChinese (Jinan, Beijing, Shenzhen)[@DMV036C12]10.027269PA, SA, EMItalian[@DMV036C118]10.81081000PA, SATunisian[@DMV036C30]25.319^a^75SATurkish[@DMV036C95]12.919147SADutch[@DMV036C159]8.8890PA, SAFrench[@DMV036C242]12.513104POIChinese (Chongqing)[@DMV036C51]13.3430SAAmerican[@DMV036C48]2.5279PA, SA FSH\>20 IU/lEnglish[@DMV036C29]32.01547POIChilean[@DMV036C180]25.41663PA, SAAmerican[^1][^2][^3] ### Numerical defects {#s3a1} The X chromosome has long been known to play an essential role in the maintenance of ovarian development and function. Females lacking an X chromosome as well as those showing an extra X chromosome are predisposed to developing POI. ### 45,X and 45,X/46,XX {#s3a2} Turner syndrome, often but not universally associated with X monosomy, leads to ovarian dysgenesis and accelerated follicular atresia. X monosomy without mosaicism is more typically found in primary amenorrhea and cases were almost universally understood to present with this phenotype. However, many early series were recruited from pediatric clinics, not among adult women. In 1975, [@DMV036C196] reported that 3% (5/178) of 45,X patients actually menstruated. 45,X/46,XX and other forms of association also are associated with secondary amenorrhea (POI). Either haploinsufficiency of pivotal genes on the X chromosome or non-specific meiotic impairment could explain the accelerated atresia of 45,X oocytes. Variability would be expected given potential heterozygosity of alleles in genes subjected to X-inactivation. ### 47,XXX {#s3a3} 47,XXX women may experience oligomenorrhea, secondary amenorrhea, and early menopause, but relative risk has not been well studied. [@DMV036C83] reported the prevalence of 47,XXX in 52 women with POI to be 3.8%, whereas in our much larger Chinese series we observed 1.5% (8/531) ([@DMV036C99]). The presence of three X chromosomes plausibly leads to meiotic disturbance and, secondarily, ovarian failure. Additionally, overexpression of genes that escape X-inactivation could cause POI in 47,XXX. Mechanisms remain to be defined ([@DMV036C211]). A confounder is that an association exists between 47,XXX and autoimmune diseases ([@DMV036C90]; [@DMV036C83]). ### X-structural abnormalities and X-autosome translocations {#s3a4} X chromosome deletions and X-autosome balanced translocations have long been observed in POI and were once the only approach available to localize causative genes. This strategy was illustrated in the 1970s by a region on the X chromosome appearing critical for the POI phenotype ([@DMV036C188]). A critical region was delineated that gave boundaries for breakpoints of X-autosome translocations associated with ovarian failure. This region extends from Xq13-Xq21 (POI2) to Xq23-q27 (POI1). It has been proposed by Rizzolio and colleagues that Xq13-Xq21 governs epigenetic regulations that down-regulate oocyte-expressed autosomal genes ([@DMV036C183], [@DMV036C184], [@DMV036C185]). Irrespective of mechanisms involved in the critical region, almost all terminal deletions originating at Xq13 are associated with primary amenorrhea, lack of breast development and complete ovarian failure ([@DMV036C201]; [@DMV036C197]). By contrast, terminal deletions arising at Xq25 or Xq26 are characterized by the more common phenotype being not primary amenorrhea but premature ovarian failure. The gene designation POI1 is applied to this region. The more distal deletions arise at Xq27 or Xq28 and exert a less severe effect on stature and reproductive function than do proximal deletions ([@DMV036C196]; [@DMV036C201]). Gene(s) in this POI-causing region are considered distinct from Fragile X mental retardation 1 (*FMR1*), located at Xq27 and premutation of which is the most common single cause of POI. Multiple genes on the X chromosome have been identified by X-autosomal translocations. These include Diaphanous-related formin 2 (*DIAPH2*, Xq22) ([@DMV036C15]), X-prolyl aminopeptidase (aminopeptidase P) 2, membrane-bound (*XPNPEP2*, Xq25) ([@DMV036C162]), Dachshund family transcription factor 2 (*DACH2*, Xq21.3) ([@DMV036C163]), Premature ovarian failure, 1B (*POF1B*, Xq21.1) ([@DMV036C126]; [@DMV036C16]), Choroideremia (*CHM*, Xq21.1) ([@DMV036C126]; [@DMV036C132]), Progesterone receptor membrane component 1 (*PGRMC1*, Xq24) ([@DMV036C132]), Collagen, type IV, alpha 6 (*COL4A6*, Xq22.3) ([@DMV036C141]) and Nuclear RNA export factor 5 (*NXF5*, Xq22.1) ([@DMV036C13]). Some will individually be discussed below. ### Autosomal rearrangements {#s3a5} Autosomal translocations---Robertsonian and reciprocal---have been observed in sporadic cases in Belgian, American, Japanese and Chinese women ([@DMV036C89]; [@DMV036C147]; [@DMV036C106]; [@DMV036C99]). Perturbations presumably confer haploinsufficiency or interrupt pivotal genes in these regions. Non-specific defective meiotic pairing or a position effect on contiguous genes is also a potential explanation ([@DMV036C197]; [@DMV036C152]). No autosomal region appears preferentially involved, long frustrating investigators seeking to use breakpoints to localize regions containing autosomal genes of relevance. Searches for autosomal regions disrupted in X-autosome translocations have similarly not proved fruitful in identifying autosomal roles in POI. Single genes causing non-syndromic POI {#s3b} -------------------------------------- Aside from regions of interest defined by chromosome deletions and translocations, the other traditional strategy to identify candidate genes in POI is to study genes whose product is known and plays a role in human folliculogenesis or shows an organ-specific effect based on murine knockout models (candidate genes). Many genes have been interrogated for these reasons. In this section we specifically review all genes for which data warrant strong consideration as a candidate gene for POI. Table [II](#DMV036TB2){ref-type="table"} and [Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1) contain available details of reported studies. Table IIVariants identified in candidate genes on the X chromosome for idiopathic and sporadic POI.GeneLocationCases (*N*)Controls (*N*)EthnicityMR^a^Sequence variationAmino acid changeFCMechanismReference*BMP15*Xp11.250214Caucasian North Africa Asia2 (4.0%)c.242A\>Gp.H81R[@DMV036C212]c.595G\>Ap.G199R100100Chinese1 (1.0%)c.985C\>Tp.R329C[@DMV036C230]300216Caucasian12 (4.0%)c.13A\>Cp.S5RYesSlightly affects transactivation of BRE-luc in COV434 granulosa cells[@DMV036C186]c.202C\>Tp.R68WMarkedly reduces mature protein production and affects transactivation of BRE-luc in COV434 granulosa cellsc.413G\>Ap.R138Hc.443T\>Cp.L148Pc.538G\>Ap.A180TNo effect on protein production or transactivation2093GermanyNone[@DMV036C122]9276ChineseNone[@DMV036C244]20354Caucasian African Asian3 (1.5%)c.443T\>Cp.L148P[@DMV036C115]c.538G\>Ap.A180Tc.468G\>ASensec.831T\>CSensec.852C\>TSense133197Indian14 (10.5%)c.181C\>Tp.R61W[@DMV036C62] and [@DMV036C92]c.182G\>Ap.R61Ec.226C\>Tp.R76CYesDecreased mature protein production, weaker Smad1/5/8 phosphorylation in COV434 cells and decreased granulosa cell proliferationc.227G\>Ap.R76Hc.538G\>Ap.A180Tc.538G\>T/c.539C\>Tp.A180F/S+Vc.588T\>Ap.N196Kc.617G\>Ap.R206HYesDecreased mature protein production, weaker Smad1/5/8 phosphorylation in COV434 cells and decreased granulosa cell proliferationc.631C\>Tp.E211Xc.661T\>Cp.W221Rc.727A\>Gp.L243Gc.381A\>GSensec.\*40dupG3′UTR166181Caucasian7 (4.2%)c.202C\>Tp.R68W[@DMV036C56], [@DMV036C57]c.538G\>Ap.A180Tc.704A\>Gp.Y235CYesDiminished GC proliferation with a dominant negative effect3851New ZealandNone[@DMV036C31]153JapaneseNone[@DMV036C209]*AR*Xq12133200Indian2 (1.5%)c.1948A\>Gp.T650A[@DMV036C149]c.1972C\>Ap.O658Kc.1885+9C\>AIntron*FOXO4*Xq13.1116143TunisianNone[@DMV036C73]*POF1B*Xq21.2223900Italian2 (0.9%)c.G1477Ap.C444Y[@DMV036C16]*DACH2*Xq21.32571110Italian2 (0.8%)c.G274Tp.R37L[@DMV036C16]c.T1111Cp.F316S*PGRMC1*Xq22-q2467153Swedish Italian1 (1.5%)c.494A\>Gp.H165RYesAttenuates ability to transduce progesterone\'s anti-apoptotic action in granulosa cells and abolishes binding capacity to CYP7A1[@DMV036C132]196200Chinese1 (0.51%)c.556C\>Tp.P186S[@DMV036C234][^4][^5] Variants occurring in evolutionary conserved regions are more likely to carry functional significance. These include missense, nonsense, insertion or deletion variants and were considered as plausible causative variants with clinical significance. Thus, perturbations should yield a severe functional defect. A nonsense mutation that results in truncated protein should lead to haploinsufficiency; a splicing site mutation or insertion/deletion should result in a frameshift that leads to a different protein product; a missense mutation may change the amino acid and cause a dominant negative effect. Those regions or genes found only by genome-wide studies, including genome-wide association study, cytogenomic study, whole-exome sequencing, and next generation sequencing (NGS), are cited separately. ### Genes on the X chromosome {#s3b1} #### Bone morphogenetic protein 15 (BMP15) (Xp11.2) {#s3b1a} *BMP15* is located on chromosome Xp11.2. The possible involvement of *BMP15* in POI pathogenesis was initially supported by evidence from animal models. Inverdale and Hanna sheep with a naturally occurring *Bmp15* mutation had increased ovulation rate and twin and triplet births in heterozygotes, but ovarian failure results from impaired follicular development beyond the primary stage in homozygotes ([@DMV036C79]). *Bmp15* knockout female mice also were subfertile, showing decreased ovulation rates, reduced litter size and decreased number of litters per lifetime ([@DMV036C240]). In humans *BMP15* was first implicated in POI by [@DMV036C56], who reported a heterozygous p.Y235C missense mutation in each of two sisters having ovarian failure. The authors presented *in vitro* evidence for a dominant negative mechanism. Other variants have been identified in Caucasian, Indian and Chinese women with POI, albeit with quite different frequencies (1.5--15%) ([@DMV036C57]; [@DMV036C62]; [@DMV036C115]; [@DMV036C122]; [@DMV036C117], [@DMV036C119]; [@DMV036C186]; [@DMV036C230]). Merely showing different frequencies between a given single nucleotide polymorphism (SNP) in POI and control is a less robust method than finding a unique perturbation with functional validation in a case ([@DMV036C244]; [@DMV036C122]). However, some variants found in higher frequency indeed show marked reduction of mature protein production ([@DMV036C186]). Of relevance is that *BMP15* is a member of the transforming growth factor (TGF) family, with dimerization occurring with other TGF proteins such as *GDF9*, to be discussed below. Most reported *BMP15* variants are, in fact, in the region corresponding to the propeptide of the protein, which is essential for dimerization and subsequent post-translational processing into biologically active proteins. #### Progesterone receptor membrane component 1 (PGRMC1) (Xq22-q24) {#s3b1b} *PGRMC1* was first described in 1998 as a putative progesterone-binding membrane receptor ([@DMV036C127]). This protein is expressed in various tissues, e.g. liver, kidney, adrenal glands, uterus and leukocytes and involves progesterone signaling in the reproductive system ([@DMV036C28]; [@DMV036C127]; [@DMV036C132]). PGRMC1 mediates progesterone\'s anti-apoptotic effects on granulosa cells ([@DMV036C67]; [@DMV036C151]; [@DMV036C127]; [@DMV036C132]). [@DMV036C132] identified a mother and daughter with POI, both of whom carried an X;autosome translocation \[t(X;11)(q24;q13)\]. Systematic mapping of the Xq breakpoint and performing RNA expression studies revealed reduced expression of *PGRMC1*. Mutation screening of 67 females with idiopathic POI identified a third patient having a missense mutation (p.H165R), located in the cytochrome b5 domain. The p.H165R mutation abolishes binding of cytochrome P450 7A1 (CYP7A1) to PGRMC1 and attenuates PGRMC1\'s ability to mediate the anti-apoptotic action of progesterone in ovarian cells. These findings suggest that mutant or reduced levels of PGRMC1 may cause POI through impaired activation of the microsomal cytochrome P450 and increased apoptosis of ovarian cells. A recent study in Chinese patients with POI identified a novel missense mutation (C.556C\>T, p. P186S), but there was no functional study to confirm a deleterious effect ([@DMV036C234]). #### Androgen receptor (AR) (Xq12) {#s3b1c} The *AR* gene encodes the androgen receptor and is involved in sex differentiation and reproduction. Its perturbation in 46,XY individuals results in the well-known sex reversed phenotype of androgen insensitivity, testosterone produced by testis exerting no effect on androgen-dependent differentiation. In the ovary, AR is expressed in developing follicles, mainly granulosa cells. Deficiency of *Ar* in female mice results in a POI-like phenotype and dysregulation of a number of major genes critical for folliculogenesis, indicating that normal folliculogenesis requires AR-mediated androgen action ([@DMV036C194]). An association between CAG repeat length in exon 1 of the *AR* gene and POI has been proposed but remains controversial ([@DMV036C25]; [@DMV036C36]; [@DMV036C206]; [@DMV036C149]). An example is a repeat of two missense mutations (p.T650A and p.O658K) in Indian women with POI ([@DMV036C149]). #### Forkhead box O4 (FOXO4) (Xq13.1) {#s3b1d} The *FOXO4* gene encodes a member of the O class of winged helix/forkhead transcription factor family (FOXO). FOXO4 is expressed in granulosa cells in mice and human, and is involved in the PI3K (phosphoinositide 3-kinase)/Akt (v-akt murine thymoma viral oncogene homolog 1)/Cdkn1b (cyclin-dependent kinase inhibitor 1B) molecular pathway, which suggests a functional role in ovarian physiology ([@DMV036C158]). Mutation screening in 116 Tunisian patients identified only one intronic variant; IVS2 + 41T\>G; therefore, *FOXO4* might not be a common cause of POI in the Tunisian population ([@DMV036C73]). #### Premature ovarian failure, 1B (POF1B) (Xq21.2) {#s3b1e} Alluded to previously, this 'gene' is actually a region, but codified by OMIM. Its significance is its location within the critical POI1 region. It was found to be interrupted by a breakpoint in an X-autosome translocation in a patient with secondary amenorrhea (POI). Subsequent mutation analysis in an Italian POI cohort (*N* = 223) only revealed 30 SNPs ([@DMV036C16]). In a Lebanese family having five sisters with POI, [@DMV036C114] established linkage to Xq21 using whole-genome SNP typing and homozygosity-by-descent mapping. Sequencing identified a homozygous p.R329Q mutation, which impaired the capacity to bind nonmuscle actin filaments, and might lead to exaggerated germ-cell apoptosis and POI. #### Dachshund family transcription factor 2 (DACH2) (Xq21.3) {#s3b1f} *DACH2*, also named dachshund family transcription factor 2, is located on Xq21.3. It was first identified by fine mapping of the disrupted region in an X;autosome translocation in POI patients ([@DMV036C163]). Subsequent mutation screening revealed two novel missense mutations---p.R37L and p.F316S---in an Italian cohort of POI patients ([@DMV036C16]). However, no subsequent evidence of involvement of *DACH2* in mammalian gonads or additional mutations in other ethnic population has been reported. #### Fragile X mental retardation 1 (FMR1) (Xq27.3) {#s3b1g} One of the commonest causes of POI is a premutation of *FMR1*, which when fully perturbed (\>200 CGG repeats) causes fragile X syndrome but paradoxically not POI. A prototype of pleiotropic single gene disorders in which POI is one component, *FMR1* is discussed in the 'Pleiotropic Single Gene Disorders Having POI' section, along with other pleiotropic genes. ### Genes on autosomes {#s3b2} In this section we will review autosomal genes for which data appear to warrant strong consideration as a candidate gene for POI. [Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1) contains available details on studies generating this conclusion. #### Growth differentiation factor 9 (GDF9) (5q31.1) {#s3b2a} Expressed in oocytes, *GDF9* is an attractive candidate gene for POI because it is, like *BMP15*, a member of the TGF gene family. Increased frequencies of certain novel variants have been detected in European, Caucasian and Asian patients ([@DMV036C59]; [@DMV036C115]; [@DMV036C111]; [@DMV036C246]), but not in Japanese and New Zealand populations ([@DMV036C209]; [@DMV036C31]). All variants were heterozygous. Recently, high-resolution array comparative genomic hybridization (CGH) (2.2 kb resolution) was applied in 26 POI Swedish cases, finding one partial *GDF9* gene duplication (475 bp) ([@DMV036C142]). Unfortunately, parents were not available to exclude a heritable copy number variant (CNV) less likely to carry significance. This pitfall is discussed further in 'Genome-Wide Studies in POI' section. Heterozygous changes could result in a dominant negative effect, quite plausible given dimerization with fellow members of the TGF gene family (e.g. BMP15). The proportion of POI due to *GDF9* perturbations is, however, unclear. If a hydrophobic amino acid replacing a hydrophilic amino acid were causative, *GDF9* perturbations could account for a substantial number (1--4%) of POI cases. #### Folliculogenesis specific bHLH transcription factor (FIGLA) (2p13.3) {#s3b2b} *FIGLA*, also named factor in the germline, alpha, is a germ-cell specific, basic helix-loop-helix (bHLH) transcription factor, that plays a crucial role in the formation of the primordial follicle and coordinates expression of zona pellucida genes. [@DMV036C247] screened 100 Chinese women with POI and identified three variants in four women: missense mutation p.A4E in two women; deletion p.G6fsX66 in one woman, resulting in a frameshift that leads to haploinsufficiency; and deletion p.140delN in a fourth woman. Functional analyses by the yeast two-hybrid assay demonstrated that the p.140delN mutation disrupted FIGLA binding to the TCF3 helix-loop-helix (HLH) domain. These findings show that a subset of Chinese women with sporadic POI harbor mutations in *FIGLA*. Recently, another novel intronic variant was found in 219 Indian POI cases ([@DMV036C215]). Further functional validation is warranted. #### Newborn ovary homeobox gene (NOBOX) (7q35) {#s3b2c} *NOBOX* is an oocyte-specific homeobox gene that plays a critical role in early folliculogenesis. The causative role was discovered by [@DMV036C179]. In mice *Nobox* deficiency disrupted early folliculogenesis and oocyte-specific gene expression. Lack of *Nobox* accelerated post-natal oocyte loss and abolished the transition from primordial to growing follicles in mice. In female mice lacking *Nobox,* follicles are replaced by fibrous tissue in a manner similar to non-syndromic ovarian failure in women. Genes preferentially expressed in oocytes, including *Pou5f1* (POU class 5 homeobox 1) and *Gdf9*, are also down-regulated in *Nobox^−/−^* mice. [@DMV036C121] showed that POI in *Nobox* deficient mice results from faulty signaling between somatic and germ line components during embryonic development. In addition, the extremely unusual presence of abnormal adherens junctions between unseparated oocytes within syncytial follicles indicates that faulty communication between somatic and germ cells is involved in, or leads to, abnormalities in the cell adhesion program. [@DMV036C167] were the first to demonstrate that a perturbation (p.R355H) in *NOBOX* was responsible for human POI. The mutation disrupted NOBOX homeodomain binding to NOBOX DNA-binding element (NBE) and had a dominant negative effect. Our functional studies demonstrated that haploinsufficiency was involved in the genetic mechanism in humans for POI. Mutations in the homeobox domain of *NOBOX* proved not to be a common explanation for POI in Chinese women (0/200) ([@DMV036C170]) but [@DMV036C23], [@DMV036C24]) subsequently reported that novel *NOBOX* loss-of-function mutations accounted for 6.2 and 5.6%, respectively, of cases in two large 'primary ovarian insufficiency' cohorts of Caucasian and African ancestry. #### Nuclear receptor subfamily 5, group A, member 1 (NR5A1); Steroidogenic factor-1(SF-1) (9q33) {#s3b2d} *NR5A1* encodes an orphan nuclear receptor that regulates transcription of an array of genes involved in reproduction, steroidogenesis and male sexual differentiation. These include anti-Mullerian hormone (*AMH*), Nuclear receptor subfamily 0, group B, member 1 (*DAX1*), Cytochrome P450, family 11, subfamily A, polypeptide 1 (*CYP11A*), steroidogenic acute regulatory protein (*StAR*), as well as genes encoding steroid hydroxylases, gonadotrophins, and aromatase. Inactivation of *Nr5a1* specifically in mouse granulosa cells causes infertility associated with hypoplastic ovaries. [@DMV036C153] identified *NR5A1* mutations as a frequent cause of 'primary amenorrhea' in 46,XY phenotypic female adolescents with a low testosterone concentration. [@DMV036C128] sequenced *NR5A1* in four families (each having at least one family member with a 46,XY disorder of sex development and another with 46,XX POI) and 25 subjects with sporadic POI, and they identified 19 different mutations in the *NR5A1* gene. Functional studies indicated that these mutations substantially impaired the transactivational activity of *NR5A1*. Subsequently additional mutations were identified in different ethnicities with low frequencies ([Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)). [@DMV036C96] sequenced the coding regions of *NR5A1* in a large, well-phenotyped cohort of 356 Dutch women with POI, finding 9 different mutations in 10 patients. Functional prediction showed low to intermediate pathogenicity for all non-conserved mutations. However, the novel p.Y5D mutation, detected in a non-domain region, was presumed to result in haploinsufficiency in Chinese patients with POI ([@DMV036C100]). #### FSH receptor (FSHR) (2p21-p16) {#s3b2e} FSH/FSHR signaling plays a key role in normal gonadal function by regulating follicular growth, estrogen production and oocyte maturation. Mutation in *FSHR* was the first autosomal molecular explanation for POI, elucidated prior to the contemporary era. [@DMV036C3] and [@DMV036C4], [@DMV036C5] ascertained 75 primary or secondary amenorrhea cases, and found homozygous mutations (c.566C\>T, p.A189V), in the extracellular portion of this G-protein receptor, in women of six Finnish families with hypergonadotrophic ovarian dysgenesis. This mutation resulted in a dramatic reduction of binding capacity and signal transduction, but with apparently normal ligand-binding affinity ([@DMV036C4]). The frequency of the c.566C\>T mutation is 0.96% in a Finnish population ([@DMV036C98]). However, subsequent screening in cohorts of different ethnicities seldom found mutations ([@DMV036C47]; [@DMV036C98]; [@DMV036C44]; [@DMV036C208]; [@DMV036C213]; [@DMV036C207]; [@DMV036C37]; [@DMV036C122]; [@DMV036C220]; [@DMV036C160]; [@DMV036C237]). Therefore, *FSHR* mutations are not uncommon in XX gonadal dysgenesis in Finland, but apparently rare elsewhere ([Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)). #### TGF, beta receptor III (TGFBR3) (1p33-p32) {#s3b2f} Human *TGFBR3* is located at 1p33-p32 and encodes the TGF-beta type III receptor. The encoded receptor is a membrane proteoglycan that often functions as a co-receptor with other TGF-beta receptor superfamily members. Two missense variants, p.E459G and p.P825L, were identified in Chinese women with idiopathic POI, both predicted to have functional and structural impacts on the TGFBR3 protein ([@DMV036C171]). Another missense mutation---p.P775S---was found in an Indian POI case ([@DMV036C61]). #### G protein-coupled receptor 3 (GPR3) (1p36.1-p35) {#s3b2g} The *GPR3* gene, located in 1p36.1-p35 and having 2 exons, is a member of the G protein-coupled receptor family. Predominantly expressed in oocytes, *GPR3* maintains meiotic arrest in antral follicles until the LH surge through pathways involved in cAMP and cGMP regulation. In *Gpr3^−/−^* mice, the majority of oocytes in antral follicles had unscheduled premature resumption of meiosis ([@DMV036C138]). A synonymous variant (c.135G\>A, p.V45V) was found in one Chinese patient ([@DMV036C252]), and another study also failed to find any potential disease-associated changes in 82 North American Caucasian women with POI ([@DMV036C112]). #### Wingless-type MMTV integration site family, member 4 (WNT4) (1p36.23-p35.1) {#s3b2h} *WNT4* encodes a secreted extracellular signaling protein that is expressed in human ovaries early in fetal development ([@DMV036C94]), and plays a critical role in female sex determination and differentiation. In the ovaries of *Wnt4* mutant mice, the rate of apoptosis was similar to that of wild type mice at birth; however, apoptotic cells progressively increased during follicular development ([@DMV036C94]). By sequencing the coding region of *WNT4* in 55 Tunisian women with POI, a synonymous variant in exon 2 (c.99G\>A, p.S33S) was identified ([@DMV036C120]). Mutational analysis was also performed in 145 Chinese women with POI with no causative variants found ([@DMV036C39]). #### Inhibins: inhibin, alpha (INHA) (2q35); inhibin, beta A (INHBA) (7p15-p13); inhibin, beta B (INHBB) (2cen-q13) {#s3b2i} Inhibin is a dimeric glycoprotein hormone. Belonging, like BMP-15 and GDF9, to the superfamily of TGF-β, inhibin is a negative regulator of FSH. Inhibin encompasses inhibin, alpha (*INHA*) (2q35), inhibin, beta A (*INHBA*) (7p15-p13), and inhibin, beta B (*INHBB*) (2cen-q13). The missense mutation c.769G\>A (p.A257T) in the *INHA* gene was more frequently found in patients with POI in New Zealand (7%) ([@DMV036C191]), India (11.2%) ([@DMV036C58]) and Italy (4.5%) ([@DMV036C133]). Increased susceptibility to POI was associated with impaired inhibin B bioactivity ([@DMV036C32]). The additional novel missense mutations c.275G\>A (p.S92N), c.525C\>G (p.H175Q) and c.545C\>A (p.A182D) were exclusively identified in Indian POI patients (1.25% for each mutation) ([@DMV036C60]). It is unclear whether polymorphisms in the *INHA* promoter result in reduced inhibin expression, but the promoter variant c.-16C\>T was significantly under-represented in patients with POI in New Zealand ([@DMV036C88]). The *INHBA* and *INHBB* gene encode inhibin βA and inhibin βB subunits. Causative mutations were not found in the two genes, however, except possibly for one synonymous mutation c.1032C\>T in the *INHBA* gene ([@DMV036C191]; [@DMV036C34]). #### POU class 5 homeobox 1 (POU5F1) (6p21.31) {#s3b2j} The *POU5F1* transcription factor gene, located on 6p21.31, is significantly down-regulated in *Nobox* knockout mice. Thus, *POU5F1* becomes a potential candidate gene for POI, a downstream target of *NOBOX*. [@DMV036C231] sequenced 175 Chinese POI cases and found one non-synonymous variant (p.P13T), a heterozygous hydrophobic to hydrophilic substitution. #### MutS homolog 4 (MSH4) (1p31) and *MSH5* (6p21.3) {#s3b2k} *MSH4* and *MSH5* belong to the DNA mismatch repair gene family, playing pivotal roles in meiotic recombination. Mammalian MSH4 and MSH5 proteins form a heterodimeric complex and exert essential functions for normal chromosome synapsis during zygotene. Disruption of *Msh4* or *Msh5* in female mice resulted in sterility, degenerated ovaries and progressive loss of oocytes due to meiotic failure ([@DMV036C53]; [@DMV036C109]). In a case--control study in a Caucasian population, in which both genes were sequenced, a heterozygous mutation p.P29S in *MSH5* was found in 2 of 41 cases. The mutation was located in the Hmsh4-binding domain of *MSH5* which could disrupt the integrity of the protein interaction between MSH5 and MSH4 ([@DMV036C131]). #### Forkhead box O3 (FOXO3) (6q21) {#s3b2l} Forkhead transcription factor *FOXO3*, located at 6q21, encodes a master regulator and potent suppressor of primordial follicle activation. Loss of *Foxo3* function in mice leads to POI due to global follicle activation ([@DMV036C124]). The frequency of variants in POI patients differs in different ethnic groups ([Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)), but in several populations the frequencies of *FOXO3* variants are not insignificant (6% in French and 13.3% in Chinese). However, the pathological role of these variants needs to be determined by functional studies. #### Cbp/p300-interacting transactivator, with Glu/Asp-rich carboxy-terminal domain, 2 (CITED2) (6q23.3) {#s3b2m} *CITED2* is essential for early embryonic development. This is evidenced by delayed differentiation of gonads in *Cited2^−/−^* mice ([@DMV036C43]). Fonseca and colleagues reported a novel missense mutation p.P202T in one of 116 Tunisian POI cases ([@DMV036C74]). Further studies in other populations are warranted. #### Spermatogenesis and oogenesis specific basic helix-loop-helix transcription factor 1 (SOHLH1) (9q34.3) and SOHLH2 (13q13.3) {#s3b2n} As germ cell specific master-master transcription factors, SOHLH1 and SOHLH2 orchestrate different oocyte-specific genes essential for early folliculogenesis. *Sohlh1/2*-deficient mice exhibit atrophied ovaries devoid of follicles due to defective primordial-to-primary follicle transition ([@DMV036C150]; [@DMV036C42]). Novel distinct heterozygous variants were identified in both *SOHLH1* and *SOHLH2* in large cohorts of women with POI of Han Chinese and Serbian origin ([@DMV036C175]; [@DMV036C250]). Plausible pathogenesis might involve disturbing the expression, transactivation or homo-/hetero-dimerization of SOHLH1 or SOHLH2 proteins. No subsequent reports exist, to our knowledge. #### Phosphatase and tensin homolog (PTEN) (10q23.3) {#s3b2o} Localized on chromosome 10q23.3, *PTEN* plays a causative role in early activation of primordial follicles by negatively regulating the PI3K pathway ([@DMV036C181]). Primordial follicles become depleted in *Pten* null mice in early adulthood, mimicking the phenotype of POI in humans. However, no causative mutation was detected in coding regions of *PTEN* gene in Japanese and Chinese women with POI ([@DMV036C195]; [@DMV036C248]). #### Nanos homolog 1, 2, 3 (Drosophila) (NANOS1, 10q26.11; NANOS2, 19q13.32; NANOS3, 19p13.13) {#s3b2p} The *NANOS* gene family is known to be required for primordial germ cell (PGC) development and maintenance. Three homologs (*NANOS1*, *NANOS2* and *NANOS3*) exist. Disruption of *Nanos1* in mice did not affect germ cell development, but knockout of *Nanos2* or *Nanos3* resulted in infertility with decreased gonad size due to loss of PGC. *Nanos2* deficiency only resulted in spermatogonia loss whereas *Nanos3* impaired PGC maintenance in both males and females ([@DMV036C216]). Mutations in *NANOS3* were investigated in 80 Chinese and 88 Caucasian women with POI ([@DMV036C168]). No causative mutations were found in coding exons. However, one potentially relevant heterozygous mutation (c.457C\>T; p.R153W) was identified in another study involving 100 Chinese POI patients ([@DMV036C239]). Functional studies showed decreased stability of NANOS3, potentially resulting in a hypomorph. And a homozygous mutation (c.358G\>A, p.E120K) was found in two sisters with primary amenorrhea from 85 Brazilian women with POI. *In vitro* and in silico functional studies revealed that this mutation impaired the ability of NANOS3 to prevent apoptosis, suggesting a mechanism for POI involving increased PGC apoptosis during embryonic cell migration ([@DMV036C187]). Taken together, these results suggest a role for *NANOS3* mutation in some cases of POI. #### Cyclin-dependent kinase inhibitor 1B (CDKN1B) (12p13.1-p12) {#s3b2q} *CDKN1B*, also known as *P27* and *KIP1*, encodes a cyclin-dependent kinase inhibitor that regulates proliferation and differentiation in many tissues. It suppresses ovarian follicle endowment and activation, and promotes follicle atresia. Premature follicle depletion occurred due to accelerated activation in *Cdkn1b* knockout mice ([@DMV036C178]). Sequence analysis of *CDKN1B* found one novel heterozygous mutation c.356T\>C (p.I119T) in one of 87 Tunisia POI patients ([@DMV036C144]). However, no variants were identified in Chinese cohorts ([@DMV036C229]; [@DMV036C249]), suggesting that mutations in *CDKN1B* are not common in POI, at least in this population. #### Anti-Mullerian hormone receptor, type II (AMHR2) (12q13) {#s3b2r} *AMHR2* encodes a receptor in the AMH pathway which plays a crucial role in the development and maintenance of reproductive organs in mammals. Polymorphism c.-482 A\>G (rs2002555) in *AMHR2* was revealed to be associated with age at menopause in interaction with parity in Dutch women, but no association was found with POI in Korean and Chinese women ([@DMV036C241]; [@DMV036C176]). Negative results in *AMHR2* were also reported in 16 Japanese women with POI ([@DMV036C226]). However, [@DMV036C176] identified two novel missense mutations (p.I209N and p.L354F) in a cohort of Chinese POI women. #### KIT ligand (KITLG) (12q22) {#s3b2s} The human *KITLG* gene, located at 12q22, encodes the ligand of a tyrosine-kinase receptor. KIT/KITLG plays a critical role during oogenesis and folliculogenesis. Mice with a deficiency in *Kitlg* manifested impaired PGCs ([@DMV036C135]). However, no perturbations were reported in the coding region of *KITLG* from 40 Caucasian POI patients ([@DMV036C91]). #### Forkhead box O1 (FOXO1) (13q14.1) {#s3b2t} *FOXO1*, another member of the forkhead family of transcription factors, is important in granulosa cell function and follicle maturation. [@DMV036C235] identified one 5′UTR mutation (c.-30C\>T) and one missense mutation (p.P84L) in 60 New Zealand and Slovenia POI patients. #### Spalt-like transcription factor 4 (SALL4) (20q13.2) {#s3b2u} SALL4, a zinc finger transcription factor, is expressed in murine oocytes. SALL4 binds to POU5F1 and could regulate its expression. Both *Sall4* and *Pou5f1* are drastically down-regulated in *Nobox^−/−^* newborn ovaries ([@DMV036C243]; [@DMV036C41]). [@DMV036C227] screened the coding regions of *SALL4* in 100 Han Chinese females with non-syndromic POI and identified two heterozygous missense mutations (p.V181M and p.T817A) in the conserved region. These may or may not be POI-associated gene variants. Further studies are needed to determine the functional effect of these variants. #### Meiotic protein covalently bound to DSB (SPO11) (20q13.31) {#s3b2v} *SPO11* is involved in meiosis, forming the double-strand breaks (DSBs) that initiate meiotic recombination. *Spo11^−/−^* mice are infertile with premature depletion of oocyte because of defective meiosis. However, no novel variants were found in 41 women with non-syndromic POI ([@DMV036C131]). It is not clear whether an association between *SPO11* mutation and sporadic POI exists in human. #### DNA meiotic recombinase 1 (DMC1) (22q13.1) {#s3b2w} Genes perturbing meiosis are logical candidates for non-syndromic POI. *DMC1* encodes a member of the superfamily of recombinases, which are important for repairing double- strand DNA breaks during mitosis and meiosis. Among 41 French women with POI, [@DMV036C131] found one POI case with homozygous mutation p.M200V. However, a subsequent screening revealed no mutation but two known SNPs in 192 Chinese women with POI ([@DMV036C232]). Pleiotropic single gene disorders in POI {#s3c} ---------------------------------------- Distinct from non-syndromic POI, pleiotropic Mendelian disorders may manifest POI as part of their phenotypic spectrum. Indeed, the most common single genetic explanation for POI is represented by such a disorder---premutation for fragile X syndrome (Table [III](#DMV036TB3){ref-type="table"}). Table IIICandidate genes responsible for Mendelian disorders that manifest POI.GeneLocationMendelian syndromeSomatic featuresReference*FMR1*Xq27.3Fragile X syndromeAttention deficits, hyperactivity, social deficits, anxiety disorder, deficits in cognitive flexibility.[@DMV036C182] and [@DMV036C203]*FOXL2*3q23Blepharophimosis-ptosis-epicanthus BPE type I syndrome, BPES IBPES type I is a complex eyelid malformation associated with POI. The major features of the eyelid malformation involve (i) narrowed horizontal aperture of the eyelids (blepharophimosis), (ii) drooping of the upper eyelid (ptosis), (iii) the presence of a fold of skin arising from the lower eyelid that runs inward and upward (epicanthus inversus), and (iv) lateral displacement of the inner canthi (telecanthus).[@DMV036C253] and [@DMV036C146]*GALT*9p13GalactosemiaCataracts, speech defects, poor growth, poor intellectual function, neurologic deficits (predominantly extrapyramidal findings with ataxia).[@DMV036C189]*AIRE*21q22.3Autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy syndrome, APECEDCandidiasis, Addison\'s disease, hypoparathyroidism, type 1 diabetes, alopecia, vitiligo, ectodermal dystrophy, celiac disease and other intestinal dysfunctions, chronic atrophic gastritis, chronic active hepatitis, autoimmune thyroid disorders, pernicious anemia.[@DMV036C70]*EIF2B*EIF2B2 -14q24.3; EIF2B4-\ 2p23.3; EIF2B5-\ 3q27.1Central nervous system leukodystrophy and ovarian failure, ovarioleukodystrophyNeurological disorder characterized by involvement of the white matter of the central nervous system. When Leukodystrophies associated with premature ovarian failure referred to as ovarioleukodystrophy.[@DMV036C134]*POLG*15q25Progressive external ophthalmoplegia, PEOManifestations range from involvement limited to the eyelids and extraocular muscles.[@DMV036C84]*NOG*17q22Proximal symphalangism, SYM1Ankylosis of the proximalinterphalangeal joints.[@DMV036C110]*PMM2*16p13PMM2-CDG CDG-1 (a previously known as congenital disorder of glycosylation type 1a)Cerebellar dysfunction (ataxia, dysarthria, dysmetria), non-progressive cognitive impairment, stroke-like episodes, peripheral neuropathy with or without muscle wasting, absent puberty in females, small testes in males, retinitis pigmentosa, progressive scoliosis with truncal shortening, joint contractures, and premature aging[@DMV036C202]*HSD17B4*\ *HARS2*\ *CLPP*\ *LARS2*\ *C10orf2*5q21\ 5q31.3\ 19p13.3\ 3p21.3\ 10q24Perrault syndrome, PSSensorineural deafness in both males and females, and neurological manifestations in some patients.[@DMV036C97], [@DMV036C139], [@DMV036C156], [@DMV036C157] and [@DMV036C155]*BLM*15q26.1Bloom syndromeChromosomal breakage leading to early onset of aging, short stature and elevated rates of most cancers.[@DMV036C65]*ATM*11q22-q23Ataxia telangiectasia, A-TProgressive cerebellar degeneration, telangiectasias, immunodeficiency, recurrent infections, insulin-resistant diabetes, premature aging, radiosensitivity, and high risk for epithelial cancers in surviving adults.[@DMV036C80] and [@DMV036C205]*WRN*8p12Werner syndromePremature aging of the skin, vasculature, and bone and elevated rates of certain cancers, particularly sarcomas.[@DMV036C68]*RECQL4*8q24.3Rothmund--Thomson syndrome, RTSCutaneous rash, sparse hair, small stature, skeletal and dental abnormalities, cataracts, premature aging, and an increased risk for cancer, especially malignancies originating from bone and skin tissue.[@DMV036C225][^6] ### Fragile X syndrome: familial mental retardation 1 (FMR1) (Xq27.3) {#s3c1} Perturbations of *FMR1* are responsible for fragile X syndrome. Clinical features include mental retardation, characteristic facial features with large ears and prominent jaw, connective tissue findings (joint laxity), large testes after puberty, and behavioral abnormalities. Fragile X syndrome occurs in males when CGG repeats number above 200. In females ∼70% of women with \>200 CGG repeats show intellectual disability ([@DMV036C52]). The incidence of fragile X syndrome in males is approximately 1 in 4000, and in females 1 in 8000 ([@DMV036C2]) The normal number of CGG repeats in *FMR1* is 32. Thereafter, there is a stage (premutation) in which 54--200 CGG repeats exist. Pathogenic effects including mental retardation and ataxia may exist, as well as POI. About 15--20% of women with a FMR1 premutation develop POI ([@DMV036C236]). Conversely, 5% of sporadic cases and 10--15% of familial cases in the Caucasian population are explained by *FMR1* premutations. For reasons that are not clear, the number of CGG repeats significantly correlates with risk of POI only within selected ranges. There is only a slightly increased risk of expansion associated with 40--79 repeats; but higher risk with 80--99 repeats, yet no further increased risk occurs after \>100 repeats, and as noted POI is not observed with the full mutation (\>200 CGG) ([@DMV036C7]). One possible explanation is that certain genes are suppressed in the 54--100 premutation range and link to POI, whereas other genes become suppressed only with higher numbers of CGG repeats. Perhaps, then, phenotypes in the two groups differ because some ovarian genes are inhibitory and others are the converse. Thus, ovarian function may be initially suppressed but later return to normal function. Given the comparatively higher frequency of premutation of *FMR1* in POI than the general population ([Supplemental Table SIII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)), *FMR1* testing has become part of the work-up for women with POI. It is formally recommended in Europe ([@DMV036C76]; [@DMV036C69]). In other populations the prevalence is lower. [@DMV036C86] reported that only 2 premutation carriers were found in 379 sporadic Chinese POI cases (0.49%); none were found in 402 controls. The frequency in Chinese women is thus considerably lower than in Caucasian women (3.3--6.7%). Frequencies are 1.56% in Japanese ([@DMV036C93]) and 4.8% in Slovenian women ([@DMV036C81]). ### Blepharophimosis-ptosis-epicanthus syndrome (BPES): forkhead box L2 (FOXL2) (3q23) {#s3c2} Blepharophimosis-ptosis-epicanthus syndrome (BPES) is a pleiotropic autosomal dominant syndrome in which *FOXL2* is perturbed and premature ovarian failure occurs ([@DMV036C46]). That *FOXL2* plays a key if not the pivotal role in ovarian development initially came from study of BPES kindreds. More than one hundred unique *FOXL2* mutations have now been described in BPES in different populations ([@DMV036C14]). By contrast, constitutional mutations are uncommon but reported in non-syndromic POI. Perhaps 2--3% of isolated POI cases have a *FOXL2* mutation ([@DMV036C50]; [@DMV036C87]) ([Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)). ### Galactosemia: galactose 1-phosphate uridyl transferase (GALT) (9p13) {#s3c3} Galactosemia is caused by deficiency of galactose 1-phosphate uridyl transferase (*GALT*). Ovarian failure is a common long-term complication in girls and women with galactosemia, first described by Kaufman and coworkers ([@DMV036C104], [@DMV036C105]), who observed POI in 12 of 18 (67%) galactosemic women. Later [@DMV036C223] reported that 17% (8/47) galactosemic women presented with ovarian failure. However, with regard to duarte galactosemia, a mild variant of GALT deficiency, no apparent ovarian dysfunction was reported ([@DMV036C10]). Pathogenesis involves excess galactose toxicity that impairs folliculogenesis, induces resistance to gonadotrophins and accelerates follicular atresia ([@DMV036C77]; [@DMV036C11]). ### Carbohydrate-deficient glycoprotein syndrome type I (CDG-Ia): phosphomannomutase 2 (PMM2) (16p13) {#s3c4} In type 1 carbohydrate-deficient glycoprotein deficiency, also named phosphomannomutase deficiency, mannose-6-phosphate cannot be converted to mannose-1-phosphate. This lipid linked oligosaccharide is necessary for formation of secretory glycoproteins. Neurologic abnormalities and ovarian failure occur ([@DMV036C55]; [@DMV036C113]). Located on 16p13, *PMM2* is typically caused by a missense mutation ([@DMV036C18]). ### Proximal symphalangism (SYM1) and multiple synostoses syndrome (SYNS1): noggin (NOG) (17q22) {#s3c5} *NOG* encodes a secreted polypeptide that binds to and inactivates members of the TGF-β superfamily (i.e. BMP2, 4, 7, 14 and GDF5). NOG is expressed in various tissues including female reproductive organs. *NOG* mutations are known to explain proximal symphalangism (SYM1) and multiple synostoses syndrome (SYNS1) ([@DMV036C82]). In addition, [@DMV036C110] described a heterozygous mutation (p.E48K) in *NOG* in a female presenting with SYM1 and also having POI. However, [@DMV036C116] concluded the relationship between *NOG* mutations and non-syndromic POI was not clear, having screened the coding sequence of *NOG* in 100 non-syndromic sporadic POI patients and identifying only one heterozygous mutation (p.G92E) ([Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)). Actually this experience mirrors the situation involving other pleiotropic genes causing POI. ### Autoimmune regulation/autoimmune polyendocrinopathy-candidiasis-ectodermal dystrophy (APECED): autoimmune regulator (AIRE) (21q22.3) {#s3c6} The *AIRE* gene can if perturbed lead to multi-system abnormalities: alopecia, vitiligo, keratopathy, malabsorption, hepatitis and mucocutaneous candidiasis. Ovarian hypoplasia often occurs, usually in the third decade ([@DMV036C224]). Many different *AIRE* perturbations have been found in this autosomal dominant disorder, not only nonsense mutations but also frame shifts. No particular mutation leads to ovarian failure as distinct from other autoimmune phenomena. *AIRE* mutations have not yet been sought in women with isolated POI (non-syndromic). ### Ovarian leukodystrophy: eukaryotic translation initiation factor (EIF2B): subunit 2 beta, 39 kDa (EIF2B2) (14q24.3); subunit 4 delta, 67 kDa (EIF2B4) (2p23.3); subunit 5 epsilon, 82 kDa (EIF2B5) (3q27.1) {#s3c7} 'Vanishing white matter' leads to variable but progressive neurological degeneration. Ovarian failure may coexist ([@DMV036C190]; [@DMV036C21]). That the causative gene (*EIF2B*) allows denatured stress-related proteins to accumulate makes plausible the relevance to oogenesis, given ongoing oocyte degeneration. [@DMV036C71] found variants in *EIF2B2*, *EIF2B4* and *EIF2B5* in seven POI cases with neurologic abnormalities, but in 93 non-syndromic POI none were found positive ([@DMV036C72]). Therefore, *EIF2B* genes have not yet been shown to be responsible for non-syndromic POI. ### Perrault syndrome: hydroxysteroid (17-beta) dehydrogenase 4, HSD17B4, 5q21; histidyl-tRNA synthetase 2, mitochondrial, HARS2, 5q31.3; leucyl-tRNA synthetase 2, mitochondrial, LARS2, 3p21.3; caseinolytic mitochondrial matrix peptidase proteolytic subunit, CLPP, 19p13.3; chromosome 10 open reading frame 2, C10orf2, 10q24 {#s3c8} Perrault syndrome is a well-recognized pleiotropic autosomal recessive disorder characterized by ovarian failure in females, progressive sensorineural deafness in both males and females, and in some patients, neurological manifestations. Only recently have the underlying genes been identified and proved to be heterogeneous. In a family of mixed European ancestry with two sisters presenting with Perrault syndrome, compound heterozygous variants - c.650A\>G (p.Y217C) and c.1704T\>A (p.Y568X), in *HSD17B4* were found. Then with similar genomic strategies with linkage analysis or whole-exome sequencing (WES), mutations in *HARS2*, *LARS2*, *CLPP* and *C10orf2* have been found in the context of Perrault syndrome ([@DMV036C156], [@DMV036C157]; [@DMV036C97]; [@DMV036C139]). All these genes are essential for normal mitochondrial function. This group of causative genes is detailed in 'Mitochondrial Genes Causing POI' Section. ### Other pleiotropic disorders {#s3c9} Table [III](#DMV036TB3){ref-type="table"} lists other nuclear genes causing syndromes in which POI is a feature. These include *POLG* associated with progressive external ophthalmoplegia (PEO, detailed in 'Mitochondrial Genes Causing POI' section), *BLM* with Bloom syndrome, *AIRE* with Ataxia telangiectasia, *WRN* with Werner syndrome and *RECQL4* with Rothmund--Thomson syndrome. A common feature of several is chromosomal breakage, best exemplified by Bloom syndrome ([@DMV036C200]; [@DMV036C199]). In aggregate these and other conditions in Table [III](#DMV036TB3){ref-type="table"} explain no more than 1% of cases of POI. From a scientific perspective however, elucidating the role these genes play in reproduction offers novel clues to integrity needed for normal ovarian differentiation. Clinicians caring for females with these syndromes should thoroughly investigate and evaluate any reported menstrual irregularities. Mitochondrial genes causing POI {#s3d} ------------------------------- Perturbations of mitochondrial genes or nuclear genes affecting mitochondria are good candidates for POI because the mature oocyte has the greatest number of mitochondria of any human cell. Mature oocytes readily accumulate mitochondria during oogenesis, mitochondrial biogenesis playing an essential role in oocyte maturation, fertilization and embryo development. Dysregulation of mitochondrial dynamics contributes to excess oxidative stress and initiation of apoptosis, thus accelerating follicle depletion. A marked quantitative decrease of mitochondrial DNA (mt DNA) in oocytes and peripheral blood cells has been well documented in women with ovarian insufficiency ([@DMV036C136]; [@DMV036C22]). Thus, any gene affecting mitochondria involving muscular and neurological disturbance is a candidate, because these systems are so dependent on mitochondrial integrity. Genes governing mitochondrial functions may be located in the nucleus, like those discussed in 'Single Genes Causing Non-syndromic POI' section, or in mitochondria itself (mt DNA). To date, those of relevance to POI have been nuclear genes. ### Progressive external ophthalmoplegia (PEO): polymerase (DNA directed), gamma (POLG) (15q25) {#s3d1} mtDNA mutations usually affect muscular and neurological systems. In progressive external ophthalmoplegia (PEO) clinical features are proximal myopathy, sensory ataxia and parkinsonism. The causative mutation lies in the gene encoding polymerase gamma (*POLG*), which is responsible for mtDNA replication and repair. In three of seven families studied by [@DMV036C129], POI cosegregated with PEO. A p.Y955C mutation in *POLG* was found in two of these three families. p.Y955C (tyrosine to cytosine) affects a highly conserved region, rending functional plausibility. Compound heterozygosity (N468D/A1105T) was observed by [@DMV036C148] in another 3-generation family in which multiple family members were affected with both PEO and POI. Mutation screening of *POLG* in *isolated* POI, has, however, yielded few perturbations. A single p.R953C mutation was found in 201 POI cases (0.5%) cumulatively reported by [@DMV036C214]. No novel mutations were identified in Italian and British women with POI ([@DMV036C148]; [@DMV036C22]; [@DMV036C64]) ([Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)). Therefore, *POLG* variation is not a common explanation for sporadic POI in the absence of clinical suspicion for other mitochondrial-associated physical signs. ### Perrault syndrome: HARS2, 5q31.3; LARS2, 3p21.3; CLPP, 19p13.3; C10orf2, 10q24 {#s3d2} As reviewed in 'Pleiotropic Single Gene Disorders Having POI' Section, ovarian failure is one of the characterized features in Perrault syndrome in females. The genes listed encode mitochondrial tRNA synthetase, chambered protease and primase-helicase and have been found to harbor mutations responsible for POI in Perrault syndrome ([@DMV036C156], [@DMV036C157]; [@DMV036C97]; [@DMV036C139]). Through WES in a consanguineous Palestinian family and a nonconsanguineous Slovenian family with Perrault Syndrome, homozygous mutation c.1565C\>A (p.T522N) and compound heterozygous mutation c.1077delT and c.1886C\>T (p.T629M) in the *LARS2* gene were identified, respectively ([@DMV036C157]). WES was also performed in a nonconsanguineous family of mixed European ancestry, showing compound heterozygous mutations c.598C\>G (p.L200V) and c.1102G\>T (p.V368L) in the *HARS2* gene ([@DMV036C156]). The homozygous mutations c.433A\>C (p.T145P), c.440G\>C (p.C147S) and c.270+4A\>G in the *CLPP* gene were observed in three consanguineous Pakistani families with Perrault Syndrome ([@DMV036C97]). In a Japanese family, affected individuals carried compound heterozygous p.R391H and p.N585S in the *C10orf2* gene essential for replication of mtDNA; meanwhile compound heterozygous mutations p.W441G and p.V507I were responsible for this disease in another family of European ancestry ([@DMV036C139]). In aggregate, the above supports a critical role for genes controlling mitochondria in the maintenance of ovarian function and their roles in non-syndromic POI warrant further investigation. Multiple malformation syndromes characterized by POI {#s3e} ---------------------------------------------------- In other syndromes, POI or primary ovarian failure (gonadal dysgenesis) is accepted as one component but the causative gene has not been found. [Supplementary Table SIV](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1) lists these conditions. Of particular interest is POI associated with cerebellar ataxia ([@DMV036C198]). Genome-wide studies in POI {#s3f} -------------------------- Contemporary genetic strategies applied to locate susceptible loci or genes causing POI have extended beyond suspected candidate gene interrogations to genome-wide approaches. Approaches include linkage analysis in families with multiple affected members, CGH for CNV, genome-wide association studies (GWAS), genome-wide sequencing of exomes (WES) and, in the future, whole genome sequencing (WGS). ### GWAS {#s3f1} In GWAS, also known as whole genome association study (WGAS), one examines many common genetic variants in different individuals to see if any variant is associated with a trait. In GWAS one searches agnostically through the entire genome to identify variants (SNPs) more common in cases than controls of similar ethnicity. No *a priori* expectations exist. Six GWAS have been performed to identify variants associated with POI, but not all adhered to accepted criteria. Sample sizes were often very small and replicates not sought (Table [IV](#DMV036TB4){ref-type="table"}). The first GWAS showed association with the PTH-responsive B1 gene (*PTHB1*) in a small discovery set of 24 women and 24 controls ([@DMV036C102]). [@DMV036C107] identified an association of an intron SNP in ADAM metallopeptidase with thrombospondin type 1 motif, 19 (*ADAMTS19*) with POI in a discovery set of only 99 Dutch women and 181 controls; no replication set existed. Laminin, gamma 1 (*LAMC1*) was then reported to be significantly associated with POI in Korea in 122 cases versus 242 controls ([@DMV036C164]). Table IVGenome-wide association studies for POI.[@DMV036C102][@DMV036C107][@DMV036C173])[@DMV036C164][@DMV036C145][@DMV036C26]EthnicityKoreanCaucasian (Dutch)ChineseKoreanDutchMiddle-EasternDiscovery set No. of cases249939124105 No. of controls241818952454 Associations in discovery set*PTHB1* at 7p14 showed strongest association.\ Ht1 GAAAG: POI-susceptible haplotype;\ Ht2 TGTGC: POI-resistant haplotype.rs246246 mapped to *ADAMTS19* intron8q22.3\ (10^−6^)22 SNPs in *LAMC1* associated with POISusceptible locus: 5q14.1-q15Susceptible loci: 7p21.1-15.3, 7q21.3-22.2Replication set No. of cases1016040098---- No. of controls8790800218---- Results of replication set*PTHB1* associated with POI; Ht1 confers susceptibility to POI.Association not confirmed.Frequencies of 9 SNPs and 1 haplotype were higher in POI than in control.--Sequencing three candidate genes *DLX5*, *DLX6* and *DSS1* did not reveal causal mutations[^7] Our group conducted the largest GWAS with an initial discovery set of 391 cases versus 895 controls; the independent replication set consisted of 400 cases and 800 controls. The most significant association was at 8q22.3 (1.6 × 10^−6^--3.86 × 10^−6^). This falls short of the canonical 10^−8^ expected to confer unequivocal significance. This region does not contain a protein coding candidate gene ([@DMV036C173]). However, it is now appreciated that 90% of significant GWAS associations are in intragenic or intergenic regions, portions of the 98.5% of the genome not coding for protein. These regions include a host of regulatory genes and networks (ENCODE). 8q22.3 may be an important yet undefined long-distance regulatory region affecting ovarian differentiation and oogenesis. Disruption might lead to ovarian failure, analogous to disruption of the region upstream of Sex determining region Y-box 9 (*SOX9*) that causes XY sex reversal in mice and humans ([@DMV036C165], [@DMV036C166]). Replication in independent cohorts needs to be performed to determine potential causative roles. GWAS have revealed multiple loci potentially associated with POI in Chinese, Korean, and Dutch women. However, in each it was difficult to implicate specific novel genes, and in none did significance exceed 10^−6^; positive findings were not always replicated. The most likely explanation for these data is limitation based on small sample size. In no study did sample number exceed 1000, thus lacking statistical power sufficient to detect a modest association when evaluating hundreds of thousands of SNPs. That POI is a rarer condition than polycystic ovary syndrome or endometriosis makes it difficult to accrue the requisite sample size possible in studying those disorders ([@DMV036C217]; [@DMV036C38]; [@DMV036C192]). ### GWAS involving family linkage analysis {#s3f2} Table [IV](#DMV036TB4){ref-type="table"} includes two genome-wide association linkage studies involving single extended families. In a Dutch POI family subjected to a genome-wide linkage scan 5q14.1-q15 was found to be a susceptibility region ([@DMV036C145]). [@DMV036C26] performed genome-wide linkage and homozygosity analysis in a large consanguineous Middle-Eastern POI-affected family showing autosomal inheritance. Two regions---7p21.1-p15.3 and 7q21.3-q22.2---were identified as candidate regions by homozygosity mapping. However, sequencing the three most plausible candidate genes in this region---*DLX5*, *DLX6* and *DSS1*---failed to reveal mutations. ### GWAS involving age of menopause {#s3f3} Insights derived from shared genetic susceptibility between POI and either age at natural menopause (AANM) or early menopause (EM) represent another potential path identifying novel entry points for unraveling genetic mechanism involved in POI. Thus, we gathered 36 SNPs shown from previous GWAS studies seeking SNPs associated with AANM or EM, plus 3 additional SNPs in *ESR1* and 2 additional SNPs in *PTHB1.* Differential association was then sought in 371 POI and 800 controls, all of Chinese Han origin ([@DMV036C174]). Three SNPs, rs2278493 in hexokinase 3 (*HK3*), rs2234693 in estrogen receptor 1 (*ESR1*) and rs12611091 in BR serine/threonine kinase 1 (*BRSK1*), showed nominally significant association with POI. Thus, a plausible relationship could exist between POI and *ESR1*, *BRSK1*, *HK3*. ### Cytogenomic studies (CNV) {#s3f4} There is increasing interest toward whole genome studies based on CNVs. Using CGH, one can efficiently search for small duplications or deletions potentially associated with a complex trait (such as POI). The purpose of identifying CNVs is that they may contain or be linked to a causative gene for the disorder studied. Seven array CGH studies have identified CNVs associated with POI ([Supplementary Table SV](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)). In 2009, [@DMV036C1] reported that 8 previously known CNVs in 99 POI French patients located near 5 potential candidate genes---Dynein, axonemal, heavy chain 5 (*DNAH5*), NLR family, apoptosis inhibitory protein (*NAIP*), Dual specificity phosphatase 22 (*DUSP22*), Nuclear protein, transcriptional regulator, 1 (*NUPR1*), and *AKT1*. However, this and all other CGH studies have the major pitfall of usually lacking information on parents. A large, *de novo*, CNV deletion is more likely to be pathogenic than an inherited CNV. Another array CGH study involving 74 German patients with POI or ovarian dysgenesis identified 44 losses or gains at autosomes and X chromosome that might explain POI ([@DMV036C123]). [@DMV036C137] identified 17 novel microduplications and 7 novel microdeletions among 89 POI patients, all but one located at autosomes. Included were two novel microdeletions causing haploinsufficiency for Synaptonemal complex central element protein 1 (*SYCE1*) and Cytoplasmic polyadenylation element binding protein 1 (*CPEB1*), genes known to cause ovarian failure in knockout mouse models ([@DMV036C20]; [@DMV036C143]). Recently, a high-resolution array CGH identified eleven unique CNVs in 11 patients with POI. Among these CNVs, a tandem duplication of 475 bp containing 3 *NOBOX*-binding elements and an E-box important for *GDF9* gene regulation in the promoter of *GDF9* is likely causative of POI ([@DMV036C142]). Using a complete X chromosome tiling path array, [@DMV036C177] found 15 novel discrete X chromosome intervals in 20/42 (48%) women with POI in the UK. However, in patients from New Zealand, [@DMV036C63] detected only two microduplications (Xp22.33 and Xq13.3) in a low frequency of 4%. [@DMV036C108] found one CNV in Xq21.3 to be associated with POI, specifically in a region where Protocadherin 11 X-linked (*PCDH11X*) and TGFB-induced factor homeobox 2-like, X-linked (*TGIF2LX*) are located. Interestingly, no deletions were found in other regions, several considered on the basis of traditional cytogenetic studies to be pivotal ([@DMV036C201]). Differing array CGH platforms and definitions for pathogenicity probably explain some of the discrepant results among the above studies. Also problematic is the limited sample size and lack of parental studies to exclude a CNV ostensibly associated with POI merely having been transmitted from a normal mother. ### WES {#s3f5} Low prevalence and impaired fecundity resulting in limited numbers of POI pedigrees without associated somatic anomalies (non-syndromic) has led to increasing application of WES, another agnostic approach. WES has identified several genes in POI not previously anticipated. In POI associated with somatic features, causative perturbations have been found by WES for *HSB17B4*, *LARS2*, *CLPP* and *C10orf2* in Perrault syndrome ([@DMV036C155], [@DMV036C157]; [@DMV036C97]; [@DMV036C139]) (detailed in 'Mitochondrial Genes Causing POI' section and Table [V](#DMV036TB5){ref-type="table"}). Up to the present, there have been six WES conducted in non-syndromic POI pedigrees. Interestingly, almost all plausible candidate genes identified were involved in meiosis and DNA repair (Table [V](#DMV036TB5){ref-type="table"}). Table VWhole-exome sequencing for syndromic or isolated POI.GeneLocationMutationFunctionCompound Het/Homo/HetReference*HSB17B4*5q21c.650A\>G (p.Y217C); c.1704T\>A (p.Y568X)Reduced expression with mutant proteinCompound heterozygous[@DMV036C155]*CLPP*19p13.3c.433A\>C (p.T145P),Alter the structure of the CLPP barrel chamber that captures unfolded proteins and exposes them to proteolysisHomozygous[@DMV036C97]c.440G\>C (p.C147S)c.270+4A\>GWeakens donor splice-site function*LARS2*3p21.3c.1565C\>A (p.T522N)Partially functional in Yeast complementation assayHomozygous[@DMV036C157]c.1077delT;\ c.1886C\>T (p.T629M)Nonfunctional in Yeast complementation assayCompound heterozygous*C10orf2*10q24c.1172G\>A (p.R391H) c.1754A\>G (p.N585S)Affect interactions of the linker domain\ Likely affect enzyme activityCompound heterozygous[@DMV036C139]c.1321T\>G (p.W441G) c.1519G\>A (p.V507I)Affect interactions of the linker domainCompound heterozygous*HFM1*1p22.2c.1686-1G\>C\ c.2651T\>G (p.I884S)Compound heterozygous[@DMV036C233]c.2206G\>A (p.G736S); c.3929_3930 delinsG, (p.P1310R fs\*41)Compound heterozygous*MCM9*6q22.31c.1732+2T\>CAbnormal splicing and truncated protein that are unable to be recruited to sites necessary for DNA damageHomozygous[@DMV036C238]c.394C\>T (p.R132\*)Repair of chromosome breaks impaired in lymphocytesHomozygous*STAG3*7q22.1c.968delC (p.F187fs\*7)Homozygous[@DMV036C27]*SYCE1*10q26.3c.613C\>T (p.Q205\*)Homozygous[@DMV036C54]*MCM8*20p12.3c.446C\>G (p.P149R)Impedes the repair of MMC-induced chromosomal breaks; inhibits MCM8 recruitment to sites of DNA damage; impairs DNA binding ability at the N-terminusHomozygous[@DMV036C6]*eIF4ENIF1*22q11.2c.1286C\>G (p.S429\*)Heterozygous[@DMV036C103][^8] #### ATP-dependent DNA helicase homolog (HFM1) (1p22.2) {#s3f5a} *HFM1*, a meiotic gene encoding DNA helicase preferentially expressed in germ-line tissues such as testis and ovary, is necessary for homologous recombination and synapsis during meiosis ([@DMV036C210]). *Hfm1*^−/−^ female mice had significantly reduced ovaries and follicle numbers and an increase in stromal cells ([@DMV036C85]). The recent report by [@DMV036C233]) identified a compound heterozygous mutation (c.1686-1G\>C and c.2651T\>G, p.I884S) in the *HFM1* gene in two affected Chinese sisters. Screening for *HFM1* mutations in a cohort of 69 Chinese women with sporadic POI identified another patient with compound heterozygous mutations (c.2206G\>A, p.G736S and c.3929_3930 delinsG, p.P1310R fs\*41). These variants were not found in 316 matched controls or databases. #### Minichromosome maintenance complex component 8 and 9 (MCM8, 20p12.3; MCM9, 6q22.31) {#s3f5b} *MCM8* and *MCM9,* recently discovered members of the highly conserved mini-chromosome maintenance proteins (MCM), are genes implicated in homologous recombination and repair of double-stranded DNA breaks. The MCM8/MCM9 complex is required for the resolution of dsDNA breaks that occur during homologous recombination in pachytene of meiosis I. Failure to resolve breaks predictably leads to oocyte death and small or absent ovaries. *Mcm8* and *Mcm9-*deficient mice are infertile and have small gonads due to germ cell depletion ([@DMV036C130]). Recently, Rajkovic and colleagues ([@DMV036C6]; [@DMV036C238]) discovered homozygous mutations in *MCM8* and *MCM9* genes in consanguineous families with POI (primary amenorrhea), using WES. One of the two mutations found in *MCM9* is c.1732+2T\>C, which resulted in abnormal alternative splicing and truncated forms of MCM9 that are unable to be recruited to sites of DNA damage. The other mutation c.394C\>T (p.R132\*) results in loss of function of MCM9. It is suspected that preferential sensitivity of germline meiosis to MCM9 functional deficiency and compromised DNA repair in the somatic component most likely account for the ovarian failure and short stature. A homozygous mutant c.446C\>G (p.P149R) found in *MCM8* inhibited recruitment of MCM8 to sites of DNA damage and led to genomic instability. ASNP rs16991615 in *MCM8* was also found associated with the age of natural menopause in the GWAS previously discussed ([@DMV036C40]). The role the novel *MCM8/MCM9* pathway plays in women with idiopathic POI needs to be explored further. #### Stromal antigen 3 (STAG3) (7q22.1) {#s3f5c} *STAG3* encodes a subunit of cohesin, a large protein complex that is essential for proper pairing and segregation of chromosomes during meiosis. A homozygous frameshift mutation resulting from a 1-bp deletion (c.968delC, p.F187fs\*7) in *STAG3* was identified in a highly consanguineous Palestinian pedigree; in which a 10-Mb region on 7q21.3--22.2 and a 3-Mb region on 7p21.1--15.3 had been previously identified to show significant linkage with the POI phenotype ([@DMV036C26], [@DMV036C27]). Deficiency of *Stag3* in female mice results in severe and early ovarian dysgenesis, with distinctive lack of oocytes and ovarian follicles. Early meiotic arrest and the centromeric chromosomal cohesion defects were observed in *Stag3^−/−^*fetal oocytes providing further evidence that *Stag3* is essential for assembly of the meiotic cohesin ring and the synaptonemal complex ([@DMV036C27]). #### Synaptonemal complex central element 1 (SYCE1) (10q26.3) {#s3f5d} The *SYCE1* gene encodes a component of the synaptonemal complex where paired chromosome homologs closely associate in meiosis (synapsis) before crossover. *Syce1*^−/−^ mice were infertile with smaller gonads, and showed loss of follicles in ovaries and postmeiotic cells in testis. Early meiosis arrest at prophase\\xE2\\x85followed by cell apoptosis could explain the phenotype ([@DMV036C20]). Different microdeletions of *SYCE1* have also been reported in Caucasian and Chinese POI patients ([@DMV036C137]; [@DMV036C251]). In a consanguineous Israeli Arab family, a homozygous nonsense mutation (c.613C\>T, p.Q205\*) in *SYCE1* was identified in two sisters with primary amenorrhea, presumably inherited in autosomal recessive fashion ([@DMV036C54]). #### Eukaryotic translation initiation factor 4E nuclear import factor 1 (eIF4ENIF1) (22q11.2) {#s3f5e} eIF4ENIF1, a transport protein, plays an important role in repressing translation through eIF4E. Both genes appear to play important roles in ovarian germ cell development ([@DMV036C219]). A heterozygous nonsense mutation (c.1286C\>G, p.S429\*) was identified in *eIF4ENIF1*, segregating with ovarian insufficiency (menopause age 29--35 years) in a large French Canadian family. No additional mutations were identified in *eIF4ENIF1* or *eIF4E* in 38 unrelated women with isolated POI. Haploinsufficiency or nuclear sequestration might disrupt the development of a normal oocyte complement ([@DMV036C54]). The whole exome studies in familial POI mentioned above mainly involve genes crucial during meiosis, such as generating and repairing DSBs, chromosome synapsis and recombination, and sister chromatid cohesion. This implies that perturbation of genes encoding proteins that regulate meiosis can result in autosomal recessive primary ovarian insufficiency in humans. Despite some negative candidate gene results with meiotic genes (*DMC1*, *SPO11* and *MSH4*) in POI, WES results still provide reasons to pursue other genes participating in meiosis and DNA repair in ubiquitous pathways. ### NGS {#s3f6} Future studies on targeted candidate genes of POI can be anticipated using NGS. A recent abstract ([@DMV036C75]) reported a study of 12 POI cases having non-syndromic POI and 176 control women whose menopause had occurred after age 50 years. A further 345 women from the same ethnic origin were stated to be recruited to assess allele frequency for potentially deleterious sequence variants. In the 12 POF cases, complete coding regions of 70 candidate genes were fully sequenced with mutations claimed in *ADAMTS19*, *BMPR2* and *LHCGR* genes. Full details are awaited. Regulatory genes and networks {#s3g} ----------------------------- Seeking POF genes has to date largely focused on coding variants, presuming plausible protein disruption. However, only 1.5% of the genome is protein-coding. Indeed, many POF-associated variants in whole genome studies map within, or in linkage disequilibrium to, intronic or intergenic regions; thus, these regions likely contain causative regulatory genes or networks. In our own POI GWAS, for example, 'gene desert' 8q22.3 was the region of most significance by association ([@DMV036C173]). Non-coding variants must be more robustly interrogated. Perturbations sought should include involving non-coding RNA (microRNA \[miRNA\], long non-coding RNA \[lncRNA\]), disruption or creation of alternative splicing or transcription factor-binding sites, and epigenetic modifications (DNA methylation patterns, chromatin modification). Large epigenetic consortia, such as ENCODE (<http://genome.ucsc.edu/ENCODE/>), Roadmap (<http://www.roadmapepigenomics.org/>) and iHEC (<http://www.ihec-epigenomes.org/>) point to approaches used to characterize the regulatory landscape of susceptibility regions for specific cell types. Unfortunately, ENCODE ([@DMV036C17]; [@DMV036C66]; [@DMV036C204]) bypassed the reproductive track (males and females), and thus an ovarian ENCODE does not yet exist. In order to generate the reproductive ENCODE, human ovarian tissue of specific cell types must be studied. Then, differences in regulatory genes between affected and normal individuals can be determined. Massive parallel sequencing using RNA-seq, or Methyl-seq will facilitate a systematic study of the transcriptome in ovaries in relation to genotypes and variants in POI. Chromatin immunoprecipitation followed by sequencing (ChIP-Seq) and RNA-Seq can also help identify novel transcriptional and epigenetic regions and, hence, mechanisms that potentially underlie POI. To analyze the complete proteomes and quantification of post-translational modifications, mass spectrometry would be the most attractive approach. Having accumulated the requisite information, network and pathway-based analyses of assembled datasets can be assembled using systems biology approaches. Cascades of transcriptional regulation can be evaluated simultaneously, identifying overlap among genes or putative networks that currently constrain key variables in model systems. The ultimate proof that an identified variant has a pathological effect traditionally relies on constitutive or conditional knock in/out or cell lines with a mutation signature. Increasingly, other approaches will be utilized. Patient-specific induced pluripotent stem cells (iPSCs) lines might offer an individually targeted genetic model system to identify and manipulate pathologic pathway. Use of iPSCs should be of particular value in assessing regulatory genes, which may have a quantitative deleterious effect rather than 'all or none' qualitative effect, observed or assumed, in missense or nonsense protein-coding mutations. Conclusion: current status of genes causing POI {#s4} =============================================== The POI---causative genes surveyed and found to date allow us to reach several conclusions. *First*, many genes have emerged as POI candidates (Fig. [1](#DMV036F1){ref-type="fig"}), but in non-syndromic POI only a minority have been proven equivocally causative by functional validation. These include *BMP15*, *PGRMC1*, and *FMR1* premutation on the X chromosome and on autosomes *GDF9*, *FIGLA*, *FSHR, NOBOX*, *NR5A1*, *NANOS3*, *STAG3*, *SYCE1*, *MCM8/9* and *HFM* 1. No perturbations have been found in a dozen other plausible candidates for which murine knockouts show ovarian failure, but this may simply reflect small sample sizes or interrogation restricted to a single ethnic group. Figure 1Schematic representation of chromosomal location of plausible genes associated with primary ovarian insufficiency. *Second*, notable differences in frequency exist among different populations. This is predictable for any genetic condition, and in POI this has already been observed in *FSHR*, *BMP15*, *NOBOX*, *FOXL2*, *TGFBR3*, *CDKN1B* and *FOXO3A* ([Supplementary Table SII](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1)). Future genetic studies should involve different ethnic groups and larger sample sizes. Clinically, caution should apply when counseling on the basis of data derived from an ethnic group different from that of the counseled individual. *Third*, causative POI genes are increasingly being shown not only to be restricted to expression in ovaries but also expressed ubiquitously. True, POF5 (*NOBOX*) and POF6 (*FIGLA*) are restricted to the ovaries. However, *PTEN* is a major regulator of the PI3K pathway involved in systemic cell proliferation, survival, migration and metabolism. PTEN also plays a vital role in the activation of primordial follicles. The spectrum of candidate genes potentially causing POI is being enriched. *Fourth*, many genes that currently appear isolated in function actually may be interrelated within yet to be defined pathways. It is logical to stratify by gene function in ostensibly distinct systems: endocrine, folliculogenesis, cell cycle, meiosis, mitochondrial, as examples. More difficult are gene-gene or protein-protein interactions, acting in ways not yet evident. This review thus unavoidably overestimates the role that protein-coding genes play to the detriment of upstream and downstream regulatory genes. Until recently, exome studies had to be restricted to sequencing individual candidate genes. Continued advances in sequencing techniques and contemporary bioinformatives will facilitate finding additional genes responsible for POI in other portions of the genome. *Fifth*, elucidating the etiology and molecular basis of POI is of paramount importance not only in understanding ovarian physiology but also in providing genetic counseling and fertility guidance. Once additional variants are detected, it will be increasingly possible to predict the age of menopause. Women having certain perturbations of POI can be offered the option of oocyte cryopreservation, with later thawing and use in assisted reproductive technology at the appropriate age. Supplementary data {#s5} ================== [Supplementary data are available at http://humupd.oxfordjournals.org/.](http://humupd.oxfordjournals.org/lookup/suppl/doi:10.1093/humupd/dmv036/-/DC1) Authors\' roles {#s6} =============== Y.Q. contributed to literature searching and drafting the article. X.J. participated in literature searching and creating figures and tables. J.L.S. contributed the study design and revised the article critically. Z.-J.C. supervised the study design and revision. Funding {#s7} ======= This work was supported by the National Basic Research Program of China (973 program-2012CB944700); the National Natural Science Foundation of China (81270662, 81471509, 81370692, 81370687); the Key Program of National Natural Science Foundation of China (81430029). Funding to pay the Open Access publication charges for this article was provided by National Basic Research Program of China (973 program-2012CB944700). Conflict of interest {#s8} ==================== None declared. Supplementary Material ====================== ###### Supplementary Data [^1]: Chromosomal 'abnormalities' means visible structural changes in karyotype that are sufficiently large to cause clinical abnormalities. Variants (e.g. prominent satellites) are not included. [^2]: CA, chromosomal abnormalities; PA, primary amenorrhea; POI, primary ovarian insufficiency; SA, secondary amenorrhea; EM, early menopause. [^3]: ^a^Including 2 46,XY gonadal dysgenesis (Swyer syndrome). [^4]: MR: mutation rate; FC: functional confirmation; BMP15: bone morphogenetic protein 15; AR: androgen receptor; FOXO4: forkhead box O4; POF1B: premature ovarian failure, 1B; DACH2: dachshund family transcription factor 2; PGRMC1: progesterone receptor membrane component-1. [^5]: ^a^Only refer to novel missense, frameshift and nonsense mutations [^6]: FMR1: Fragile X mental retardation 1; FOXL2: forkhead box L2; GALT: galactose 1-phosphate uridyl transferase; AIRE: autoimmune regulator; EIF2B: eukaryotic translation initiation factor; POLG: polymerase (DNA directed), gamma; NOG: noggin; PMM2: Phosphomannomutase 2; HSD17B4: Hydroxysteroid (17-beta) dehydrogenase 4; HARS2: Histidyl-tRNA synthetase 2, mitochondrial; CLPP: caseinolytic mitochondrial matrix peptidase proteolytic subunit; LARS2: leucyl-tRNA synthetase 2, mitochondrial; C10orf2: Chromosome 10 open reading frame 2; BLM: Bloom syndrome, RecQ helicase-like; ATM: ATM serine/threonine kinase; WRN: Werner syndrome, RecQ helicase-like; RECQL4: RecQ protein-like 4. [^7]: PTHB1: Bardet--Biedl syndrome 9 (BBS9); ADAMTS19: ADAM metallopeptidase with thrombospondin type 1 motif, 19; LAMC1: laminin, gamma 1; DLX5, 6: distal-less homeobox 5 and 6; DSS1: split hand/foot malformation (ectrodactyly) type 1. [^8]: HSD17B4: Hydroxysteroid (17-beta) dehydrogenase 4; CLPP: Caseinolytic mitochondrial matrix peptidase proteolytic subunit; LARS2: leucyl-tRNA synthetase 2, mitochondrial; C10orf2: Chromosome 10 open reading frame 2; HFM1: ATP-dependent DNA helicase homolog; MCM8, MCM9: minichromosome maintenance complex component 8 and 9; STAG3: stromal antigen 3; SYCE1: synaptonemal complex central element; eIF4ENIF1: eukaryotic translation initiation factor 4E nuclear import factor 1.

Introduction ============ Hepatic encephalopathy frequently complicates both acute liver failure and end-stage chronic liver disease \[[@B1]-[@B3]\]. The pathophysiological mechanisms of hepatic encephalopathy are poorly understood, although alterations in both the cerebral microcirculation and neuronal function associated with an excessive amount of toxic circulating substances not metabolised by the liver have been implicated \[[@B4]\]. High ammonia levels \[[@B5]\] and an imbalance between aromatic and branched amino acids are some of the mechanisms involved \[[@B6]-[@B9]\]. Thus, high levels of phenolic aromatic amino acids (tyrosine and phenylalanine) have been associated with the development of encephalopathy in patients with liver diseases. Moreover, the ratio between branched chain amino acids and phenolic aromatic amino acids has been suggested to correlate with the degree of encephalopathy \[[@B8]\]. A new \'detoxifying\' method, based on albumin dialysis or Molecular Adsorbent Recirculating System (MARS), has been launched in the past decade \[[@B10],[@B11]\]. This procedure effectively decreases hepatic encephalopathy in patients with liver failure \[[@B12]-[@B14]\]. However, the mechanisms responsible for this improvement, and their relationship with circulating levels of amino acids, are poorly identified. Therefore, the aim of the current study was to assess the effect of albumin dialysis on hepatic encephalopathy and on the circulating levels of amino acids and ammonia in patients with severe liver failure. Materials and methods ===================== The study was carried out in a series of nine patients with biopsy proven severe alcoholic hepatitis (six with cirrhosis) defined by a total serum bilirubin level above 10 mg/dl and a prothrombin index lower than 50%, and in a control group of four patients with primary biliary cirrhosis and resistant pruritus. The results of the effect of albumin dialysis on pruritus in these latter patients have already been published \[[@B15]\]. These four patients had circulating albumin, bilirubin and prothrombin index within the normal range. Patients with alcoholic hepatitis were not treated with corticosteroids before and during albumin dialysis and they were under standard medical therapy. No specific treatments for hepatic encephalopathy were used during albumin dialysis. The study protocol conforms to the ethical guidelines of the Declaration of Helsinki and was approved by the Hospital Clínic, Barcelona ethics committee. Patients were included after giving informed written consent (next-of-kin assent in encephalopathic patients). The study evaluates 23 seven-hour sessions of albumin dialysis performed in the patients with severe alcoholic hepatitis and eight seven-hour sessions performed in the control patients with resistant pruritus. All the patients were haemodynamically stable, did not require additional respiratory or circulatory support and had no evidence of severe infection or multiorgan failure. The following parameters were assessed before and after each treatment: standard liver function tests, arterial ammonia levels, and the concentration of total amino acids, branched amino acids, phenolic aromatic amino acids and tryptophan by high-performance liquid chromatography. Briefly, the blood samples were immediately transferred to chilled heparinised tubes, placed on ice and within five minutes were centrifuged at 3000 rpm for 15 minutes at 4°C and then kept frozen and stored at -80°C until analysis. Amino acids were separated by reversed phase high-performance liquid chromatography after phenyl-iso-thiocyanate derivatisation, to obtain a phenyl-thio-carbamyl derivative (Pico-Tag method, Waters, Milford, MA, USA) and subsequent ultraviolet detection in an automatic high-performance liquid chromatography system and data processing software. The internal standard was methionine sulfone. The coefficient of variation was less than 5% for all amino acid measurements. Hepatic encephalopathy ---------------------- Patients were evaluated by the same physician (LC) before and after each dialysis session for the presence and severity of hepatic encephalopathy according to the West Haven Criteria for semiquantitative grading of mental state \[[@B16]\]. The number connection test was assessed as a measure of cognitive motor abilities \[[@B17],[@B18]\]. The patient has to order numbers printed on a piece of paper consecutively from 1 to 25, as quickly as possible. Errors are not enumerated, but patients are instructed to return to the preceding correct number and then carry on. The test score is the time the patient needs to perform the test, including the time needed to correct the errors. Albumin dialysis treatment -------------------------- Extracorporeal albumin dialysis was performed with MARS (Gambro Lundia AB, Lund, Sweden). It is an extracorporeal liver support device, using a hollow-fibre dialysis column in which the blood of the patient is dialysed across an albumin-coated membrane, while at the same time maintaining a constant flow of albumin-rich dialysate in the extracapillary compartment. Substances with a molecular weight greater than 50 KD are not removed from the plasma by the system \[[@B19]\]. The albumin-enriched dialysate, containing 10 to 20% human serum albumin, is recirculated and its binding sites regenerated by online perfusion through a charcoal column and an anion-exchanger column, and simultaneously dialysated against a bicarbonate-buffered solution using a standard dialysis machine. Therefore, the albumin-bound toxins can be removed through MARS and the water-soluble toxins through haemodialysis. In the current study each dialysis session was performed using a double-lumen catheter in a femoral vein for blood access, and the albumin-enriched dialysate contained 600 ml of 20% human serum albumin. The extracorporeal blood flow and MARS flow were maintained at 250 mL/minute and the dialysate was maintained at body temperature to avoid cooling of the patient. A continuous infusion of heparin at a dose of 1500 to 4000 IU/hour was used as an anticoagulant when necessary. Statistical analyses -------------------- Data are expressed as mean ± standard error of the mean (SEM). The unpaired students\' t-test was used to compare data from patients and controls, and paired students\' t test or Wilcoxon test were used when appropriate to analyse differences before and after the procedure. A probability level of 5% was regarded as statistically significant. Results ======= Baseline clinical and analytical abnormalities for patients with alcoholic hepatitis and primary biliary cirrhosis with pruritus are shown in Table [1](#T1){ref-type="table"}. Patients with alcoholic hepatitis had severe liver function impairment as compared with patients with pruritus. Thus, all the patients with alcoholic hepatitis had a Maddrey\'s discriminant function higher than 32 (mean ± SEM: 74.5 ± 6.0) and the Model for End-stage Liver Disease score was 24.3 ± 4.1. Hepatic encephalopathy was detected in five patients (grade I in four patients and grade II in one patient) and the number connection test was also significantly higher in these patients with alcoholic hepatitis. Additionally, seven patients with alcoholic hepatitis had ascites. Total amino acid, as well as phenolic aromatic amino acid, concentrations were significantly higher in patients with alcoholic hepatitis than in patients with pruritus, as were tryptophan levels. No significant differences were found in baseline branched amino acid levels between the two groups of patients but the Fischer ratio was significantly lower in patients with alcoholic hepatitis. Circulating ammonia levels were higher in patients with alcoholic hepatitis as compared with patients with resistant pruritus. Thus, ammonium was elevated in four patients with alcoholic hepatitis and normal in the cases with pruritus. Ammonium was above the normal values (9 to 33 μM/L) before treatment in 12 of the 23 sessions (52.2%) performed in patients with alcoholic hepatitis. ###### Clinical and biochemical data of patients with alcoholic hepatitis and patients with resistant pruritus before albumin dialysis Patients with severe alcoholic hepatitis n = 9 Patients with resistant pruritus n = 4 p = ----------------------------- ------------------------------------------------ ---------------------------------------- ------- Age (years) 48.2 ± 2.0 52.0 ± 4.2 n.s. Male/Female 6/3 1/3 Ascites 7 0 Encephalopathy 5 0 NCT (minute) 3.68 ± 0.97 (5) 1.01 ± 0.09 0.05 Bilirubin (mg/dl) 22.8 ± 2.8 1.08 ± 0.05 0.005 AST (u/L) 96 ± 11 52 ± 7 0.031 ALT (u/L) 48 ± 4 93 ± 40 n.s. AP (u/L) 398 ± 53 999 ± 544 0.031 γGT (u/L) 202 ± 65 449 ± 426 n.s. Albumin (g/L) 24.4 ± 1.03 36.5 ± 1.2 0.005 Prothrombin index (%) 33.2 ± 2.8 91.5 ± 8.5 0.005 Creatinine 1.7 ± 0.8 0.8 ± 0.1 n.s. Serum Na 130 ± 3 138 ± 2 0.01 Serum K 3.5 ± 0.1 3.9 ± 0.2 0.03 Haemoglobin (g/L) 10.5 ± 0.5 11.8 ± 1.1 n.s. Leucocytes (cell/×10^9^) 12.9 ± 2.4 8.6 ± 3.7 n.s. Platelets (cell/×10^9^) 105 ± 14 264 ± 51 0.005 Total amino acids (μM/L) 2692 ± 274 1850 ± 93 0.01 Branched chain AA (μM/L) 253 ± 37 331 ± 20 n.s. Phenolic aromatic AA (μM/L) 186 ± 26 111 ± 8 0.04 Fischer index 1.32 ± 0.15 3.01 ± 0.12 0.005 Tryptophan (μM/L) 28.2 ± 2.8 39.7 ± 2.7 0.02 Ammonium (μM/L) 45.2 ± 11.6 23.2 ± 4.7 n.s. AA = amino acids; ALT = alanine aminotransferase AP = alkaline phosphatase; AST = aspartate aminotransferase; γGT = gamma-glytamyl transpeptidase; K = potassium; Na = sodium; n.s. = not significant; NCT = number connection test. Eight episodes of encephalopathy were recorded before treatment in the nine patients with alcoholic hepatitis. Albumin dialysis was associated with a significant improvement in the degree of hepatic encephalopathy (p = 0.02), and no encephalopathy was present after treatment in five of the eight episodes. The number connection test decreased in 11 of the 15 cases with alcoholic hepatitis with assessments before and after treatment. No significant changes were observed in the test before and after treatment in patients with pruritus. Marked attenuations in circulating bilirubin and a significant improvement in albumin concentration were also observed after dialysis in these patients, along with clear recoveries in creatinine and electrolyte concentrations. Haemoglobin and platelets decreased significantly after treatment. No significant changes were observed in patients with pruritus except for a significant decrease in the platelet count (Table [2](#T2){ref-type="table"}). The improvement in hepatic encephalopathy was not associated with changes in creatinine, sodium or potassium levels after treatment. ###### Clinical and biochemical data of patients with alcoholic hepatitis and patients with resistant pruritus before and after the seven-hour sessions of albumin dialysis Patients with alcoholic hepatitis n dialysis = 23 Patients with resistant pruritus n dialysis = 8 ----------------------------- --------------------------------------------------- ------------------------------------------------- -------- ------------- ------------- ------ Before After p Before After p Encephalopathy (score) 0.36 ± 0.11 0.16 ± 0.09 0.02 0 0 n.s. NCT (minute) 3.28 ± 0.43\* 2.8 ± 0.31 n.s. 1.17 ± 0.11 1.03 ± 0.08 n.s. Bilirubin (mg/dl) 20.6 ± 1.5 16.3 ± 1.5 0.001 1.01 ± 0.1 1.2 ± 0.1 n.s. Albumin (g/L) 24.2 ± 0.6 25.5 ± 0.7 0.03 37.4 ± 0.7 38.2 ± 1.4 n.s. Prothrombin index (%) 34.4 ± 1.9 30.3 ± 2.2 0.03 92.0 ± 4.8 84.1 ± 5.9 n.s. Creatinine 1.3 ± 0.3 0.7 ± 0.1 0.008 0.85 ± 0.05 0.78 ± 0.12 n.s. Serum Na 134 ± 1.0 138 ± 0.6 0.0003 138 ± 0.4 138 ± 0.4 n.s. Serum K 3.7 ± 0.1 4.0 ± 0.1 0.05 4.1 ± 0.1 4.5 ± 0.2 0.04 Haemoglobin (g/L) 10.0 ± 0.4 9.4 ± 0.4 0.002 11.8 ± 0.4 11.4 ± 0.6 n.s. Leucocytes (cell/×10^9^) 11.9 ± 1.4 11.1 ± 1.4 0.06 8.2 ± 1.2 7.3 ± 0.6 n.s. Platelets (cell/×10^9^) 78.6 ± 7.0 56.0 ± 6.0 0.0000 234 ± 19 187 ± 19 0.01 Total amino acids (μM/L) 2491 ± 152 2229 ± 114 0.01 2080 ± 159 2354 ± 158 n.s. Branched AA (μM/L) 243 ± 18 233 ± 12 n.s. 399 ± 44 443 ± 47 n.s. Phenolic aromatic AA (μM/L) 193 ± 17 165 ± 10 0.04 123 ± 10 145 ± 11 0.05 Fischer index 1.32 ± 0.08 1.47 ± 0.05 0.01 3.20 ± 0.16 3.08 ± 0.24 n.s. Tryptophan (μM/L) 28.4 ± 1.9 26.0 ± 1.8 n.s. 45.0 ± 2.6 42.9 ± 4.8 n.s. Ammonium (μM/L) 46.2 ± 6.6 41.1 ± 5.5 n.s. 22.2 ± 4.1 21.5 ± 3.2 n.s. \*Assessed in 15 of the 23 treatments. AA = amino acids; K = potassium; Na = sodium; n.s. = not significant; NCT = number connection test. Total amino acid concentrations diminished significantly in patients with alcoholic hepatitis and no changes were observed in the patients with pruritus who served as the control group (Table [2](#T2){ref-type="table"}). Circulating branched amino acids were not significantly modified by albumin dialysis, whereas phenolic aromatic amino acids decreased markedly in patients with alcoholic hepatitis and increased in patients with pruritus (Figure [1](#F1){ref-type="fig"}). Hence, the Fischer ratio increased significantly in patients with alcoholic hepatitis (from 1.32 ± 0.08 to 1.47 ± 0.05; p \< 0.01) but decreased in controls (from 3.20 ± 0.16 to 3.08 ± 0.24; p = not significant). Albumin dialysis resulted in a 17% increase in the Fischer ratio in patients with alcoholic hepatitis and a 3% decrease in patients with resistant pruritus (p \< 0.05; Figure [2](#F2){ref-type="fig"}). These changes were more pronounced when corrected by the baseline albumin levels (data not shown). No significant changes were observed in either the circulating levels of tryptophan (from 28 ± 2 μM to 26 ± 1 μM) or ammonia (from 46 ± 6 mM to 41 ± 5 mM). In the alcoholic hepatitis group the changes observed in amino acid levels and the Fischer ratio were more prominent in patients with hepatic encephalopathy before treatment, baseline albumin concentration below 24 g/l, and in those in whom albumin dialysis resulted in a decrease in total bilirubin greater than 20% with respect to pre-treatment levels (Figure [3](#F3){ref-type="fig"}). The amino acid profile was not significantly modified by albumin dialysis in the group of patients with resistant pruritus. {#F1} {#F2} {#F3} Six patients with alcoholic hepatitis and the four patients with pruritus were discharged from the hospital after the MARS treatment. The other three patients with alcoholic hepatitis died 16, 36 and 46 days after treatment. The mean hospital stay of patients with alcoholic hepatitis was 21.1 ± 4.1 days (median: 16 days). The three-month, six-month and one-year survival rate of patients with alcoholic hepatitis were 55.5%, 44.4% and 33.3%, respectively. All the patients with pruritus were alive four years after treatment. Discussion ========== Hepatic encephalopathy is a common complication in patients with liver failure, including acute decompensated patients with advanced cirrhosis of different aetiologies, and in patients with severe alcoholic hepatitis. Actually, hepatic encephalopathy is a feature clearly associated with a bad prognosis in patients with this condition \[[@B20]-[@B23]\]. The pathophysiology of hepatic encephalopathy is unknown, and different hypotheses have emerged over the years, but portal hypertension and the consequences of liver dysfunction are included in all the proposals, although hepatic encephalopathy may develop into acute liver failure with no significant portal hypertension. There is evidence suggesting that hepatic encephalopathy results from the accumulation of neurotoxic or neuroactive substances in the brain, including ammonia, manganese, aromatic amino acids, mercaptans, phenols, short-chain fatty acids and others. Prevention and treatment of hepatic encephalopathy relies on the reduction of circulating ammonia either by a reduction in gut production using disaccharides or antibiotics or by increasing its metabolites \[[@B24]\]. There is also some evidence suggesting that increasing the Fischer\'s ratio may improve hepatic encephalopathy, and this was the basis of the utilisation of branched amino acid supplements in patients with encephalopathy \[[@B25]-[@B30]\]. Another way to remove the substances potentially related to hepatic encephalopathy is the utilisation of liver assist devices in order to eliminate or reduce the neurotoxins generated in liver failure. The results of the current study clearly indicate that albumin dialysis is able to induce favourable effects in patients with severe alcoholic hepatitis. Actually, the degree of hepatic encephalopathy improved in all the cases after albumin dialysis. This positive effect was associated with significant changes in the levels of circulating amino acids, and particularly with changes in phenolic aromatic amino acids, which decreased in patients with alcoholic hepatitis but no changes were observed in the group of patients treated with albumin dialysis for resistant pruritus. As a consequence, the Fischer ratio increased significantly in these patients, but not in the patients with pruritus who had a different baseline amino acid profile. Similar data on the effects of albumin dialysis on the amino acid profile have been reported in patients with acute and acute-on-chronic liver failure \[[@B31]\]. In the current study, the effects of albumin dialysis on Fischer\'s index were particularly prominent in patients with hepatic encephalopathy, baseline albumin levels below 24 g/l, and in those in which the procedure was apparently most efficient as measured by the decrease in bilirubin levels. Another interesting finding is the fact that the amelioration in the index of hepatic encephalopathy was not related to improvements in variables associated with renal dysfunction, such as creatinine, or the correction in the electrolyte imbalance, therefore, strengthening the effect of albumin dialysis but not of the standard renal dialysis that is able to correct the electrolyte disturbances. In regard to this, recent data indicate that albumin dialysis using the MARS device is an effective approach for improving hepatic encephalopathy in patients with different clinical conditions, but particularly in patients with acute-on-chronic liver failure \[[@B13],[@B14],[@B32]\]. The inability of albumin dialysis to correct ammonia and tryptophan levels is an intriguing finding of this study, thus indicating that improvement in hepatic encephalopathy does not depend exclusively on the normalisation or decrease of these molecules. In this regard, different *in vivo*and *in vitro*studies have suggested that albumin dialysis may improve hepatic encephalopathy by decreasing the circulating levels of both ammonia and tryptophan \[[@B33],[@B34]\]. However, in the current study tryptophan levels were normal in all patients, whereas baseline ammonia was elevated in about 50% of patients. The mechanism by which albumin dialysis decreases phenolic aromatic amino acids in patients with severe alcoholic hepatitis is unknown. However, it could be speculated that the procedure is able to dialyse and adsorb the free circulating levels of these amino acids in patients who have low albumin concentrations, with their serum albumin probably completely saturated and unable to bind or transport more substances. However, further studies must be performed to answer this. The lack of effect of albumin dialysis in significantly modifying the amino acid profile in primary biliary cirrhosis patients with bilirubin, prothrombin index and albumin levels within normal ranges may sustain this explanation. Actually, in these patients albumin dialysis did not induce marked changes in the amino acid profile, but rather a trend to increase the phenolic amino acids after treatment was observed. On the other hand, it should be taken into account that the effect of albumin dialysis on amino acid profile was more relevant in patients who experienced higher bilirubin decrease after treatment, thus suggesting that the favourable effect on hepatic encephalopathy results from the removal of phenolic amino acids and that the procedure may cause some effect on the removal of other known substances involved in the pathogenesis of encephalopathy. Accordingly, it has been speculated that albumin dialysis is able to remove nitric oxide or a number of cytokines and chemokines probably involved in the pathogenesis of hepatic encephalopathy \[[@B35],[@B36]\]. Thus, the ability of albumin dialysis to remove other toxins and pro-inflammatory stimuli such as lipopolysaccharides and lipid peroxidation end-products may have implications for limiting the inflammatory response that could be implicated not only in renal impairment and circulatory dysfunction but also in the pathophysiology of hepatic encephalopathy in patients with severe liver failure \[[@B37]\]. Recent data support the importance of infection and inflammation even in minimal alteration of cognitive function in patients with liver failure and features of systemic inflammatory response syndrome \[[@B37]\]. Conclusions =========== In summary, the results of this study demonstrate that treatment of patients with severe alcoholic hepatitis with albumin dialysis improves hepatic encephalopathy, and that this favourable effect results from the correction of the abnormal amino acid profile, basically by decreasing phenolic aromatic amino acids. These results may explain at least in part the hopeful effects of albumin dialysis on hepatic encephalopathy observed in different trials, although further studies are warranted to define the clinical and biochemical effects of this new procedure in the treatment of patients with acute or acute-on-chronic liver failure. Key messages ============ • Albumin dialysis results in favourable effects in patients with severe alcoholic hepatitis, because hepatic encephalopathy improved in all cases. • Total amino acid and phenolic aromatic amino acids diminished and the Fischer ratio increased in patients with alcoholic hepatitis treated with MARS. • Changes in amino acid levels and the Fischer ratio were more prominent in patients with hepatic encephalopathy, low albumin concentration and greater bilirubin extraction with MARS treatment. • Although the mechanisms by which albumin dialysis decreases phenolic aromatic amino acids in patients with severe alcoholic hepatitis remain unknown, it could be speculated that MARS is able to dialyse and adsorb the free circulating levels of these amino acids in patients who have low albumin concentration, with their serum albumin probably completely saturated and unable to bind or transport more substances. Abbreviations ============= MARS: Molecular Adsorbent Recirculating System; SEM: standard error of the mean. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= AP conceived the study, participated in its design and coordination and performed the statistical analysis. RD participated in the study design and carried out the laboratory analytics. LC, JMS, JC, AE and AM treated the patients, participated in the study design and helped to draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ The study was supported in part by grants from the Instituto de Salud Carlos III (C03/02 and PI 05/1285), the Ministerio de Educación (SAF2005-03649) and by the Centro de Investigaciones Biomédicas en Red de Enfermedades Hepáticas y Digestivas (CIBERehd).

Introduction ============ Acute lymphoblastic leukemia (ALL) is a common hematological malignancy with an incidence of 1.7 per 100,000 per year across all age groups. Although the median age at diagnosis overall is 14 years, the peak incidence occurs in early childhood (2--5 years of age).[@b1-ijn-8-4361] ALL comprises approximately 30 percent of all childhood cancers and has a male preponderance.[@b2-ijn-8-4361],[@b3-ijn-8-4361] A diagnosis of ALL is usually made by documenting the presence of lymphoblasts in peripheral blood and/or \>25% lymphoblasts in the bone marrow. Common presenting features are nonspecific, including fever, bone pain, lymphadenopathy, and anemia, and bleeding or bruising secondary to thrombocytopenia.[@b1-ijn-8-4361] The precise pathogenic events leading to the development of ALL are still unknown, although it is likely to arise from complex interactions between prenatal and postnatal exogenous and endogenous exposures, genetic susceptibility, and chance. Major cytogenetic and molecular genetic abnormalities seen in ALL include gene rearrangements and dysregulation, hyperdiploidy (\>50 chromosomes), hypodiploidy (\<44 chromosomes), and chromosomal translocations, of which t(12;21)(p13;q22) encoding *ETV6-RUNX1* is the most common. Certain chromosomal abnormalities in leukemic lymphoblasts disrupt genes that regulate normal hematopoiesis and lymphoid development, activate oncogenes, or constitutively activate tyrosine kinases. Several of these chromosomal rearrangements are significantly associated with clinical outcome and are used in the classification and risk stratification of leukemia.[@b1-ijn-8-4361],[@b4-ijn-8-4361] One archetypical genetic abnormality implicated in leukemogenesis is the Philadelphia chromosome (Ph), which arises from a balanced translocation between the long arms of chromosome 9 and 22 (t\[9;22\]\[q34;q11\]), resulting in the fusion of the B-cell receptor (*BCR*) and the nonreceptor tyrosine kinase *ABL1* genes.[@b5-ijn-8-4361] The reciprocal translocation results in the *BCR--ABL1* fusion product, leading to constitutive activation of ABL1 kinase following juxtaposition of BCR. The Ph is the characteristic cytogenetic feature of chronic myeloid leukemia (present in \>90% of patients), but also occurs in ALL, with an age-related incidence ranging from 3% of patients under 20 years to 21% of cases over 50 years.[@b6-ijn-8-4361] Chemotherapy combined with tyrosine--kinase inhibitors such as imatinib mesylate (IM) induces complete remission (CR) in \>90% of Ph-positive (Ph+) adults and children, many with undetectable minimal residual disease, translating to an overall survival rate of 50% in adults and event-free survival (EFS) of 88% in children.[@b7-ijn-8-4361],[@b8-ijn-8-4361] In contrast, the CR and overall survival rates for adults with Ph+ ALL in the pre-IM era were \<70% and 20%, respectively,[@b9-ijn-8-4361] whilst the EFS for children was \<45%.[@b7-ijn-8-4361] In Ph-negative (Ph−) ALL, CR rates are 96%--99% in children and 78%--92% in adults.[@b1-ijn-8-4361] However, there is a greater discrepancy in EFS, with rates approaching 90% for children compared with only 30%--40% in adults.[@b1-ijn-8-4361] Standard treatment for ALL (Ph+ and Ph−) differs in adults and children, largely due to better tolerance of intensive multiagent chemotherapy in children. In both age groups, however, vincristine (VCR) is a key component of therapy. Vincristine =========== VCR is a lipophilic amine, first introduced as an anticancer therapy over 45 years ago.[@b10-ijn-8-4361],[@b11-ijn-8-4361] VCR is a cell cycle-dependent compound that directly binds to tubulin, causing microtubule depolymerization, M-phase arrest, and apoptosis in mitotic cells.[@b12-ijn-8-4361] At low concentrations, VCR induces reversible mitotic arrest with little effect on morphology or polymerization of spindle microtubules.[@b13-ijn-8-4361]--[@b15-ijn-8-4361] In contrast, higher VCR doses and long-term VCR exposure are associated with microtubule depolymerization-induced cytotoxicity.[@b14-ijn-8-4361]--[@b18-ijn-8-4361] In addition, VCR impedes tumor blood flow, inducing tumor necrosis.[@b19-ijn-8-4361] Although the role of microtubules in this process has not been fully elucidated, the efficacy of VCR for treating hemangiomas with high epithelial cell content[@b20-ijn-8-4361] (thus high tubulin expression levels) suggests that the VCR mechanism of action against microtubule polymerization may play a role in the inhibition of tumor angiogenesis.[@b21-ijn-8-4361] VCR also affects intracellular transport processes, which are thought to contribute less to its antineoplastic activity than to its modulation of microtubule polymerization.[@b19-ijn-8-4361] However, it is the perturbance of these biochemical pathways that is predicted to mediate VCR-induced autonomic and peripheral sensory--motor polyneuropathy, a dose-limiting side effect of VCR.[@b22-ijn-8-4361] The neurotoxic effects of VCR, mediated by impaired microtubule function leading to blockade of axon transport and subsequent axonal degradation, have significantly impaired the use of high-dose VCR in the treatment of neoplastic disease.[@b22-ijn-8-4361] As a result, VCR doses are generally capped at 2 mg.[@b22-ijn-8-4361] Therefore, there has been a recent impetus to enhance the therapeutic activity of VCR with liposomal encapsulation systems to increase the VCR dose whilst limiting free-drug-associated toxicity. Liposomes ========= Lipids are naturally occurring amphipathic small molecules that are immiscible in aqueous solutions. When dispersed in aqueous solutions, the presence of a hydrophilic polar head group and hydrophobic apolar tail ([Figure 1](#f1-ijn-8-4361){ref-type="fig"}) induces a steric organization of phospholipids. As such, phospholipids spontaneously form bilayer membranes encapsulating ions or molecules present in the solution in which they are formed ([Figure 1](#f1-ijn-8-4361){ref-type="fig"}).[@b23-ijn-8-4361] This functions to minimize the exposure to, and interaction of, the hydrophobic aliphatic chain with water. The amphiphilic nature of lipids induces the formation of liposomes, and contributes to the structural integrity of cell membranes as well as compartmentalization of cells into functional membrane-bound organelles. Liposomes are multilamellar or unilamellar bilayer microspheres composed of lipids encapsulating an aqueous solution.[@b24-ijn-8-4361] Importantly, liposome-enveloped substances are protected from enzyme-mediated degradation or inactivation by the immune system. Furthermore, liposomal encapsulation of drugs aims to enhance plasma concentration and circulation half-life, increase transport and accumulation within specific target tissues, and minimize toxicity.[@b25-ijn-8-4361] These characteristics have led to the exploitation of liposomes in drug delivery. For example, liposomal encapsulation of VCR sulfate was recently approved by the US Food and Drug Administration as a therapeutic strategy for the treatment of Ph− ALL. Early studies investigated the encapsulation of VCR within egg phosphatidylcholine (EPC)/cholesterol or distearoylphosphatidylcholine (DSPC)/cholesterol liposomes, both at a concentration ratio of 55%:45%.[@b18-ijn-8-4361] The inclusion of cholesterol acts to stabilize the liposome, preventing vesicle destabilization and concomitant drug release.[@b25-ijn-8-4361] The drug uptake process can be driven by a pH gradient whereby the internal pH of the liposome is acidic at 4.0 while the external is slightly basic at pH 7.5.[@b26-ijn-8-4361] Uptake efficiencies of up to 98% are achieved utilizing pH gradients, although disparity between the two liposomal formulations can occur. For example, VCR rapidly accumulates in EPC/cholesterol at 21°C, leading to \>90% drug encapsulation after 30 minutes while entrapment of VCR in DSPC/cholesterol liposomes was only 17% under the same conditions. However, increasing the experimental temperature to 60°C enhanced entrapment to \>98% after just 10 minutes.[@b26-ijn-8-4361] Just as the liposomal formulation affects VCR uptake, it can affect drug release. Under biologically relevant conditions, EPC/cholesterol liposomes release 96% of VCR into 37°C whole blood after just 24 hours. Due to the high membrane permeability of EPC liposomes, they are not commonly used in drug formulations today.[@b25-ijn-8-4361] In contrast, DSPC/cholesterol liposomes release approximately 80% of VCR under the same conditions.[@b27-ijn-8-4361] Importantly, the ability of the liposome to retain VCR correlated with the stability of the pH gradient, indicating that maintenance of the pH gradient magnitude across the vesicle membrane may be key to developing slow-release formulations of VCR from liposomes. In vivo studies mimicked findings from in vitro work whereby VCR encapsulated in EPC/cholesterol liposomes leaked rapidly into the plasma following intravenous injection into mice.[@b27-ijn-8-4361] Although substitution of EPC with DSPC enhanced the longevity of VCR retention within the liposome, improvements in in vivo retentions have also been achieved through manipulation of the intraliposomal pH concentration[@b28-ijn-8-4361],[@b29-ijn-8-4361] or addition of the ganglioside GM1[@b28-ijn-8-4361] or 5% polyethylene glycol.[@b30-ijn-8-4361] The addition of polyethylene glycol chains provides a steric barrier around the liposome, which is predicted to protect the liposome from clearance by the patient's mononuclear phagocyte system, following opsonization of the liposome, and minimize interaction with serum proteins.[@b31-ijn-8-4361] This, in turn, enhances circulation time and may alter the biodistribution of the liposome, to enhance tumor-specific liposome aggregation and functional interaction between the liposome and tumor cells.[@b25-ijn-8-4361] Similarly, substitution of DSPC with sphingomyelin has been utilized in the encapsulation of VCR and exhibits the best retention properties such that up to 75% of encapsulated VCR remains within the liposome 24 hours following intravenous injection into mice in vivo.[@b28-ijn-8-4361],[@b32-ijn-8-4361] As such, sphingomyelin liposomes are utilized in the encapsulation of VCR sulfate employed for the treatment of ALL and non-Hodgkin's lymphoma. Liposomal vincristine sulfate injection ======================================= Vincristine sulfate liposomal injection (VSLI) (Marqibo®; Talon Therapeutics, Inc., South San Francisco, CA, USA) is a sphingomyelin and cholesterol-based nanoparticle formulation of VCR sulfate. It was designed to deliver a larger dose of VCR directly to tumor cells via encapsulation within an aqueous core of nanoparticles comprising sphingomyelin and cholesterol liposomes, thereby avoiding undue neurotoxicity. A recent landmark study of VSLI monotherapy in adults with multiply relapsed or refractory Ph− ALL demonstrated an overall response rate of 35% and a composite CR rate of 20%.[@b33-ijn-8-4361] In August 2012, the US Food and Drug Administration approved VSLI for the treatment of Ph− ALL in adult patients with progressive disease or second or greater relapse. VSLI is a nanoparticle VCR formulation that encapsulates the compound in an aqueous core within a sphingomyelin/cholesterol liposome designated the nomenclature of Optisome™ (Talon Therapeutics, Inc.). Preclinical studies show that the sphingomyelin/cholesterol VCR formulation improves tumor drug exposure by enhancing drug delivery and tissue targeting.[@b32-ijn-8-4361] Data from pharmacokinetic (PK) studies revealed that clearance of liposome-encapsulated VCR was slower than that for free VCR and was thought to contribute to the higher plasma concentrations observed over a longer time period for liposomal VCR.[@b34-ijn-8-4361],[@b35-ijn-8-4361] For example, a preclinical model comparing conventional, aqueous VCR, DSPC/cholesterol, and sphingomyelin/cholesterol liposomal formulations showed that the encapsulated drug exhibited a significantly larger area under the concentration curve (AUC) (measure of bioavailability), a longer mean plasma residence time, and a lower volume of distribution.[@b35-ijn-8-4361] Importantly, the progressive in vivo accumulation of VCR in tissues, based on maximum concentration (Cmax), demonstrates a preference for mononuclear phagocyte system tissues such as spleen, liver, lymph nodes, and bone marrow,[@b36-ijn-8-4361] correlating to those tissues most affected by leukemic burden. Moreover, the specific efficacy of liposomal-encapsulated VCR is highlighted by the high concentration of VCR in tissue compared with the relatively low concentration of VCR within the plasma, suggesting that liposomal formulations result in little release of VCR in the peripheral blood but target encapsulated drug to the tumor.[@b35-ijn-8-4361] Indeed the in vivo tumor AUC value for sphingomyelin/cholesterol was over 120 times higher compared with nonliposomal VCR, and two-fold that of DSPC/cholesterol formulations.[@b35-ijn-8-4361] In human VCR studies, as for most anticancer drugs, intrapatient and interpatient PK profiles vary markedly[@b22-ijn-8-4361] For example, in a recent study of children, the median clearance, AUC, and Cmax of VCR were 482 mL/min/m^2^ (range 132--698 mL/min/m^2^), 49.7 mcg/L•h (16.5--143.1 mcg/L•h), and 3.5 mcg/L (1.0--31.2 mcg/L), respectively[@b37-ijn-8-4361] It is important to highlight that comparison between VSLI and VCR PK studies is difficult due to dose capping resulting from the neurotoxicities associated with free VCR compound. In contrast, VSLI studies are often reported with the absence of dose capping. The enhanced tissue targeting and drug delivery observed with liposomal formulations of VCR permit the administration of higher doses of VSLI compared with conventional administration of free compound.[@b38-ijn-8-4361] For example, the conventional VCR dose of 1.5 mg/m^2^ has a cap of 2.0 mg, limiting the dose for patients with a body surface area \>1.33 m^2^. However, there is no dose cap for the VSLI formulation.[@b39-ijn-8-4361] In the Phase II Relapsed Acute Lymphoblastic Leukemia (RALLY) trial for VSLI treatment of Ph− ALL patients, the individual dose of VSLI ranged from 3.1 mg to 5.5 mg, up to 2.8-fold higher than the standard VCR dose. The weekly dose was fixed at 2.25 mg/m^2^,[@b33-ijn-8-4361],[@b39-ijn-8-4361] resulting in a cumulative dose range of 3.5--70.1 mg, much greater than that attainable with conventional VCR therapy.[@b33-ijn-8-4361],[@b39-ijn-8-4361] Although the absence of dose capping makes comparison of PK between the VCR and VSLI formulations difficult, a recent study in patients with advanced solid tumors suggests that encapsulation of VCR improves PK. Patients treated with a single 2.0 mg/m^2^ dose of the encapsulated VSLI formulation exhibited increased plasma AUC and decreased clearance rates compared with those treated with 2.0 mg/m^2^ of conventional VCR.[@b34-ijn-8-4361] Interestingly, variation of VSLI dose (1.5, 2.0, or 2.3 mg/m^2^) or number of doses (1.5 or 1.8 mg/m^2^ weekly for four cycles) did not significantly alter the PK of VSLI in these patients.[@b34-ijn-8-4361] However, it should be noted that the patients in this study were Chinese and exhibited lower AUC and Cmax at a VSLI dose of 2.0 mg/m^2^ compared with previous studies,[@b40-ijn-8-4361],[@b41-ijn-8-4361] suggesting that pharmacogenomics may contribute to the PK of VSLI. Efficacy -------- The first published Phase II clinical study of single-agent VSLI reported 16 adults (median age 35 years, range 23--64 years) with recurrent or refractory ALL, in whom the estimated likelihood of achieving CR was 30%--40% for 19% of the studied patients, and \<10% for the remaining patients.[@b42-ijn-8-4361] Pre-existing grade 1--2 neuropathy was permitted for the trial. VCR sulfate was encapsulated with sphingomyelin and cholesterol liposomes with sodium phosphate to produce a sphingosomal VCR mixture at a dose of 0.16 mg/mL. This VSLI mixture was administered by intravenous infusion over 60 minutes within 8 hours of preparation at a dose of 2.0 mg/m^2^. This dose was repeated every 14 days in the absence of rapid disease progression or dose-limiting toxicities with the median number of doses of VSLI being two (range one to fve). Dose decrements of 0.2 mg/m^2^ were implemented for nonhematologic toxicities of grade 3--4 severity. Premedication with antiemetics and adjuvant aperients were used to decrease adverse effects. Two patients were unevaluable for response due to the coadministration of other chemotherapy agents. Of the 14 evaluable patients, two (14%) had an overall objective response. This comprised one CR after three VSLI doses and a partial response with two VSLI doses. Both of these patients were Ph+. Another six Ph+ patients did not show a response. The remaining six evaluable patients were resistant to treatment with VSLI, although two showed transient reductions in bone marrow infiltrate prior to developing progressive disease.[@b42-ijn-8-4361] In the landmark Phase I trial of VSLI in advanced, relapsed, or refractory Ph− adult ALL, 65 patients aged ≥18 years were treated, with several durable responses seen.[@b33-ijn-8-4361] Patients with residual persistent grade 1 or nonpersistent grade 2 or higher VCR-related neuropathy were included. Patients received weekly VSLI at 2.25 mg/m^2^ with no dose capping, administered intravenously over 60 minutes. The primary efficacy end point was the proportion of patients who achieved CR or CR with incomplete recovery of peripheral blood neutrophil counts or platelet counts (CRi). An overall response rate of 35% was found, with 20% achieving either a CR or a CRi. CR/CRi was achieved in 25% of patients with an untreated relapse and 14% of those with relapse previously refractory to single-agent or multiagent antileukemic therapy. There was an associated survival beneft of 7 months in responders compared with 3 months in nonresponders.[@b33-ijn-8-4361] These results are particularly significant given that this was a population of patients with heavily pretreated and advanced disease. Furthermore, 12 (19%) VSLI-treated patients proceeded to hematopoietic stem cell transplantation.[@b33-ijn-8-4361] Safety and tolerability ======================= VSLI appears to be well tolerated and safe, even at doses exceeding the 2 mg maximal dose commonly applied to free VCR. Clinical experience with specific toxicities is summarized as follows. Neuropathy ---------- In the Phase II trial of VSLI reported by Thomas et al,[@b42-ijn-8-4361] 16 patients received a median number of two doses, at a median dose of 3.8 mg (range 2.9--4.2 mg). Neurological assessments were performed prior to each dose of VSLI, with minimal neurotoxicity seen. Two patients were found to have grade 1 peripheral neuropathy after two and four doses of VSLI, both with a prior history of VCR-related peripheral neuropathy. Grade 2 orthostasis and intermittent headaches were seen in one patient. No significant unexpected toxicity was observed in the remaining patients.[@b42-ijn-8-4361] O'Brien et al[@b33-ijn-8-4361] reported the use of VSLI doses of 2.25 mg/m^2^ with no dose capping, with patients receiving a median of four doses of VSLI (range one to 18) and a median individual dose of 4.12 mg (range 3.14--5.51 mg). Neuropathy-associated adverse events were reported in 86% of the 65 patients evaluated. Grade 3 peripheral neuropathy-related events combined were reported in 23% of patients. However, 77% of patients had reported neuropathy-related signs or symptoms prior to the commencement of VSLI, consistent with prior VCR exposure. Higher grades of peripheral neuropathy were related to VSLI dose and response. Overall, VSLI produced no new or unexpected toxicities and showed a toxicity profile comparable with standard VCR at its labeled dose.[@b33-ijn-8-4361] Autonomic neuropathy and constipation ------------------------------------- In addition to peripheral neuropathy, VCR has also been shown to cause autonomic neuropathy with secondary delayed gastric emptying, constipation, and bladder dysfunction.[@b43-ijn-8-4361] Constipation has been reported in 34%--44% of patients with the use of VSLI. However, most of these have been of grade 1 and 2 severity.[@b33-ijn-8-4361],[@b44-ijn-8-4361] In the landmark ALL study, only 3% of patients experienced constipation of grade 3 severity.[@b33-ijn-8-4361] Vesicant effect --------------- Another significant risk of free VCR is its potential to cause tissue necrosis upon accidental extravasation. This risk has been shown to be significantly reduced with the use of liposomal VCR, with no gross inflammatory response seen on subcutaneous VSLI in mouse models. As extravasation following standard intravenous administration occurs in up to 1%--2% of chemotherapy infusions, this decreased risk of tissue necrosis is potentially clinically relevant.[@b18-ijn-8-4361] Hepatic dysfunction ------------------- Bedikian et al[@b41-ijn-8-4361] studied the use of VSLI in seven patients with malignant melanoma and abnormal liver function. The VSLI was administered at a dose of 1.0 mg/m^2^ every 2 weeks, based on the recommended 50% dose reduction for conventional VCR in subjects with impaired liver function. Grade 3 adverse events were seen in three (43%) patients. Nausea and constipation were common adverse events and were usually mild. Grade 1 neuropathy, presenting as numbness in fingers and toes, was seen in 43% of patients.[@b41-ijn-8-4361] Drug--drug interactions ----------------------- As VCR is usually given as combination chemotherapy in the treatment of ALL, it is important to be mindful of the potential adverse drug interactions. Particular considerations are the potential of the liposomal drug to change the PK of a coadministered free drug, and the potential for a free drug to affect the behavior of the liposomal carrier and encapsulated drug. The interactions between combination therapy with VSLI and mitoxantrone hydrochloride, idarubicin hydrochloride, daunorubicin hydrochloride, and doxorubicin hydrochloride have been assessed in vitro.[@b45-ijn-8-4361] The addition of free daunorubicin was found to cause release of 26% of encapsulated VCR within 2 hours, followed by a 99% increase in uptake of daunorubicin into the liposomes.[@b45-ijn-8-4361] The only agent studied in vivo in mouse models was idarubicin, which was not found to impact on the release of VCR from liposomes in the plasma compartment, in contrast to causing a rapid release of approximately 30% of encapsulated VCR in in vitro models.[@b45-ijn-8-4361] Idarubicin was, however, found to have altered PK when administered shortly after VSLI injection, with increased free idarubicin concentrations at 15 and 60 minutes postadministration. Thus, it has been recommended that VSLI be administered at a time point when other free drug concentrations are low, to minimize drug--drug interactions.[@b45-ijn-8-4361] VCR is known to undergo oxidization in the liver to one metabolite, M1, by the cytochrome P450 (CYP) group of enzymes -- in particular, selective metabolism by CYP3A5. This may be clinically important due to the expression of CYP3A5 polymorphisms, with subsequent inter-racial differences in expression and potential effects on VCR efficacy and toxicity.[@b22-ijn-8-4361] There are no specific interactions reported with the liposomal formulation of VCR, but other known interactions with free VCR would still apply to the use of VSLI, including drugs known to inhibit these CYP3A enzymes or P-glycoprotein and induce CYP3A enzymes ([Table 1](#t1-ijn-8-4361){ref-type="table"}).[@b22-ijn-8-4361] Intrathecal administration -------------------------- Inadvertent intrathecal administration of VCR is universally fatal.[@b46-ijn-8-4361] It is expected that this catastrophic complication would also apply to VSLI. Thus, standard precautions to reduce the risk of erroneous administration should remain a priority. Pediatric experience ==================== There are limited data on the use of VSLI in the pediatric setting, although early results from an ongoing trial on VSLI in children and adolescents are promising. In this single-center, Phase I dose-escalation study of VSLI in patients aged 2--18 years, six with ALL and three with solid tumors, patients received weekly VSLI for 4 weeks at doses of 1.75 mg/m^2^ and 2.24 mg/m^2^ with no dose capping.[@b47-ijn-8-4361] One patient required dose de-escalation at the third VSLI dose due to neuropathy, and no patients were removed from the study due to neurotoxicity. Overall, the use of VSLI was demonstrated to be safe, with a similar toxicity profile in children and adults.[@b47-ijn-8-4361] Patient-focused perspectives ============================ The ultimate goal of any new anticancer therapy is to improve response with reduced toxicity. The potential for improved efficacy with reduced peripheral and autonomic neurotoxicity would be a major advantage compared with free VCR. Although uncommon, free VCR-related peripheral neuropathy can be debilitating, and autonomic neuropathy can result in acute colonic pseudo-obstruction.[@b48-ijn-8-4361] Conclusion ========== There are currently eight open clinical trials of VSLI registered on ClinicalTrials.gov, three of which are for patients with ALL ([Table 2](#t2-ijn-8-4361){ref-type="table"}). Importantly, a number of studies in ALL and other malignancies are evaluating VSLI with multiagent chemotherapy. Given the importance of VCR for both Ph− and Ph+ ALL, leukemia trials of VSLI are also now including Ph+ patients. Clinical experience to date suggests that VSLI is well tolerated at doses greater than those currently used for free VCR. Whether this translates to improved outcomes remains to be seen. For ALL, particularly pediatric ALL where outcomes are generally excellent and chemotherapy regimens are complex, reduction in VCR-related toxicity may be a more realistic aim for VSLI. Nonetheless, tolerable dose escalation of VCR may still contribute to improved clinical responses, particularly in high-risk patients. Although the pursuit of molecularly targeted therapeutics for cancer remains critical, novel small molecules such as kinase inhibitors are rarely curative as single agents. Conventional cytotoxic chemotherapy will continue to form the backbone of anticancer treatment for the foreseeable future, so strategies to improve the therapeutic window of existing (and proven) anticancer drugs, such as encapsulation of VCR in liposomes, remain an important part of cancer drug discovery. Amanda M Smith is supported by the Children's Health Foundation Queensland. Andrew S Moore is supported by the National Health and Medical Research Council (Australia) and the Children's Health Foundation Queensland. **Disclosure** Andrew S Moore has received competitive research funding from Pfizer Inc., by way of a Pfizer Australia Cancer Research Grant. The other authors report no conflicts of interest in this work. {#f1-ijn-8-4361} ###### Clinically important drug interactions with vincristine Drug Effect on vincristine concentration Mechanism of interaction ------------------- ------------------------------------- ------------------------------------- Aprepitant Variable CYP3A4 inhibition then induction Azole antifungals Increase CYP3A inhibition Nifedipine Increase CYP3A and P-glycoprotein inhibition Cyclosporin A Increase CYP3A and P-glycoprotein inhibition Erythromycin Increase CYP3A inhibition HAART Variable CYP3A inhibition and induction Corticosteroids Decrease CYP3A induction Carbamazepine Decrease CYP3A4 induction Phenytoin Decrease CYP3A induction **Notes:** Reproduced with permission of wiley and Sons. Moore A, Pinkerton R. vincristine: can its therapeutic index be enhanced? *Pediatr Blood Cancer*. 2009;53(7): 1180--1187.[@b22-ijn-8-4361] **Abbreviations:** CYP, cytochrome P450; HAART, highly-active antiretroviral therapy. ###### Currently open clinical trials of vincristine sulfate liposomal injection Identifer Title Disease Phase Age ------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------- --------------- --------------------- NCT01439347[@b49-ijn-8-4361] A Phase 3 Study to Evaluate Marqibo® in the Treatment of Subjects ≥60 Years Old with Newly Diagnosed ALL Ph− ALL III ≥60 years NCT01319981[@b50-ijn-8-4361] Hyper-CVAD with Liposomal Vincristine in Acute Lymphoblastic Leukemia ALL (Ph− and Ph+) II ≥18 years NCT00873093[@b51-ijn-8-4361] Bortezomib and Combination Chemotherapy in Treating Young Patients with Relapsed Acute Lymphoblastic Leukemia or Lymphoblastic Lymphoma ALL and lymphoblastic lymphoma II 1--31 years NCT01222780[@b52-ijn-8-4361] To Evaluate the Safety, Activity and Pharmacokinetics of Marqibo in Children and Adolescents with Refractory Cancer Multiple I and II 2--21 years NCT01478542[@b53-ijn-8-4361] OPTIMAL \>60, Improvement of Therapy of Elderly Patients with CD20+ DLBCL Using Rituximab Optimized and Liposomal Vincristine CD20+ B-non-Hodgkin's lymphoma III 61--80 years NCT01096368[@b54-ijn-8-4361] Maintenance Chemotherapy or Observation Following Induction Chemotherapy and Radiation Therapy in Treating Younger Patients With Newly Diagnosed Ependymoma Ependymoma III 1--21 years NCT01055314[@b55-ijn-8-4361] Temozolomide, Cixutumumab, and Combination Chemotherapy in Treating Patients with Metastatic Rhabdomyosarcoma Metastatic Rhabdomyosarcoma Not specified 1 month to 49 years NCT00506142[@b56-ijn-8-4361] Safety and Efficacy of Marqibo in Metastatic Malignant Uveal Melanoma Metastatic melanoma II ≥18 years **Abbreviations:** ALL, acute lymphoblastic leukemia; CD, cluster of differentiation; CVAD, cyclophosphamide, vincristine, doxorubicin (adriamycin), dexamethasone; DLBCL, diffuse large B-cell lymphoma; Ph, Philadelphia chromosome; −, negative; +, positive.

Introduction {#sec1-1} ============ Hydatidosis caused by the cestode *Echinococcus granulosus* is a disease reported globally, however, it is endemic in Asia,\[[@ref1][@ref2]\] the Middle East, North Africa and Australia. It is much more frequent in Mediterranean countries.\[[@ref3]\] The adult taenia lives in the small intestine of carnivores such as dogs, foxes and wolves as the main host. Taenia eggs are excreted in the feces of these animals and either, directly or by contamination of plants, fruits, or vegetables, enters the gut of herbivores where the enzymes dissolve the eggs and the larvae are released. The parasite again enters the main host as these contaminated animals are eaten by carnivores. Man is an intermediate or accidental host of this parasite, and is infected by consuming contaminated plants or vegetables, or by contact with dogs.\[[@ref3]\] *Echinococcus* produce cysts that implant in different organs. Intraventricular hydatid cyst is extremely rare. Although patients may remain asymptomatic for many years or have only mild nonspecific symptoms, they might experience life-threatening complications. Because the cyst grows very slowly, the disease is diagnosed with long delay after infection.\[[@ref3]\] Case Report {#sec1-2} =========== A 15-year-old female child presented with a short history of frontal headache, projectile vomiting, and intermittent fever since 10 days. She deteriorated in the last 2 days and was brought with altered sensorium to our hospital. No signs of meningism were seen. A provisional clinical diagnosis of meningoencephalitis was made. Her general condition rapidly deteriorated with worsening sensorium, which needed intubation and ventilatory support. Emergency computed tomography (CT) of the brain \[[Figure 1](#F1){ref-type="fig"}\] revealed a nonenhancing lesion having a tiny speck of calcification in the body of right lateral ventricle. Asymmetric dilatation of the right lateral ventricle was seen with an entrapped posterior body and temporal and occipital horns. Midbrain compression was also noted. A provisional diagnosis of intraventricular cystic lesion was made and magnetic resonance imaging (MRI) was suggested. {#F1} MRI revealed multiple spherical lesions of varying sizes and varying signal intensity predominantly in the posterior body and temporal horn of the right lateral ventricle \[Figures [2](#F2){ref-type="fig"} and [3](#F3){ref-type="fig"}\]. Dilatation of the posterior body and occipital and temporal horns of the right lateral ventricle was noted. The temporal horn was seen descending through the tentorial hiatus into the quadrigeminal plate cistern compressing the midbrain. The 4^th^ ventricle was normal. Although very rare, diagnosis of a ruptured hydatid cyst was made with daughter cysts in the right lateral ventricle causing ventricular "trapping." {#F2} {#F3} Because the patient had an acute onset of symptoms and imaging revealed an entrapped right lateral ventricle with tentorial herniation, she was taken up for emergency surgery. A right parietal craniotomy was done. The temporal horn was accessed and the cystic mass was removed successfully by gentle irrigation of the cleavage plane between the cyst wall and the ventricular interface with saline. After cyst removal, the ventricle was irrigated several times with 3% hypertonic saline. Ventricular drains were inserted. Gross pathology demonstrated pearly white cystic lesions of varying sizes, containing clear fluid. Culture sensitivity revealed cysts with brood capsules with protoscolices. Histopathology revealed pieces of hooklets and inner layer of cyst wall, which was consistent with our diagnosis \[Figure [4A](#F4){ref-type="fig"} and [B](#F4){ref-type="fig"}\]. {#F4} Postoperatively prior to extubation a repeat MRI was done \[[Figure 5](#F5){ref-type="fig"}\], which revealed complete clearance of the cysts. Her postoperative course was uneventful, and she was discharged 10 days after surgery. The patient was continued on anthelmintic albendazole. {#F5} Discussion {#sec1-3} ========== Echinococcosis is also known as hydatidosis or hydatid disease. Six species have been recognized, four of which are of public health concern.\[[@ref4]\] Hydatid disease is commonly caused by the parasite *E. granulosus*. It is a zoonosis in which the primary host is a dog and other carnivores and sheep are the intermediate host, whereas humans are the accidental/intermediate host who do not complete the life-cycle of the parasite. Humans become infected by the ingestion of food or milk or water contaminated by dog feces containing the ova of the parasite. The greatest prevalence of cystic echinococcosis in human and animal hosts is found in countries of the temperate zones. Intracranial hydatid disease is rare, with reported incidence of 1--2% of all cases with hydatid disease;\[[@ref2][@ref5]\] cerebral involvement is more common in children. The incidence of intracranial hydatid in India is 0.2%, and is endemic in the Kurnool district of Andhra Pradesh, Madurai district of Tamil Nadu and Punjab.\[[@ref6]\] The typical cerebral hydatid cyst presents as a well-defined solitary cystic lesion in the usually intraparenchymal and supratentorial compartment in the middle cerebral artery territory especially in the parietal lobes, although they can be seen in any location including the skull vault, extradural compartment, intraventricular, over the meningeal surface, posterior fossa and brainstem.\[[@ref5][@ref7]\] The cerebral hydatid cysts are usually slow growing; the growth rate has been variably reported between 1.5 and 10 cm/year.\[[@ref8][@ref9]\] They may reach a considerable size before the patient becomes symptomatic. As a rule, the hydatid cyst of the brain tends to be solitary and spherical.\[[@ref10]\] Intracranial hydatid cysts may also be classified as primary or secondary.\[[@ref11]\] Primary intracranial hydatid infection is caused by embryos bypassing hepatic and pulmonary barriers and occur as a result of direct infestation of the larvae in the brain without demonstrable involvement of other organs. These are the most common types and are solitary. Secondary cysts are usually multiple which may follow embolization of the ruptured cardiac cyst or spontaneous, traumatic, and surgical rupture of a primary cyst in other organs. They lack brood capsule and scolices.\[[@ref12]\] It is commonly thought that *E. granulosus* lesions in the central nervous system are typically secondary, however, a number of studies have indicated that brain and vertebral involvement may more frequently be primary.\[[@ref13][@ref14]\] It is proposed that the high rates of secondary involvement reported previously may have been due to delayed diagnosis and insufficient treatment during the pre-CT era.\[[@ref13]\] The case we present supports this view in that no evidence of any other organ involvement was found by imaging studies or laboratory tests. Patients with intracranial hydatids usually present with nonspecific signs and symptoms, most common ones being a headache, papilledema, vomiting, and symptoms of raised intracranial pressure due to interference with pathway of cerebrospinal fluid can be seen.\[[@ref15][@ref16][@ref17]\] Focal symptoms usually depend on the site and size of the lesion. Our patient showed multiple well-defined round intraventricular lesions of varying signal intensity, few of which were bright on T1, not showing signal loss on the fat suppressed images or blooming on the gradient echo images. These cysts were thought to contain proteinaceous fluid. No enhancement was seen excluding other infective/parasitic as well as other cystic and neoplastic lesions. A hypointense capsule seen in the body of the right lateral ventricle on T2-weighted images prompted diagnosis of intraventricular hydatid. The capsule showed a break in continuity-suggestive of a ruptured cyst. The presence of daughter cysts on MRI is one of the rare but pathognomonic features.\[[@ref17]\] It is likely that the hydatid cyst was present in the lateral ventricle for a long time and its spontaneous rupture made the patient symptomatic. We reviewed the literature and came across few case reports on intraventricular hydatid cysts in adults as well as children. All these cases presented with varying manifestations of raised ICT,\[[@ref1]\] however, none of them presented with rupture liberating the daughter cysts in the lateral ventricle. Furthermore, previously reported cases demonstrated few cysts in the ventricles, however, our case report highlights existence and successful surgical clearance of multiple daughter cysts from lateral ventricle with uneventful postoperative course. This highlights the importance of having high clinical suspicion as the prognosis following surgical intervention is good. It is known that human parasitic zoonoses sometimes cause diagnostic and therapeutic problems more so when located at atypical sites. Despite newly developed neuroradiologic and serologic methods, these entities are rarely diagnosed preoperatively.\[[@ref5]\] Preoperative diagnosis of intraparenchymal hydatid cysts can be made by CT and confirmed by an MRI scan. Other investigations such as ultasonography of the abdomen and pelvis, echocardiography and X-ray chest are necessary to differentiate primary and secondary hydatid cysts. Surgically, intact cyst excision is the ideal treatment. Medical treatment with albendazole seems to be beneficial both pre and postoperatively. In an unruptured cyst, pericystic hydraulic method (Dowling--Orlando technique) gives better results in removing these cysts intact.\[[@ref7]\] The definitive diagnosis can be made by histopathologic examination. Conclusion {#sec1-4} ========== Intraventricular hydatid cysts although rare can prove to be dangerous if not diagnosed and treated earlier. Slow growth rate and appearance of symptoms in later stage of the disease adds to the morbidity and mortality. Therefore, keeping high index of suspicion in a person residing or hailing from endemic area is important for a correct preoperative diagnosis. Imaging techniques and histopathology aid to reach a correct diagnosis. Preventing infection in humans depends on the education to improve hygiene and sanitation.\[[@ref1]\] Financial support and sponsorship {#sec2-1} --------------------------------- Nil. Conflicts of interest {#sec2-2} --------------------- There are no conflicts of interest.

1.. Introduction ================ Doxorubicin, an anthracycline antibiotic, is widely used to treat a number of cancers \[[@b1-ijms-14-21629]--[@b3-ijms-14-21629]\], including as breast and lung cancers \[[@b4-ijms-14-21629]--[@b6-ijms-14-21629]\]. The first anthracyclines were isolated from the pigment-producing *Streptomyces peucetius* var. *caesius* in the 1960s and named doxorubicin (DOX) and daunorubicin (DNR) \[[@b7-ijms-14-21629]\]. They still remain one of the most effective chemotherapeutic antitumor agents \[[@b8-ijms-14-21629]\], which effect is based on intercalation into DNA helix \[[@b9-ijms-14-21629]\]. In addition, they inhibit the activity of enzyme topoisomerase II that prevents DNA repairing \[[@b10-ijms-14-21629]--[@b12-ijms-14-21629]\]. Doxorubicin (DOX) also acts by stabilizing a reaction, in which DNA strands are cut and covalently connected to the tyrosine residues of topoisomerase II, eventually impeding DNA resealing. Topoisomerase II-induced DNA damage is followed by a growth arrest in the G~1~ and G~2~ phases and apoptosis \[[@b13-ijms-14-21629]\]. Anthracyclines exert their cytotoxic effect also by generating reactive oxygen species (ROS) \[[@b14-ijms-14-21629]\], such as H~2~O~2~ and superoxide anion radical \[[@b15-ijms-14-21629]\]. These pro-oxidant properties of DOX have a potential to induce cell death through an oxidative damage of mitochondria \[[@b16-ijms-14-21629]\]. On the other hand, doxorubicin may cause several side effects \[[@b17-ijms-14-21629],[@b18-ijms-14-21629]\], which are mainly evidenced by serious deteriorations of the cardiac muscle including dilated cardiomyopathy \[[@b19-ijms-14-21629]--[@b21-ijms-14-21629]\], congestive heart failure \[[@b22-ijms-14-21629],[@b23-ijms-14-21629]\], arrhythmias \[[@b24-ijms-14-21629]\] and also myelotoxicity \[[@b7-ijms-14-21629]\]. All these effects significantly limit clinical use of DOX. Despite the extensive studies of the cardiotoxicity of DOX at cellular, biochemical, molecular, and genetic levels, it has not satisfactorily been elucidated yet \[[@b7-ijms-14-21629],[@b11-ijms-14-21629]\]. Most likely, it is a multifactorial process where alterations in cellular structure \[[@b25-ijms-14-21629]\], formation of ROS that attack the non-target structures \[[@b26-ijms-14-21629],[@b27-ijms-14-21629]\], and induction of apoptosis \[[@b20-ijms-14-21629],[@b28-ijms-14-21629]\] play the important roles. Therefore, new strategies for decreasing the cardiotoxicity of DOX are looking for. Amino acids exert a cardioprotective effect in ischemia and other cardiac disorders. They play a crucial role in the cardiac metabolism as a source of acetyl-CoA, and contribute to the production of NADH and FADH~2~ and conversion of glutamine and glutamate to free radical scavengers \[[@b29-ijms-14-21629],[@b30-ijms-14-21629]\]. However, the relevance of amino acid metabolism in the general population suffering from heart diseases remains still poorly elucidated \[[@b31-ijms-14-21629]\]. The *in vitro* ion-exchange liquid chromatographic (IELC) and spectrophotometric studies of interactions between fundamental amino acids contained in myocardium with major representative of the anthracycline cytostatics doxorubicin and comparison of content and representation of amino acids in myocardium before and after exposure to doxorubicin were the most important aims of this study. We also determined the breaking points, the critical amount of DOX that is sufficient for formation of mutual complexes for each amino acid. 2.. Results and Discussion ========================== 2.1.. Amino Acid Profile of Chicken Myocardium ---------------------------------------------- The purpose of the study was to investigate the influence of DOX on major amino acids present in myocardium. Proline, taurine, glutamic acid, arginine, aspartic acid, leucine, glycine, valine, alanine, isoleucine, threonine, lysine and serine in the downward trend were determined as the most common amino acids in the amino acid profile obtained by IELC. Their concentrations ranged from 1 μmol mL^−1^ for serine to 14 μmol mL^−1^ for proline ([Figure 1](#f1-ijms-14-21629){ref-type="fig"}), where values of concentrations were obtained as the averages from ten independent measurements. These values ranged in amounts similar to values of some amino acids that were determined in human heart in the study by Weitzel *et al.* \[[@b32-ijms-14-21629]\]. All these amino acids were subsequently used for monitoring possible interactions with DOX. 2.2.. Spectrophotometric Analysis of Amino Acids-Doxorubicin Interactions ------------------------------------------------------------------------- To investigate DOX-induced interactions with amino acids, we primarily used an UV-VIS spectrophotometric method. Marked interactions of lysine, β-alanine, valine, and arginine respectively were evident from the obtained spectra. The spectra of interactions of DOX with other amino acids pointed at no interactions at the constant concentration of doxorubicin when compared different concentrations of AA ([Figures 2](#f2-ijms-14-21629){ref-type="fig"} and [3](#f3-ijms-14-21629){ref-type="fig"}). Doxorubicin exhibited the maximum at λ = 480 nm, which corresponds to findings published in several studies \[[@b33-ijms-14-21629]--[@b35-ijms-14-21629]\]. Lysine ([Figure 2A](#f2-ijms-14-21629){ref-type="fig"}), whose amino group is highly reactive and often participates in enzymatic reactions \[[@b29-ijms-14-21629]\], β-alanine ([Figure 2E](#f2-ijms-14-21629){ref-type="fig"}), serine ([Figure 3G](#f3-ijms-14-21629){ref-type="fig"}), valine ([Figure 3I](#f3-ijms-14-21629){ref-type="fig"}) and arginine ([Figure 3M](#f3-ijms-14-21629){ref-type="fig"}) showed significantly stronger interaction with DOX with the increasing concentration (in a concentration dependent manner; highest effects observed at the concentration of 1000 μg mL^−1^) and thus indicated the highest binding affinity for the formation of complex with DOX. This phenomenon was monitored up to the concentration of 3 μmol mL^−1^ (serine), and 12 μmol mL^−1^ (lysine, arginine, β-alanine, valine and aspartic acid). Other amino acids showed interactions with DOX in much higher concentrations applied as it is apparent from doxorubicin peaks. Trends confirming the above mentioned facts can be seen in insets (a) in [Figures 2](#f2-ijms-14-21629){ref-type="fig"} and [3](#f3-ijms-14-21629){ref-type="fig"}, which indicate relations between the concentration of amino acid interacting with DOX and its subsequent effect on absorbance of DOX. Decreasing trend in the series serine, lysine, β-alanine, valine, and arginine points at higher interaction rates and at a higher content of individual amino acids. The insert curves showed no significant changes in the case of all other amino acids. In addition, we also processed differential spectra of DOX-AA interactions. These results show real forms of interaction output and its real wavelength (insets (a) in [Figures 2](#f2-ijms-14-21629){ref-type="fig"} and [3](#f3-ijms-14-21629){ref-type="fig"}). When doxorubicin subjected to interaction with serine, lysine, arginine, β-alanine, valine and aspartic acid, shifts of maximum wavelengths within the range from 464 to 465 nm were observed. Similar wavelength shift was observed also in the case of proline that changed wavelength of doxorubicin to λ = 462 nm. Other amino acids exhibited relatively small shifts of DOX wavelength, but the differences were present at all of them. These changes mention the amendment to structural changes of molecule of DOX, which is capable to form a complex with amino acid. Curves in insets (a) in [Figures 2](#f2-ijms-14-21629){ref-type="fig"} and [3](#f3-ijms-14-21629){ref-type="fig"} point at weak interaction between DOX, proline and glycine. This interaction is not influenced by concentrations of these amino acids. Low effects of AA concentrations were also observed in the case of leucine, isoleucine, and threonine. Interestingly, among these amino acids the smallest wavelength changes of doxorubicin were also observed. These amino acids were shown as the least accessible for the interaction with DOX. Using the spectrophotometric method we proved that doxorubicin may interact with some amino acids, the basic stones of all myocardial proteins, substrates for the synthesis of proteins, and products of their degradation. Some connections with these products were also found by Taetrneyer *et al.* \[[@b36-ijms-14-21629]\]. 2.3.. Analysis of Amino Acids-Doxorubicin Interactions by IELC -------------------------------------------------------------- To gain more detailed insight into the mechanisms, in terms of how doxorubicin interacts with the amino acids, we carried out IELC analysis. For this purpose, different concentrations of DOX were subjected to interaction with the constant concentration of each AA. As illustrated in [Figures 4](#f4-ijms-14-21629){ref-type="fig"} and [5](#f5-ijms-14-21629){ref-type="fig"}, we similarly observed apparent effect of DOX on amino acids serine, lysine, β-alanine, valine, and arginine respectively. Surprisingly, we also observed its effect on other amino acids, *i.e.*, proline, glycine, taurine, threonine, aspartic acid, glutamic acid, isoleucine, and leucine. This effect was manifested even at relatively low concentrations of doxorubicin in the range between 1 μmol mL^−1^ for serine and 84 μmol mL^−1^ for glutamic acid (see Chapter *Breaking points of amino acids*). Amino acids may act as important signaling molecules \[[@b31-ijms-14-21629]\], especially BCAAs are effective activators of the mammalian target of rapamycin (mTOR) signalling cascade \[[@b37-ijms-14-21629]\], which is directly involved in cardiac hypertrophy in pathways of regulation of proteosynthesis \[[@b38-ijms-14-21629],[@b39-ijms-14-21629]\]. The main function of mTOR is the stimulation of cell growth and anabolism through increasing protein and lipid synthesis via activation of S6K (S6 kinase), 4E-BP (4E-binding protein), and SREBP (sterol response element binding protein) \[[@b40-ijms-14-21629],[@b41-ijms-14-21629]\]. The limited availability of myocardial proteins and the potential to lose the function may induce structural alterations resulting in the formation of free radicals or in changes in antioxidant status \[[@b42-ijms-14-21629]--[@b44-ijms-14-21629]\]. Accumulation of free radicals may play a crucial role in depletion of adenosine triphosphate and subsequent opening of the non-specific mitochondrial permeability transition pores (mPTPs) \[[@b45-ijms-14-21629]\] allowing molecules smaller than 1.5 kDa to penetrate through the mitochondrial pores and change mitochondrial membrane potential. All these effects, especially the loss of mitochondrial membrane potential, leads to the release of molecules with pro-apoptotic potential (e.g., cytochrome c) into cytosol \[[@b46-ijms-14-21629],[@b47-ijms-14-21629]\], which results in the degradation of mitochondria, loss of myofilaments and progressive atrophy of myofibrils \[[@b48-ijms-14-21629]\]. Additionally, we also investigated retention times of individual amino acids. Whereas most of the amino acids analyzed maintained their retention time under the influence of doxorubicin without significant changes, retention time of proline was significantly influenced by DOX in the concentration-dependent manner ([Figure 4B](#f4-ijms-14-21629){ref-type="fig"}). Retention time of proline without DOX added was established at 32.05 min. The highest concentration of doxorubicin (1000 μg mL^−1^) led to a shortening of retention time to 30.35 min. Similar changes in retention time were observed in the case of glutamic acid, where the signal was observed after 13.7 min without DOX and after 12.29 min after adding of 1000 μg mL^−1^ of doxorubicin ([Figure 5M](#f5-ijms-14-21629){ref-type="fig"}). Changes in retention time point at a possible formation of certain AA-DOX complex that has very similar properties as corresponding amino acid, but slightly shifted its retention time. 2.4.. Effect of Doxorubicin on Breaking Points of Amino Acids ------------------------------------------------------------- Results of IELC analysis showed that doxorubicin interacts with amino acids, especially at low concentrations. Due to this fact, we carried out mathematical analysis of the breaking points for individual amino acids to determine the lowest concentration of DOX that causes noticeable effect on AA ([Figure 6A](#f6-ijms-14-21629){ref-type="fig"}), expressed as the lowest concentration of DOX required for formation of the complex with AA. The lowest breaking point has been determined for serine at the concentration of 1 μg mL^−1^ of doxorubicin. The highest one has been shown for glutamic acid (84 μg mL^−1^). The breaking points of all analyzed amino acids are shown in [Table 1](#t1-ijms-14-21629){ref-type="table"}. As it is clear from [Table 1](#t1-ijms-14-21629){ref-type="table"}, doxorubicin possesses the ability to interact with amino acids in concentrations lower than we expected. Predictive value of breaking points according to cardiotoxicity is considerable, but it is important to reveal the real amino acids composition of heart. This information may be further applied as a simple mathematical calculation revealing how much of doxorubicin has potential to influence the major amino acids---especially the most vulnerable ones. When compared with spectrophotometric analysis, serine showed the lowest breaking point, however, amino acids serine, lysine, arginine, β-alanine, and valine have very low values of breaking point; thus, these amino acids are the most accessible for interaction with DOX. Valine belongs to the important group of BCAAs that act as the activators of mTOR signalling pathway \[[@b49-ijms-14-21629],[@b50-ijms-14-21629]\]. The depletion of these amino acids may result in the alterations in the function of mTOR with subsequent influencing of synthesis of proteins. In addition, we assembled the summarizing output from UV/VIS spectrophotometric analysis ([Figure 6B](#f6-ijms-14-21629){ref-type="fig"}). These results are similar to the breaking points shown in [Figure 6A](#f6-ijms-14-21629){ref-type="fig"} regarding to a willingness of amino acids to interact (form a complex) with DOX. When comparing these two outputs coming from different analyses, in the case of the amino acids, which are influenced by the lowest DOX concentration simultaneously, it is valid that a low concentration of amino acid is sufficient to be influenced by the constant concentration of DOX. Serine at the concentration of 3 μmol mL^−1^ exhibited noticeable effect of DOX. At valine, lysine, arginine, β-alanine and aspartic acid sufficient concentration for the same effect was detected to 12 μmol mL^−1^. Threonine and leucine were influenced by DOX at the concentrations of 50 μmol mL^−1^, taurine at 75 μmol mL^−1^ and proline, glycine, isoleucine and glutamic acid at 100 μmol mL^−1^. The concentrations are indicative for verifying of the interaction trends and are based on the initial concentrations used for UV/VIS spectrophotometric analysis. 2.5.. Impact of Doxorubicin on Amino Acids ------------------------------------------ According to Manocha and Margaritis, and Yoo and Park \[[@b35-ijms-14-21629],[@b51-ijms-14-21629]\], DOX is a positively charged amphoteric molecule, containing in its sugar moiety (daunosamine) a protonable amino group and in its aglycone part, two deprotonable phenolic groups ([Figure 7](#f7-ijms-14-21629){ref-type="fig"}). At physiological pH or in deionized distilled water, the amino group gets protonated (as NH~3~^+^) and provides positive charge to the DOX molecule \[[@b35-ijms-14-21629]\]. Hence, it is most likely that electrostatic interactions are established between positively charged DOX and negatively charged amino acids, which results in formation of the DOX-AA complexes in dependence on binding affinity of AA. The ability of doxorubicin to form a linkage through the amide bond has been already described in several studies \[[@b52-ijms-14-21629]--[@b54-ijms-14-21629]\]. Therefore, we hypothesize that DOX-AA complexes are formed in the physiological environment found in the human body and this fact affects non-target cytotoxicity of doxorubicin. 2.6.. Comparison of Amino Acid Profile of Myocardium before and after Application of Doxorubicin ------------------------------------------------------------------------------------------------ We carried out scans of both hearts, untreated and treated with DOX, to confirm its presence in myocardium ([Figure 8A,B](#f8-ijms-14-21629){ref-type="fig"}). Distribution of DOX (red highlighting) mainly in heart apex and estuary of aorta is well evident (red highlighting) in [Figure 8B](#f8-ijms-14-21629){ref-type="fig"}. Compared with control, reduced levels of amino acids occurring in myocardium after exposure to doxorubicin were observed. Proline (19.94% disparity), taurine (19.03%), and glutamic acid (32.54%) were the least affected amino acids by DOX. On the other hand, serine (85.18%) and lysine (83.16%) followed by valine (78.51%) and β-alanine (77.02%) were the most affected amino acids in myocardium ([Figure 8C](#f8-ijms-14-21629){ref-type="fig"}). These data support our findings from previous measurements about increased reactivity of above mentioned amino acids and their ability to interact easily with doxorubicin. 3.. Experimental Section ======================== 3.1.. Chemicals and pH Measurement ---------------------------------- Working solutions as buffers or standard solutions of amino acids and DOX were prepared daily by a diluting the stock solutions. Amino acids, DOX standards and others were purchased from Sigma Aldrich (St. Louis, MO, USA) in ACS purity, unless noted otherwise. All solutions were prepared in deionized water obtained using a reverse osmosis equipment Aqual 25 (Aqual s.r.o., Brno, Czech Republic). The deionized water was further purified by using an apparatus Direct-Q 3 UV Water Purification System equipped with an UV lamp from Millipore (Billerica, MA, USA). The resistance was established to 18 MΩ cm^−1^. The pH was measured using a pH meter WTW inoLab (Weilheim, Germany). 3.2.. Preparation of Myocardium Samples and DOX Fluorescence Detection ---------------------------------------------------------------------- For acquisition of the profiles of amino acids, ten chicken hearts were obtained (Diema s.r.o., Frydek-Mistek, Czech Republic). From each heart (*n* = 10), 10 mg of tissue was equally removed, weighed and added to 0.5 mL of 6 M HCl. Sample was subsequently subjected to digestion in a microwave reaction system Anton Paar (Anton Paar GmbH, Graz, Austria) using the following conditions: power-80, Ramp 15 min, Hold 90 min, Max 120 °C, Max pressure 25 bar, Rotor-XF-100-6. Thereafter, the digested sample was diluted 10 times with dilution buffer composed of thiodiglycol 5 mL L^−1^, citric acid 14 g L^−1^, sodium chloride 11.5 g mL^−1^ and centrifuged using a Microcentrifuge 5417R (Eppendorf AG, Hamburg, Germany) under 25,000 g at 4 °C for 10 min. The samples prepared like this were diluted with a neutralizing solution (6 M NaOH in a dilution buffer) again in ratio 1:1 and analyzed on an analyzer of amino acids (Model AAA-400, Ingos, Prague, Czech Republic). Ten chicken hearts were thereafter injected with 50 μL doxorubicin dissolved in physiological saline solution to the final concentration of 1000 μg mL^−1^. For confirmation of the presence of DOX in the myocardium, a Carestream *In Vivo* Xtreme Imaging System (Carestream Health, Inc., Rochester, NY, USA) was used to detect the fluorescence of doxorubicin after one hour-lasting exposure. Parameters were set as it follows: excitation wavelength---480 nm, emission wavelength---600 nm, exposure time---2 s, binning---2 × 2, *f*-stop---1.1, field of view---7.2 × 7.2 cm. For analysis of amino acids profile in myocardium, samples were prepared in the same way as was described above in this chapter. Analysis was then carried out at on an analyzer of amino acids (Model AAA-400, Ingos, Prague, Czech Republic) using conditions described in chapter "*Determination of content of amino acids in myocardium and analysis of interactions using IELC*". To compare content of amino acids before and after application of DOX, the differences were expressed as percentage disparities. 3.3.. Preparation of Amino Acid-Doxorubicin Sample for Interaction Study ------------------------------------------------------------------------ The results showing the most abundant amino acids in chicken hearts were further used to monitor interactions with DOX. The stock solutions of amino acids (AA) and doxorubicin were prepared daily in the concentration of 1 mg mL^−1^ by diluting with ACS water. The final concentrations of AA and DOX were prepared by diluting with ACS water from the stock solutions unless noted otherwise. The concentrations of DOX of 8; 16; 32; 64; 125; 250; 500 and 1000 μg mL^−1^ interacting with the constant concentration of amino acids of 100 μg mL^−1^ were used for IELC evaluation of AA-DOX interactions. The concentrations of individual amino acids of 1; 2; 3; 6; 12; 25; 50 and 100 μmol mL^−1^ were used to obtain absorption spectra of AA-DOX interactions. The absorption spectra were obtained after 24 h of interaction at 25 °C with doxorubicin in concentration of 250 μg mL^−1^. 3.4.. Determination of Content of Amino Acids in Myocardium and Analysis of Interactions Using IELC --------------------------------------------------------------------------------------------------- Firstly, IELC was used to determine AA content in myocardium before and after DOX application. An ion-exchange liquid chromatography (Model AAA-400, Ingos, Prague, Czech Republic) with post column derivatization by ninhydrin and an absorbance detector in the visible light range (VIS) was used. A glass column with inner diameter of 3.7 mm and 350 mm length was filled manually with strong cation exchanger in the sodium cycle LG ANB with approximately 12 μm particles and 8% porosity. The column was tempered on the 60 °C. The double channel VIS detector with an inner cell of 5 μL volume was set to two wavelengths: 440 and 570 nm. Solution of ninhydrin was prepared in 75% *v*/*v* methylcelosolve (Ingos, Prague, Czech Republic) and in 2% *v*/*v* 4 M acetic buffer (pH 5.5). Tin chloride (SnCl~2~) was used as a reducing agent. Prepared solution of ninhydrin was stored under inert atmosphere (N~2~) in dark at 4 °C. Elution of amino acid was done by a buffer containing 10.0 g of citric acid, 5.6 g of sodium citrate, and 8.36 g of NaCl per liter of solution and pH was 3.0. Flow rate was 0.25 mL min^−1^. Reactor temperature was set on 120 °C. For dilution of samples, a dilution buffer was used (composition: thiodiglycol 5 mL L^−1^, citric acid 14 g L^−1^, sodium chloride 11.5 g mL^−1^). For monitoring of AA-DOX interactions the same parameters of analysis were used instead of the time of analysis, which depended on amino acid determined. 3.5.. Spectrophotometric Analysis --------------------------------- Absorption spectra of interactions between amino acids and doxorubicin were carried out on a spectrophotometer SPECORD 210 (Analytik Jena AG, Jena, Germany) within the range from 220 to 800 nm with 1 nm step. For analysis, an UV semi-micro plastic cuvette with 1 cm optical path (Brand GMBH, Wertheim, Germany) was used. Cell compartment was thermostated to 25 °C by a thermostat Julabo (Labortechnik, Wasserburg, Germany). Absorption spectra were recorded after 24 h of the interaction and evaluated by the program WinASPECT version 2.2.7.0 (Analytik Jena AG, Jena, Germany). Finally, differential spectra of DOX-AA interactions were processed according to the formula: spectrum DOX \- AA \- spectrum DOX 3.6.. Determination of Breaking Points of Amino Acids ----------------------------------------------------- For the function calculating the breaking points of amino acids, following variables were defined: *a* for absorbance (*y*-axis), *l* for lowest point on *y*-axis, *u* for uppermost point on *y*-axis, *k* for slope of curve, *i* for log breakpoint position and *c* for concentration of doxorubicin. Using these variables, the relation of doxorubicin concentration and absorbance can be expressed as it follows: a = l \- u \- l 1 \+ 10 k ( i \- c ) To fit the curve, variables *l*, *u*, *k* and *i* were calculated using the least squares method. Consequently, breakpoint (*b*) was subsequently calculated according to: b = 10 i Macro in Microsoft Excel using a solver tool was used to create the fit of the curve and to calculate the breaking points of individual amino acids. 3.7.. Descriptive Statistics ---------------------------- Mathematical analyses of the experimental data and their graphical interpretation were realized by the Microsoft Office tools (MS Excel^®^, MS Word^®^, and MS PowerPoint^®^). All results were expressed as a mean ± standard deviation (S.D.) unless noted otherwise. The detection limits (3 signal/noise, S/N) were calculated according to Long and Winefordner \[[@b55-ijms-14-21629]\], whereas N was expressed as a standard deviation of noise determined in the signal domain unless stated otherwise. Differences with *p* \< 0.05 were considered significant and were determined by using of one way ANOVA test (particularly Scheffe test), which was applied for means comparison. 4.. Conclusions =============== In our study, we determined that doxorubicin induces formation of complexes with amino acids in myocardium. This fact supports the well-known knowledge about the cardiotoxicity of doxorubicin. Despite the fact that some hypotheses about the mechanism of the anthracycline-induced cardiotoxicity have been established, it has not yet been sufficiently explained. An effect of doxorubicin on amino acids may be an important factor involved in this multifactorial and very complicated process. Possible formations of complexes may play important role in the adverse effects of doxorubicin; however, this phenomenon must be further investigated. We also carried out comparison of the quantitative representation of amino acids in myocardium before and after application of doxorubicin. We observed significant reduction of levels of all amino acids in myocardium after exposure to doxorubicin. These findings lead us to knowledge that amino acids play an important role in the cardiotoxicity of doxorubicin in a dose-dependent manner. Initial amino acid composition of heart may hypothetically play an essential role in resistance of heart to doxorubicin. Nevertheless, sensitive BCAAs are very much needed for mTOR managed proteosynthesis. The solution might be found in nutrition supplements providing branched chain amino acids, and thus protecting the proper function of protein synthesis, but their effectiveness would have to be tested. It is important to reveal if DOX induced damage leads to alterations of contractility or cardiac conduction and if there is any chance how to recognize the progressive and reversible damage. From these purposes we have the future plans to carry out *in vivo* experiments, further uncovering this phenomenon. Understanding the pathophysiology of cardiac dysfunction associated with anthracyclines is important for prediction, treatment, and prevention of these adverse side effects of chemotherapy. Financial support from CYTORES GACR P301/10/0356, CEITEC CZ.1.05/1.1.00/02.0068, by MH CZ-DRO, University Hospital Motol, Prague, Czech Republic 00064203 and Liga proti rakovine LPR 2014 is highly acknowledged. 4E-BP : 4E-binding protein AA : Amino acids BCAAs : Branched-chain amino acids DNR : Daunorubicin DOX : Doxorubicin IELC : Ion-exchange liquid chromatography mPTPs : Mitochondrial permeability transition pores mTOR : Mammalian target of rapamycin ROS : Reactive oxygen species S6K : S6 kinase SREBP : Sterol response element binding protein. The authors declare no conflict of interest. {#f1-ijms-14-21629} {#f2-ijms-14-21629} {#f3-ijms-14-21629} {#f4-ijms-14-21629} {#f5-ijms-14-21629} {#f6-ijms-14-21629} {#f7-ijms-14-21629} {#f8-ijms-14-21629} ###### Overview of the breaking points expressing the lowest concentration of doxorubicin that influences noticeably the amino acids analysed. Ser---serine, Val---valine, Ala---β-alanine, Lys---lysine, Asp---aspartic acid, Thr---threonine, Leu---leucine, Pro---proline, Tau---taurine, Ile---isoleucine, Gly---glycine, Glu---glutamic acid, BP---breaking point. Amino acid Ser Val Ala Arg Lys Asp Thr Leu Pro Tau Ile Gly Glu ----------------------- ----- ----- ----- ----- ----- ----- ----- ----- ----- ----- ----- ----- ----- BP (μg mL^−1^ of DOX) 1 8 12 13 25 26 31 59 60 64 76 77 84

Introduction {#s0001} ============ Over the past decade, biologics have significantly changed the management of inflammatory Crohn's disease (CD)[@CIT0001] and ulcerative colitis (UC)[@CIT0002]. However, access to treatment with biologics is frequently limited by their high cost[@CIT0003]. The introduction of less costly biosimilars may increase patient access to well-established therapies while improving healthcare affordability, which is particularly important in lower GDP countries. Biosimilar infliximab (CT-P13) was the first monoclonal antibody biosimilar approved in the European Union, and has also received regulatory approval in Australia, Canada, Japan, and the US[@CIT0004]. Early clinical trials to demonstrate the therapeutic equivalence of biosimilar infliximab to reference infliximab were conducted in patients with ankylosing spondylitis and rheumatoid arthritis[@CIT0005]. The results of these studies supported regulatory approval of biosimilar infliximab across all indications for which the parent biologic was approved, including inflammatory bowel disease (IBD)[@CIT0004]. Additional observational data are, thus, required to confirm the risk:benefit profile of biosimilar infliximab in patients with IBD under everyday use in clinical practice. The aim of this study was to evaluate the effectiveness and safety of biosimilar infliximab administered in a real-life setting in adults with active CD or UC in three selected Eastern European countries. Methods {#s0002} ======= Study design and inclusion criteria {#s0003} ----------------------------------- This multi-centre, observational, non-interventional study was conducted at sites in Romania, the Czech Republic, and Bulgaria. Participating sites were medical centres experienced in the biological treatment of patients with CD and UC. Investigators were selected for participation if they had the available patient population (representative of the target patient population), and were able to conduct the observational study according to applicable legal and regulatory requirements of their respective country. Eligible patients were adults greater than 18 years of age with moderate-to-severe active CD (not responding to corticosteroids and/or immunosuppressive agents) or moderate-to-severe active UC (inadequate response to conventional therapy) who were eligible for treatment with infliximab and were prescribed biosimilar infliximab. Exclusion criteria were: known hypersensitivity to infliximab, other murine proteins, or any component of the formulation; moderate-to-severe heart failure (New York Heart Association Class III/IV); presence of fever \>38 °C on the day of administration; received a live or live-attenuated vaccine within 8 weeks of screening or were scheduled to receive a live or live-attenuated vaccine before study end (killed vaccines were allowed); received an investigational agent for CD or UC within 5 half-lives or 90 days of screening, whichever was greater; currently receiving biologic therapy indicated for treatment of CD or UC (patients experiencing an inadequate response to previous biologic therapy and switching to biosimilar infliximab were allowed to participate); current history of chronic or active infection with hepatitis B (HBsAg), hepatitis C, or infection with human immunodeficiency virus 1 or 2; current diagnosis of tuberculosis or other severe chronic infection (e.g. sepsis, abscess or opportunistic infection, or invasive fungal infection such as histoplasmosis), or a past diagnosis without sufficient documentation of complete resolution following treatment; and patients who, in the opinion of their general practitioner or investigator, should not participate in the study. Patients were treated with biosimilar infliximab as per the approved product label[@CIT0009]^,^[@CIT0010]. Patients received biosimilar infliximab at baseline, and at Weeks 2, 6, 12, 14, 22, and 30 (end of treatment). Evaluations were conducted at enrolment (baseline) and at Week 30. In the event of early withdrawal from treatment, the patient's next routine follow-up visit was the study termination visit and included efficacy evaluations. The study was approved by the respective national Ethics Committees, and all patients provided written informed consent prior to participation. Assessment criteria {#s0004} ------------------- The primary efficacy endpoint was the percentage of patients with a response to biosimilar infliximab at Week 30. The Crohn's Disease Activity Index (CDAI) was used to evaluate effectiveness in CD patients. Clinical response was defined as a decrease of \>70 points from the baseline CDAI score; clinical remission was defined as a CDAI score of 150 or less[@CIT0011]. The partial version (without endoscopy) of the Clinical Activity Index (pCAI) was used to evaluate effectiveness in UC patients. Clinical response was defined as a decrease of at least 3 points from the baseline pCAI score and an absolute sub-score for bleeding not greater than 1 point; clinical remission was defined as a total pCAI score of 2 points or lower[@CIT0012]^,^[@CIT0013]. Secondary efficacy endpoints were changes from baseline to Week 30 in mean CDAI and pCAI scores, and change from baseline to Week 30 in quality-of-life (QoL) scores assessed using the Short Inflammatory Bowel Disease Questionnaire (SIBDQ)[@CIT0014]. Safety was assessed according to early withdrawals and adverse events (AEs) monitoring. Adverse events were categorized according to their severity (mild, moderate, or severe) and likely relationship to treatment (unrelated, possibly related, probably related, or definitely related). Treatment-related events were categorized as adverse drug reactions (ADRs). Statistical analysis {#s0005} -------------------- Descriptive statistics were used. Continuous parameters were summarized using mean and standard deviation (SD). Qualitative data were reported as absolute (*n*) and relative (%) frequency distributions. Paired *t*-tests were used to compare mean change in disease activity from baseline to end of treatment. A *p*-value \<0.05 was considered statistically significant. Analyses were conducted in the safety population, which included all patients who received at least one dose of biosimilar infliximab. For analyses of disease activity, the last available observation carried forward (LOCF) method was used for patients with no effectiveness measurements at Week 30 due to early termination. For analyses of response and remission, patients who discontinued treatment early were considered to be non-responders/non-remitters. The observed case (OC) method was used for analysis of QoL. Based on patient enrolment (*n* = 85), the precision estimate of the 95% confidence interval (CI) for the primary outcome measure (clinical response rate) was ±10.6%. CIs for binomial proportions were calculated using the Agresti-Coull method. Results {#s0006} ======= Patient characteristics {#s0007} ----------------------- Demographics and clinical characteristics of patients are summarized in [Table 1](#t0001){ref-type="table"}. A total of 85 patients were enrolled in the study, 38 (44.7%) with CD and 47 (55.3%) with UC. The mean age of the cohort was 40.7 ± 15.0 years and 58.8% of patients were male. The majority of patients with CD (*n* = 24; 63.2%) or UC (*n* = 44; 93.6%) were new to infliximab treatment. ###### Baseline demographic and clinical characteristics (safety population). Characteristic Crohn's disease (*n* = 38) Ulcerative colitis (*n* = 47) All patients (*n* = 85) ---------------------------------- ---------------------------- ------------------------------- ------------------------- **Gender (M:F), *n* (%)** 17 (44.7):21 (55.3) 33 (70.2):14 (29.8) 50 (58.8):35 (41.2) **Age, years, mean ± SD** 38.1 ± 15.0 42.8 ± 14.9 40.7 ± 15.0 **Body weight, kg, mean ± SD** 67.2 ± 17.3 73.2 ± 14.5 70.5 ± 16.0 ***De novo* treatment, *n* (%)** 24 (63.2) 44 (93.6) 68 (80.0) The median follow-up was 210 (range = 43--282) days. The mean dose of biosimilar infliximab at Week 30 was 5.0 ± 0.6 mg/kg. Most patients (*n* = 67; 78.8%) received the six scheduled infusions of biosimilar infliximab. Patient disposition {#s0008} ------------------- A total of 67 patients (CD: 27; UC: 40) completed the study as per protocol. Eighteen patients withdrew from biosimilar infliximab treatment prematurely. Of these, three patients discontinued treatment because of AEs (including one patient with a life-threatening infusion reaction), eight patients were withdrawn due to therapeutic failure, four patients were withdrawn due to low baseline CDAI or pCAI scores, two patients were withdrawn at their request, and one patient was lost to follow-up. Effectiveness {#s0009} ------------- Treatment response is summarized in [Table 2](#t0002){ref-type="table"}. At Week 30, the response rate was 65.8% (95% CI = 49.8--78.9) in patients with CD and 55.3% (95% CI = 41.2--68.6) in patients with UC. Corresponding remission rates were 47.4% (95% CI = 32.5--62.7) and 48.9% (95% CI = 35.3--62.8), respectively. ###### Response and remission rates at the end of treatment in patients with Crohn's disease or ulcerative colitis. Outcome Crohn's disease[^a^](#TF2){ref-type="table-fn"} (*n* = 38) Ulcerative colitis[^b^](#TF3){ref-type="table-fn"} (*n* = 47) All (*n* = 85) ------------------------------------------------------ ------------------------------------------------------------ --------------------------------------------------------------- ---------------- Response % (95% CI[^c^](#TF4){ref-type="table-fn"}) 65.8 (49.8--78.9) 55.3 (41.2--68.6) 60.0 Remission % (95% CI[^c^](#TF4){ref-type="table-fn"}) 47.4 (32.5--62.7) 48.9 (35.3--62.8) 48.2 Patients who terminated the study before Week 30 were considered to be non-responders/non-remitters. Clinical response was defined as a decrease of \>70 points from the baseline Crohn's Disease Activity Index (CDAI) score; clinical remission was defined as a CDAI score of 150 or less[@CIT0011]. Clinical response was defined as a ≥3 point decrease from the baseline partial Clinical Activity Index (pCAI) score and an absolute sub-score for bleeding not greater than 1 point; clinical remission was defined as a total pCAI score of 2 points or lower[@CIT0012]^,^[@CIT0013]. Calculated using the Agresti-Coull method. In patients with CD, the mean CDAI score decreased from 359 ± 175 at baseline to 190 ± 170 at the end of treatment (LOCF), indicating a significant improvement (*p* \< 0.0001) in disease activity with biosimilar infliximab ([Figure 1](#F0001){ref-type="fig"}). {#F0001} In patients with UC, the mean pCAI score decreased from 9.8 ± 3 at baseline to 3.1 ± 4.0 at the end of treatment (LOCF), indicating a significant improvement (*p* \< 0.0001) in disease activity with biosimilar infliximab ([Figure 2](#F0002){ref-type="fig"}). {#F0002} In the combined CD and UC patient population, QoL improved significantly (*p* \< 0.0001) during the observation period. The mean SIBDQ score increased from 40.0 ± 11.3 at baseline to 56.2 ± 9.3 at the end of treatment (OC), as shown in [Figure 3](#F0003){ref-type="fig"}. {#F0003} Safety {#s0010} ------ A total of 473 infusions of biosimilar infliximab were administered and 31 AEs (6.6%) were recorded. Two events (0.4%) were considered by investigators to be definitely related to treatment. Both events were infusion reactions of moderate severity, one of which led to treatment discontinuation. One AE was considered possibly related to treatment, and the remaining 28 AEs were considered unrelated to treatment with biosimilar infliximab. Of three treatment-related ADRs, two were of moderate and one was of mild intensity. Discussion {#s0011} ========== The results of this study show that, in a real-life clinical setting in three selected Eastern European countries, biosimilar infliximab was effective and safe in patients with CD or UC. After 30 weeks' treatment, 66% of the overall population had a clinical response and 48% were in remission. Highly significant improvements from baseline to Week 30 were observed in disease activity and QoL. Biosimilar infliximab was well tolerated. ADRs with a definite causal relationship with treatment as judged by investigators were limited to two events of moderately severe infusion reactions. Subsequent to regulatory approval of biosimilar infliximab across all indications, numerous observational studies reported on its effectiveness and safety during use in everyday practice. Komaki *et al*.[@CIT0015] reviewed 11 observational studies involving 829 patients with CD or UC who were either induced with or switched to biosimilar infliximab. Consistent with our findings, pooled clinical response rates at 24--30 weeks were 77% (CD) and 77% (UC), and pooled clinical remission rates were 60% and 52%, respectively. Gisbert and Chaparro[@CIT0016] reviewed 24 studies in 1326 patients with IBD who had been switched from originator infliximab to biosimilar infliximab. These authors reported a weighted mean disease control rate of 88%, with no increased risks of immunogenicity or other safety concerns. The therapeutic equivalence of biosimilar infliximab observed after switching is not unexpected, as serum concentrations were shown to be non-inferior to those with originator infliximab in patients with CD or UC in clinical remission[@CIT0017]. In the past few years, several other groups have added to the weight of evidence suggesting that biosimilar infliximab is effective and safe as induction or switch therapy in patients with IBD[@CIT0018]. Although the number of patients (*n* = 17) in the current study with previous exposure to anti-TNF agents was too low to conduct a meaningful sub-group analysis, we were interested to compare outcomes in this sub-group with those in the overall population. Among 14 evaluable patients, nine (53%) responded to treatment, of which six (35%) were in remission. Although no data were collected about withdrawal of the initial anti-TNF in these patients, it has been reported that response to a second anti-TNF is better if the reason for switching is intolerance rather than primary or secondary failure[@CIT0023]. Immunogenicity is a recognized risk associated with use of biologic agents in the management of chronic inflammatory disease. The risk has been shown to vary with different biologic/biosimilar agents, including infliximab[@CIT0024]. In one study of the use of biosimilar infliximab in 384 patients with IBD in Hungary and the Czech Republic, there were 28 reports of infusion reactions (7.3%). Patients with prior anti-TNF exposure or antidrug antibody positivity were significantly more likely to develop such reactions[@CIT0025]. A similar incidence of infusion reactions was reported in the PROSIT cohort (8.8%, 71 of 810 IBD patients)[@CIT0018]. The incidence of infusion reactions in the current study (two of 84 patients, 2.4%) was less than that in these earlier reports, perhaps reflecting the lack of previous exposure to anti-TNFα treatment in the majority (80.0%) of the population. Potential exists for biosimilar infliximab to generate costs savings sufficient to meet the needs of many more patients with IBD. A budget impact analysis in six European countries, including Romania and the Czech Republic, showed that interchange of medication from infliximab to biosimilar infliximab in patients with CD may result in budgetary savings adequate to treat an estimated 700--1500 additional patients[@CIT0026]. Severs *et al*.[@CIT0027] designed a stochastic economic model to simulate the introduction of biosimilars for IBD in the Netherlands over a 5-year time horizon. These authors concluded that, while the magnitude of the economic impact would ultimately depend on factors such as local pricing, procurement policies, and the willingness of physicians to switch patients to biosimilars, the potential cost savings would be in the vicinity of 30%. In 2015, annual direct drug cost savings in five European countries through the introduction of biosimilar infliximab were projected to range from 10--30%[@CIT0028]. The study is limited by its observational design, which can introduce selection bias, and by the relatively modest sample size and short observation period. On the other hand, the study provided insight into the real-world management of patients with IBD in a naturalistic setting. Another limitation is the lack of objective efficacy measures such as biomarkers and endoscopy; however, as these evaluations are rarely performed in everyday practice, the amount of available patient data was insufficient to conduct meaningful analyses. Overall, we consider that the close alignment between our findings and those of large pooled analyses strengthens the perception that biosimilar infliximab is an effective and safe alternative to reference infliximab in patients with IBD, and has the potential to fulfil the unmet needs of more patients. Conclusions {#s0012} =========== In the real-life setting, biosimilar infliximab administered for 30 weeks improved disease activity and was well tolerated in patients with CD or UC. In addition, biosimilar infliximab had a strong positive impact on patients' QoL. Together with other studies reporting on the effectiveness and tolerability of biosimilar infliximab in patients with chronic inflammatory conditions including IBD, the weight of evidence suggests that biosimilar infliximab is an effective and cost-effective replacement for originator infliximab. Transparency {#s0013} ============ Declaration of funding {#s0014} ---------------------- This paper was not funded. Declaration of financial/other relationships {#s0015} -------------------------------------------- The authors have no relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript. This includes employment, consultancies, honoraria, stock ownership or options, expert testimony, grants or patents received or pending, or royalties. JDA peer reviewers on this manuscript have no relevant financial or other relationships to disclose. Acknowledgements ================ Statistical analysis was provided by Gábor Csukly of BioData Kft, Budapest, Hungary. Medical writing assistance was provided by Kerry Dechant of Content Ed Net (Budapest, Hungary), with funding from Egis Pharmaceuticals PLC (Budapest, Hungary). Previous presentation {#s0016} ===================== This article was presented as a poster at the 13th Congress of the European Crohn's and Colitis Organization (ECCO), 14--17 February 2018, in Vienna, Austria.

{#sp1 .459} {#sp2 .460}

Introduction {#mgg3282-sec-0005} ============ Campomelic dysplasia (CD; MIM \#114290) is a semilethal skeletal malformation syndrome characterized by macrocephaly, mid‐face hypoplasia, Robin sequence, hip dislocation, bowed femora and tibiae (i.e., campomelia), talipes, and respiratory distress. Radiological features include hypoplastic scapulae, small chest, 11 rib pairs, undermineralized thoracic pedicles, kyphosis or scoliosis, pelvic malformations, bowing of femora and tibiae, hypoplastic fibulae, and short first metacarpals (Mansour et al. [1995](#mgg3282-bib-0013){ref-type="ref"}; Unger et al. [2013](#mgg3282-bib-0025){ref-type="ref"}). About 10% of cases missing the eponymous feature of campomelia are referred to as acampomelic CD (ACD). An additional feature in CD is the XY sex reversal that affects about two‐thirds of karyotypic males. Campomelic dysplasia has an overall mortality rate of \~75% in neonates, increasing to \~90% by 2 years of age (Mansour et al. [1995](#mgg3282-bib-0013){ref-type="ref"}). ACD cases appear to have a much better prognosis, with \~65% alive at ≥1 year (Mansour et al. [2002](#mgg3282-bib-0014){ref-type="ref"}). CD/ACD individuals who survive the neonatal period may have a number of complications, the most common of which include recurrent apnea/respiratory tract infection, progressive kyphoscoliosis, short stature, dislocation of the hips, conductive hearing loss, and mild to moderate learning difficulties (Mansour et al. [2002](#mgg3282-bib-0014){ref-type="ref"}). Campomelic dysplasia results from de novo heterozygous *SOX9* mutations. The majority of the mutations are distributed over the entire coding region of the *SOX9* gene and include missense, nonsense, frameshift, and splice mutations, while a few CD cases are due to large deletions covering *SOX9*. As these mutations are predicted to result in loss‐of‐function alleles, CD is regarded as a haploinsufficiency syndrome (Unger et al. [2013](#mgg3282-bib-0025){ref-type="ref"}). In addition, chromosomal rearrangements including translocations, inversions, and deletions that do not affect the gene body but instead interrupt the *SOX9* regulatory domain, which extends over more than 1 Mb upstream of *SOX9*, can also result in CD of varying severity. In these cases, removal of long‐range tissue‐specific enhancers is assumed to affect *SOX9* expression (Gordon et al. [2009](#mgg3282-bib-0006){ref-type="ref"}). Material and Methods {#mgg3282-sec-0006} ==================== Human subjects {#mgg3282-sec-0007} -------------- This is a single case report. Appropriate informed consent was obtained from the patient\'s parents. Generation of constructs {#mgg3282-sec-0008} ------------------------ The 5′UTR of the *SOX9* gene (GenBank [NM_000346.3](NM_000346.3)) from wild‐type and mutant DNA was amplified by PCR, cloned into pcR2.1 TOPO (Invitrogen, Karlsruhe, Germany) and subsequently transformed into competent Dh5*α* cells. Plasmids were isolated and forward (5′‐ACT GCT GTG CTG TGA TTG GCG GGT GGC TCT AAG‐3′) and reverse (5′‐CTA GGG CCC TTG GTT GCC CGG GGC CGG GGC AGG GGG CTG G‐3′) primers were used to create a 10 bp frameshift. All constructs, wild type, mutant and frameshift, were cloned in‐frame 5′ to the c‐myc epitope tag using the *HindIII* and *ApaI* site in the pcDNA3.1/myc‐His/B expression vector (Invitrogen). A partial *SOX9* Exon 1‐containing vector (pcDNA3.1/Sox9/1‐304) was digested with *ApaI*, and the resulting 711 bp fragment was introduced into the *ApaI* site next to the epitope of each construct. The constructs were verified by automated sequencing (ABI Prism 3100 Genetic Analyzer, Applied Biosystems). In vitro transcription‐translation {#mgg3282-sec-0009} ---------------------------------- For the generation of fusion proteins, the TNT T7 Coupled Reticulocyte Lysate System (Promega, Mannheim, Germany) was used together with ^35^S‐methionine (Amersham, Freiburg, Germany) following the manufacturer\'s protocol. For this purpose, 1 μg of each construct and 0.5 μg of control plasmid MKKS‐pGBKT7 (encoding a C‐myc‐tagged MKKS protein of 62 kD) were used. After completed TNT reaction, 20 μL of 6 × SDS‐PAGE sample buffer was added to 5 μL of the samples, boiled for 5 min and separated by SDS‐PAGE on a 12% acrylamid gel. The gel was treated with fixation buffer and Amplify Solution (Amersham) and exposed overnight at room temperature using Kodak BioMax MS film (Kodak, Rochester, NY, USA). Transfection and immunoblotting {#mgg3282-sec-0010} ------------------------------- COS‐7 cells were cultured in RPMI supplemented with 10% FCS and 100 μg/mL pen/strep. For expression experiments, approximately 1 × 10^6^ cells were seeded into 100 mm dishes and transfected at 60--80% confluency. Using Lipofectamine™ (Invitrogen), 6 μg of each construct was cotransfected with a control plasmid (4 μg). Each transfection was performed at least in triplicate and with two different DNA preparations of each construct. Cells were harvested 24 h posttransfection, lysed in 300 μL RIPA buffer, sonified and centrifuged at 10000 rpm for 1 min. Cell lysates were resolved by SDS‐PAGE on a 12% acrylamide gel and analyzed by immunoblotting using primary mouse anti‐c‐myc‐antibody (1:1000, BD Bioscience, Heidelberg, Germany) at 4°C over night, followed by a secondary antibody (1:5000) for 1 h at RT. The signals were monitored by incubating with ECL reagent (Amersham) for 1 min. A Fuji Medical Super RX X ray film (Fuji, Düsseldorf, Germany) was exposed according to the signal intensity. Results {#mgg3282-sec-0011} ======= Clinical course and diagnostic workup of patient {#mgg3282-sec-0012} ------------------------------------------------ The patient and her healthy twin brother were born at 40 weeks gestation by cesarean section. Parents are healthy and unrelated and conceived the twins by IVF. Apgar scores for the patient were 1^9^ and 5^9^. Birth weight was 2.44 kg (\<2nd centile) and OFC 35.4 cm (50th--98th centile). She had a dusky episode at 10 minutes which resolved with oxygen but showed increasing respiratory distress requiring intubation and ventilation by day 3. A tracheostomy tube was inserted because of persisting requirements for ventilation. Bronchoscopy showed tracheomalacia with extrinsic compression. An aortopexy was performed to relieve compression caused by the right brachiocephalic artery. She required gastrostomy because of failure to thrive. Referral was made to the genetics department at Glasgow at age 7 months. On examination, she had macrocephaly, a flat nasal root, epicanthic folds, a long philtrum, micrognathia, and low‐set ears (Fig. [1](#mgg3282-fig-0001){ref-type="fig"}A,B). Skeletal survey revealed hypoplastic scapulae and cervical kyphosis, thoracic scoliosis with undermineralized thoracic pedicles, 11 rib pairs, very short ischia, and unossified inferior pubic rami (Fig. [1](#mgg3282-fig-0001){ref-type="fig"}C); straight femora, tibiae, and fibulae with delayed ossification of the femoral epiphyses (Fig. [1](#mgg3282-fig-0001){ref-type="fig"}D); enlarged cranial vault (Fig. [1](#mgg3282-fig-0001){ref-type="fig"}E); and short metacarpal I and moderately short middle and distal phalanges (Fig. [1](#mgg3282-fig-0001){ref-type="fig"}F). She had a dilated left renal pelvis, a small patent foramen ovale and mild sensorineural hearing loss. Karyotype is 46,XX. The patient therefore has 3/5 radiological and 5/12 clinical features used as diagnostic criteria for CD (Mansour et al. [1995](#mgg3282-bib-0013){ref-type="ref"}). A clinical diagnosis of acampomelic campomelic dysplasia, ACD, was made. {#mgg3282-fig-0001} To alleviate problems with respiration, VEPTR (vertical expandable prosthetic titanium rib) implants were applied at age of 5 years, which were subsequently expanded by repeated operations. At age 12, a vertebral column resection with removal of two apical vertebrae, anterior column support and long posterior fusion from the second thoracic to the third lumbar vertebra was performed. About 2 years later, she developed late infection and implant removal was required. She carried a tracheostomy tube and needed assisted ventilation at night until age 14, when the tracheostoma was successfully closed. At her present age of 15 years, she still has hearing problems and requires tympanostomy tubes, needs orthopedic shoes, and has a strong wheat allergy. She attends an international school, where she performs rather well. Molecular genetic results {#mgg3282-sec-0013} ------------------------- As sequence analysis of the *SOX9* coding region and exon/intron boundaries failed to reveal a mutation in the patient, the 372 bp 5′ untranslated region (UTR) was scrutinized following amplification with forward (5′‐ACT GCT GTG CTG TGA TTG GCG‐3′) and reverse (5′‐CTT CTC CTG CTC GTC GGT CAT C‐3′) primers. A heterozygous mutation c.‐185G\>A in the 5′ UTR was found. This sequence alteration gives rise to a novel upstream AUG (uAUG) translation initiation codon 185 nucleotides before the normal AUG start codon (the A corresponding to nucleotide +1). The uAUG creates a novel upstream open reading frame (uORF) of 62 codons which is out‐of‐frame with the normal ORF and which terminates just after the wild‐type start codon (Fig. [2](#mgg3282-fig-0002){ref-type="fig"}A). The patient\'s parents and the healthy twin brother were homozygous G/G at this position, indicating a de novo mutation in the patient. The G\>A transition is not found in 100 control chromosomes, dbSNP, or the ExAC consortium database (Lek et al. [2016](#mgg3282-bib-0011){ref-type="ref"}) nor in 30 CD or CD‐like cases without a *SOX9* coding region mutation. Paternity was confirmed by microsatellite marker analysis. Interestingly, the flanking sequence of the upstream AUG translation start codon matches the Kozak consensus (Kozak [1986](#mgg3282-bib-0007){ref-type="ref"}) much better than the sequence preceding the wild‐type AUG start codon (4/7 vs. 1/7) (Fig. [2](#mgg3282-fig-0002){ref-type="fig"}B). {#mgg3282-fig-0002} Expression studies {#mgg3282-sec-0014} ------------------ To analyze the translation initiation potential of the upstream AUG, we generated c‐myc‐epitope‐tagged partial SOX9 expression constructs in pcDNA3.1/myc‐His/B bearing either wild‐type or mutant 5′‐UTR sequences. Adopting a previously described strategy (Liu et al. [1999](#mgg3282-bib-0012){ref-type="ref"}), a frameshift construct with a 10 bp insert within the upstream ORF was also made to obtain an in‐frame peptide initiating at the uAUG and extending up to the c‐myc tag. The frameshift construct allowed us to evaluate the translational activity at the mutant and the wild‐type start codon. Conceptual translation from the wild‐type construct results in a 29 kD protein with the c‐myc‐tag, whereas translation from the mutant construct yields the 29 kD wild‐type protein and a 7 kD protein lacking the epitope tag. Translation from the frameshift construct predicts the wild‐type protein as well as a novel, longer 36 kD protein, both with the epitope tag (Fig. [3](#mgg3282-fig-0003){ref-type="fig"}A). {#mgg3282-fig-0003} In vitro transcription‐translation confirmed that the wild‐type construct directed expression of the 29 kD protein. The mutant construct yields less protein, indicating a residual translational activity at the wild‐type start codon. The frameshift construct drives translation of both the 29 kD wild‐type and the 36 kD novel protein. The stronger signal corresponding to the novel protein indicates preferential translation from the mutant AUG (Fig. [3](#mgg3282-fig-0003){ref-type="fig"}B). To confirm that the mutant AUG directs translation also in eukaryotic cells, the same constructs were transfected into COS‐7 cells, and expression was monitored by western blot analysis with an anti‐c‐myc‐antibody. The results (Fig. [3](#mgg3282-fig-0003){ref-type="fig"}C) were in complete agreement with those obtained by in vitro transcription‐translation shown in Fig. [3](#mgg3282-fig-0003){ref-type="fig"}B. As internal control, a construct driving expression of a 62 kD c‐myc‐tagged MKKS protein was used (Fig. [3](#mgg3282-fig-0003){ref-type="fig"}B,C). We conclude that the c.‐185G\>A mutation has functional consequences for translation initiation at the normal AUG in vitro, and by extrapolation reduces expression of the native SOX9 protein in normal tissues. Discussion {#mgg3282-sec-0015} ========== As shown by in vitro transcription‐translation and transient transfection assays, the de novo mutation in the *SOX9* 5′ UTR results in preferential initiation of translation at the uAUG at the expense of efficient translation initiation at the wild‐type AUG start codon. The fact that generation of wild‐type SOX9 protein from the mutant allele of this ACD patient is not entirely abolished indicates that the situation is not complete haploinsufficiency. A genotype/phenotype correlation has repeatedly been postulated based on the observation that mutant *SOX9* alleles with supposedly residual function lead to the milder, acampomelic form of CD (Scherer et al. [2000](#mgg3282-bib-0020){ref-type="ref"}; Staffler et al. [2010](#mgg3282-bib-0023){ref-type="ref"}). This assumption is strengthened by the fact that ACD cases are overrepresented in the group of *SOX9* translocation cases and in the group with *SOX9* missense mutations. In both groups, the affected *SOX9* alleles are not obvious loss‐of‐function alleles and thus can be assumed to retain residual activity. Two recent publications provide comprehensive lists of *SOX9* mutations in CD/ACD. The first lists about 70 mutations in the *SOX9* coding region, of which 11 are from ACD cases, almost exclusively missense mutation in the DNA‐binding HMG domain of the SOX9 transcription factor (Table S3; Mattos et al. [2015](#mgg3282-bib-0016){ref-type="ref"}). The other report summarizes chromosomal rearrangements upstream and downstream of *SOX9*, with 13 and nine rearrangements reported for ACD and CD cases, respectively (Table 1; Fonseca et al. [2013](#mgg3282-bib-0003){ref-type="ref"}). Three additional ACD mutations have meanwhile been published, two missense mutations (Gopakumar et al. [2014](#mgg3282-bib-0005){ref-type="ref"}; Preiksaitiene et al. [2016](#mgg3282-bib-0019){ref-type="ref"}) and one translocation (Walters‐Sen et al. [2014](#mgg3282-bib-0026){ref-type="ref"}). Despite this large number of reported *SOX9* mutations in ACD, the functional consequences of these mutations have only occasionally been tested experimentally. For five missense mutations, four in the HMG domain (M113V, R152P, H165Q, and H169Q) and one in the dimerization domain (A76E), residual DNA‐binding (Meyer et al. [1997](#mgg3282-bib-0017){ref-type="ref"}) and also transactivation of SOX9 target genes has been demonstrated (Sock et al. [2003](#mgg3282-bib-0022){ref-type="ref"}; Staffler et al. [2010](#mgg3282-bib-0023){ref-type="ref"}; Matsushita et al. [2013](#mgg3282-bib-0015){ref-type="ref"}). Pop et al. ([2005](#mgg3282-bib-0018){ref-type="ref"}) studied a cohort of 10 CD cases, including several cases with longer survival and two ACD cases, which were all heterozygous for the nonsense mutation Y440X that truncates the C‐terminal transactivation domain spanning residues 402--509 of SOX9. Cell transfection experiments revealed that the mutant SOX9 protein retained some transactivation activity on SOX9 target genes (Pop et al. [2005](#mgg3282-bib-0018){ref-type="ref"}). In addition, in a t(13;17) ACD translocation case surviving for at least 26 years (case 2 in Tommerup et al. [1993](#mgg3282-bib-0024){ref-type="ref"}), expression of both *SOX9* alleles has been demonstrated in lymphoblasts from the patient (Wirth et al. [1996](#mgg3282-bib-0027){ref-type="ref"}). The documentation that some protein is made from the normal ORF of the mutant *SOX9* allele in our patient gives further support to the notion that more mildly affected, surviving CD/ACD patients like the present case may carry mutant *SOX9* alleles which are hypomorphic rather than complete null alleles. The G\>A mutation creates an AUG triplet in the 362 nucleotide *SOX9* 5′ UTR that is normally devoid of AUGs. That this mutant upstream AUG triplet functions as an effective initiator codon results from two facts. Firstly, mRNAs are usually translated by a scanning mechanism, whereby the ribosome enters at the capped 5′ end of the mRNA and migrates linearly until it encounters the first AUG codon, which is the uAUG in the present case. Secondly, the uAUG triplet resides in a sequence context that is more favorable for translation initiation than that at the wild‐type AUG. While the latter matches the GCCRCC[AUG]{.ul}G Kozak consensus only at one position, the sequence around the uAUG matches it at four positions. The purine (R) 3 nucleotides before the AUG in the Kozak sequence and the G following the AUG make the strongest contributions to an optimal translation initiation (Kozak [1986](#mgg3282-bib-0007){ref-type="ref"}, [1997](#mgg3282-bib-0008){ref-type="ref"}). Interestingly, the uAUG has a G residue 3 nucleotides before the AUG, while the normal AUG has a C at this position; in either case, an A follows the AUG. Although the uAUG is the initiation codon closest to the 5′ end of the mRNA, some translation still occurs at the wild‐type AUG. This possibly results from a reinitiation mechanism, which allows access to an internal start codon. Because reinitiation occurs only in the forward direction, the strongest constraint is imposed by an upstream ORF that overlaps the internal start site leading to low‐level translation (Kozak [2002](#mgg3282-bib-0009){ref-type="ref"}), as in the present case. The second possibility is leaky scanning, whereby ribosomes bypass the first AUG and initiate translation farther downstream, usually when the first AUG codon resides in a suboptimal context (Kozak [2002](#mgg3282-bib-0009){ref-type="ref"}). Although the sequence at the mutant uAUG has a better match to the Kozak consensus than the wild‐type AUG, it is still suboptimal, as it has an A and not a G residue following the AUG. Thus, either or both of these possible mechanisms may be responsible for the observed low‐level initiation at the *SOX9* start codon. A mutation in the 5′UTR creating a functional uAUG codon has first been described in autosomal dominant hereditary melanoma. There, a novel uAUG in *CDKN2A* led to decreased translation from the wild‐type AUG and to predisposition to melanoma. As in our case, the short out‐of‐frame ORF initiated from the uAUG overlaps the *CDKN2A* start codon and resulted in profound inhibition of translation initiation at the normal AUG (Liu et al. [1999](#mgg3282-bib-0012){ref-type="ref"}). Mutations creating uAUGs have subsequently been described and functionally tested in 12 other human disorders (reviewed by Barbosa et al. [2013](#mgg3282-bib-0001){ref-type="ref"}). In all these cases, the uORFs generated by the novel uAUGs reduced protein expression from the normal ORF to 30% or less. Two further examples of novel uAUGs compromising translation at the normal start site were reported: in the *IFITM5* gene in osteogenesis imperfecta (Cho et al. [2012](#mgg3282-bib-0002){ref-type="ref"}; Semler et al. [2012](#mgg3282-bib-0021){ref-type="ref"}), and in the *GRHPR* gene in primary hyperoxaluria type II (Fu et al. [2015](#mgg3282-bib-0004){ref-type="ref"}). Together with these reports, our study emphasizes that for mutation screening in human disorders, the 5′ UTR should be routinely analyzed for potentially pathogenic variants generating novel uAUGs and reading frames. In conclusion, we report the first CD/ACD case with a de novo base substitution in the 5′ UTR of *SOX9* that generates a functional upstream start codon and an upstream ORF that is out‐of‐frame with the main coding sequence, leading to reduced but not completely abolished expression of the normal ORF. Our study corroborates the notion that more mildly affected surviving CD/ACD patients carry mutant *SOX9* alleles with residual expression of SOX9 wild‐type protein. Conflict of Interest {#mgg3282-sec-0017} ==================== None declared. We are most grateful to the patient and her family for participating in this study.

Introduction ============ Although effective treatment can resolve acute postsurgical pain in many patients, 30%--50% of individuals who undergo common invasive surgeries suffer persistent postoperative pain.[@b1-jpr-10-763] The nociceptive processes under these conditions are not fully understood, and our knowledge on the triggers of persistent postsurgical pain is limited. Identifying these factors (ie, targets) will aid the development of more adequate non-invasive therapies.[@b2-jpr-10-763] Studies of pain axes in the spinal cord have shown that mitogen-activated protein kinases (MAPKs) contribute to the initiation and maintenance of post-surgical pain. Inactivation of phosphorylated MAPK p38 (p-p38) successfully reduced pain-related behaviors in rodents, following plantar incision surgery[@b3-jpr-10-763],[@b4-jpr-10-763] and treated postoperative pain in humans.[@b5-jpr-10-763] The role of MAPKs is also instrumental in chronic pain conditions. Phosphorylation of p38 or extracellular signal-regulated kinase (ERK)-1/2, another MAPK, is necessary for the development and perpetuation of mechanical hypersensitivity, as observed in models of chronic neuropathic[@b6-jpr-10-763] and acute-to-chronic postoperative pain.[@b3-jpr-10-763],[@b7-jpr-10-763],[@b8-jpr-10-763] Although p38 inhibition has demonstrated some efficacy in neuropathic pain in humans,[@b9-jpr-10-763],[@b10-jpr-10-763] its role in this condition remains unclear.[@b11-jpr-10-763] Even though the pathophysiological events in the spinal cord are important for the maintenance of postoperative pain,[@b3-jpr-10-763],[@b7-jpr-10-763],[@b8-jpr-10-763] cellular and molecular processes at the site of the surgical incision are also relevant. Studies on injured tissue in rodents, following paw incision, have linked the up-regulation of neurotrophin and cytokine genes with postoperative pain.[@b12-jpr-10-763] Molecular mediators such as nerve growth factor,[@b13-jpr-10-763] lactic acid,[@b14-jpr-10-763] prostaglandins,[@b15-jpr-10-763] and complement factor,[@b16-jpr-10-763] in the incised tissue are also related to postoperative pain. Moreover, inhibition of p-p38 and phosphorylated ERK-1/2 (p-ERK-1/2), in the mouse peripheral tissue, reduced nociceptive behaviors in a model of phorbol 12-myristate 13-acetate-induced inflammatory pain.[@b17-jpr-10-763] Increased levels of a downstream MAPK signaling product, interleukin 6, at the incision site, are associated with postoperative pain in humans[@b18-jpr-10-763] and with the development of hyperalgesia in murine pain models.[@b19-jpr-10-763]--[@b21-jpr-10-763] This study proposes that a dysregulation of the peripheral mechanisms that promote MAPK deactivation underlies the transition from acute to persistent postoperative pain. MAPK phosphatases (MKP) are the major negative regulators of MAPKs. Specifically, MKP-3 preferentially dephosphorylates p-ERK-1/2, whereas MKP-1 favors p-p38 dephosphorylation.[@b22-jpr-10-763],[@b23-jpr-10-763] However, their substrate preference varies between cell types, anatomical locations, or pathological conditions.[@b7-jpr-10-763],[@b24-jpr-10-763] It has been demonstrated that MKPs can act as potential analgesic targets.[@b25-jpr-10-763],[@b26-jpr-10-763] This study showed that impaired MKP-3 signaling in the spinal cord leads to persistent mechanical hypersensitivity, through dysregulation of p-p38 and p-ERK-1/2.[@b7-jpr-10-763] Moreover, it demonstrated that spinal induction of MKP-1 prevents neuropathic mechanical hypersensitivity in rats.[@b27-jpr-10-763] In addition to their role in the central nervous system, MKP-1 and MKP-3 mediate peripheral inflammatory processes.[@b28-jpr-10-763],[@b29-jpr-10-763] However, the contribution of MKPs at the peripheral sites in pain conditions has not been investigated. This study aimed to understand the role of MKPs and their target MAPKs at the site of the surgical incision in nociceptive behavior in mice. Based on previous studies,[@b7-jpr-10-763],[@b26-jpr-10-763] it is hypothesized that the dysregulation of peripheral MKP-3 will result in persistent ERK-1/2 and p38 phosphorylation and persistent mechanical allodynia, in the mouse paw incision model. Materials and methods ===================== Animals ------- Wild-type (WT, B6129SF2/J) and MKP-3 knockout (KO, B6129X1-Dusp6 TM1Jmol/J) male and female (50--60 days old) mice were used (The Jackson Laboratory, Bar Harbor, ME, USA). MKP-3 KO mice were reported to be deficient in MKP-3 protein in the spleen, heart, lungs, and thymus tissues.[@b30-jpr-10-763] It was shown that MKP-3 KO mice lack MKP-3 protein in the spinal cord.[@b7-jpr-10-763] The authors confirmed that these mice also lack MKP-3 in the paw tissue ([Figure 1](#f1-jpr-10-763){ref-type="fig"}). The animals were housed individually and maintained in light-dark cycles for 12 h with ad libitum access to food and water. All animal procedures were done in accordance with the Guidelines for Animal Experimentation of the International Association for the Study of Pain and were approved by the Institutional Animal Care and Use Committee at Dartmouth College. All the animals (KO or WT) used for behavioral experiments or that received treatment or that were used for tissue collection and molecular assessments were distributed using a randomized allocation to different groups. The method used was simple, choosing random numbers that coincide with the identification number of the animals from a list including all the animals of the given groups. Surgery: paw incision model of postoperative pain ------------------------------------------------- All the animals were anesthetized with isoflurane (4% induction, 2% maintenance) in oxygen, delivered through a nose cone. Plantar aspect of the left hind paw was sterilized with providone-iodine before and after surgeries. Approximately 0.5 cm longitudinal incision was made through skin and fascia of the plantar aspect of the left hind paw from the heel to the base of the toes, using a No. 10 scalpel blade and sterile technique. The plantaris muscle and ligaments were elevated and stretched for 6--8 s (they were not incised). The wound was cleaned and apposed with inverted 6.0 silk sutures (Ethicon Inc., Somerville, NJ, USA). The animals were allowed to recover from the surgery for 1 h and maintained with ad libitum access to food and water.[@b31-jpr-10-763],[@b32-jpr-10-763] Behavioral testing ------------------ Mechanical allodynia, measured as 50% withdrawal threshold, was assessed in the ipsilateral paw to surgery using von Frey filaments (0.07-, 0.17-, 0.40-, 0.60-, 1.04-, 1.37-, and 2.0-g bending force) (Stoeling, Wood Dale, IL, USA) and calculated using an up-down statistical method.[@b3-jpr-10-763],[@b33-jpr-10-763] In both WT and KO animal groups, baseline behavior was tested in naïve mice (no surgery), as well as 1, 5, 7, and 12 days after surgery (n=6 per group; 6 males in each group). The animals were followed up until postoperative day 12 since it had been previously shown that paw incision-induced inflammation, measured as paw circumference, returns to baseline levels on day 12 after surgery.[@b7-jpr-10-763] An additional group of MKP-3 KO mice (n=3) was used to assess the role of MAPKs; 50% withdrawal threshold was measured before the surgery, 12 days after surgery, and 0.5, 1, and 2 h after intraplantar injection of an MEK inhibitor (PD98059) that blocks phosphorylation of ERK-1/2, administered on day 12. Two experimenters (blind to treatment conditions) performed two sets of experiments; the data were combined for analysis. Cellular infiltration --------------------- Animals were anesthetized and perfused transcardially using 0.1 M phosphate-buffered saline, followed by 4% formaldehyde. Tissue from the operated paw was removed in a consistent manner: the center of the incision wound was localized, and a No. 10 blade was used to cut the soft tissue 1.5 mm apart from the center, rostrally, and caudally. The plantar bones determined the depth of the collected tissue; the tissue was removed from the paw by an additional longitudinal cut above the bone of the paw. A 3 mm piece of plantar paw tissue was placed in 30% sucrose solution for 48--72 h, at 4°C. The tissue was mounted and frozen in optimal cutting temperature compound (Sakura Finetek, Torrence, CA, USA) at −80°C. Hematoxylin and Eosin (H&E) staining was performed on 20 μm paw tissue sections, cut using Leica CM1859 Cryostat, according to standard protocols. The stained tissues were dehydrated and sealed with Vectashield mounting medium (Vector Labs, Burlingame, CA, USA). Images were obtained using Q-fired cooled CCD camera attached to an Olympus microscope BX60. H&E staining was performed on two groups of animals (n=3 male animals per group, WT and MKP-3 KO), prior to (naïve) and on different days after surgery (days 1, 5, 7, and 12). Three tissue sections per animal per time point were evaluated for the number of infiltrating cells. Three areas of 0.1 mm^2^ per tissue section were scored and averaged. The percentage of polymorphonuclear (PMN) and mononuclear cells was assessed morphologically. One experimenter (blind to genotype conditions) performed this analysis. Drugs and intraplantar injections --------------------------------- PD98059, an MEK inhibitor used to block p-ERK-1/2 expression (Sigma-Aldrich Co., St. Louis, MO), was diluted in dimethylsulfoxide and saline (1:1, vehicle). PD98059 (25 nmol, 8 μL), or vehicle, was injected intraplantary to the left hind paw of MKP-3 KO mice (ipsilateral to surgery), on postoperative day 12. The concentration of the drug was chosen based on a previous study[@b7-jpr-10-763] and other pain mouse models.[@b34-jpr-10-763] Mechanical allodynia was assessed before surgery and on postoperative day 12, prior to PD98059 injection, as well as 0.5, 1, and 2 h following drug administration. The experimenter performed these studies blind to treatment conditions (n=9, six males and three females in the vehicle treated group; n=8, five males and three females in the MEK inhibitor-treated group). Western blot was performed on six paw tissue samples, five male mice and one female mouse, from each treatment group. Western blot ------------ The animals were euthanized using decapitation under isoflurane anesthesia (4% in oxygen). The plantar tissues of the left hind paw (ipsilateral) were removed using the method described in the "Cellular Infiltration" section. A 3 mm piece of tissue was obtained and processed; tissue was sonicated in PBS, with 1:1000 protease inhibitors (Sigma-Aldrich Co.). Protein concentrations were determined by DC assay (Bio-Rad Laboratories Inc., Hercules, CA, USA). A 50 μg protein sample was loaded onto sodium dodecyl sulfate (SDS) polyacrylamide electrophoresis gel (10%; Bio-Rad Laboratories Inc.). The protein was transferred to a 0.2 μm nitrocellulose membrane (Bio-Rad Laboratories Inc.) and processed as previously described.[@b7-jpr-10-763] Primary antibodies used were as follows: mouse anti-p-ERK-1/2 (1:1000), rabbit anti-p-p38 (1:1000), mouse anti-ERK-1/2 (1:1000), rabbit anti-p38 (1:500) (Cell Signaling Technology, Inc., Danvers, MA, USA), rabbit anti-MKP-1 (1:500; Santa Cruz Biotechnology Inc., Dallas, TX, USA), and rabbit anti-MKP-3 (1:3000; Cell Signaling Technology). β-Actin was used as loading control and stained using mouse anti-β-actin primary antibody (1:3000; Abcam, Cambridge, MA, USA). Following primary antibody incubation (16--20 h at 4°C), the blots were washed in tris-buffered saline (TBS) + 0.05% Tween 20 and incubated for 1 h at room temperature in one of the following secondary antibodies: goat anti-mouse HRP-conjugated antibody (1:3000; Bethyl Laboratories Inc., TX, USA) or goat anti-rabbit horseradish peroxidase (HRP)-conjugated antibody (1:3000; Pierce, Rockford, IL, USA). Subsequently, the blots were washed and visualized using Super Signal West Femto Maximum Sensitivity Substrate (Pierce). Imaging was conducted using Syngene G-Box system and software (Synoptics, Frederick, MD, USA). Following visualization, the membranes were incubated for 25 min at 37°C in 5% sodium azide TBS buffer and re-probed with another primary and secondary antibody as described previously.[@b7-jpr-10-763] In some cases (MKP-3 and MKP-1 Western blots), β-actin blots were run simultaneously on a different membrane using the same samples and same protein amounts in both membranes. Signal intensity of the protein of interest was assessed using Image J software, and the data were quantified by dividing the intensity of the signal of the protein of interest with the corresponding signal intensity of β-actin. Quantifications of p-ERK-1/2 and p-p38 levels were done by normalizing p-ERK-1/2, ERK-1/2, p-p38, and p38 against β-actin, after which p-ERK-1/2 and p-p38 data were quantified as relative intensity of the particular signal, divided by the previously normalized value of total ERK-1/2 and total p38, respectively. For the MKP-1, MKP-3, p-ERK, and p-p38 time course quantification studies in WT and MKP-3 KO groups, two independent experimenters (blind to genotype conditions) performed these analyses, and both the sets of data were combined (n=3 male mice per group). Statistical analysis -------------------- Statistical analyses were performed using GraphPad Prism 5 (GraphPad Software Inc., La Jolla, CA, USA), and all data are expressed as mean ± standard error of mean (SEM). The size of the samples was decided based on previous publications by these authors, and therefore, no a priori statistical power calculation was conducted.[@b7-jpr-10-763],[@b24-jpr-10-763] Normality was tested using Shapiro--Wilk test. Two data sets from the MEK inhibitor (PD98059) group and one data set from the vehicle group were identified as outliers (2-fold standard deviation) and therefore excluded from the analysis in the behavioral experiments. Behavior data were analyzed using a repetitive measurement two-way analysis of variance (ANOVA) followed by Bonferroni post hoc test; N=6. The values for cellular infiltration measurements after the paw incision did not follow a normal distribution and were analyzed as non-parametrical data. Kruskal--Wallis followed by a Dunn test was performed to determine the changes in cellular infiltration against the baseline values (naïve mice) in each group. Mann--Whitney *U* test was performed to compare the effects on cellular infiltration between the KO and WT groups at each time point; N=6. The Western blot analyses between groups were determined by one-way ANOVA followed by Dunnett's post-test using the naïve group as the control, or two-way ANOVA followed by Bonferroni's post-test, as appropriate, N=3. Comparison of the relative expression of p-ERK-1/2 in PD98059-treated versus vehicle-treated group in KO mice was performed using Mann--Whitney *U* test for non-parametrical data (N=6). A *P*-value \<0.05 was considered statistically significant. Results ======= Relationship between MKP-3 and pain-related behaviors in the mouse model of acute postoperative pain ---------------------------------------------------------------------------------------------------- In both WT and MKP-3 KO mice, paw incision surgery led to an equal level of mechanical hypersensitivity on postoperative day 1 ([Figure 1A](#f1-jpr-10-763){ref-type="fig"}). Although the 50% withdrawal threshold in WT mice returned to baseline levels on postoperative day 12, mechanical allodynia in MKP-3 KO mice continued until 7--12 days after surgery ([Figure 1A](#f1-jpr-10-763){ref-type="fig"}). As previously demonstrated, MKP-3 KO mice did not exhibit overtly altered behavior when compared with WT mice.[@b7-jpr-10-763] Development and grooming were comparable between the two murine strains. It was confirmed that MKP-3 KO mice did not express MKP-3 protein in the paw tissue and that MKP-1 protein expression was consistent between WT and MKP-3 KO mice ([Figure 1B](#f1-jpr-10-763){ref-type="fig"}), thus excluding potential compensatory mechanisms. Role of peripheral inflammation in mechanical allodynia in WT and MKP-3 KO mice ------------------------------------------------------------------------------- This study evaluated the cellular infiltration levels in the paw tissue 1--7 days after surgery and compared it with cellularity at baseline and at day 12 after surgery. H&E histological staining in WT mice showed a significant increase in the number of infiltrating cells on postoperative day 5 after surgery, which remained elevated at day 7 after surgery ([Figure 2A and B](#f2-jpr-10-763){ref-type="fig"}). In contrast, in MKP-3 KO mice, the number of infiltrating cells significantly increased on postoperative day 1, in which a peak effect and a significant difference compared to the WT group was observed. The number of cells in the KO mice declined initially but remained significantly elevated on postoperative days 5 and 7 ([Figure 2A and B](#f2-jpr-10-763){ref-type="fig"}). However, these values were not different from the WT group values ([Figure 2A and B](#f2-jpr-10-763){ref-type="fig"}). The degree of cellular infiltration returned to basal levels 12 days after surgery in both the groups. When different populations of PMN and mononuclear cells were evaluated, it was found that the percentage of PMN cells in the MKP-3 KO mice (86.9±2.8%) was significantly higher than that in WT mice (77.1±5.1%) 1 day after surgery (*P*=0.02). A decrease in the percentage of PMN cells was found at the later points after paw incision in both MKP-3 KO and WT mice, but no changes in the percent of PMN cells were found between the groups on postoperative days 5 (MKP-3 KO =31.4±5.7%, WT =22.9±3.3% PMN cells, *P*=0.20) and 7 (MKP-3 KO =18.5±5.2%, WT =13.1±2.3% PMN cells, *P*=0.10). The proportion of mono-nuclear cells was smaller than the proportion of PMN cells on postoperative day 1 in both MKP-3 KO and WT mice, as expected. However, a lower percentage of mononuclear cells was obseverd on postoperative day 1 in the MKP-3 KO group (13.1±4.0%) compared to the WT group (22.9±5.1%, *P*=0.02). Conversely, a predominant proportion of mono-nuclear cells was observed on postoperative days 5 (MKP-3 KO =68.6±5.73%, WT =77.1±4.6% PMN cells, *P*=0.14) and 7 (MKP-3 KO =81.5±5.2%, WT =86.9±2.28% PMN cells, *P*=0.10) in both MKP-3 KO and WT groups, but no significant differences were found between these groups. p-ERK-1/2 and p-p38 expression levels in the paw tissue of WT and MKP-3 KO mice ------------------------------------------------------------------------------- Since MAPKs are the natural substrates of MKP-3, and MAPKs determine at least in part the pro-inflammatory phenotype of immune cells, the phosphorylation level of p-ERK and p-p38 was studied in the paw soft tissue (surgical area). In WT mice, levels of p-ERK-1/2 protein were significantly elevated on postoperative days 1 and 5, and no significant differences were observed between day 7 or 12 after surgery and the naïve animals ([Figure 3A and B](#f3-jpr-10-763){ref-type="fig"}). Similarly, expression levels of p-p38 protein significantly increased on postoperative days 1--5, and no statistical differences were observed between day 7 or 12 after surgery and the naïve groups ([Figure 3A and C](#f3-jpr-10-763){ref-type="fig"}). These observations are consistent with the decrease in paw edema[@b7-jpr-10-763] and cellular infiltrates, as well as with the resolution of mechanical allodynia in WT mice, on days 7--12 following surgery. In MKP-3 KO mice, p-ERK-1/2 protein levels were significantly elevated on postoperative day 1 and remained consistently elevated all through postoperative day 12, relative to the protein levels in naïve MKP-3 KO mice ([Figure 3A and D](#f3-jpr-10-763){ref-type="fig"}). On the contrary, p-p38 expression patterns in MKP-3 KO mice after paw incision surgery resembled those of WT mice, whereas p-p38 levels increased on postoperative days 1--7; they subsided by day 12 after surgery, relative to naïve mice ([Figure 3A and E](#f3-jpr-10-763){ref-type="fig"}). MKP-3 and MKP-1 expression levels in the paw tissue of WT and MKP-3 KO mice --------------------------------------------------------------------------- In order to understand the temporal MAPK dynamics observed in the soft tissue of the operated paw, temporal expression of MKP-3 in WT mice and the expression of MKP-1, the other major regulator of MAPKs, were studied. In WT mice, MKP-3 expression levels in the paw tissue increased on postoperative days 5--7 ([Figure 4A and C](#f4-jpr-10-763){ref-type="fig"}), in accordance with the declining expression levels of p-ERK-1/2 and p-p38 at these time intervals ([Figure 3](#f3-jpr-10-763){ref-type="fig"}). Although MKP-1 expression in the WT paw tissue initially declined on postoperative day 1, MKP-1 protein levels increased 7 days after surgery ([Figure 4A and D](#f4-jpr-10-763){ref-type="fig"}), potentially contributing to the return of p-p38 and p-ERK-1/2 expression to basal levels on postoperative day 12. In MKP-3 KO mice, MKP-1 expression levels in the paw tissue increased on postoperative day 1, but returned to baseline levels on days 5--12 after surgery, relative to naïve mice ([Figure 4B and D](#f4-jpr-10-763){ref-type="fig"}). The difference in MKP-1 expression levels between WT and MKP-3 KO mice, on postoperative day 1, could be due to a compensatory mechanism in MKP-3 KO mice following paw incision. Interestingly, the high expression of p-p38 on postoperative day 1 is not in accordance with the elevation of MKP-1 protein levels in MKP-3 KO mice, despite p-p38 being the primary MKP-1 substrate. This suggests that other factors might contribute to the regulation of p-p38 at the early stages after surgery, for example, MKP-3. Effects of inhibiting ERK-1/2 phosphorylation in the paw tissue of MKP-3 KO mice -------------------------------------------------------------------------------- This study observed that p-ERK-1/2 expression levels remained elevated in the paw tissue of MKP-3 KO mice on postoperative day 12, contrary to the unchanged p-p38 levels, relative to naïve mice ([Figure 3D and E](#f3-jpr-10-763){ref-type="fig"}). To test a potential causative association between persistent p-ERK and persistent allodynia, the effect of inhibiting MEK was studied, which subsequently prevents ERK phosphorylation, in the paw tissue, on mechanical allodynia in MKP-3 KO mice, 12 days after surgery. It was also observed that local intradermal administration of a small molecule MEK inhibitor, PD98059 (25 nmol), partially, but significantly, reversed incision-induced persistent mechanical allodynia 1--2 h after injection, compared to the vehicle-treated group ([Figure 5A](#f5-jpr-10-763){ref-type="fig"}). To confirm the inhibitory effects of PD98059 over p-ERK-1/2, its expression levels after treatment were measured. A partial but significant reduction in p-ERK-1/2 expression was observed in the paw tissue 2 h after treatment, compared to the vehicle ([Figure 5B and C](#f5-jpr-10-763){ref-type="fig"}). These data strongly suggest a causative relationship between peripheral p-ERK expression and persistent postoperative hypersensitivity observed in MKP-3 KO mice. Discussion ========== This study demonstrates that lack of MKP-3 protein in mice leads to a long-lasting phosphorylation of MAPK ERK-1/2 in the peripheral tissue following a surgical intervention (paw incision), which results in persistent postoperative mechanical allodynia. The major novelty of this study is therefore the discovery of the important role that MKP-3 plays in the periphery in the transition from acute to chronic postoperative mechanical hypersensitivity. It has been previously shown that dysregulation of MKPs in the spinal cord is associated with chronic neuropathic pain, such that MKP-1 and MKP-3 induction in the spinal cord reduced mechanical allodynia in the rat model of neuropathic pain.[@b26-jpr-10-763],[@b27-jpr-10-763] In a more recent study, a mouse model of acute postoperative pain was used, that is, paw incision surgery, to demonstrate that MKP-3 also plays a role in the transition from acute to chronic postoperative mechanical allodynia through mechanisms in the spinal cord.[@b7-jpr-10-763] Using the same model, this study aimed to understand the role of MKP-3 in the mechanisms governing persistent mechanical hypersensitivity in the peripheral tissue, as a more clinically relevant, accessible, and feasible treatment site. This study confirms that lack of MKP-3 does not alter nociceptive behaviors in mice in response to mechanical stimulation under normal conditions. Moreover, the absence of MKP-3 does not influence the extent of acute mechanical allodynia (postoperative day 1). Nonetheless, lack of MKP-3 influences the transition from acute to chronic postoperative hypersensitivity. These data indicate that the function of MKP-3 in the peripheral injured tissue is minimal in nociceptive basal tone and the early stages of postoperative hypersensitivity. However, under normal conditions, peripheral MKP-3 seems to play a key role in the resolution phase of postsurgical mechanical allodynia by dephosphorylating p-ERK. This is supported by the temporal pattern of expression of peripheral MKP-3 and p-ERK following paw incision in WT animals. For example, expression levels of paw p-ERK decline as postoperative allodynia resolves in WT mice (postoperative days 7--12). This temporal pattern correlates well with the expression of MKP-3 in the injured tissue in these mice, which increases on postoperative days 5--7 and returns to baseline on day 12. This assumption is confirmed by the fact that the lack of MKP-3 (KO mice) results in a persistent p-ERK expression in the injured tissue. These findings show that MKP-3 controls p-ERK dephosphorylation in the incision area and thus promotes the resolution of postoperative hypersensitivity in acute postoperative pain. The current study also uncovers a differential substrate preference of MKP-3 between the periphery and the spinal cord in mice. Even though MKP-3 can modulate different MAPKs, MKP-3 preferentially dephosphorylates p-ERK in most systems.[@b22-jpr-10-763],[@b23-jpr-10-763] It has been previously demonstrated that MKP-3 in the spinal cord modulates both p-ERK and p-p38 to a similar extent,[@b7-jpr-10-763] and furthermore that MKP-3 preferentially dephosphorylates m-JNK (another major MAPK) in the rat brain-derived astrocytes.[@b24-jpr-10-763] Together, these data strongly suggest that the substrate preference of MKP-3 is determined by the relevance of a given MAPK in a given tissue. This is consistent, for example, with the prominent role of p-JNK in astrocytic function.[@b35-jpr-10-763],[@b36-jpr-10-763] Herein, this study observed that the temporal pattern of expression of peripheral p-p38 following paw incision was not altered in MKP-3 KO mice when compared to WT mice. In other words, p-p38 is dephosphorylated at the time of resolution of postoperative allodynia even in the absence of MKP-3. Therefore, the current results demonstrate that MKP-3 preserves its selectivity for p-ERK at the periphery, since the lack of MKP-3 (KO mice) resulted in persistent expression of p-ERK that paralleled the persistent mechanical hypersensitivity in these mice. This study confirms the functional association between the lack of MKP-3, persistent p-ERK expression in the peripheral injured tissue, and the maintenance of postoperative mechanical hypersensitivity using an MEK inhibitor. Accordingly, phosphorylation of ERK in the peripheral tissues during inflammation contributes to the pain-related behaviors in several rodent models.[@b37-jpr-10-763]--[@b39-jpr-10-763] Since the findings indicate that persistent p-ERK expression on postoperative day 12 is not associated with cell infiltration (resolved at this time point), these data suggest that this MAPK phosphorylation takes place in non-immune cells, namely nerve fibers or skin cells. It has been demonstrated that an elevation of p-ERK expression in epidermal and dermal peripheral nociceptors (C-fibers) is responsible for capsaicin-induced thermal and mechanical hypersensitivity.[@b40-jpr-10-763],[@b41-jpr-10-763] An increase in p-ERK-1/2 and p-p38 expression as well as a functional association between the two have been observed in keratinocytes of psoriatic lesions in humans[@b42-jpr-10-763]--[@b44-jpr-10-763] and in mouse keratinocytes under pruritogenic conditions.[@b45-jpr-10-763] Furthermore, p-ERK expression is enhanced in human keratinocytes in in vitro models of acute and chronic wounds.[@b46-jpr-10-763],[@b47-jpr-10-763] The phosphorylation of ERK can also be induced by pro-algesic factors such as epidermal growth factor.[@b47-jpr-10-763] Since ERK phosphorylation induces the production of pro-algesic factors by non-neuronal cells that sensitize peripheral nociceptors, which in turn are characterized by neuronal ERK phosphorylation,[@b6-jpr-10-763] it is likely that the results are explained by a pronociceptive feedback loop of skin cell and peripheral nociceptor ERK phosphorylation. The findings of this study indicate that, although some aspects of the neuroinflammatory process might resolve in a timely fashion, following a peripheral insult, they might not be indicative of a complete resolution of the peripheral mechanisms underlying the transition from acute to chronic pain. It is plausible that the magnitude of the early cellular infiltration contributes to the persistent elevation of p-ERK in the periphery (nociceptors and/or skin cells) and subsequently to the transition from acute to chronic pain in MKP-3 KO mice. Immunohistochemistry staining to determine the specific repertoire of immune cells might better elucidate the role of immune cell infiltration in the resolution of postoperative pain. Whether the MAPK/MKP signaling pathway pattern could be used as a biomarker to identify potential patients at risk of developing chronic postoperative pain, or those already suffering from it, remains to be studied. It is certainly a testable hypothesis (minimal required invasiveness for assessment) that deserves further study. These studies in human tissue would move forward the field to set a rationale for potential studies with already available p-ERK inhibitors (some already evaluated in clinical trials for osteoarthritis or cancer).[@b48-jpr-10-763],[@b49-jpr-10-763] This study also suggests that p-ERK, even though it promotes a cellular pro-inflammatory phenotype, does not seem to play a role in cellular infiltration. On the other hand, p-p38 seems to be functionally associated with the initiation and resolution of cell infiltration, because of the similar pattern of appearance of these phenomena in both WT and MKP-3 KO groups. These hypotheses are in line with previous observations, in which the intraarticular inhibition of p-p38 or p-JNK, but not p-ERK, reduced cellular infiltration into the synovium, in a carrageenan-induced arthritis rat model.[@b50-jpr-10-763] Therefore, this study does not rule out the potential role of peripheral p-p38 in the transition from acute to chronic pain. Similarly, the role of MKP-1 (which preferentially dephosphorylates p-p38) in the peripheral mechanisms of chronic pain cannot be ruled out. Elevated MKP-1 expression in MKP-3 KO mice on postoperative day 1 could suggest a compensatory mechanism triggered by surgical incision and the initial events of postoperative hypersensitivity. The high levels of p-p38 expression in the presence of elevated MKP-1expression in the early stages of acute postoperative pain in MKP-3 KO animals suggest that other factors, besides MKP-1, might be involved in p38 regulation. Based on the observation that p-p38 dephosphorylation associates with the resolution of inflammation, it is speculated that inflammatory mediators could be involved in p38 modulation. The role of MAPKs, such as p-ERK, in chronic pain models is well recognized.[@b6-jpr-10-763],[@b51-jpr-10-763] Activation of MAPKs is required for effective production of pro-algesic factors that sensitize peripheral nociceptors and subsequently perpetuate MAPK phosphorylation, leading to central sensitization and establishment of chronic pain.[@b27-jpr-10-763] The induction of an effective MKP activity could break this cycle and interrupt the molecular signals driving the transition to chronic pain. It has been demonstrated that novel approaches could be utilized to develop gene therapies, for example, induction of MKP expression to modulate cell phenotype and treat chronic pain.[@b27-jpr-10-763] The induction of MKPs can also be achieved by pharmacological approaches. For example, glucocorticoids[@b52-jpr-10-763] or cannabinoids[@b26-jpr-10-763] produce anti-inflammatory effects by inducing MKP-1 and/or MKP-3. Curiously, the use of glucocorticoids in surgical patients seems to reduce the use of analgesics in postoperative pain.[@b53-jpr-10-763],[@b54-jpr-10-763] Although this study focused on peripheral MKP-3 and MAPK molecular changes in the context of postoperative pain resolution, it recognizes that multiple factors, such as genetics,[@b55-jpr-10-763],[@b56-jpr-10-763] preoperative pain and physiological states,[@b57-jpr-10-763] psychological vulnerability,[@b58-jpr-10-763] and surgical techniques,[@b59-jpr-10-763] contribute to post-surgical pain and its persistence. It warrants further investigation on the potential association of these factors with the abovementioned pathways. Conclusion ========== This study revealed that MKP-3/p-ERK-1/2 signaling in the periphery greatly influences the transition from acute to chronic postoperative pain. The current data further support the notion that biochemical and regulatory properties of MKP-3 render it a promising target for preventing and treating chronic postoperative pain. Moreover, studies by the authors uncover the possibility of intervening at the site of surgery, which provides a more feasible therapeutic site than the central nervous system.[@b60-jpr-10-763] This study was supported by the Rita Allen Foundation & American Pain Society 2011 Pain Grant (EAR-S), Hitchcock Foundation Award 2011--2012 (EAR-S), and James O Freedman Undergraduate Presidential Scholars Program (BSL). The authors thank Ms Beth Wilkerson, Presbyterian College School of Pharmacy, for edits and revision. **Disclosure** The authors report no conflicts of interest in this work. ![50% paw withdrawal threshold (**A**) in naïve mice and mice that underwent paw incision surgery, measured on days 1, 5, 7, and 12 in wild-type (WT, open circles \[○\] n=6) and in MKP-3 KO mice (closed circles \[●\] n=6). Data were analyzed using two-way ANOVA + Bonferroni post hoc analysis. \**P*\<0.05, WT vs MKP-3 KO. Representative Western blot images (**B**) of hind paw plantar tissue MKP-3 and MKP-1 (β-actin was used as the loading control) in WT and MKP-3 KO naïve mice. Data are expressed as mean ± SEM.\ **Abbreviations:** WT, wild type; KO, knockout; ANOVA, analysis of variance; MKP-3, Mitogen-activated protein kinase phosphatase-3; SEM, standard error of mean.](jpr-10-763Fig1){#f1-jpr-10-763} {#f2-jpr-10-763} {#f3-jpr-10-763} {#f4-jpr-10-763} {#f5-jpr-10-763} [^1]: These authors contributed equally to this work.

Professor John M. Fitzpatrick (1948--2014) contributed to the manuscript before passing away. Introduction {#bju13123-sec-0001} ============ The development and function of the prostate gland is dependent on androgen regulation via the androgen receptor (AR) signalling pathway. The AR is a 110 kDa member of the steroid‐receptor family and contains four modular domains: the ligand‐binding domain (LBD), the hinge region, the DNA‐binding domain (DBD) and the N‐terminal domain (NTD) (Fig. [1](#bju13123-fig-0001){ref-type="fig"} [1](#bju13123-bib-0001){ref-type="ref"}, [2](#bju13123-bib-0002){ref-type="ref"}, [3](#bju13123-bib-0003){ref-type="ref"}) [1](#bju13123-bib-0001){ref-type="ref"}. The C‐terminal LBD mediates activation of transcription by the binding of ligand [1](#bju13123-bib-0001){ref-type="ref"}. The DBD facilitates binding to DNA and the hinge region is important for nuclear localisation [1](#bju13123-bib-0001){ref-type="ref"}. The NTD is responsible for regulation of transcription [1](#bju13123-bib-0001){ref-type="ref"}. {ref-type="ref"}, [2](#bju13123-bib-0002){ref-type="ref"}, [3](#bju13123-bib-0003){ref-type="ref"}. Adapted with permission from Hu et al. [2](#bju13123-bib-0002){ref-type="ref"} and Quigley et al. [3](#bju13123-bib-0003){ref-type="ref"}. NLS, nuclear localisation signal.](BJU-117-215-g001){#bju13123-fig-0001} Within a normal prostate, the AR \[regulated by testosterone and dihydrotestosterone (DHT)\], has a homeostatic function to balance the rate of cell proliferation with the rate of apoptosis. In prostate epithelial and stromal cells, testosterone is converted into the more active form, DHT, by 5α‐reductase [2](#bju13123-bib-0002){ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}. Testosterone or DHT binds to the LBD of the AR, causing a conformational change and activation of the receptor. This leads to homodimer formation and translocation of the AR from the cytoplasm to the nucleus. Within the nucleus, the AR binds to androgen response elements of DNA, recruiting co‐factors (co‐activators or co‐repressors) that regulate transcription of target genes, e.g. PSA. In malignant prostate cells, the AR signalling pathway drives uncontrolled growth and the balance between the rate of cell proliferation and the rate of apoptosis is lost (Fig. [2](#bju13123-fig-0002){ref-type="fig"}) [4](#bju13123-bib-0004){ref-type="ref"}. The AR signalling pathway plays a key role in all phases of prostate cancer, from disease initiation to disease progression, including metastatic transformation and spread [4](#bju13123-bib-0004){ref-type="ref"}, [5](#bju13123-bib-0005){ref-type="ref"}. {ref-type="ref"}. DHEA, dehydroepiandesterone; T, testosterone.](BJU-117-215-g002){#bju13123-fig-0002} Different stages within the prostate cancer disease continuum can be clinically defined by the presence or absence of detectable metastases and whether testosterone concentrations are in the castrate range [6](#bju13123-bib-0006){ref-type="ref"}. Treatment of prostate cancer, at any stage in the disease continuum, is determined by tumour characteristics, PSA level, extent of tumour spread, estimated life‐expectancy of the patient, whether or not symptoms are present, and progression on previous treatments [7](#bju13123-bib-0007){ref-type="ref"}. For most patients with metastatic prostate cancer, androgen‐deprivation therapy (ADT) with LHRH agonists or antagonists, or orchidectomy, constitutes first‐line therapy [7](#bju13123-bib-0007){ref-type="ref"}, [8](#bju13123-bib-0008){ref-type="ref"}. Medical ADT with LHRH agonists or antagonists can involve monotherapy or can be combined with antiandrogens (e.g. bicalutamide) that block the effect of any residual testosterone. This is known as combined androgen blockade. ADT strategies can produce a dramatic improvement of symptoms, but invariably prostate cancer stops responding to androgen suppression and progresses. The average time to development of castration‐resistant prostate cancer (CRPC) after commencement of ADT is 2--3 years [9](#bju13123-bib-0009){ref-type="ref"}, [10](#bju13123-bib-0010){ref-type="ref"}. While this state has previously been referred to as 'hormone refractory' or 'androgen independent', advancements in the understanding of disease progression and the continued involvement of the AR have led to this state being termed 'castration‐resistant prostate cancer' [10](#bju13123-bib-0010){ref-type="ref"}. CRPC is defined by serial rises in PSA values despite castrate levels of testosterone, and/or evidence of disease progression on imaging studies [11](#bju13123-bib-0011){ref-type="ref"}. Compared with hormone‐sensitive prostate cancer, the prognosis for patients with CRPC is poor and survival is reduced [12](#bju13123-bib-0012){ref-type="ref"}. In clinical trials, the median survival of patients with CRPC varies between 9 and 30 months [12](#bju13123-bib-0012){ref-type="ref"}. This review highlights the continued importance of AR signalling in the treatment of patients with CRPC and discusses the clinical benefits associated with inhibition of the AR signalling pathway maintained even in later stages of the disease. Evidence Acquisition {#bju13123-sec-0002} ==================== A literature search was conducted, using PubMed and congress abstracts, to identify articles relating to the role of AR signalling in CRPC and therapies that inhibit the AR signalling pathway. Key words included *AR signalling, mCRPC, abiraterone* and *enzalutamide*. Evidence Synthesis {#bju13123-sec-0003} ================== Continued AR Signalling in CRPC {#bju13123-sec-0004} ------------------------------- Significant progress has been made in the understanding of the underlying cancer biology of CRPC. The AR signalling pathway is reactivated in CRPC and is a key driver of tumour progression in this condition. This is demonstrated by levels of PSA, which is a protein known to be regulated by AR signalling, which is present in high levels in most patients with CRPC [13](#bju13123-bib-0013){ref-type="ref"}. In addition, many genes known to be regulated by the AR signalling pathway are also expressed in CRPC [13](#bju13123-bib-0013){ref-type="ref"}. Various molecular mechanisms have been proposed to explain how prostate tumour cells continue to use the AR signalling pathway for growth despite castrate levels of testosterone (Fig. [3](#bju13123-fig-0003){ref-type="fig"}) [2](#bju13123-bib-0002){ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}, [13](#bju13123-bib-0013){ref-type="ref"}, [14](#bju13123-bib-0014){ref-type="ref"}, [15](#bju13123-bib-0015){ref-type="ref"}, [16](#bju13123-bib-0016){ref-type="ref"}. The most commonly reported mechanisms of continued AR signalling are extragonadal steroidogenesis and AR gene amplification. {ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}, [16](#bju13123-bib-0016){ref-type="ref"}, [33](#bju13123-bib-0033){ref-type="ref"}. T, testosterone.](BJU-117-215-g003){#bju13123-fig-0003} Androgen synthesis by the adrenal glands is known to continue despite medical or surgical castration [2](#bju13123-bib-0002){ref-type="ref"}. Intratumoral testosterone/DHT production is also implicated in the pathogenesis of CRPC. In addition, the synthesis of androgen from circulating precursors occurs within prostate tumour cells, reactivating AR signalling in CRPC despite serum levels of testosterone in the castrate range [5](#bju13123-bib-0005){ref-type="ref"}. The incidence of AR gene amplification in untreated primary prostate cancer is low [14](#bju13123-bib-0014){ref-type="ref"}. In comparison, studies have shown that 50--85% of tumours in CRPC have an increased AR gene copy number causing upregulation of AR gene expression. This leads to overexpression of AR in the cytoplasm of CRPC cells and consequent sensitisation of the tumour to low levels of androgens, which confers a survival and growth advantage upon tumours [14](#bju13123-bib-0014){ref-type="ref"}, [15](#bju13123-bib-0015){ref-type="ref"}, [17](#bju13123-bib-0017){ref-type="ref"}. This suggests that AR gene amplification and AR overexpression contribute to the progression of prostate cancer to a castration‐resistant state [14](#bju13123-bib-0014){ref-type="ref"}, [18](#bju13123-bib-0018){ref-type="ref"}. AR overexpression also appears to promote the conversion of first‐generation antiandrogens, e.g. bicalutamide, to full agonists [14](#bju13123-bib-0014){ref-type="ref"}. Raised levels of AR splice variants have also been found in some CRPC tissue specimens [16](#bju13123-bib-0016){ref-type="ref"}. Transcriptional active splice variants have also been detected in normal prostate tissue [16](#bju13123-bib-0016){ref-type="ref"}. AR splice variants are a result of alternative splicing of the AR primary gene transcript. Up to seven different AR splice variants that lack the LBD have been identified in a range of human prostate tissues [16](#bju13123-bib-0016){ref-type="ref"}. In particular, expression levels of AR‐V7 are found to be elevated 20‐fold in CRPC when compared with hormone‐naïve prostate cancer [16](#bju13123-bib-0016){ref-type="ref"}, [19](#bju13123-bib-0019){ref-type="ref"}. Although AR splice variants are associated with the progression of prostate cancer, the exact role they play in CRPC is complex and not yet fully understood. Some AR splice variants that lack the LBD have been shown to be constitutively active, suggesting that AR signalling could occur in the complete absence of ligand binding [16](#bju13123-bib-0016){ref-type="ref"}, [19](#bju13123-bib-0019){ref-type="ref"}. This mechanism of continued AR signalling may drive resistance to therapies that rely on binding to the LBD for their activity. AR splice variants are more common when residual androgen levels are low, suggesting that AR splice variants may arise as a result of ADT [20](#bju13123-bib-0020){ref-type="ref"}, [21](#bju13123-bib-0021){ref-type="ref"}. Constitutively active AR splice variants are also associated with increased expression of N‐cadherin, a protein that plays an important role in cell adhesion, which in turn induces epithelial--mesenchymal transition (EMT) [22](#bju13123-bib-0022){ref-type="ref"}. EMT is a physiological process whereby epithelial cells are converted into mesenchymal cells during embryonic development as part of organ development [22](#bju13123-bib-0022){ref-type="ref"}. After embryogenesis, it was assumed that EMT is 'switched off'; however, it is now thought that EMT is reactivated by malignant cells [22](#bju13123-bib-0022){ref-type="ref"}. This transformation may be responsible for the formation of metastatic tumours via dissemination of cells from the primary tumour site to tissue sites such as the liver, lungs or bone marrow [22](#bju13123-bib-0022){ref-type="ref"}. AR gene mutations constitute another genetic alteration involved in the progression to CRPC. The incidence of AR gene mutations has been found to increase with cancer stage, with ≈10% of CRPC tumours found to have such mutations [2](#bju13123-bib-0002){ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}. Most of the AR mutations characterised to date occur in the LBD and may alter ligand binding. This may allow activation of the AR signalling pathway by alternative ligands, such as some first‐generation antiandrogens and steroids [23](#bju13123-bib-0023){ref-type="ref"}. The AR LBD mutations may provide a mechanistic explanation for the development of resistance to antiandrogen therapy [4](#bju13123-bib-0004){ref-type="ref"}. Another proposed biological mechanism for the reactivation of AR signalling in CRPC is cross‐talk between signal transduction pathways [2](#bju13123-bib-0002){ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}. Two or more signal transduction pathways often cross‐talk via the activation or inhibition of downstream signalling molecules or target genes common to all pathways [2](#bju13123-bib-0002){ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}. Cross‐talk between growth factor signalling pathways may influence the phosphorylation of AR and AR co‐regulators to stimulate AR activity [2](#bju13123-bib-0002){ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}. Growth factor kinase signalling pathways, including phosphoinositide‐3‐kinase (PI3K), protein kinase B (AKT), extracellular signal‐regulated kinase (ERK) and mammalian target of rapamycin (mTOR), have been shown to stimulate AR target gene expression in the absence of an AR ligand [2](#bju13123-bib-0002){ref-type="ref"}, [4](#bju13123-bib-0004){ref-type="ref"}. Studies have shown that the PI3K/AKT signalling pathway is upregulated in 30--50% of prostate cancer cases, with the associated loss of phosphatase and tensin homolog (PTEN) function shown to enhance CRPC disease development [24](#bju13123-bib-0024){ref-type="ref"}. The loss of PTEN function increases the expression and signalling of the pro‐inflammatory chemokine ligand 8 (CXCL8) in prostate cancer cells. CXCL8 signalling is associated with the activation of additional signalling pathways, leading to upregulation of anti‐apoptotic proteins and, ultimately, the survival of cancer cells [25](#bju13123-bib-0025){ref-type="ref"}. The AR Signalling Pathway and Current Treatment Options in mCRPC {#bju13123-sec-0005} ---------------------------------------------------------------- As our understanding of the biological mechanisms underlying disease progression expands, the treatment landscape in CRPC has been evolving. Cytotoxic chemotherapy agents, such as docetaxel and cabazitaxel, have shown survival benefits in patients with mCRPC [26](#bju13123-bib-0026){ref-type="ref"}, [27](#bju13123-bib-0027){ref-type="ref"}, [28](#bju13123-bib-0028){ref-type="ref"}. Taxanes (docetaxel and cabazitaxel) cause apoptosis of prostate cancer cells by stabilising their microtubule network [28](#bju13123-bib-0028){ref-type="ref"}, [29](#bju13123-bib-0029){ref-type="ref"}. The stabilisation of microtubules inhibits their disassembly, preventing cell division and promoting cell‐cycle arrest [28](#bju13123-bib-0028){ref-type="ref"}, [29](#bju13123-bib-0029){ref-type="ref"}. An indirect consequence of the stabilisation of microtubules by taxanes is inhibition of AR nuclear translocation [30](#bju13123-bib-0030){ref-type="ref"}. Docetaxel was the first agent to show improvement in overall survival (OS), in addition to palliative benefits in patients with mCRPC who had progressed on ADT [28](#bju13123-bib-0028){ref-type="ref"}. Cabazitaxel has shown an OS benefit in patients with mCRPC whose disease has progressed during or after treatment with docetaxel [26](#bju13123-bib-0026){ref-type="ref"}. Current understanding of prostate cancer has identified the AR signalling pathway as a logical target for the treatment of CRPC. Available therapies that inhibit the AR signalling pathway include AR blockers (bicalutamide, nilutamide and flutamide), androgen biosynthesis inhibitors (ketoconazole and abiraterone) and AR signalling inhibitors (enzalutamide). Other available treatments with different mechanisms of actions include sipuleucel‐T, an autologous cellular immunotherapy and radium‐223, a radiopharmaceutical that targets bone metastases. Agents that do not affect the AR signalling pathway are not covered in the present review. ### AR blockers {#bju13123-sec-0006} Traditionally, antiandrogens (bicalutamide, nilutamide or flutamide) have been added to ADT at PSA progression, with the objective of achieving a more complete androgen blockade. Mainly used in advanced prostate cancer, these agents provide modest survival benefits of ≈3% improvement in survival at 5 years [31](#bju13123-bib-0031){ref-type="ref"}. After long‐term use of antiandrogens combined with ADT, some patients will respond to the selective discontinuation of the antiandrogen. This is known as 'antiandrogen withdrawal syndrome', which is characterised by decreasing PSA levels and regression of the tumour on discontinuation of the antiandrogen [32](#bju13123-bib-0032){ref-type="ref"}, indicating that antiandrogens can serve as AR agonists under specific circumstances. AR overexpression occurs in most cases of CRPC and is perhaps the most common mechanism whereby antiandrogens gain agonist activity [14](#bju13123-bib-0014){ref-type="ref"}, [17](#bju13123-bib-0017){ref-type="ref"}. *In vitro* studies of AR overexpressing cells indicate that first‐generation antiandrogens induce changes in the AR that continue to allow nuclear translocation, DNA binding and co‐activator recruitment at variable efficiencies [14](#bju13123-bib-0014){ref-type="ref"}, [33](#bju13123-bib-0033){ref-type="ref"}. Up‐regulation of co‐activators may also allow activation of wild‐type AR by some antiandrogens, thus demonstrating agonist activity [4](#bju13123-bib-0004){ref-type="ref"}, [34](#bju13123-bib-0034){ref-type="ref"}. In the presence of AR gene mutations, some antiandrogens may confer agonist activity [4](#bju13123-bib-0004){ref-type="ref"}. ### Androgen biosynthesis inhibitors {#bju13123-sec-0007} The androgen biosynthesis inhibitor ketoconazole has been suggested to have limited clinical efficacy in CRPC [35](#bju13123-bib-0035){ref-type="ref"}. However, its widespread use in this condition is now restricted due to significant side‐effects and the need to co‐administer with steroids. Abiraterone irreversibly and selectively blocks cytochrome P450 17A1 (CYP17A1), indirectly inhibiting production of androgens from the testes, adrenal glands and from the prostate tumour itself [36](#bju13123-bib-0036){ref-type="ref"}. Elevated mineralocorticoid levels due to CYP17 blockade by abiraterone require co‐administration with prednisone to suppress adrenocorticotrophic hormone and reduce the adverse events (AEs) associated with mineralocorticoid excess. Abiraterone has shown efficacy over placebo in both chemotherapy‐naïve and post‐docetaxel patients with mCRPC [37](#bju13123-bib-0037){ref-type="ref"}, [38](#bju13123-bib-0038){ref-type="ref"}. In the chemotherapy‐naïve setting, the AEs of fatigue, arthralgia, peripheral oedema, grade 3 or 4 mineralocorticoid‐related AEs, and abnormalities on liver‐function testing were reported more frequently in the abiraterone‐prednisone group than in the prednisone‐alone group [38](#bju13123-bib-0038){ref-type="ref"}. A similar tolerability profile was reported for abiraterone‐prednisone in the post‐docetaxel study [37](#bju13123-bib-0037){ref-type="ref"}. ### AR signalling inhibitors in mCRPC {#bju13123-sec-0008} The AR signalling pathway remains implicated throughout the prostate cancer disease continuum and reactivation of the AR signalling pathway is thought to be a key driver of CRPC progression [5](#bju13123-bib-0005){ref-type="ref"}. The underlying molecular mechanisms of CRPC progression are considered to affect tumour growth and some have been shown to potentiate agonist activity of first‐generation antiandrogens, such as the AR blocker bicalutamide [14](#bju13123-bib-0014){ref-type="ref"}, [33](#bju13123-bib-0033){ref-type="ref"}. The AR signalling pathway is a logical target for novel therapies in CRPC. AR blockers, e.g. bicalutamide, have provided the starting point for development of AR signalling inhibitors. These therapies would need to be potent AR inhibitors capable of avoiding significant agonist activity. Enzalutamide is the first approved AR signalling inhibitor (also described as an AR inhibitor), with a novel mechanism of action, that distinguishes it from both androgen biosynthesis inhibitors (e.g. abiraterone) and first‐generation antiandrogens (e.g. bicalutamide) [39](#bju13123-bib-0039){ref-type="ref"}, [40](#bju13123-bib-0040){ref-type="ref"}. Enzalutamide was rationally designed by conducting a chemical synthesis programme to identify novel chemical structures that would be potent AR inhibitors in CRPC without significant agonist activity [41](#bju13123-bib-0041){ref-type="ref"}. Based on structure‐activity relationships and optimisation of half‐life and oral bioavailability, enzalutamide was selected for further preclinical and clinical studies [41](#bju13123-bib-0041){ref-type="ref"}, [42](#bju13123-bib-0042){ref-type="ref"}. This approach to drug discovery has provided a new model for the rational design and development of AR signalling inhibitors. Other AR signalling inhibitors currently being investigated include ODM‐201 and ARN‐509. ODM‐201 has been shown to inhibit AR nuclear translocation in preclinical studies and 86% of patients had a ≥50% PSA level decrease with a 1 400‐mg dose of ODM‐201 in a Phase I/II study [43](#bju13123-bib-0043){ref-type="ref"}, [44](#bju13123-bib-0044){ref-type="ref"}. ARN‐509 targets the AR signalling pathway by binding to AR and a Phase I study has demonstrated a ≥50% PSA level decrease in 46.7% of patients [45](#bju13123-bib-0045){ref-type="ref"}, [46](#bju13123-bib-0046){ref-type="ref"}. The Phase III SPARTAN trial will evaluate the efficacy and safety of ARN‐509 in patients with non‐metastatic CRPC at high risk of progression [47](#bju13123-bib-0047){ref-type="ref"}. Direct AR Inhibition with Enzalutamide {#bju13123-sec-0009} ====================================== Extensive *in vitro* studies have shown that enzalutamide targets the AR signalling pathway at three key stages, exerting its effect by blocking binding of androgens to AR, by inhibiting nuclear translocation of activated AR, and by impairing binding of activated AR with DNA (Fig. [4](#bju13123-fig-0004){ref-type="fig"}) [16](#bju13123-bib-0016){ref-type="ref"}, [33](#bju13123-bib-0033){ref-type="ref"}. In preclinical CRPC models, enzalutamide has been shown to competitively inhibit androgen binding to the receptor by binding with five‐ to eight‐times higher affinity than bicalutamide, but without partial agonist activity [33](#bju13123-bib-0033){ref-type="ref"}. Molecular modelling suggests that enzalutamide sits in the ligand binding domain in a manner that is distinct from bicalutamide and offers mechanism for the partial agonist activity of bicalutamide [48](#bju13123-bib-0048){ref-type="ref"}. AR localisation studies have shown that enzalutamide inhibits nuclear translocation of the AR and impairs DNA binding and activation in preclinical CRPC models [33](#bju13123-bib-0033){ref-type="ref"}. {ref-type="ref"}, [33](#bju13123-bib-0033){ref-type="ref"}.](BJU-117-215-g004){#bju13123-fig-0004} Enzalutamide Clinical Studies {#bju13123-sec-0010} ----------------------------- The theoretical benefit of enzalutamide, designed to interrupt the androgen signalling mechanism in prostate cancer cells, has been shown to translate into clinical responses in patients with mCRPC [49](#bju13123-bib-0049){ref-type="ref"}, [50](#bju13123-bib-0050){ref-type="ref"}. A Phase I/II, open‐label, uncontrolled, dose‐escalation study in patients with progressive CRPC with or without detectable metastases, demonstrated the antitumour activity of enzalutamide, both in patients who had previously received chemotherapy and those who had not [51](#bju13123-bib-0051){ref-type="ref"}. At the end of the study, maximal PSA level decrease did not differ significantly by prior chemotherapy status [51](#bju13123-bib-0051){ref-type="ref"}. The clinical efficacy and safety of enzalutamide in patients with mCRPC has been shown in two Phase III randomised clinical trials, thus further supporting the AR signalling pathway as a therapeutic target in mCRPC [49](#bju13123-bib-0049){ref-type="ref"}, [50](#bju13123-bib-0050){ref-type="ref"}. In the Phase III AFFIRM trial, 1 199 men with mCRPC who had received prior docetaxel‐based chemotherapy were randomly assigned to receive either enzalutamide (160 mg as a single oral dose, once‐daily) or placebo in a 2:1 ratio [50](#bju13123-bib-0050){ref-type="ref"}. Concomitant therapy with steroids was permitted but not required [50](#bju13123-bib-0050){ref-type="ref"}. At the time of the interim analysis (after 520 deaths), a significant benefit in the primary endpoint of OS was seen, with a 37% reduction in risk of death favouring enzalutamide compared with placebo (hazard ration \[HR\] = 0.63, 95% CI 0.53--0.75; *P* \< 0.001, median OS 18.4 vs 13.6 months, respectively). This effect was consistent across all patient subgroups analysed. Enzalutamide was associated with significant improvements in all secondary efficacy endpoints compared with placebo, including radiographic progression‐free survival (rPFS), time to PSA progression, PSA response rate, and soft‐tissue response rate [50](#bju13123-bib-0050){ref-type="ref"}. Enzalutamide was also associated with significantly better patient‐related outcomes compared with placebo (i.e. health‐related quality of life, time to first skeletal‐related event and pain) [50](#bju13123-bib-0050){ref-type="ref"}, [52](#bju13123-bib-0052){ref-type="ref"}. Additional secondary analyses have shown consistent benefits in OS, rPFS and time to PSA progression with enzalutamide in elderly (≥75 years) and younger (\<75 years) patients [53](#bju13123-bib-0053){ref-type="ref"}, in patients with different levels of disease severity assessed by baseline PSA levels, and in both North American‐ and European‐treated patients [54](#bju13123-bib-0054){ref-type="ref"}, [55](#bju13123-bib-0055){ref-type="ref"}. In the Phase III PREVAIL trial, 1 717 chemotherapy‐naïve patients with progressive mCRPC who had failed on ADT were randomly assigned to receive either enzalutamide (160 mg as a single oral dose, once‐daily) or placebo in a 1:1 ratio [49](#bju13123-bib-0049){ref-type="ref"}. The patient population included ≈12% of patients with visceral metastases, typically not included in CRPC trials. At the time of the planned interim analysis (after 540 deaths) enzalutamide treatment resulted in significant benefits in both co‐primary endpoints vs placebo, significantly reducing the risk of radiographic progression by 81% (HR 0.19, 95% CI 0.15--0.23; *P* \< 0.001) or death by 29% (HR 0.71, 95% CI 0.60--0.84; *P* \< 0.001). The rPFS and OS benefits were seen across all subgroups including age, baseline pain intensity, geographical region, and type of disease progression at entry [49](#bju13123-bib-0049){ref-type="ref"}. In addition, enzalutamide was associated with significant benefits over placebo across all secondary outcome measures (i.e. time to chemotherapy initiation, time to PSA progression, reduction in PSA level, objective soft‐tissue response, and quality of life maintenance) [49](#bju13123-bib-0049){ref-type="ref"}. In both the AFFIRM and PREVAIL studies, enzalutamide was generally well tolerated [49](#bju13123-bib-0049){ref-type="ref"}, [50](#bju13123-bib-0050){ref-type="ref"}. In AFFIRM, the most common AEs reported more frequently in the enzalutamide group compared with placebo included fatigue (34% vs 29%), diarrhoea (21% vs 18%), and hot flushes (20% vs 10%) [50](#bju13123-bib-0050){ref-type="ref"}. The rate of discontinuations due to AEs was low in both the enzalutamide and placebo groups (8% vs 10%) [50](#bju13123-bib-0050){ref-type="ref"}. Of 800 patients treated with enzalutamide, five (0.6%) had a seizure, whereas no seizures occurred in patients receiving placebo [50](#bju13123-bib-0050){ref-type="ref"}. One additional patient was identified with an event termed syncope with features suggestive of a seizure, and another patient was diagnosed with a seizure after the interim analysis cut‐off date [52](#bju13123-bib-0052){ref-type="ref"}. Enzalutamide was not associated with liver or cardiac toxicity [50](#bju13123-bib-0050){ref-type="ref"}. In the PREVAIL trial in patients with chemotherapy‐naïve mCRPC, enzalutamide showed a generally favourable tolerability profile, despite the reporting period for enzalutamide being more than twice that for placebo [49](#bju13123-bib-0049){ref-type="ref"}. AEs that occurred in ≥20% of patients receiving enzalutamide at a rate that was at least 2% higher than that in the placebo group were fatigue (36% vs 26%), back pain (27% vs 22%), constipation (22% vs 17%), and arthralgia (20% vs 16%). Cardiac AEs were seen in 10% of patients receiving enzalutamide compared with 8% of placebo patients. Hypertension was more commonly seen in the enzalutamide group than in the placebo group (13% vs 4%). Two seizures were reported during the study, one in each treatment arm. The seizure in the enzalutamide treatment arm occurred after the data cut‐off date. Sequencing of Hormonal Therapies {#bju13123-sec-0011} ================================ While the number of treatments available across the prostate cancer disease continuum that target or have an effect on the AR continues to grow, treatment decisions in the management of patients with mCRPC are becoming more complex. There have been several retrospective reports in the literature, with some studies involving patients initially included in early access or compassionate use programmes prior to commercial availability of enzalutamide in various countries. Typically they were based on a few heavily pre‐treated patients, and none of the studies were designed to specifically investigate sequential treatment in patients with mCRPC in the post‐chemotherapy setting (Table [1](#bju13123-tbl-0001){ref-type="table-wrap"}) [56](#bju13123-bib-0056){ref-type="ref"}, [57](#bju13123-bib-0057){ref-type="ref"}, [58](#bju13123-bib-0058){ref-type="ref"}, [59](#bju13123-bib-0059){ref-type="ref"}, [60](#bju13123-bib-0060){ref-type="ref"}, [61](#bju13123-bib-0061){ref-type="ref"}, [62](#bju13123-bib-0062){ref-type="ref"}, [63](#bju13123-bib-0063){ref-type="ref"}, [64](#bju13123-bib-0064){ref-type="ref"}, [65](#bju13123-bib-0065){ref-type="ref"}, [66](#bju13123-bib-0066){ref-type="ref"}, [67](#bju13123-bib-0067){ref-type="ref"}, [68](#bju13123-bib-0068){ref-type="ref"}, [69](#bju13123-bib-0069){ref-type="ref"}, [70](#bju13123-bib-0070){ref-type="ref"}, [71](#bju13123-bib-0071){ref-type="ref"}. In this population of heavily pre‐treated patients with advanced mCRPC, the effectiveness of enzalutamide appeared to be attenuated by prior treatment with abiraterone, with a generally smaller proportion of patients experiencing ≥50% decline in PSA level with enzalutamide after prior treatment with abiraterone (Table [1](#bju13123-tbl-0001){ref-type="table-wrap"}) compared with patients treated with enzalutamide post‐docetaxel in the AFFIRM Phase III study (54%) [50](#bju13123-bib-0050){ref-type="ref"}. None of the reports identified any reliable predictors of response to enzalutamide. Similarly, subsequent treatment with abiraterone in patients who progressed on enzalutamide treatment also only experienced modest responses (Table [1](#bju13123-tbl-0001){ref-type="table-wrap"}). However, significant cabazitaxel activity was reported in patients with mCRPC progressing after abiraterone or enzalutamide [71](#bju13123-bib-0071){ref-type="ref"}. ###### Retrospective studies of hormonal sequencing with enzalutamide in patients with mCRPC Reference Sequence of hormonal therapies Type of study Number of patients ≥50% decline in PSA with second treatment, % Key study notes -------------------------------------------------------------- --------------------------------------------------------------------------------------------- ----------------------------- -------------------- ----------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Docetaxel, abiraterone → enzalutamide Badrising et al. [56](#bju13123-bib-0056){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Retrospective 61 21 Enzalutamide had modest clinical activity in patients with mCRPC who previously received docetaxel and abiraterone; PSA response to docetaxel and abiraterone did not predict PSA response to enzalutamide Bournakis et al. [57](#bju13123-bib-0057){ref-type="ref"} Abiraterone and/or orterenol (25 of 35 patients) prior to enzalutamide Patient access programme 35 40 (six out of 15 evaluable patients) Enzalutamide benefited a subset of patients resistant to prior abiraterone treatment Scholz et al. [58](#bju13123-bib-0058){ref-type="ref"} Docetaxel, abiraterone prior to enzalutamide Retrospective 66 NR; 29% had \>30% decline in PSA Enzalutamide had activity in a heavily pre‐treated population of men resistant to abiraterone and docetaxel Caffo et al. [59](#bju13123-bib-0059){ref-type="ref"} Docetaxel prior to second‐ and third‐line agent (abiraterone, cabazitaxel, or enzalutamide) Retrospective 260 20 (70 out of 260 patients received enzalutamide as third‐line agent) No difference in clinical outcomes of abiraterone, cabazitaxel, or enzalutamide as third‐line treatment was seen, regardless of the previous treatment Schrader et al. [60](#bju13123-bib-0060){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Retrospective 35 29 Patients progressing after abiraterone achieved only a modest response rate with enzalutamide; a small but significant number of patients showed significant benefit from sequential treatment Bianchini et al. [61](#bju13123-bib-0061){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Retrospective 39 13 Limited activity of enzalutamide was reported in the post‐docetaxel and post‐abiraterone patient population Thomsen et al. [62](#bju13123-bib-0062){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Retrospective 24 17 Previous abiraterone therapy was associated with a less marked reduction in PSA level following enzalutamide treatment, compared with reported results in randomised studies Brasso et al. [63](#bju13123-bib-0063){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Compassionate use programme 137 18 Modest PSA level responses and improved survival was seen with enzalutamide in patients progressing after chemotherapy and abiraterone Schmid et al. [64](#bju13123-bib-0064){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Compassionate use programme 35 10 Consecutive use of enzalutamide and abiraterone after taxane‐based chemotherapy shows a modest clinical activity Thomson et al. [65](#bju13123-bib-0065){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Expanded access programme 23 39 Enzalutamide appears to show only modest activity after failure of docetaxel and abiraterone Sandhu et al. [66](#bju13123-bib-0066){ref-type="ref"} Abiraterone prior to enzalutamide Retrospective study 23 17 (four out of 23 patients) Sequential enzalutamide in patients with CRPC post‐abiraterone showed only modest activity Vera‐Badillo et al. [67](#bju13123-bib-0067){ref-type="ref"} Docetaxel and abiraterone prior to enzalutamide Retrospective study 26 27 Modest clinical activity seen with enzalutamide treatment who progressed on docetaxel and abiraterone Cheng et al. [68](#bju13123-bib-0068){ref-type="ref"} Abiraterone and/or docetaxel prior to enzalutamide Retrospective study 310 24 (65 out of 274 patients with prior abiraterone and/or docetaxel) Enzalutamide activity was blunted after abiraterone and after docetaxel; among patients with primary resistance to abiraterone, a subset was sensitive to subsequent enzalutamide Enzalutamide → abiraterone or cabazitaxel Noonan et al. [69](#bju13123-bib-0069){ref-type="ref"} Enzalutamide prior to abiraterone + prednisone Retrospective 30 3 Abiraterone treatment was associated with a modest response rate and brief duration of effect in patients who progressed after enzalutamide Loriot, et al. [70](#bju13123-bib-0070){ref-type="ref"} Docetaxel and enzalutamide prior to abiraterone + prednisolone Retrospective 38 8 Abiraterone plus prednisolone had modest antitumour activities in patients with CRPC pre‐treated with docetaxel and enzalutamide Pezaro et al. [71](#bju13123-bib-0071){ref-type="ref"} Docetaxel, abiraterone and/or enzalutamide prior to cabazitaxel Retrospective 59 39 Significant cabazitaxel activity was reported in patients with CRPC progressing after docetaxel and abiraterone or enzalutamide NR, not reported. John Wiley & Sons, Ltd Mechanisms of Resistance to Enzalutamide {#bju13123-sec-0012} ======================================== In general, studies that have investigated the sequencing of hormonal therapies have shown that with any treatment a reduced response is expected with each subsequent line of therapy. Recent research, primarily in patients who progressed on chemotherapy and abiraterone, suggests that one marker for resistance to AR‐targeted therapies is AR‐V7, a constitutively active AR splice variant [72](#bju13123-bib-0072){ref-type="ref"}. Based on the currently available data, detection of AR‐V7 in circulating tumour cells from patients with CRPC may be associated with resistance to enzalutamide and abiraterone [72](#bju13123-bib-0072){ref-type="ref"}. In a small Phase II prospective study evaluating the effect of enzalutamide in blood and bone marrow, the presence of an AR‐V7 variant was associated with primary resistance to enzalutamide [73](#bju13123-bib-0073){ref-type="ref"}. In addition, *in vivo* studies have shown that cabazitaxel remained highly effective in enzalutamide‐resistant tumours in castrated mice and demonstrated superior antitumour activity compared with docetaxel. These findings suggest that there could be cross‐resistance in the AR pathway between enzalutamide and docetaxel, but not with cabazitaxel, in CRPC [74](#bju13123-bib-0074){ref-type="ref"}. The results of the studies summarised in this section should be interpreted with caution, as they present analyses of small, retrospective case series, typically involving populations of patients previously progressing on more than two previous therapies and in different stages of mCRPC. None of these reports were based on studies specifically designed to assess the efficacy of sequential treatments. Larger, prospective sequencing and combination studies are required to assess the impact of treatments and to define optimal sequencing strategies for patients with mCRPC as the treatment paradigm continues to evolve. Conclusions {#bju13123-sec-0013} =========== Progress is being made in the treatment of mCRPC with increased understanding of the underlying molecular mechanisms involved in CRPC disease progression. This improved understanding is being applied to rationally designed pharmacological treatment options with novel mechanisms of action, e.g. enzalutamide, for patients with mCRPC. Enzalutamide inhibits three steps in the AR signalling pathway and positive results of Phase III clinical trials validate the AR signalling pathway as a therapeutic target in CRPC [49](#bju13123-bib-0049){ref-type="ref"}, [50](#bju13123-bib-0050){ref-type="ref"}. Enzalutamide is administered orally, can be taken with or without food, and does not require concomitant administration of steroids. Enzalutamide has not shown signs of liver toxicity and does not require specific monitoring [50](#bju13123-bib-0050){ref-type="ref"}. Prostate cancer is a markedly heterogeneous disease, with potential for multiple mechanisms of resistance to castration. Despite the clinical benefits obtained with currently available drugs for mCRPC, patients will ultimately become resistant and disease progression will eventually occur. Better understanding of drug resistance mechanisms and evaluation of treatment combinations are areas of ongoing research that have, and will continue to, become more important as the treatment options for CRPC evolve and expand. Funding/Support and Role of the Sponsor {#bju13123-sec-0015} ======================================= The authors participated in reviewing the paper. Enzalutamide is co‐developed by Medivation, Inc., and Astellas Pharma, Inc. This article was reviewed by each company for medical and scientific accuracy, but the authors had full control of the content and incorporation of comments from the companies was left up to their discretion. Author Contributions {#bju13123-sec-0016} ==================== J.S takes responsibility for the integrity and accuracy of the paper. J.S has read and approved the final version of this manuscript. J.M.F. passed away during development of this manuscript. Conflicts of Interest {#bju13123-sec-0017} ===================== J.S. has received speaker honoraria from Astellas, Sanofi Aventis, Ipsen, and Janssen Oncology. ADT : androgen‐deprivation therapy AE : adverse event AKT : protein kinase B AR : androgen receptor (m)CRPC : (metastatic) castration‐resistant prostate cancer CXCL8 : chemokine ligand 8 CYP17A1 : cytochrome P450 17A1 DBD : DNA‐binding domain DHT : dihydrotestosterone EMT : epithelial--mesenchymal transition HR : hazard ratio LBD : ligand‐binding domain NTD : N‐terminal domain OS : overall survival PI3K : phosphoinositide‐3‐kinase PTEN : phosphatase and tensin homolog rPFS : radiographic progression‐free survival The authors would like to thank Amineh Zafarani, PhD, at Ogilvy Healthworld and Joy Ramos, PhD, at Complete HealthVizion for assistance with writing and revising the draft manuscript based on detailed discussion and feedback from both authors, and Lauren Smith at Complete HealthVizion for copyediting assistance. Writing and copyediting assistance was funded by Astellas Pharma, Inc. and Medivation, Inc.

{.90} {.91} {.92}

1. Introduction {#sec0005} =============== Despite significant attention given to animal diseases in the last half-century, our understanding of disease processes, and how to manage them at the livestock--wildlife interface, remains limited ([@bib0575]). The increasing role of wildlife in the emergence of livestock and human diseases is due to multiple changes occurring within wildlife, livestock, and human populations, as well as at the livestock--wildlife interface ([@bib0335]). Human driven land use change---which frequently includes encroachment into wildlife habitat---continues to increase along with more intensified livestock production practices ([@bib0195], [@bib0550]). Alteration of wildlife population demographics, such as larger deer populations, increases the potential for contact and pathogen transmission at the livestock--wildlife interface ([@bib0575]). All of these changes work to create new interfaces between livestock and wildlife ([@bib0265], [@bib0210], [@bib0575]), potentially exacerbating pathogen transmission processes between them. Globally, the role of wildlife in livestock diseases is expected to increase ([@bib0635]) in conjunction with human population growth, which is expected to reach 9 billion by 2030. This will create increased demand for animal protein thereby increasing livestock populations ([@bib0015]). The demand will further increase potentially infectious contacts between livestock and wildlife leading to an increased potential for new zoonotic diseases to emerge. All of these challenges will require an improved understanding of the ecology of pathogens at the livestock--wildlife interface along with development of tools and mitigations to manage these pathogens. Historically, managing diseases affecting both livestock and wildlife as a single, linked system in North America, has presented several obstacles. Conflicting agency and institutional missions, program goals, and cultural differences that limit the potential for developing comprehensive mitigation of pathogen transmission contribute to hampering efforts in this area. Nevertheless, research and policy at the livestock--wildlife interface has received increased attention in recent years with the number of scientific publications in English journals addressing this topic rising dramatically ([Fig. 1](#fig0005){ref-type="fig"} ). This is driven, though not exclusively, by a rapid increase in the number of zoonotic disease events associated with wildlife in the latter part of the 20th century ([@bib0215], [@bib0535], [@bib0210]). Three-fourths of all emerging infectious diseases (EIDs) of humans are zoonotic with most originating in wildlife ([@bib0690], [@bib0335]). A large proportion (77%) of livestock pathogens---and an even higher proportion (91%) of carnivore pathogens---infect multiple hosts including wildlife ([@bib0145]). Therefore, diseases that arise from the livestock--wildlife interface are of paramount importance and must be an area of focus for animal health authorities ([@bib0635]).Fig. 1The number of publications in English language journals identified in Scopus database with the words "wildlife" and "parasite" or "disease" in the title, abstract, or key words. One example, Nipah virus---classified as an emerging infectious pathogen---recently moved from its natural host (fruit bats) to domestic swine, causing disease and mortality in both swine and local agricultural workers and resulting in economic losses ([@bib0235]). The 1999 Nipah virus outbreak in Malaysia destroyed the Malaysian swine industry while the associated human fatalities simultaneously created massive public panic ([@bib0235]). This newly recognized virus was carried by fruit bats for decades and emerged as a result of newly occurring habitat destruction, climatic changes, and the encroachment of food--animal production into wildlife domains ([@bib0240]). Although little discussed, pathogen transmission at the livestock--wildlife interface is frequently bi-directional ([@bib0065]). In contrast to conventional thinking, livestock have introduced several pathogens, such as bovine brucellosis and tuberculosis bacterium, to naïve wildlife populations in North America. These two pathogens are found in at least five wildlife populations ([@bib0700], [@bib0680]) and create significant challenges for disease control at the livestock--wildlife interface. In some instances, spillover events from livestock into wildlife impact conservation of species of concern ([@bib0215], [@bib0500], [@bib0325], [@bib0185]). An example is the transmission and introduction of bovine brucellosis and tuberculosis from livestock into native wood bison (*Bison bisonathabascae*) populations in Canada, which has created a conservation challenge for the species ([@bib0695], [@bib0500]). Another well-publicized example is the introduction of brucellosis into native bison and elk populations of the Yellowstone ecosystem in 1917 ([@bib0440], [@bib0460]). This resulted in a wildlife management challenge due to conflicts between livestock and bison preservation. The presence of brucellosis poses continued risk for transmission back into livestock creating biological, social, and policy challenges ([@bib0185], [@bib0180]). Obstacles faced by wildlife managers and livestock authorities for mitigating contact between wildlife and livestock has resulted in significant efforts to develop technology that reduces contact and is economically feasible. However, development of effective tools that can be readily deployed has been met with a host of challenges. Many devices prove to be ineffective or only effective for a short duration ([@bib0760], [@bib0740]). The most successful tools have involved fencing technology (e.g. high fence, wire mesh, electrified high-tensile steel wire, or polytape) that reduces contact between wildlife and livestock feed ([@bib0745], [@bib0755]). However, fencing suffers from limitations such as the need for relatively frequent maintenance. More recently, research has focused on the use of historic tools such as livestock protection dogs to prevent contact between livestock and wildlife. In some cases these traditional tools have proven to be the most effective ([@bib0750], [@bib0735]). In addition to the challenges faced in developing effective mitigation tools is gaining social acceptance of their use by farmers, which is fundamental in successfully using these tools ([@bib0095]). However, there remains a need for identifying new economically feasible tools that wildlife and livestock managers can deploy to reduce contact at the livestock--wildlife interface. Improving our understanding of the biological and anthropogenic processes that promote contact between wildlife and livestock is critical for limiting pathogen transmission at this interface. Given the frequently bi-directional nature of pathogen transmission, cooperation is required between livestock owners, animal health officials, and wildlife managers if control efforts are to be successful. Conflicts will undoubtedly continue to challenge wildlife managers and livestock authorities seeking solutions, which can only be found through the creation of new partnerships and the strengthening of existing ones that bridge the gap between wildlife and livestock agencies at all levels. Here we conduct a systematic review of the English scientific literature to evaluate the status of diseases and pathogens at the livestock--wildlife interface in the United States. Specifically, the goals of the literature review were three fold: first, to evaluate domestic animal diseases currently found in the United States where wildlife may play a role; second, to identify critical issues faced in managing these diseases at the livestock--wildlife interface; and third, to identify potential technical and policy strategies for addressing these issues. We highlight two examples of emerging diseases at the livestock--wildlife interface in North America, which pose management challenges and offer an opportunity for comprehensive disease management by facilitating cross-agency and state-federal partnering. 2. Status of diseases at the livestock--wildlife interface in the United States {#sec0010} =============================================================================== In the United States, there are currently 86 avian, ruminant, swine, and lagomorph diseases reportable to the OIE. Of those, 53 are listed as present in the United States ([@bib0035]). Our review of these pathogens identified 42 (79%) which have a potential wildlife component associated with the transmission, maintenance, or life cycle of the pathogen, and 21 (40%) are known to be zoonotic ([Table 1](#tbl0005){ref-type="table"}, [Table 2](#tbl0010){ref-type="table"} ). Of these 42 pathogens, 12 (29%) have an arthropod vector involved in the transmission while the remaining 71% involve direct or indirect transmission. Sixteen (38%) of the diseases present in the United States affect multiple species of livestock, all of these have a wildlife component, and 81% are zoonotic. Of the OIE reportable diseases affecting cattle, 6 out of 8 are present in the United States and have a wildlife component; 3 have zoonotic potential. A wildlife component has been identified for 10 out of 11 (91%) OIE reportable avian diseases with 3 of these recognized as zoonotic.Table 1Number of OIE reportable diseases present in the United States and number with a known potential wildlife component associated with the transmission, maintenance, or life cycle of the causative agent.Established free (absent)Known present (sporadic or limited distribution)TotalWildlife componentZoonoticAvian3 (1)9 (2)151091%327%Cattle4 (2[\*](#tblfn0005){ref-type="table-fn"})814675%338%Equine46 (1)11343%114%Lagomorphs1 (1)22100%Multiple1014 (2)2616100%1381%Sheep and goat46 (1)11343%114%Swine4 (1)272100%

Total OIE diseases29 (4)53864279%2140%[^1]Table 2OIE reportable livestock diseases present in the United States with a known wildlife component.DiseaseAffected livestockWildlife hostCitationTransmission modeAnthraxRuminants^c^, horses^c-sc^, swine^c-sc^All mammals susceptible, environmental reservoirs([@bib0315])DirectAujeszky\'s diseaseDomestic swine^c^, cattle^c^, sheep^c^, goats^c^, horses^o^Feral Swine^r^, wild mammals^s^([@bib0280], [@bib0170], [@bib0360], [@bib0640])Direct, indirectAvian chlamydiosisDucks^c^, turkeys^c^, chickens^o^Gulls^r^, ducks^r^, herons^r^, egrets^r^, pigeons^r^, blackbirds^r^, grackles^r^, house sparrows^r^, killdeer^r^, raptors^s^, shorebirds^s^, migratory birds^s^([@bib0725], [@bib0710], [@bib0640])Direct, indirectAvian infectious bronchitisChickens^c^Wild birds^u^([@bib0330], [@bib0485])Direct, indirectAvian influenzaChickens^c^, turkeys^c^, ducks^c^, geese^c^, game birds^c^Numerous wild birds^r^, many mammals susceptible^a^([@bib0155], [@bib0520])Direct, indirectAvian mycoplasmosis (*Mycoplasma gallisepticum, Mycoplasma synoviae*)Chickens^c^, turkeys^c^, game birds^c^, ducks^c^, geese^c^House finches^a^, American goldfinches^a^, purple finches^a^, eastern tufted titmice^a^, pine grosbeaks^a^, evening grosbeaks^a^, others^a^([@bib0710], [@bib0415], [@bib0405], [@bib0640])Direct, indirectBluetongueSheep^c^, goats^c^, cattle^sc^Wild ovine species^a^, cervids^a^, water buffalo^a^, pronghorn^a^,([@bib0790], [@bib0645], [@bib0590], [@bib0640], [@bib0305])Arthropod-borneBovine anaplasmosisCattle^c^Cervids^r^([@bib0800], [@bib0375])Arthropod-borneBovine babesiosis[\*](#tblfn0010){ref-type="table-fn"}Cattle^c^White-tailed deer^s^, water buffalo^s^, African buffalo^s^, reindeer^s^([@bib0640], [@bib0615], [@bib0115])Arthropod-borneBovine genital campylobacteriosisCattle^c^Numerous^r^([@bib0790])Direct, indirectBovine tuberculosisPrimarily cattle^c^White-tailed deer^r^, feral swine^r^, numerous spillover hosts([@bib0510], [@bib0105], [@bib0790], [@bib0640])Direct, indirectBovine viral diarrheaCattle^c^, camelids^c^, bison^o^White-tailed deer^r^, mule deer^s^, caribou^s^, pronghorn^s^, elk^s^, moose^s^, bison^s^([@bib0545], [@bib0810], [@bib0790], [@bib0220])Direct, indirectBrucellosis (*Brucella abortus*)Cattle^c^, sheep^c^, horses^o^Bison^r^, water buffalo^r^, elk^r^, feral swine^u^, numerous spillover hosts([@bib0530], [@bib0810], [@bib0790], [@bib0640])Direct, indirectBrucellosis (*Brucella ovis*)[\*\*](#tblfn0015){ref-type="table-fn"}Sheep^c^Red deer^a^([@bib0580], [@bib0640])Direct, indirectBrucellosis (*Brucella suis*)Domestic swine^c^, horses^o^Feral swine^r^, European hare^r^, caribou^r^, reindeer^r^, rodents^r^, numerous spillover hosts([@bib0260], [@bib0530], [@bib0790], [@bib0160], [@bib0640])Direct, indirectContagious agalactiaSheep^c^, goats^c^, cattle^o^, camelids^o^Spanish ibex^u^, roe deer^u^, red deer^u^([@bib0730]), ([@bib0640])Direct, indirectEchinococcosis/hydatidosisSheep^c^, cattle^c^Carnivore sp. including canids^r^, and felids^r^, cervids^u^, rodents^u^, lagomorphs^u^, muskrats^u^([@bib0385], [@bib0670], [@bib0665], [@bib0715], [@bib0640])IndirectEpizootic hemorrhagic diseaseCattle^c-sc^, sheep^sc,o^White-tailed deer^r^, mule deer^s^, pronghorn^s^, other wild ruminant species^s^([@bib0020])Arthropod-borneEquine encephalomyelitis (Eastern and Western)Equids^c^, occasional reports in cattle, sheep, camelids and pigsBirds^r^, rodents^r^, jackrabbits^r^, white-tailed deer^s^, numerous species^s^([@bib0570], [@bib0230], [@bib0365], [@bib0685], [@bib0610], [@bib0640])Arthropod-borneEquine influenzaEquids^c^Wild birds^r^, numerous other species^s^([@bib0480])Direct, indirectEquine piroplasmosisEquids^c^Uncertain([@bib0340], [@bib0640])Arthropod-borneEquine rhinopneumonitisEquids^c^Numerous species^u^([@bib0355])Direct, indirectFowl choleraPoultry^c^Wild birds^r^([@bib0710], [@bib0555], [@bib0085], [@bib0075])Direct, indirectInfectious bovine rhinotracheitis/infectious pustularvulvovaginitisCattle^c^Several implicated^u^([@bib0810], [@bib0355])Direct, indirectInfectious bursal diseaseChickens^c^, turkeys^sc^, ducks^sc^, guinea fowl^sc^, ostriches^sc^Game birds^r^, Waterfowl^r^([@bib0710], [@bib0110], [@bib0025])Direct, indirectLeptospirosisCattle^c-sc^, sheep^c-sc^, goats^c-sc^, pigs^c-sc^, horses^c-sc^, all mammals^c-sc^Rodents^r^, raccoons^r^, skunks^r^, opossum^r^, nutria^r^, others^s^([@bib0810], [@bib0790])Direct, indirectMaedi-visnaSheep^c^, goats^c^Wild ruminants^u^([@bib0720], [@bib0640])Direct, indirect is rareMarek\'s diseaseChickens^c^, Turkeys^o^, Quail^o^Galliformes^r^([@bib0135])Direct, indirectMyxomatosisLagomorphs^c^Lagomorphs^r^([@bib0790], [@bib0225])Arthropod-borneNewcastle disease[\*\*\*](#tblfn0020){ref-type="table-fn"}Chickens^c^, turkeys^c-sc^, game birds^c-sc^, ducks^sc^, geese^sc^, pigeons^c-sc^Wild birds^r^, exotic birds^r^([@bib0710], [@bib0100], [@bib0620], [@bib0150], [@bib0640])Direct, indirectParatuberculosisCattle^c^, sheep^c^, goats^c^Wild ruminants^r^, rabbits^r^, numerous wild mammals^u^([@bib0790], [@bib0165], [@bib0275], [@bib0640], [@bib0050])Direct, indirectPorcine reproductive and respiratory syndromeSwine^c^Feral swine^r^([@bib0790], [@bib0160])Direct, indirectPullorum diseaseChickens^c^, turkeys^c^, pheasants^c^, other poultry^o^Waterfowl^r^, numerous wild bird species^u^([@bib0710], [@bib0630])Direct, indirectQ feverCattle^c-sc^, sheep^c-sc^, goats^c-sc^Numerous species including mammals^r^, birds^r^, and reptiles^r^([@bib0810], [@bib0640])Direct, indirect, arthropod borneRabbit hemorrhagic diseaseDomestic *Oryctolaguscuniculus*^c^Wild *Oryctolaguscuniculus*^c^([@bib0790], [@bib0640])Direct, indirectRabiesAll mammals susceptibleRaccoons^r^, coyotes^r^, fox^r^, bats^r^, skunks^r^, mongoose^r^, bobcats^u^, others^s^([@bib0655], [@bib0370], [@bib0640])DirectTransmissible gastroenteritisSwine^c^Feral swine^r^([@bib0790])Direct, indirectTrichinellosisSwine^c-sc^Carnivores^a^, feral swine^a^, rodents^a^, bears^a^, others^a^([@bib0490])Direct, indirectTularemiaSheep^c^, horses^o^, pigs^o^Lagomorphs^r^, muskrats^r^, rodents^r^, mink^s^, prairie dogs^s^, others^s^([@bib0005], [@bib0790], [@bib0640], [@bib0320], [@bib0465])Direct, indirect, arthropod-borneVesicular stomatitisCattle^c^, swine^c^, equids^c^, camelids^sc^, sheep^sc^, goats^sc^Numerous wildlife species susceptible including mammals and birds, reservoir hosts unknown([@bib0790], [@bib0640], [@bib0770])Direct, indirect, arthropod-borneWest NileEquids^c^, domestic geese^c^Wild birds^r^, other species^u^([@bib0195], [@bib0710], [@bib0640])Arthropod-borneDiseases of importance that are not OIE listed.Chronic Wasting DiseaseDomestic cervids^c^Wild cervids^r^([@bib0790], [@bib0795], [@bib0055], [@bib0785], [@bib0285])Direct, indirectMalignant Catarrhal FeverCattle^c^, bison^c^, swine^c^, sheep^sc^, goats^sc^Wildebeest^r^, oryx^s^, ibex^s^, cervids^s^, wild ovine and caprinespecies^r^([@bib0790], [@bib0640])DirectPlagueDomestic mammals^c-sc^Prairie dogs^r^, chipmunks^r^, ground^r^squirrels^r^, other rodents^r^, carnivores^s^, numerous other species^a^([@bib0605])Direct, indirect, arthropod-borneTrichomoniasis (*Trichomonasgallinae*)Poultry^c^, doves^c^, pigeons^c^Pigeons^r^, doves^r^, falcons^a^, hawks^a^, others^a^([@bib0710])Direct, indirect[^2][^3][^4] Of the avian, ruminant, and swine diseases, 21 are currently actively managed in the United States with 11 of these having a Federal eradication or control program ([Table 3](#tbl0015){ref-type="table"} ). Thirteen (62%) of these actively managed diseases have a wildlife component and at least 6 (bovine tuberculosis, paratuberculosis, brucellosis, avian influenza, rabies, and cattle fever tick \[vector control\]) have a wildlife reservoir that is a recognized impediment to eradication due to continued spillover to domestic populations. Of these diseases, 2 (bovine tuberculosis and brucellosis) have foci of infection in wildlife as a result of spillover from livestock---further complicating eradication programs.Table 3Diseases actively managed in the United States and corresponding wildlife component.DiseaseNational or agency programPrimary domestic speciesWildlife componentAvian influenzaControlPoultryYesBluetongueMultipleYes (Arthopod-borne)Bovine spongiform encephalopathyCattleBovine TuberculosisEradicationMultipleYesBrucellosis (*Brucella abortus*)EradicationMultipleYesBrucellosis (*Brucella suis*)EradicationMultipleYesCattle Fever Tick (vector only)EradicationCattleYes (Arthopod-borne)Chronic wasting diseaseEradicationCervidsYesClassical swine feverSwineYesContagious equine metritisEquineEquine herpesvirusEquineEquine infectious anemiaEradicationEquineEquine piroplasmosisEquineUncertain (Arthopod-borne)Equine viral arteritisEquineParatuberculosis (Johnes)ControlMultipleYes (Arthopod-borne)Pseudorabies (Aujeszky\'s disease)EradicationMultipleYesRabiesEradicationMultipleYesScrapieEradicationSheep, GoatsVesicular stomatitisMultipleYes (Arthopod-borne)West NileMultipleYes (Arthopod-borne) Specific estimates of direct and indirect costs to livestock and recreational hunting industries, and to governmental agencies resulting from pathogen transmission at the livestock--wildlife interface, are elusive; however, some estimates are available for specific diseases. Reestablishment of bovine *Babesia* sp. to its historic range in North America via adaptation of *Babesia* sp. vectors to white-tailed deer would cost approximately \$1.2 billion to the cattle industry ([@bib0010]). In Michigan, the loss of bovine tuberculosis accredited-free status is estimated to result in total agriculture and livestock losses of approximately \$12 million per year ([@bib0310]). Furthermore, the Michigan Department of Natural Resources spent an estimated \$15 million on defining the extent of the disease in wildlife and initial management steps alone ([@bib0510]) and to date has spent an estimated \$23 million ([@bib0505]) on control, surveillance, and management of the disease. Rabies---an important zoonotic disease with significant public health, agricultural, and ecological impacts---is known to impose a financial burden on countries around the world. The Centers for Disease Control and Prevention estimates that the United States spends in excess of \$300 million annually on rabies prevention, detection, and control ([@bib0030]) with more than \$130 million spent on wildlife vaccination alone ([@bib0650]). Avian influenza, which has a well-documented wild waterfowl reservoir, continues to plague the domestic poultry industry in the United States with estimated outbreak associated losses ranging from \$5 to \$212 million ([@bib0125], [@bib0595]). Estimated impacts to the United States in the event of an epizootic avian influenza pandemic are at least \$71 billion ([@bib0450], [@bib0045]). Other livestock diseases with wildlife reservoirs including brucellosis, bovine viral diarrhea, and several poultry diseases are associated with significant losses in livestock production. 3. Structured approach to livestock--wildlife disease management {#sec0015} ================================================================ Concepts for integrated and adaptive management systems for EIDs at the livestock--wildlife interface are proposed by multiple authors ([@bib0705], [@bib0765]). Many countries, have developed passive and active surveillance systems for EID events in wildlife. Some of the earliest systems were developed in Denmark (1930s) and Sweden (1940s) however surveillance systems are established in Norway, Finland, France, United Kingdom, Italy, Spain, Switzerland, and the United States ([@bib0470], [@bib0815]). The United Kingdom ([@bib0600], [@bib0420], [@bib0290], [@bib0295]) has developed a program to implement integrated risk management and EID monitoring systems for wildlife. These nascent emerging systems have common themes, which may be adaptable to the United States. In the existing literature, five interdependent aspects of disease management are suggested as being necessary for successfully addressing disease issues at the livestock--wildlife interface: (1) horizon scanning (issue identification); (2) risk analysis and assessment; (3) risk mitigation; (4) surveillance and monitoring; and (5) disease control and management. These components, described as integrating sequentially with feedback loops, incorporate learning about the system. As information about the disease agent is improved, management is adapted thereby improving actions performed in the other components ([Fig. 2](#fig0010){ref-type="fig"} ). This process of adaptive management has been well described in the ecological and wildlife management literature ([@bib0345], [@bib0435]), but concepts related to adaptive management have only recently been proposed as a method for managing disease systems ([@bib0705], [@bib0765]).Fig. 2Conceptual model of adaptive disease management at the livestock--wildlife interface. 4. Horizon scanning {#sec0020} =================== Rapid identification of new and emerging infectious diseases (horizon scanning) in wildlife is critical to protecting animal agriculture and human health. There is mounting concern over the zoonotic potential, and subsequent wide-ranging socioeconomic impacts, associated with wildlife-borne EIDs ([@bib0335]). Recent examples of EIDs emerging from wildlife include Nipah virus in swine ([@bib0140]), severe acute respiratory syndrome (SARS) in humans ([@bib0585]), and H5N1 HPAI in domestic poultry, wild birds, and humans ([@bib0255]). In addition, new issues continue to emerge with well-documented disease systems, such as bovine tuberculosis and brucellosis influencing agricultural systems and wildlife management in North America ([@bib0525], [@bib0505]). Other EID\'s of concern to agriculture are certain to emerge in the future, some of which may disperse rapidly across broad geographic scales ([@bib0145], [@bib0635]). The risks of existing and new EID\'s to disperse rapidly highlight the need for robust systems for early identification of pathogens, which may have important health, social, economic, or other management consequences. 5. Risk analysis {#sec0025} ================ Risk analysis is an often broadly used term referring to risk characterization, risk communication, and risk management, which provides support for decision making and policy in the face of uncertainty ([@bib0675]). In the case of animal disease, risk analysis is an important tool used to identify and characterize the potential risks posed by implementation of policy or by specific events such as importation of livestock. Risk analyses form the foundation from which animal health policy is established. However, for diseases at the livestock--wildlife interface, quantitative risk assessments are often difficult. Challenges for conducting quantitative risk assessments often result from incomplete information related to the disease status of wildlife or limited understanding of the potential contact between wildlife and livestock leading to pathogen transmission. In addition accurate quantitative data describing the spatial distribution, movement, population structure, and population density are typically unavailable limiting inference to the population or understanding population level risk factors. Data quantifying important epidemiologic parameters necessary for describing disease risk such as contact rates, disease status of wildlife, wildlife population size, or biological process of the pathogen in wildlife are often unstudied or poorly understood. Risk assessments often assess the risk of pathogen transmission from wildlife to livestock ([@bib0190]). However, for many diseases of livestock in North America (e.g. bovine tuberculosis, brucellosis) the initial transmission event is from livestock to wildlife, which in some cases results in the establishment of a wildlife reservoir for the pathogen posing continued risks to livestock ([@bib0700], [@bib0680]). For these reasons the most successful and useful risk analyses consider the bi-directional nature of transmission and address questions of risk using statistical methods to explicitly incorporate uncertainty. In addition, studies that estimate contact between livestock and wildlife to understand potential for pathogen transmission are needed. 6. Risk mitigation {#sec0030} ================== Mitigating transmission risk between livestock and wildlife has received considerable attention ([@bib0745], [@bib0735], [@bib0765]). The ability to eliminate livestock pathogens from North American wildlife populations has been rare and when successful required extensive culling of wildlife. An example is the eradication of foot-and-mouth disease from the United States in 1925 which required the culling of 22,000 deer from the Stanislaus National Forest in California ([@bib0790]). Wildlife removal strategies can have unintended consequence, which was exemplified in the United Kingdom where wildlife behavior was changed as a result of culling increasing the risk of bovine tuberculosis transmission to cattle ([@bib0805]). In addition protected wildlife can complicate control or eradication efforts to control disease ([@bib0460]). Eradication efforts requiring the culling of large numbers of wildlife are likely untenable in the United States today, thus preventing establishment of livestock diseases in wildlife populations is a central pillar of long-term risk mitigation strategies. Implementing risk mitigations may offer the greatest potential for reducing economic and social impacts resulting from shared diseases. This often involves modifying animal husbandry practices to reduce contact between livestock and wildlife---including modified livestock housing, which reduces contact with peri-domestic wildlife or altered feeding practices, which reduces available forage for wildlife. Other risk mitigations include tools that prevent direct contact between wildlife such as frightening devices, fencing, or livestock protection dogs ([@bib0760], [@bib0745], [@bib0735]). However, the development and implementation of these tools comes with their own set of challenges. Successful implementation often includes changing social behaviors of livestock producers and developing new tools to manage risk mitigation which are cost effective and efficacious over the long term. Other risk mitigations may include identifying and reducing or eliminating risky management practices---such as allowing contact between livestock and wildlife which may foster the emergence of new pathogens in the United States. These may include translocation of wildlife or domestic and international wildlife trade. 7. Surveillance and monitoring {#sec0035} ============================== The need to develop comprehensive surveillance systems that integrate livestock, wildlife, and human components has been suggested ([@bib0475]). Robust surveillance systems in wildlife and at the livestock--wildlife interface to provide early detection of newly emerging EIDs or spillover and spillback of pathogens between livestock and wildlife is essential. Developing a comprehensive national monitoring system for EIDs in wildlife that is logistically and fiscally sustainable could yield economic benefits for livestock health management as a whole by reducing indemnity costs associated with spillover of disease from wildlife to livestock or by helping prevent spillover from livestock to wildlife through early detection. The objectives of such a system could be enhanced by close integration with existing livestock and wildlife health programs to guide "when", "where", and "how" surveillance is conducted. In addition, existing programs would benefit from closer working relationships between wildlife biologists, ecologists, epidemiologists, and veterinarians to improve efforts focused on reducing pathogen transmission ([@bib0080]). One obstacle to developing long-term, comprehensive surveillance efforts at the livestock--wildlife interface is inconsistent funding for these activities ([@bib0390], [@bib0660]). Funding has typically been in response to emergency directives (e.g. HPAI H5N1 surveillance) and focused for a short period until the threat is perceived to no longer exist. This has, predictably, generated problems for developing a comprehensive national infrastructure that can be maintained over the long-term. In addition, there are disease systems that have plagued agriculture for decades, such as bovine tuberculosis, that do not obtain sufficient levels of funding to fully address risks for introduction into new wildlife hosts such as feral swine ([@bib0680]). Another challenge faced by wildlife surveillance systems is that they often rely on hunter observations and reports, which are focused on game species. This increases the difficulty of identifying emergence of disease in non-game species. Finally, due to challenges associated with working across agency departmental boundaries, such as reduced communication, differing priorities, perceived competition in missions, and cultural differences wildlife surveillance efforts often remain less than fully coordinated which reduces their overall benefit. 8. Disease control and management {#sec0040} ================================= Once a pathogen is identified at the livestock--wildlife interface, active management and control of the disease agent is often the only method for reducing impacts to human health, agriculture, and recreational hunting industries ([@bib0080]). Integrated strategies that bring wildlife, human, and agricultural agencies together offer the greatest opportunity for success. Management of diseases at the livestock--wildlife interface often requires long-term engagement using a combination of altered livestock husbandry practices, active disease suppression in wildlife, and prevention of transmission using mitigation techniques. 9. Inter-agency and cross-sector collaborations and partnership {#sec0045} =============================================================== If surveillance and risk management activities at the livestock--wildlife interface are to be successful, we must recognize the complex nature of current and emerging diseases. These diseases can involve different health jurisdictions, socio-economic dimensions, and a wide range of stakeholders (i.e. livestock industry, conservation organizations, recreational hunters, etc.). We must promote strategic collaboration and partnerships across various disciplines, sectors, departments, ministries, institutions, and organizations at country, regional, and international levels ([@bib0070], [@bib0250]). With The recent focus on "One Health", which recognizes that human, animal (both domestic and wild) and ecosystems are tightly linked, successful management of disease requires an integrated approach where efforts are focused in concert across these domains ([@bib0350], [@bib0775]). In response to the One Health focus several countries have developed specific plans to address wildlife health as it relates to human and domestic animal health ([@bib0600], [@bib0295]). However, obstacles still remain in developing robust systems which integrate across the domestic, wild animal, and human domains. In most countries, sector-specific institutions have clear roles, responsibilities, and budgets---but mechanisms for cross-sector collaboration typically do not exist. Developing collaborations often proves difficult even mandated from the highest levels of government, as exemplified by continued outbreaks of highly pathogenic avian influenza in several countries ([@bib0250]). The United States suffers from similar limitations, due in part to the bicameral regulatory and legal authority for oversight of livestock and wildlife. States have clear ownership of wildlife; Federal regulatory authorities do not always extend to control disease in livestock to manage the disease in wildlife. Thus, the effective control of disease incursions from wildlife to livestock requires State and Federal livestock management agencies to foster positive working relationships with wildlife agencies. Unfortunately, such relationships frequently have not been developed resulting in a decision making process on livestock disease management in which wildlife appear as an afterthought, when often they are integral to disease maintenance and spread. Involving all relevant stakeholders (i.e. livestock industry, wildlife conservation groups, wildlife health authorities, livestock health authorities, etc.) in the development of regional or ecosystem-level livestock disease management planning, from the beginning of the process, increases the likelihood of success ([@bib0410]). One example of ongoing challenges animal health authorities face is found in the management of brucellosis in the Yellowstone ecosystem. Controversy has surrounded the management of brucellosis in the bison and elk within the ecosystem. These issues have often pitted Federal, State, agricultural, and wildlife agencies against one another. Several have noted that one of the most important constraints to managing brucellosis in Yellowstone is jurisdictional inertia, or the unwillingness of agencies to relinquish their existing domains of territorial control ([@bib0380], [@bib0430]). This example underscores the need for wildlife, human, and agricultural agencies to develop strong working relationships prior to emergence of disease. Integrated approaches to prevention before observing outbreaks in both wildlife and livestock may offer an opportunity for agencies to foster working relationships prior to a crisis. Development of clear mechanisms and agreements will enhance collaboration and interaction at all levels and should include incorporation of the roles and mandates of the various institutions and agencies involved. Often the agreements and working relationships that are established occur only at the highest levels of the organization resulting in little benefit to those working to implement program objectives. Opportunities for professional interactions and working relationships needs to be created and supported at the field level in addition to the administrative level ([@bib0250]). 10. Opportunities for success {#sec0050} ============================= Historically, integrated cross-disciplinary collaboration between livestock and wildlife agencies has been a challenge. However, many programs managing animal health diseases could benefit significantly from increased communication and collaborations that combine program objectives and activities across agency jurisdictions. While challenging from a political and cultural perspective, the outcome could be beneficial and would enhance the ability to quickly identify and respond to new and emerging disease issues. Integrating State and Federal livestock and wildlife agencies into the disease program planning process could reap future rewards. Below we illustrate the potential for cross-sector collaboration using two disease eradication programs---Cattle fever tick eradication and bovine tuberculosis eradication---facing challenges presented by the livestock--wildlife interface. Other disease eradication and management programs that address issues associated with the diseases and pathogens listed in [Table 3](#tbl0015){ref-type="table"} would also likely benefit from increased collaboration across livestock, wildlife, and human agencies at both State and Federal levels. 11. Discussion {#sec0055} ============== 11.1. Cattle fever tick eradication program {#sec0060} ------------------------------------------- Bovine babesiosies, caused by hematoprotozoan parasites of the genus *Babesia*, is globally among the most significant tick-borne disease of cattle ([@bib0780], [@bib0425]). In North America, the most important vectors of bovine babesiosis are *Rhipicephalus microplus* and *R. annulatus*---collectively known as cattle fever ticks. Cattle fever ticks were extirpated from the United States in 1960 after a nearly 60-year eradication campaign ([@bib0270], [@bib0090]). The eradication campaign exploited the perceived narrow host range of cattle fever ticks in combination with highly effective and now banned acaricides, which allowed the program to focus almost exclusively on the treatment of cattle ([@bib0090]). Reestablishment of cattle fever ticks and bovine *Babesia* to their historic range in North America is estimated to cost \$1.2 billion in control efforts and cattle production losses ([@bib0010]). As a result, animal health authorities and livestock producers consider mitigating this risk a priority. In recent years, there have been increasing infestations of cattle fever ticks on cattle along the Texas--Mexico border ([@bib0200]). Historically considered to be highly host specific for cattle, there is increasing evidence that white-tailed deer and other ungulates are suitable hosts for cattle fever ticks ([@bib0560]) with infested deer found in locations absent of cattle ([@bib0120]). In Texas, cattle fever ticks have been recovered from free-ranging and captive-exotic ungulates including axis deer, fallow deer, elk, red deer, aoudad sheep, and nilgai antelope ([@bib0455]). Due to the potential ineffectiveness of treating tick infestations in cattle with currently approved methods, such as mandatory removal of cattle from affected pastures for a period of time (i.e. pasture vacation) and treatment of cattle with acaricides the treatment of white-tailed deer and other wildlife has become necessary. A recent study indicates that cattle fever ticks have a high degree of genetic fluidity, which may allow them to adapt to new host species and therefore provide a potential pathway for reestablishment in the United States via wildlife hosts ([@bib0205]). White-tailed deer are also increasingly being recognized as a potential reservoir for the *Babesia* species (*B. bigemina*, *B. divergens*, and *B. bovis*) which cause clinical disease in cattle ([@bib0200]). Surveys for *Babesia* in northern Mexico and Texas have identified molecular and serological evidence for the presence of *B. bigemina* and *B. bovis* in white-tailed deer and in nilgai antelope populations ([@bib0120], [@bib0130]). These changes in the host--pathogen system, and gaps in the understanding of cattle fever tick ecology and the host range of *Babesia*, require the formulation of more effective control strategies that include both wildlife and livestock. To effectively address these challenges, State and Federal Agencies representing both livestock and wildlife authorities need to partner to develop policy that integrates surveillance and risk mitigations across both cattle and wildlife populations. An historic limitation of the program has been the nearly exclusive focus on controlling cattle fever ticks on cattle ([@bib0395]). Recently the program has begun to deploy mitigations to control ticks on wildlife; however, the program is limited by a lack of operational tools to mitigate infestations on wildlife and a regulatory framework that would integrate management of the disease across wildlife and livestock authorities. While challenging, this offers an exciting opportunity to develop effective strategies and methods to address surveillance at the livestock--wildlife interface and to develop new mitigations that reduce the risk of infestation. 11.2. Bovine tuberculosis eradication program {#sec0065} --------------------------------------------- Bovine tuberculosis (bTB), identified in nine geographically distinct wildlife populations in North America and Hawaii, is endemic in at least four populations, including members of the Bovidae, Cervidae, and Suidaefamilies ([@bib0680]). The emergence of bTB in North American wildlife poses a serious and growing risk for livestock and human health and for the recreational hunting industry. Experience in many countries, including the United States and Canada, has shown that while bTB can be controlled when restricted to livestock species, it is almost impossible to eradicate this disease once it has spread into ecosystems with free-ranging maintenance hosts. Recent epidemiological models suggest that once bTB is introduced, the probability of becoming established in a wildlife population once introduced is at least 10% ([@bib0565]). Spillover into wildlife---and establishment of new foci of infection in wildlife---would be costly to the cattle industry and animal health authorities. In addition, new foci of wildlife infection would complicate eradication efforts. Therefore, preventing spillover of *Mycobacterium bovis* into wildlife may be the most effective way to mitigate economic costs of bTB. Historically, wildlife control efforts for bTB have focused solely on potential spillover into wild cervid species. However, *M. bovis* has been isolated from free-ranging swine (i.e. wild boar and feral swine) in at least 15 countries ([@bib0400], [@bib0175], [@bib0245], [@bib0515], [@bib0040], [@bib0625], [@bib0540]). New evidence from Mediterranean ecosystems supports the role of wild swine as maintenance hosts of bTB---sustaining infection and transmitting the pathogen to other species ([@bib0040], [@bib0495]). Circumstances favoring bTB transmission between wildlife and livestock in the Mediterranean include artificial increases in wild game populations stimulated by a robust hunting industry, feeding and baiting of wildlife, and intensive cattle grazing in proximity to wild swine ([@bib0300]). All of these characteristics likely apply to conditions in North America. Particularly worrisome is the recent appearance of feral swine in the state of Michigan where the potential exists for interaction with bTB-infected white-tailed deer and cattle. Regions of the southern United States also pose a risk where high densities of feral swine, an established hunting industry, significant baiting and feeding of wildlife, and introductions of bTB infected cattle from Mexico continue to occur ([@bib0680]). Furthermore recent evidence indicates that *M. bovis* may be present in free ranging white-tailed deer in northern Mexico. One study report the presence of *M. tuberculosis* complex identified using amplification of DNA from a tissue by PCR ([@bib0060]). The authors also report histopathology consistent with *M. bovis* infection observed in white-tailed deer. Another study reported the frequent detection of antibodies against mycobacterium antigens in a cross-sectional survey of white-tailed deer in Northern Mexico ([@bib0445]). While the risks posed by wildlife have been recognized, current investigations and response to potential spillover events from cattle to wildlife (cervid or swine), where disease is exceedingly more difficult to control or eradicate is inconsistently managed. Few standards are in existence which establish best practices for investigating potential spillover into wildlife hosts. Developing national policies and working relationships across agencies responsible for domestic and wildlife health at the State and Federal level would have long-term benefits for preventing the risk of introduction of bTB into new wildlife host populations. 12. Conclusions {#sec0070} =============== Nearly 80% of the pathogens present in the United States have a potential wildlife component. To successfully manage and control these pathogens at the livestock--wildlife interface will require the development of cross-discipline collaborations and establishing common goals between agencies and organizations that in some cases have rarely worked together. We believe the principles of adaptive management offer the greatest opportunities to formulate a framework from which collaborations can be developed to manage diseases at the livestock--wildlife interface. EID monitoring systems for wildlife that incorporate and implement integrated risk management in an adaptive management framework offer the best opportunity for success. In addition, new and creative funding mechanisms that bring livestock and wildlife animal health authorities along with livestock industry and wildlife stakeholders together will need to be created. Despite these social and policy challenges, there remain opportunities to develop new collaborations---along with the development of new technologies---to mitigate disease risks at the livestock--wildlife interface. We believe that two diseases eradication programs---bovine tuberculosis and cattle fever tick---offer the best opportunity to apply these principles and demonstrate success. We would like to acknowledge the insightful comments and critical review of early versions of this manuscript by Dr. Steve Sweeney, Dr. Tom DeLiberto, Dr. Reginald Johnson, Dr. Kathe Bjork, Dr. Tracey Lynn, and Mr. Allan Nelson. We also would like to recognize the diligent efforts of Ms. Mary Foley for supporting our continued literature and library science inquiries and Ms. Carol LoSapio for her editing contribution. We also thank two anonymous reviewers for their critical and insightful comments on this manuscript. [^1]: Bovine spongiform encephalopathy (BSE) is not considered free or present in the United States but rather a controlled risk. [^2]: Bovine babesiosis is not present in cattle in the United States however the causative agent has been reported in wildlife and a vector eradication program exists. [^3]: *B. ovis* has been found to cause poor semen quality in red deer but abortions have not been reported. The role potential role of red deer is still in doubt. [^4]: The United States is considered free from new castle disease in poultry however new castle disease is present in free ranging species and is included here for completeness. ^c^ = clinical ^sc^ = subclinical ^c-sc^ = may be clinical or subclinical ^o^ = occasional reports ^r^ = reservoir ^s^ = spillover ^a^ = affected species (not a true reservoir, nor a spillover host) ^u^ = uncertain

Introduction {#s0001} ============ Each year 33.8 million Infants and children become inflicted with lower respiratory disease after RSV infection, leading to 3.4 million hospitalizations with severe disease complications.[@cit0001] Furthermore, an inability of natural infection to mount robust long-lived immunity can leave immunosenescence populations, such as the elderly, vulnerable to respiratory complications after exposure. Currently the treatment option outside of high resource countries is limited to supportive care, immunoprophylaxis with the neutralizing antibody palivizumab being cost-prohibitive. Though considered a high medical need there is no approved RSV vaccine available. However, there is heavy investment into vaccine development. The high level of vaccine development activity is revealed on the PATH ([www.PATH.org](http://www.PATH.org)) snapshot of the current landscape. It reveals 40 candidates in preclinical development, and 14 at various clinical stages. Even though the most advanced candidate -- a RSV-F targeting VLP-based vaccine, recently reported disappointing topline efficacy results in a phase 3 study in elderly adults (9.15.2016 Novavax press release) there are many other promising vaccine candidates under development on a variety of platforms including, live-attenuated/chimeric, whole-inactivated, particle-based, subunit and gene-based vector. This commentary article will focus on nucleic acid-based vaccines targeting RSV. *In vivo* delivery of nucleic acid-based vaccines {#s0002} ================================================= At the beginning of the 1990\'s Wolf and colleagues reported *in vivo* protein expression after intramuscular injection of plasmid DNA or mRNA into mice.[@cit0003] It was this discovery that marked the beginning of the use of nucleic acids encoding antigens as a form of vaccination. While the instability of mRNA limited its use, plasmid DNA offered a very promising new vaccine platform. pDNA was stable, it could be produced both rapidly and in bulk, the transgene could be designed to encode antigen of choice, and the pDNA-vectored antigen could be delivered multiple times to boost immunity (pDNA itself is not immunogenic, and the issues associated with anti-vector immunity can be avoided). Studies in small animals revealed an attractive profile of both immunogenicity and safety.[@cit0004] However, initial studies in larger animals and humans disappointed, lower levels of immunogenicity were observed than those predicted from small animal models.[@cit0006] One major reason cited for this inconsistency was inefficiency of *in vivo* gene delivery. In response to this investigators in the field began developing both physical (electroporation (EP), ultrasound, gene gun) and chemical (lipids, polymers) *in vivo* delivery strategies to enhance the passage of pDNA into the host cell. [@cit0007] In recent years electroporation has become the go to delivery aide for nonviral gene delivery. Studies have consistently shown 100--1000 fold enhanced *in vivo* gene expression upon the employment of electroporation to protocols delivering naked pDNA.[@cit0008] EP gene transfer operates by inducing transient perturbations in the cell membrane and an electrical gradient which promotes the passage pDNA into the cell. Importantly, employment of EP into DNA vaccine protocols has significantly enhanced immune responses in both small and large animals, permitting protection against pathogen in challenge models. [@cit0013] Multiple DNA vaccine trials are successfully employing this technology to elicit robust host immune responses, and clinical efficacy of this platform has now been reported.[@cit0018] RSV nucleic acid-based vaccines 1.0 {#s0003} =================================== With an impressive safety profile, ability to stimulate humoral and cellular immune responses, and the capability of the investigator to design the vaccine to express only the desired antigen target, DNA vaccines may be an ideal platform to tackle RSV. Additionally, DNA vaccines exhibit the capacity to drive potent immune responses skewed towards Th1, which is a desirable trait considering the lung inflammation associated with the VED responses after FI-RSV vaccination have been attributed to dysregulated Th2 responses.[@cit0020] In 1998 Li and colleagues designed a DNA vaccine to target the RSV fusion (F) glycoprotein and demonstrated intramuscular immunization elicited strong Th1 responses, neutralizing antibodies and cytotoxic T cells in mice, and also achieved protection from disease challenge.[@cit0021] Many RSV vaccines have been designed to target the F protein, which is a confirmed target for neutralizing antibody and CTL responses in human. [@cit0022] The FDA-approved immunoprophylactic monoclonal Palivizumab targets antigen site 2 on the RSV F fusion protein.[@cit0023] Another vaccine target is the G glycoprotein, which is less well conserved than the F glycoprotein across the RSV subgroups.[@cit0025] While initial studies with non-DNA vaccine platforms suggested RSV G antigen responses to be polarized towards Th2,[@cit0026] and thus promoting atypical lung inflammation after live RSV exposure, in contrast vaccine studies using DNA revealed a more balanced Th1/Th2 in the cotton rat model.[@cit0028] Cotton rats are considered the gold standard small animal model to study RSV infection, being susceptible to non-adapted RSV and displaying many features of human lung pathology.[@cit0029] RSV nucleic acid-based vaccines 2.0 {#s0004} =================================== Almost 20 years has passed since the first description of RSV nucleic acid-based vaccines, but no candidate is in the clinic. For the reasons discussed above concerning difficulties involved in scaling-up and retaining immunogenicity form small animals to large animals and humans, has hampered the RSV DNA vaccine field\'s progress. However, since the first wave of DNA RSV vaccines several important advances have occurred in the field. These include codon optimization,[@cit0030] further understanding and rational design to specific regions or protein structure (pre- or post-fusion F glycoprotein) of the RSV antigen, [@cit0031] and delivery of the vaccine.[@cit0008] For DNA-based vaccines it is the improved delivery strategies which have made this platform relevant again. Electroporation has been employed to deliver DNA-based RSV vaccines to achieve robust immune responses in both small and large animals. [@cit0035] In an intramuscular (IM) delivery protocol Grunwald *et al.* demonstrated enhanced levels of immunogenicity of a RSV-F DNA vaccine with the addition of IM EP compared to conventional IM delivery in rhesus macaques.[@cit0035] The group also described increased immunogenicity upon delivery of the vaccine to the skin with EP. Our group has recently investigated both RSV-F and RSV-G DNA vaccines delivered with EP to the muscle or skin. Results revealed stronger T cell and neutralizing antibody responses after RSV-F compared to RSV-G DNA vaccination in experimental models including mice, Wistar rats, cotton rats and nonhuman primates[@cit0034] (and manuscript in preparation). Both RSV-F and RSV-G DNA vaccines delivered with EP protected from lower respiratory disease after RSV challenge in mice and cotton rats. While these studies demonstrate proof of concept, the method of EP and site of delivery must be appropriate to the disease indication. IM EP is an invasive procedure and as such is likely suited to certain vaccine targets in an adult population. With this in mind we have advanced a non-invasive delivery strategy with improved tolerability profiles. Our current focus is on surface EP (SEP) as a means to deliver a DNA RSV vaccine. This EP approach does not penetrate the live skin layers and operates at low voltages to produce only a shallow electrical field to target the pDNA vaccine to the epidermis. The epidermis is highly enriched in dendritic cells and is a site which permits dose sparing. We recently reported full protection against lower respiratory disease in the cotton rat RSV/A challenge model after a single low dose of a DNA based pRSV-F vaccine delivered at the skin using the SEP device.[@cit0034] We demonstrated the ability of this strategy to elicit robust immune responses. In contrast to the formalin-inactivated RSV (FI-RSV) vaccine, post mortem histological examination of the lung tissue revealed no enhanced lung inflammation upon virus challenge after DNA vaccination and thus no evidence of vaccine-enhanced disease, supporting vaccine safety. This dataset supports the advancement of a DNA vaccine candidate combined with an appropriate delivery platform to target vulnerable populations, including infants. Arrival of the mRNA platform {#s0005} ============================ In addition to the established pDNA platform, another branch of the nucleic acid vaccine family, RNA based vaccines, are beginning to show promise. [@cit0039] Although burdened by issues surrounding instability and manufacturing difficulties, many RNA vaccines researchers are making significant progress in overcoming these hurdles. Impressive preclinical and clinical immunogenicity data is now being reported on a variety of disease targets.[@cit0039] In 2013, Geall and colleagues reported immunogenicity and protection against disease challenge in cotton rats after 2 rounds of IM immunization with an RSV- naked self-amplifying RNA vaccine. Upon formulation of the RNA vaccine with lipid nanoparticles (LNPs), there was significantly enhanced immunogenicity and reduction in lung viral titers after viral challenge. Historically as was the case with DNA vaccines, naked delivery of mRNA appears to be suboptimal and data suggests that delivery optimization is required. LNPs have been shown successfully employed to enhance *in vivo* delivery of RNA vaccines.[@cit0043] Currently CureVac has a protamine-complexed mRNA RSV vaccine candidate in preclinical development.[@cit0041] Conclusions {#s0006} =========== The field of nucleic acid-based vaccines has seen significant scientific and clinical progress since their inception in the early 1990\'s. Specifically, significant clinical efficacy data for EP-enhanced DNA vaccines has now been generated, and promising early development data for RNA vaccines looks encouraging. With improved and optimized delivery platforms, the historical hurdle of low vaccine immunogenicity has been addressed, and ground breaking clinical efficacy is now being reported.[@cit0018] Though a nucleic acid-based RSV- vaccine has yet to reach the clinic, multiple promising candidates are in the development stages, both DNA[@cit0033] and mRNA-based.[@cit0040] Preclinical studies have revealed the desirable Th1 responses and strong protection in relevant animal models. With the advances in design and development of appropriate delivery platforms to immunize vulnerable populations we would expect clinical trials of these candidates to start soon. Disclosure of potential conflicts of interest ============================================= No potential conflicts of interest were disclosed.

Background {#Sec1} ========== Plants depend on the absorption of water from soil and its subsequent transport to all other plant parts. Water moves inside the plant body through apoplastic, transcellular, and symplastic pathways. The symplastic pathway transports water across membranes \[[@CR1]\] and is generally mediated by members of an ancient family of water channels called aquaporins (AQPs), which are part of the major intrinsic protein (MIP) superfamily \[[@CR2]\]. Efficient cell-to-cell water movement through the plant is controlled by AQPs in different physiological contexts \[[@CR3]\]. In addition to water uptake into roots, AQPs also function in water homeostasis in leaves \[[@CR4], [@CR5]\]. Moreover, AQPs are involved in controlling water movement for tissue expansion \[[@CR6], [@CR7]\] and have regulatory roles in processes such as fruit development \[[@CR8]\] and cell enlargement in *Arabidopsis thaliana* roots, hypocotyls, leaves, and flower stems \[[@CR6]\], and ripening of grape berries \[[@CR9]\]. AQPs are predicted to consist of six membrane-spanning segments with two cytoplasmic termini. AQPs contain Asn-Pro-Ala (NPA) motifs located in two short, fold-back alpha helices following the second (loop B, LB) and fifth (loop E, LE) trans-membrane helices. Each AQP monomer contains two hemi-pores, which fold together to form the water channel. *Arabidopsis* encodes 35 different AQPs \[[@CR10]\], whereas there are 66 AQPs in *Glycine max* \[[@CR11]\], 31 in *Zea mays* \[[@CR12]\], 33 in *Oryza sativa* \[[@CR13]\], 54 in *Populus trichocarpa* \[[@CR14]\] and 47 in *Solanum lycopersicum* \[[@CR8]\]. Based on sequence similarity and subcellular localization, higher plant AQPs have been classified into five subfamilies, namely the plasma membrane intrinsic proteins (PIPs), the tonoplast intrinsic proteins (TIPs), the NOD26-like intrinsic proteins (NIPs), the small basic intrinsic proteins (SIPs), and the X (or unrecognized) intrinsic proteins (XIPs) \[[@CR15]\]. The NIP subfamily is named for the founding member, soybean (*Glycine max*) nodulin-26 (*Gm*NOD26), which is an abundant AQP expressed in the peribacteroid membrane of N~2~-fixing symbiotic root nodules. It was initially thought that the NIP proteins were found only in the nodules of nitrogen-fixing legumes \[[@CR16]\]. However, NIP proteins were later found in many non-leguminous plants including *Arabidopsis* \[[@CR17]\], and rice \[[@CR13]\]. The SIP subfamily is conserved in all plant species, but is not well characterized to date. The XIPs form a phylogenetically distinct subfamily and have been found in moss, fungi and dicot plants \[[@CR15]\]. *Arabidopsis* encodes 35 different AQPs \[[@CR10]\], 66 AQPs in *Glycine max* \[[@CR11]\], 31 in *Zea mays* \[[@CR12]\], 33 in *Oryza sativa* \[[@CR13]\], 54 in *Populus trichocarpa* \[[@CR14]\] and 47 in *Solanum lycopersicum* \[[@CR8]\]. AQPs also appear to be involved in responses to abiotic stresses like drought, salt, and cold stresses in various plants. Seven members of the PIP1 subfamily of rice are responsive to cold stresses \[[@CR18]\]. Moreover, *Tricticum aestivum* TIP2 regulates the responses of plants to abiotic stresses (salt and drought) via an ABA-independent pathway(s) \[[@CR19]\]. In *Arabidopsis*, *PIP2;5* is up-regulated during cold exposure, and PIP subfamily genes are responsive to drought and salt stresses \[[@CR20]\]. In addition, *NtAQP1* is involved in improving water use efficiency, hydraulic conductivity, and yield production under salt stress in tobacco \[[@CR21]\]. By contrast, there is limited information whether AQPs function plant defenses against biotic stresses like attacks from fungal, bacterial and viral pathogens. In this work, we carried out a genome-wide expression profiling of the *AQP* gene family in *Brassica rapa* to characterize which genes were responsive to biotic and abiotic stresses. *Brassica rapa* is a species of the genus *Brassica*, which is economically important worldwide. We performed comprehensive *in silico* analyses of gene classifications, chromosomal distribution, synonymous and non-synonymous substitution rates, syntenic relationships, evolutionary divergence, subcellular localization, gene duplication, phylogenetic analysis, exon--intron organization, conserved motifs, and predicted functions of AQPs in *B. rapa*. We further determined the gene expression pattern of PIP subfamily members in *B. rapa* plants in response to abiotic stresses (cold, drought, salinity, water logging) and ABA treatment. We also analyzed PIP subfamily expression under biotic stress (infection with *Fusarium oxysporum* f.sp. *conglutinans*), and assessed AQP protein similarity to stress response-related proteins from other plants. Results {#Sec2} ======= Identification and in silico functional analysis of *B. rapa* aquaporin genes {#Sec3} ----------------------------------------------------------------------------- To identify all *AQP* genes in *B. rapa*, we searched SWISSPROT of the BRAD (<http://brassicadb.org/brad/>) \[[@CR22]\] and annotations of microarray data for cold-treated *B. rapa* (Chiifu & Kenshin), removing any duplicates. A total of 61 gene sequences encoding putative members of the AQP family were identified in *B. rapa*. Domain searches using SMART confirmed that 59 of the putative *AQP* genes in *B. rapa* encoded predicted MIP and trans-membrane domains. In agreement with this result, protein sequence similarity analysis of all 61 sequences using blastp (protein-protein BLAST) showed that all but the two protein sequences lacking functional MIP and trans-membrane domains were most similar to proteins of AQPs. Based on these findings, we concluded that there are 59 functional *AQP* genes in *B. rapa*, which we named based on nomenclature used in other plants and guided by sequence similarity and phylogenetic analysis. Tao et al. \[[@CR23]\] previously reported 53 *AQP* genes in *B. rapa*, and our analysis found these, along with six more *AQP* genes. Additional file [1](#MOESM1){ref-type="media"}: Table S1 lists the chromosomal position, ORF length and orthologous genes, as well as predicted protein length, iso-electric point and molecular weight for each of these 59 *B. rapa AQP* genes. These 59 AQP proteins of *B. rapa* showed a high level of sequence similarity to AQP proteins from different plant species. *In silico* functional analysis showed that the six newly identified *AQP* genes are likely involved in water transport in the plant body and leaves and in also root development (Additional file [2](#MOESM2){ref-type="media"}: Table S2). Most of the BrAQP proteins were highly similar to AQPs involved in water and solute transportation or fruit development in different plant species. Six, five and two of BrAQP proteins shared the highest degree of identity with proteins responsible for pod colour, tissue-specific expression and root development, respectively, in other plant species (Additional file [2](#MOESM2){ref-type="media"}: Table S2). Interestingly, the majority of BrPIP subfamily proteins showed high identity to abiotic stress-related AQP proteins from a wide range of plants (Additional file [2](#MOESM2){ref-type="media"}: Table S2). Therefore, we have selected BrPIP subfamily for details expression analysis. Out of 59 identified BrAQPs, 25 were most similar to abiotic stress (freezing, salt and drought)- and ABA-related AQP proteins in different plant species. Twenty out of those 25 belonged to the BrPIP subfamily are directly related to abiotic and ABA- stress responsive. Therefore, we concluded that PIP subfamily members among the BrAQP proteins are the most likely to be involved in water and solute transport in response to various abiotic stresses. Sequence analysis of *BrAQP* genes {#Sec4} ---------------------------------- Table [1](#Tab1){ref-type="table"} summarizes the aromatic/Arg (ar/R) selectivity filter (H2, H5, LE1 and LE2), Froger's positions (P1 to P5), and the prediction of domains, subcellular localization, NPA motifs, and genome fractionation (subgenome) for the 59 AQP protein sequences. With the exception of BrPIP2;2b all of the predicted BrAQP proteins contained two conserved NPA motifs, in LB and LE. Each member of predicted BrSIP subgroup member contained unusual third amino acids in the motifs, with the alanine replaced by threonine, cysteine, leucine or valine. By contrast, *BrNIP1;2a, BrNIP1;2b, BrNIP6;1a* and *BrNIP6;1b* encoded motifs with a variable third residue in which alanine was replaced by glycine and valine. Meanwhile, *BrNIP5;1a* and *BrNIP5;1b* encoded dissimilar amino acids in both NPA motifs, where alanine was replaced with serine and valine, respectively. Based on our subcellular localization predictions, all members of the NIP, SIP and PIP subfamilies of *B. rapa* appear to be present in the cell membrane. However, members of TIP subfamily were predicted to be positioned on vacuoles, with BrTIP 5;1 located in both vacuole and cell membrane (Table [1](#Tab1){ref-type="table"}).Table 1Subgenome position, conserved amino acid residues (NPA motif, Ar/R filter, Froger\'s position), the prediction of transmembrane and MIP domains and subcellular localization of *B. rapa* AquaporinsGene nameSub genomeNPA motifAr/R selectivity filterFroger's Position (P1 - P5)TMH + MIPSubcellular localizationLBLEH2H5LE1LE2P1P2P3P4P5*BrSIP1;1a*LFNPTNPAITPIIAAYW6 + 1CM*BrSIP1;1b*MF1NPTNPAITPIIAAYW6 + 1CM*BrSIP1;2*LFNPCNPAVTPIIAAYW6 + 1CM*BrSIP2;1a*MF1NPLNPASKGAFVAYW6 + 1CM*BrSIP2;1b*MF2NPLNPASKGAFVAYW6 + 1CM*BrSIP2;1c*LFNPVNPASKGAFVAYW6 + 1CM*BrNIP1;2a*LFNPANPGWVARFSAYI6 + 1CM*BrNIP1;2b*MF1NPANPGWVARFSAYI6 + 1CM*BrNIP2;1a*LFNPANPAWVARFSAYI6 + 1CM*BrNIP2;1b*LFNPANPAWVARFSAYI6 + 1CM*BrNIP3;1a*MF1NPANPAWIARFSAYI6 + 1CM*BrNIP3;1b*MF2NPANPAWIARFSAYI6 + 1CM*BrNIP4;1*MF2NPANPAWVARFSAYI6 + 1CM*BrNIP4;2a*LFNPANPAWVARFSAYI6 + 1CM*BrNIP4;2b*MF2NPANPA-VARFSAYI4 + 1CM*BrNIP4;2c*MF2NPANPA\--ARFSAYI3 + 1CM*BrNIP5;1a*MF2NPSNPVAIGRFTAYL6 + 1CM*BrNIP5;1b*MF1NPSNPVAIARFTAYL6 + 1CM*BrNIP6;1a*MF1NPANPVAIARFTAYL6 + 1CM*BrNIP6;1b*LFNPANPVAIARFTAYL6 + 1CM*BrNIP7;1*LFNPSNPAAVGRYSAYM6 + 1CM*BrTIP1;1*MF1NPANPAHIAVTAAYW6 + 1V*BrTIP1;2a*LFNPANPAHIAVTAAYW6 + 1V*BrTIP1;2b*MF1NPANPAHIAVTAAYW6 + 1V*BrTIP1;3*LFNPANPAHIAVTSAYW6 + 1V*BrTIP2;1a*LFNPANPAHIGRTSAYW6 + 1V*BrTIP2;1b*MF2NPANPAHIGRTSAYW6 + 1V*BrTIP2;1c*MF1NPANPAHIGRTSAYW5 + 1V*BrTIP2;2*LFNPANPAHIGRTSAYW6 + 1V*BrTIP2;3a*LFNPANPAHIGRTSAYW6 + 1V*BrTIP2;3b*MF1NPANPAHIGRTSAYW5 + 1V*BrTIP3;1a*MF1NPANPAHIARTAAYW6 + 1V*BrTIP3;1b*LFNPANPAHIARTAAYW6 + 1V*BrTIP3;2a*LFNPANPAHMARTASYW6 + 1V*BrTIP3;2b*MF1NPANPAHMARTASYW6 + 1V*BrTIP4;1*MF1NPANPAHIARTSAYW6 + 1V*BrTIP5;1*LFNPANPANVGCVAAYW6 + 1V and CM*BrPIP1;1a*LFNPANPAFHTRQSAFW6 + 1CM*BrPIP1;1b*MF1NPANPAFHTRQSAFW6 + 1CM*BrPIP1;2a*MF1NPANPAFHTRQSAFW6 + 1CM*BrPIP1;2b*LFNPANPAFHTRQSAFW6 + 1CM*BrPIP1;3a*MF2NPANPAFHTRQSAFW6 + 1CM*BrPIP1;3b*LFNPANPAFHTRQSAFW6 + 1CM*BrPIP1;4*MF1NPANPAFHTRQSAFW6 + 1CM*BrPIP1;5*MF1NPANPAFHTRQSAFW6 + 1CM*BrPIP2;1*LFNPANPAFHTRQSAFW6 + 1CM*BrPIP2;2a*MF2NPANPAFHTRQSAFW6 + 1CM*BrPIP2;2b*LFNPA-F\-\--Q\--FW5 + 1CM*BrPIP2;3a*MF2NPANPAFHTRQSAFW6 + 1CM*BrPIP2;3b*LFNPANPAFHTRQSAFW5 + 1CM*BrPIP2;4a*MF2NPANPAFHTRQSAFW6 + 1CM*BrPIP2;4b*MF1NPANPAFHTRQSAFW6 + 1CM*BrPIP2;4c*LFNPANPAFHTRQSAFW6 + 1CM*BrPIP2;5a*LFNPANPAFHTRQSAFW6 + 1CM*BrPIP2;5b*MF2NPANPAFHTRQSAFW6 + 1CM*BrPIP2;6*MF2NPANPAFHTRQSAFW6 + 1CM*BrPIP2;7a*MF2NPANPAFHTRMSAFW6 + 1CM*BrPIP2;7b*LFNPANPAFHTRMSAFW6 + 1CM*BrPIP2;7c*MF1NPANPAFHTRMSAFW6 + 1CMBlue colour letters denote unusual amino acids in NPA motifs. *CM* Cell membrane, *V*Vacuole*LF* Less Fractioned subgenome, *MFs (MF1 and MF2)* More Fractioned subgenomes, *LB* Loop B, *LE* Loop E, {two half helices (LB and LE)}, *NPA* Asparagine, Proline, Alanine, *AQP* contain 6 TM helices (H1 to H6), *H2* Helice 2, *H5* Helice 5, *LE1*Loop E1, *LE2* Loop E2, *Ar/R* Aromatic/Arginine, *TMH* Transmembrane helice The ar/R selectivity filter and five Froger's positions of the BrNIP subfamily members were quite divergent compared to those of the other subfamilies (Table [1](#Tab1){ref-type="table"} and Additional file [3](#MOESM3){ref-type="media"}: Figure S1a \~ 1d). The predicted polypeptides of the SIP subfamily were divided into two groups (SIP1 and SIP2) and showed 22.6--91.1% identity within the subfamily, but 72.1--91.1% identity within the groups. The ar/R filter and five Froger's positions P1 to P5 of the SIP subfamily were well conserved in all sites. The 16 putative TIP subfamily members were divided into 5 groups and showed 68.2--94.8% identity within groups (Additional file [4](#MOESM4){ref-type="media"}: Table S3). Phylogenetic analysis of BrAQP proteins {#Sec5} --------------------------------------- The phylogenic tree was constructed based on the multiple sequence alignment of 59, 45 and 35 putative full-length BrAQP, SiAQP and AtAQP proteins, respectively (Fig. [1](#Fig1){ref-type="fig"}). The BrAQPs were classified into four subfamilies (PIP, TIP, NIP and SIP) corresponding to the *Arabidopsis* grouping defined by Quigley et al. \[[@CR10]\]. The six newly identified *B. rapa* genes were distributed in PIP, NIP and TIP subfamilies, with each subfamily containing 2 members. Accordingly, these new members are named as BrNIP4;2b, BrNIP4;2c, BrPIP2;2b, BrPIP2;3b, BrTIP2;1c and BrPIP2;3b. Among the subfamilies, PIP had the most BrAQPs and contained 22 members, relative to the 16, 15 and 6 members of the TIP, NIP and SIP subfamilies, respectively. Members of XIP subfamily were totally absent in *B. rapa* (Fig. [1](#Fig1){ref-type="fig"}).Fig. 1Phylogenetic analysis of aquaporin proteins identified in *B. rapa,* Arabidopsis and tomato. Based on relatedness to characterized proteins, aquaporins were classified as plasma membrane intrinsic proteins (PIPs) in the *blue tree*, tonoplast intrinsic proteins (TIPs) in the pink tree, nodulin 26-like intrinsic proteins (NIPs) in the red tree, small basic intrinsic proteins (SIPs) in the green tree and X intrinsic proteins (XIP) in the deep green tree*. At, Sl, Br* denote *Arabidopsis, tomato and B. rapa* and *red*, *blue* and *black* letters denote *Arabidopsis,* tomato *and B. rapa* aquaporin proteins, respectively Chromosomal locations and gene duplications of *BrAQP* genes {#Sec6} ------------------------------------------------------------ We conducted *in silico* analysis to determine the localization of *AQP* genes in 10 chromosomes of *B. rapa* using gene mapping software (Fig. [2a](#Fig2){ref-type="fig"}). The most *AQP* genes were found in chromosome 3 (17.0%) and the fewest were found in chromosome 8 (3.4%) (Fig. [2d](#Fig2){ref-type="fig"}). The physical locations of the *BrAQP* genes in the *B. rapa* genome reflected the diversity and complexity of this gene family. The PIP subfamily genes were distributed on all chromosomes except chromosome 6, and TIP subfamily genes were found in all chromosomes except chromosomes 8 and 10. Other than chromosomes 6, 9 and 10, there were NIP group genes in each chromosome. Genes in the SIP subfamily were present only on chromosomes 1, 4, 5, 7, 9 and 10 (Fig. [2a](#Fig2){ref-type="fig"}). Genome triplication has occurred since divergence of the *Brassica* genus from the ancestor of *A. thaliana* between five and nine million years ago (MYA) \[[@CR24]\]. The *B. rapa* genome consists of three differentially fractionated sub-genomes, namely the least fractionated (LF), medium fractionated (MF1), and most fractionated (MF2). The 59 *BrAQPs* were fractionated into three subgenomes (i.e., LF, MF1, and MF2), including 26 (44%) in LF, 19 (32%) in MF1, and 14 (24%) in MF2 (Fig. [2c](#Fig2){ref-type="fig"} and Table [1](#Tab1){ref-type="table"}). In addition, we reconstructed the *B. rapa* genome containing 24 conserved chromosomal blocks (labelled A--X) according to previous reports \[[@CR25]\]. The colour coding of these blocks was based on their positions in a proposed ancestral karyotype (AK1-8) \[[@CR25]\]. Most of the 59 *BrAQP* genes belonged to AK3 (18%), followed by AK1 and AK7 (15%), while only 8% of *BrAQP* genes were assigned to AK2 (Fig. [2b](#Fig2){ref-type="fig"}).Fig. 2Distribution of *BrAQP* genes on 10 chromosomes. **a** The 24 (A to X) ancestral blocks and three sub-genomes were plotted, based on the report of Schranz et al. \[[@CR25]\]. **b** The percentages of *BrAQP* genes on ancestral blocks. **c** The percentages of *BrAQP* genes on the least fractionated (LF), medium fractionated (MF1) and most fractionated (MF2) subgenomes. **d** The percentages of BrAQP genes on each chromosome The arrangement of *BrAQP* genes in the *B. rapa* genome implies that some genetic events have affected this gene family during evolution. The distribution of the *AQP* gene family has likely been influenced by processes such as segmental duplication, tandem duplication, and polyploidization \[[@CR26], [@CR27]\]. In addition, genome triplication events might have played a key role in the expansion of *AQP* gene family in *B. rapa.* We found evidence of at least two tandem duplication events (*BrNIP4;1* vs. *BrNIP4;2b*, *BrNIP4;2b* vs. *BrNIP4;2c*) with total of 43 segmental duplications in the *BrAQP* gene family (Table [2](#Tab2){ref-type="table"}, Fig. [3](#Fig3){ref-type="fig"}). Estimation of the Ka/Ks ratios (synonymous and nonsynonymous substitutions per site) was done to assess the selection constraints among duplicated *BrAQP* gene pairs. In these analyses, Ka/Ks ratios \<1, 1 and \>1 indicate negative or purifying selection, neutral selection and positive selection, respectively \[[@CR28]\]. All *BrAQP* duplicated gene pairs showed a Ka/Ks ratio of \<1, suggesting that these genes evolved under strong negative or purifying selection pressure in *B. rapa*. These results suggest that purifying selection has played an important role in the functional divergence of *BrAQP* genes. We calculated the divergence time of *BrAQP* genes and found that these gene duplications began approximately 9.39 million year (mya) ago and ended at 0.38 mya ago (Table [2](#Tab2){ref-type="table"}), which indicates that the divergence time of the *AQP* genes in *B. rapa* occurred after the triplication events (i.e., 5 \~ 9 MYA) \[[@CR29]\].Table 2Estimated Ka/Ks ratios of the duplicated *BrAQP* genes with their divergence time in *B. rapa*Duplicated gene pairsKsKaKa/KsDuplication typePurify selectionTime (mya)BrSIP1;1b (MF1)vs.BrSIP1;1a (LF)0.1910.0340.18SegmentalYes0.64BrSIP2;1b (MF2)vs.BrSIP2;1a (MF1)0.2410.0590.24SegmentalYes0.80BrSIP2;1b (MF2)vs.BrSIP2;1c (LF)0.3070.0840.27SegmentalYes1.02BrNIP1;2b (MF1)vs.BrNIP1;2a (LF)0.3140.0060.02SegmentalYes1.05BrNIP3;1a (MF1)vs.BrNIP3;1b (MF2)0.4210.0510.12SegmentalYes1.40BrNIP4;1 (MF2)vs.BrNIP4;2b (MF2)0.3760.0680.18TandemYes1.25BrNIP4;2b (MF2)vs.BrNIP4;2c (MF2)0.3380.0770.23TandemYes1.13BrNIP5;1b (MF1)vs.BrNIP5;1a (MF2)0.2820.0060.02SegmentalYes0.94BrNIP6;1b (LF)vs.BrNIP6;1a (MF1)0.2830.0340.12SegmentalYes0.94BrPIP1;1a (LF)vs.BrPIP1;1b (MF1)0.2290.0120.05SegmentalYes0.76BrPIP1;1a (LF)vs.BrPIP1;2b (LF)0.7270.0180.02SegmentalYes2.42BrPIP1;1a (LF)vs.BrPIP1;3b (LF)0.9250.0910.10SegmentalYes3.08BrPIP1;1b (MF1)vs.BrPIP1;3b (LF)0.8670.0850.10SegmentalYes2.89BrPIP1;1b (MF1)vs.BrPIP1;2b (LF)0.7680.0120.02SegmentalYes2.56BrPIP1;2b (LF)vs.BrPIP1;2a (MF1)0.2240.0270.12SegmentalYes0.75BrPIP1;2b (LF)vs.BrPIP1;3a (MF2)1.0130.0540.05SegmentalYes3.38BrPIP1;2b (LF)vs.BrPIP1;3b (LF)0.9480.0660.07SegmentalYes3.16BrPIP1;3a (MF2)vs.BrPIP1;4 (MF1)0.6720.0410.06SegmentalYes2.24BrPIP1;3b (LF)vs.BrPIP1;3a (MF2)0.1140.0170.15SegmentalYes0.38BrPIP1;3b (LF)vs.BrPIP1;4 (MF1)0.7050.0470.07SegmentalYes2.35BrPIP2;1 (LF)vs.BrPIP2;2b (LF)0.6930.1520.22SegmentalYes2.31BrPIP2;1 (LF)vs.BrPIP2;2a (MF2)0.7860.1600.20SegmentalYes2.62BrPIP2;2b (LF)vs.BrPIP2;2a (MF2)0.3770.0260.07SegmentalYes1.26BrPIP2;3a (MF2)vs.BrPIP2;3b (LF)0.4100.0230.06SegmentalYes1.37BrPIP2;3a (MF2)vs.BrPIP2;5a (LF)1.1600.1950.17SegmentalYes3.87BrPIP2;3a (MF2)vs.BrPIP2;4c (LF)2.8170.1170.04SegmentalYes9.39BrPIP2;3b (LF)vs.BrPIP2;2a (MF2)0.3510.0060.02SegmentalYes1.17BrPIP2;3b (LF)vs.BrPIP2;1 (LF)0.7340.0260.04SegmentalYes2.45BrPIP2;3b (LF)vs.BrPIP2;4c (LF)1.3250.0910.07SegmentalYes4.42BrPIP2;4a (MF2)vs.BrPIP2;4b (MF1)0.1060.0200.19SegmentalYes0.35BrPIP2;4a (MF2)vs.BrPIP2;4c (LF)0.1720.0120.07SegmentalYes0.57BrPIP2;4b (MF1)vs.BrPIP2;4c (LF)0.1420.0200.14SegmentalYes0.47BrPIP2;5a (LF)vs.BrPIP2;5b (MF2)0.3740.0280.07SegmentalYes1.25BrPIP2;7a (MF2)vs.BrPIP2;7c (MF1)0.1760.0350.20SegmentalYes0.59BrPIP2;7a (MF2)vs.BrPIP2;7b (LF)0.4150.0320.08SegmentalYes1.38BrPIP2;7b (LF)vs.BrPIP2;7c (MF1)0.3040.0110.04SegmentalYes1.01BrTIP1;1 (MF1)vs.BrTIP1;2b (MF1)0.5900.0930.16SegmentalYes1.97BrTIP1;1 (MF1)vs.BrTIP1;2a (LF)0.5900.0610.10SegmentalYes1.97BrTIP1;2a (LF)vs.BrTIP1;2b (MF1)0.9640.1670.17SegmentalYes3.21BrTIP2;1a (LF)vs.BrTIP2;1b (MF2)0.1330.0350.26SegmentalYes0.44BrTIP2;1a (LF)vs.BrTIP2;1c (MF1)0.1700.0290.17SegmentalYes0.57BrTIP2;1b (MF2)vs.BrTIP2;1c (MF1)0.1340.0170.13SegmentalYes0.45BrTIP2;3a (LF)vs.BrTIP2;3b (MF1)0.5140.0410.08SegmentalYes1.71BrTIP3;1a (MF1)vs.BrTIP3;1b (LF)0.3370.0510.15SegmentalYes1.12BrTIP3;2a (LF)vs.BrTIP3;2b (MF1)0.3540.0170.05SegmentalYes1.18*LF* less fractioned subgenome, *MF* more fractioned subgenome (MF1 and MF2), *Ks* the number of synonymous substitutions per synonymous site, *Ka* the number of nonsynonymous substitutions per nonsynonymous site, *MYA* million years ago Fig. 3Microsynteny analysis of *AQP* genes among *B. rapa, B. oleracea* and *A. thaliana*. The chromosomes from the three species are indicated in different colors; red, green and yellow colors represent *B. rapa, A. thaliana* and *B. oleracea* chromosomes, respectively. Black lines denote duplicated *BrAQP* genes on 10 *B. rapa* chromosomes Microsynteny relationships {#Sec7} -------------------------- To investigate evolutionary history and relationships, a microsynteny map was constructed using orthologous gene pairs of the *AQP* genes among *B. rapa*, *B. oleracea* and *A. thaliana* (Fig. [3](#Fig3){ref-type="fig"}). Based on this analysis, 39 orthologous gene pairs between *B. rapa* and *A. thaliana* were identified, whereas 72 orthologous gene pairs were found between *B. rapa* and *B. oleracea* (Fig. [3](#Fig3){ref-type="fig"}). This result suggests that *BrAQP* genes are more closely related to those of *B. oleracea* and *A. thaliana.* We found 45 duplications of *BrAQP* genes. Out of 45 pairs, 43 were segmental and 2 pairs were identified as tandem duplications, which is denoted with a black line in Fig. [3](#Fig3){ref-type="fig"}. For clarity, we have also depicted only the *BrAQP* duplicated gene pairs in *B. rapa* chromosomes (Additional file [5](#MOESM5){ref-type="media"}: Figure S2). Motif and exon-intron distribution {#Sec8} ---------------------------------- Conserved motifs among each subfamily were identified using MEME software and compared for providing further support of the grouping of *BrAQPs*. Most BrAQP proteins of the same subfamily had similar motifs, with motifs 1 & 2 present in all subfamilies (Additional file [6](#MOESM6){ref-type="media"}: Figure S3). The protein sequences of all *BrAQPs* shared high similarity; thus, out of the 10 motifs, most (1, 2, 3, 4, 5, 6, 7 and 9) were found in all PIP subfamily members except *BrPIP2;3b* and *BrPIP2;4c*, which were lacked of motif 5, and *BrPIP1;2a,* which had no motif 4 (Additional file [6](#MOESM6){ref-type="media"}: Figure S3). Motifs 1, 2, 3, 6 and 10 were common to both TIP and NIP subfamily members, although *BrTIP2;1c, BrTIP2;3b, BrNIP4;2b,* and *BrNIP4;2c* did not contain motif 10. A unique motif (motif 8) was found in TIP group members, and motif 6 was found only in subfamily SIP1. The best possible match sequence for each motif is presented in Additional file [7](#MOESM7){ref-type="media"}: Table S4. The intron--exon structures of the *B. rapa AQPs* were analyzed using the GSDS program. Most members of the PIP subfamily had three introns, while four members had two introns and two members had four introns. In the TIP subfamily, eight members had two introns and seven members had one intron, but only one gene had no intron. All BrNIP family members had 2 to 4 introns; 7 out of 15 members had 3 introns, another 7 members had 4 introns, and only 1 had 2 introns. BrSIPs formed a small subfamily of *BrAQP* in which all members had two introns (Additional file [8](#MOESM8){ref-type="media"}: Figure S4). Microarray expression analysis in response to cold and freezing stress {#Sec9} ---------------------------------------------------------------------- Expression patterns of the 59 *BrAQP* genes were determined using our previously published microarray data set, wherein two contrasting *B. rapa* inbred lines cold-tolerance Chiifu and cold susceptible Kenshin, were treated with different temperatures (22 °C, 4 °C, 0 °C, −2 °C and −4 °C) \[[@CR30]\]. The two lines (Chiifu and Kenshin) responded differently in microarray expression. Chiifu originated in temperate regions, whereas Kenshin originated in tropical and subtropical regions. At low temperature, Kenshin shows severe injury while Chiifu does not \[[@CR31]\]. Moreover, Kenshin has been used as a breeding stock to develop heat-tolerant plants \[[@CR32]\]. We created a heat map based on differential microarray transcript values and to examine expression pattern of *BrAQP* genes in response to temperature treatments in two inbred lines (chiifu and kenshin) of *B. rapa* (Fig. [4](#Fig4){ref-type="fig"}). In the heat map, expression patterns of *BrAQP* genes were divided into seven clusters (Cl-1 to Cl-7). Most *BrPIP* genes were present in Cl-1, Cl-2, Cl-4 and Cl-6. The *BrPIP* genes in Cl-1, Cl-4 and Cl-6 showed higher expression in Chiifu than in Kenshin in response to both cold and freezing temperatures. Five *BrPIP* genes in Cl-2 showed higher expression in Kenshin than in Chiifu under normal conditions (22 °C). However, Cl-2 and Cl-3 *BrAQP* genes exhibited higher expression in Kenshin than in Chiifu in response to both cold and freezing temperatures. *BrSIP2;1b* did not show a significant response in any temperature treatment, whereas *BrSIP2;1a* did not respond to freezing temperatures. As a whole, we concluded that the majority of *BrPIP* subfamily genes were highly induced in Chiifu by cold and freezing treatment compared to in Kenshin. These results indicate that *BrPIP* subfamily genes might play an important role in the cold and freezing tolerance of Chiifu. On the contrary, a few *BrPIP* and those of other *BrAQP* subfamilies showed higher expression in Kenshin in response to cold and freezing temperature; those genes might be related to the cold and freezing susceptibility of Kenshin.Fig. 4Differential expression profiles of *BrAQP* genes in different temperatures. C and K indicate Chiifu and Kenshin, respectively, which were treated under five temperatures: control (C1&K1), 4 °C (C2 & K2), 0 °C (C3 & K3), −2 °C (C4 & K4), and **-** 4 °C (C5 & K5). Expression clusters are shown in the left (Cl1--Cl7) and gene names are at the right. Color legend at right represents differential expression in microarray data Expression profiles of BrPIP genes in various organs {#Sec10} ---------------------------------------------------- The expression of 22 *PIP* genes in different organs of *B. rapa* plants (roots, stems, leaves, and flower bud) was analyzed by qPCR and semi-quantitative RT-PCR (Fig. [5](#Fig5){ref-type="fig"}, Additional file [9](#MOESM9){ref-type="media"}: Figure S5). Eighteen PIP genes (*BrPIP1;1b, 1;2a, 1;2b, 1;3a, 1;4, 1;5, 2;1, 2;2a, 2;3a, 2;4a; 2;4b; 2;4c, 2;5a; 2;5b, 2;6, 2;7a 2;7b,* and *2;7c*) were expressed in all tested organs but *BrPIP1;1b* and *2;3a* were only slightly expressed in flower buds. Two genes (*2;2b and 2;3b*) were abundant in all of the tested organs except flower bud. *BrPIP1;1a* was highly expressed in roots and leaves and slightly expressed in stem but absent in flower buds. By contrast, *BrPIP2;5a,* and *BrPIP2;5b* were highly expressed in roots and flower buds but slightly expressed in stems and leaves. *BrPIP1;3a, 1;3b, 1;4, 1;5, 2;4a, 2;5a, 2;5b, 2;6, 2;7a, 2;7b, and 2;7c* were highly expressed in flower buds compared to other organs. However, *BrPIP1;1a, 1;2b, 2;1, 2;2a, 2;2b, 2;3a, 2;3b* and *2;4c* were more abundantly expressed in roots compared to other tested parts (Fig. [5](#Fig5){ref-type="fig"}). In most of the cases, qPCR and RT-PCR results were consistent, although slightly different results were found for *BrPIP2;4a, 2;4b, 2;5a, 2;5b* and *2;6* (Fig. [5](#Fig5){ref-type="fig"}, Additional file [9](#MOESM9){ref-type="media"}: Figure S5).Fig. 5Expression profiles of *BrPIP* genes in various tissues as determined by qPCR analyses. Expression of the indicated genes was determined in roots, stems, leaves, and flower buds Stress-responsive expression analysis {#Sec11} ------------------------------------- Crop loss due to abiotic stresses decrease average yields of most important crops and threatens food security worldwide \[[@CR33]\]. Therefore, identification of stress-responsive genes is an important basic step towards developing stress tolerant cultivars. Accordingly, we analyzed the expression of *BrPIP* subfamily genes for responsiveness to cold, drought, salt, water logging and ABA in *B. rapa* plants via qPCR using specific primers (Additional file [10](#MOESM10){ref-type="media"}: Table S5). As in the analysis of microarray data described above, two inbred lines of *B. rapa*, Chiifu and Kenshin, were used to detect the responses of *BrPIP* genes expression due to cold stress. All of the *BrPIP* genes showed higher expression in Chiifu compared to Kenshin except *BrPIP2;4b*, which did not show any higher expression change due to cold treatment either in Chiifu or in Kenshin compared to the control (Fig. [6a](#Fig6){ref-type="fig"}). Out of 22 *BrPIP* genes, 14 were differentially expressed in response to cold stress at different time points. The majorities of the genes were down-regulated at the beginning of the cold treatment, but began to be up-regulated after 4 h and continue to increase in expression up to 12 h of time course. Thereafter, the same genes were down-regulated until the end of the time courses (Fig. [6a](#Fig6){ref-type="fig"}). In Chiifu, *BrPIP1;1a, BrPIP1;4, BrPIP1;5* and *BrPIP2;6* genes showed about 3-, 8-, 10- and 41- fold higher expression at 12 h, respectively, and *BrPIP2;7c* showed about 10-fold higher expression at the 4 h time point compared to the 0 h time point. The fold changes of the expression of those genes were significantly (p ≤ 0.01) different from each other at the mentioned time points (Fig. [6a](#Fig6){ref-type="fig"}). By contrast, the majority of *PIP* genes showed down-regulation in Kenshin upon cold treatment. Only a few *PIP* genes such as *BrPIP1;3b, 1;5, 2;5b; 2;7a* and *2;7b* showed differential expression in response to cold stress in Kenshin, and their expression levels were very low. In Kenshin, *BrPIP2;6* and *BrPIP2;7c* exhibited about 10- and 2-fold higher expression at the 12 h time point compared to the control and their expression subsequently started to decreases however the expression differences between those two genes were statistically significant (p ≤ 0.01; Fig. [6a](#Fig6){ref-type="fig"}).Fig. 6Expression analysis of *BrPIP* genes under abiotic stresses using real-time quantitative RT-PCR. The relative expression levels of *BrPIP* genes under treatment with (**a**) cold, (**b**) drought, (**c**) salinity (**d**) ABA (**e**) waterlogging or (**f**) *Fusarium oxysporum* f.sp. *conglutinans* infection. The error bars represent the standard error of the means of three independent replicates. Variance analysis and the Tukey tests were carried out to determine differences among effects on different time courses due to abiotic and biotic stresses for all genes, where different letters indicate the significant difference with *p* ≤ 0.05 We next used Kenshin for qRT-PCR assays to elucidate the responses of *BrPIP* genes to drought stress. Differential expression of *BrPIP1;4, 2;4a, 2;4b, 2;5a, 2;6* and *2;7a* were observed during drought and the differences of the expression were significant (p ≤ 0.01) among the genes (Fig. [6b](#Fig6){ref-type="fig"}). *BrPIP2;4b, 2;5a* and *2;6* showed up-regulation up to 12 h, but *BrPIP1;4,* and *2;4a,* showed up-regulation up to 4 h and were subsequently down-regulated to the end of the time courses (Fig. [6b](#Fig6){ref-type="fig"}). Meanwhile, *BrPIP2;7a* showed down-regulation at the initial stage of stress and was gradually up-regulated thereafter, whereas *BrPIP1;3b* showed up-regulation at the beginning of drought (1 h) but was subsequently down-regulated. The rest of the *BrPIP* genes were down-regulated soon after drought stress and remained consistent throughout the stress period. These results are in agreement with those for plasma membrane *AQPs* in response to abiotic stresses in *Arabidopsis thaliana* \[[@CR17]\]. The majority of the *BrPIP* genes were significantly up-regulated during salt-stress (p ≤ 0.01). *BrPIP1;3a, 1;3b, 2;4a, 2;4b, 2;7b* and *2;7c* were up-regulated and showed the highest expression at 24 h and then were down-regulated. *BrPIP1;2a, 1;2b, 1;4, 1;5*, *2;3b* and *2;4c* were alternately up- and down-regulated throughout the treatment time course (Fig. [6c](#Fig6){ref-type="fig"}). Under salt stress, *BrPIP2;1, 2;2a* and *2;2b* showed down-regulation at 1 h but exhibited higher expression at 4 h; thereafter they were gradually down-regulated up to the end of time courses. By contrast, *BrPIP2;3a* expression reached a peak at 4 h and remain unchanged up to 24 h, followed by a radical down-regulation at 48 h. *BrPIP2;5a* showed slight down-regulation at 1 h followed by up-regulation (up to 12 fold compared to the control) at 12 h, but again started down-regulation to the end of the time course (Fig. [6c](#Fig6){ref-type="fig"}). *BrPIP2;6* and *2;7a* were down-regulated at the beginning of salt stress and continues to 12 h; thereafter they suddenly exhibited higher expression at 24 h. During salt stress, *BrPIP1;3b, 2;4b, 2;6, 2;7a* and *2;7c* showed about 8-, 14-, 4-, 5- and 26- fold higher expression compared to the control at 24 h, respectively, while *2;5a* showed 12- fold higher expression at 12 h and those expression fold changes were statistically significant (p ≤ 0.01; Fig. [6c](#Fig6){ref-type="fig"}). The *BrPIP* gene expression under salt stress treatment was similar to that of plasma membrane *AQPs* in *A. thaliana* under abiotic stresses \[[@CR17]\]. Abscisic acid (ABA) is an important phytohormone that plays a vital role in plant growth and development as well as in responses to a wide range of stresses. As shown in Fig. [6d](#Fig6){ref-type="fig"}, most of the *BrPIP* genes were up-regulated in response to ABA treatment and showed their highest expression at 24 h. A small number of *BrPIP* genes (*BrPIP1;1a,* and *BrPIP2;3a*) exhibited higher expression at 4 h, while *BrPIP1;2a* and *BrPIP2;7c* peaked at 1 h and decreased thereafter. *BrPIP2;1, BrPIP2;2a, BrPIP2;4b, BrPIP2;6,* and *BrPIP2;4a* genes showed the highest expression at the 24 h time point*.* By contrast, *BrPIP1;2b, BrPIP2;3b* and *BrPIP2;5a* were down-regulated throughout the ABA treatment. *BrPIP1;5* exhibited about 8- fold higher expression at 48 h and *BrPIP 2;4a* showed about 14- fold higher expression at 24 h; the expression change of those genes was statistically significant (p ≤ 0.01) compared to other genes in the same time courses (Fig. [6d](#Fig6){ref-type="fig"}). In the case of water logging stress, all *BrPIP* genes except *BrPIP2;4a* exhibited down-regulation compared to control. Some *BrPIP* genes showed increasing expression from 12 h to the end of treatment, but their relative expression remained below that of the control (Fig. [6e](#Fig6){ref-type="fig"}). Expression of *BrPIP* genes under biotic stress {#Sec12} ----------------------------------------------- We also analyzed the responses of *BrPIP* genes to biotic stress treatment using *Fusarium oxysporum* f.sp*. conglutinans*, which specifically attacks *Brassica* species and causes wilt diseases. Upon artificial infection by this pathogen, 4 out of the 22 *BrPIP* genes showed significantly higher expression (p ≤ 0.01; Fig. [6f](#Fig6){ref-type="fig"}). *BrPIP1;3b, BrPIP2;6* and *BrPIP2;1* displayed about 4.5-, 2- and 1.5- fold higher expression at 4 dai (days after infection), respectively. *BrPIP2;2a* exhibited about 6- fold higher expression at 11 dai compared to mock-treated plants (Fig. [6f](#Fig6){ref-type="fig"}). These results suggest that *BrPIP1;3b, BrPIP2;6, BrPIP2;1* and *BrPIP2;2a* may be involved in responses to *F. oxysporum* f.sp*. conglutinans* infection. Discussion {#Sec13} ========== *AQP* genes are ubiquitously important in higher plants because of their function as water and/or small neutral solute transporters in plant body. Precise gene annotation is an important starting point for future functional studies of this family. The *AQP* gene family has 35 members in *Arabidopsis* and 47 members in tomato \[[@CR8]\]. Meanwhile, we have found 59 *AQPs* in *B. rapa* and carried out *in silico* functional analysis, which showed that most of the PIP subfamily proteins shared a high degree of identity with abiotic stress-related AQP proteins from other plant species. Proteins of another three subfamilies (SIP, NIP and TIP) exhibited similarity to *AQPs* in crop plants involved in water and solute transport in leaves and fruits during fruit development, pod development, root development, nutrient uptake and arsenic transportation. All of the members of PIP, NIP and SIP subfamily and most of the TIP subfamily members contained the same ar/R selectivity filter and Froger's positions. In some cases, these were different in TIP subfamily which is consistent with previous research \[[@CR34]\]. The ar/R selectivity filter and Froger's positions in the BrTIP subfamily members were quite divergent compared to those of the other subfamilies, indicating that they have different solute permeability. Nineteen members of the *BrPIP* subfamily showed high similarity to both water flow and abiotic stress-related *PIP* genes from other plant species, whereas three showed high similarity to proteins involved in water flow between the pollen and stigma papillae, and abiotic stress-related *PIP* genes from other plant species (Additional file [2](#MOESM2){ref-type="media"}: Table S2). We therefore concluded that AQPs of *B. rapa* are likely involved in water and solute transport and that *BrPIP* subfamily members might be involved in abiotic stress responses as well. We analyzed the relative expression patterns of 59 *BrAQP* genes using a whole-genome microarray dataset obtained upon treatment at various temperatures (22, 4, 0, −2, and −4 ° C) in two inbred lines of *B. rapa*; Chiifu and Kenshin \[[@CR31]\]. Thereafter, *BrPIP* subfamily genes were selected based on their variation in transcript abundance compared to the control, and analyzed for responsiveness to temperature treatments in those two contrasting *B. rapa* inbred lines (Fig. [4](#Fig4){ref-type="fig"}). The results indicated that *BrPIP* genes might play a vital role in abiotic stress responses in *B. rapa.* On the other hand, the *BrPIP* subfamily members were highly conserved, indicating their probable involvement in similar biological functions. From an evolutionary viewpoint, gene number increases can be due to gene duplication events, including tandem and segmental duplication \[[@CR35]\]. Gene duplication may play the driving role in the evolution of gene families and genetic systems \[[@CR36]\]. Here, we identified 43 segmental duplicated gene pairs and two pairs tandemly duplicated genes (Table [2](#Tab2){ref-type="table"}), suggesting that segmental duplication was the main contributor to the expansion of this gene family. We analyzed the evolutionary history of this family and calculated the Ka, Ks and Ka/Ks ratios of duplicated gene pairs. Interestingly, all gene pairs had Ka/Ks ratios \<1 (Table [2](#Tab2){ref-type="table"}), indicating that the *BrAQP* gene family has undergone large-scale purifying selection. The evolutionary timescale of *B. rapa* was estimated based on the synonymous substitution rate \[[@CR37]\], revealing that the divergence time of the duplicated *BrAQP* genes spanned 0.38 to 9.39 million years, which suggests that duplication-based divergence of the *BrAQP* family members in *B. rapa* occurred after the triplication events (i.e., 5 \~ 9 MYA) \[[@CR27]\]. Our microsynteny analysis showed that there are 39 and 72 orthologous gene pairs between *B. rapa* / *A. thaliana* and *B. rapa* / *B. oleracea*, respectively (Fig. [2](#Fig2){ref-type="fig"}). Based on our organ-specific expression analysis, all *BrPIP* genes are expressed at different levels in at least one of the tested organs of *B. rapa* plants. *BrPIP1;1a, 1;2a; 2;2a,* and *2;3a* were more abundantly expressed in roots compared to other tested organs; which is consistent with previous findings \[[@CR4], [@CR17], [@CR20]\]. *BrPIP1;2b, 1;3a, 1;4, 2;2a,* and *2;3a* were abundantly expressed in stem while *BrPIP1;1a, 1;2b* and *2;2a* were highly expressed in leaves, like their *Arabidopsis* counterparts. Previous reports have been suggested that *AQP* genes are expressed in all plant tissues and are involved in growth and development and responses to environmental stress conditions \[[@CR5]\]. This abundantly expressed *BrPIP* genes in roots, stem and leaves might be related to different cellular controls of water flow. However, *BrPIP1;2a, 1;2b, 1;3a, 1;4, 2;5b 2;6; 2;7a, 2;7b* and *2;7c* were typically more expressed in flower buds of *B. rapa* plants (Fig. [4](#Fig4){ref-type="fig"}). Pollen absorbs water from the stigma surface before it germinates \[[@CR38]\]. According to Marin-Olivier et al. \[[@CR39]\] water flows from stigma papillae to the pollen, and this may be dependent on *AQP* genes, although they are not directly related to pollen grain germination. Our results provide candidate abundantly expressed *BrPIP* genes in flower, which may play a role in the control of pollen rehydration, which is an essential step for the success of pollination. Our expression analysis showed that *BrPIP* genes are expressed differently upon various abiotic stress treatments. In response to cold stress, all *BrPIP* genes showed down-regulation, except *BrPIP, 1;3b, 1;5, 2;4a, 2;6, 2;7a, 2;7b* and *2;7c* in Kenshin (Fig. [6a](#Fig6){ref-type="fig"}). Interestingly, *BrPIP2;6* showed 10-fold higher expression compared to the control at 12 h in Kenshin. By contrast, most of the *BrPIP* genes showed up regulation in Chiifu and exhibited higher expression at 12 h. All of the genes showed several-fold higher expression in Chiifu compared to Kenshin. In summary, the *BrPIP* genes were more highly induced than any other group of *BrAQP* genes in response to cold or freezing stress. These results are expected due to the origin of two lines, where Chiifu is cold tolerant and Kenshin is cold susceptible \[[@CR40]\]. Plasma membrane *AQP* genes have been reported to play roles under both low and freezing temperatures in rice \[[@CR18]\]. *AQP* genes also function to maintain homeostasis and water balance under stress conditions \[[@CR41]\]. The expression of specific *AQPs* is high in guard cells \[[@CR42], [@CR43]\]; therefore, it seems that *AQPs* play a role in water movement in guard cells, and regulate stomatal movement. Under low temperature conditions, leaf stomata of cold-sensitive plants remain open but those of cold-tolerant plants close rapidly \[[@CR44], [@CR45]\] and maintain cell turgor pressure. All *BrPIP* genes showed higher expression in cold-tolerant Chiifu than in cold-susceptible Kenshin lines. Therefore, we speculate that *BrPIP* genes might be involved in maintenance of water balance in the cell and cell turgor pressure during cold stress. We found that the majority of *BrPIP* genes were significantly down-regulated during drought stress treatment (Fig. [6b](#Fig6){ref-type="fig"}). Mittler et al. \[[@CR46]\] reported that quick accumulation of reactive oxygen species (ROS) leads to damage of the cell membrane and oxidation of proteins, lipids, and DNA during drought stress. Down-regulation of *BrPIP* gene expression during drought stress may reduce membrane water permeability and cellular water conservation during dehydration periods. In agreement with our findings, the *MIP* genes in *Nicotiana glauca* \[[@CR47]\] and *PIP* genes in *Arabidopsis* \[[@CR20]\] were down-regulated under drought stress. By contrast, very few *BrPIP* genes displayed up-regulation and showed higher expression at 4 or 12 h (Fig. [6b](#Fig6){ref-type="fig"}). Notably, *BrPIP2;4a* and *2;4b* exhibited 4- and 7-fold higher expression, respectively, compared to the control. In addition, over-expression of *AQP7* in tobacco plants and *MaPIP1;1* in banana plants reduced membrane injury compared to wild-type plants under drought stress \[[@CR48], [@CR49]\]. These results indicate that up-regulated *BrPIP* genes might participate in avoiding membrane injury under drought stress. Muries et al. \[[@CR50]\] reported that 3 *AQP*s genes showed low expression in roots and were highly expressed in leaves and/or flowers, and remained stable or were up-regulated under drought. This result indicated that the *AQP* genes that are down regulated under normal condition can be highly expressed in drought stress in roots. This pattern might be due to the existence of post transcriptional mechanisms regulating PIP trafficking to the plasma membrane to overcome the drought via decreasing injury of the membrane. Therefore, it is necessary to take root samples in addition to leaf samples under drought stress conditions for expression profiling of *BrPIP* genes in order to make decisive conclusions for development of drought tolerant cultivars. Otherwise, the transcriptional down-regulation of PIP genes upon drought stress could also be observed on the protein level \[[@CR51]\]. Under salt stress, all of the *BrPIP* genes were up-regulated except *BrPIP1;1a* and *BrPIP1;1b.* However, most of the *BrPIP* genes showed initial down-regulation and subsequent up-regulation, and highest expression was observed at 24 h (Fig. [6c](#Fig6){ref-type="fig"}). During salt stress, the initial down- and subsequent up-regulation of *BrPIP* gene expression indicate that these genes likely function in limiting water loss at the early stage and subsequent water uptake to maintain homeostasis in the cell. Early down-regulation and subsequent up-regulation of *AQP* gene expression has also been observed in microarray analysis of the two rice cultivars \[[@CR52]\] and *Arabidopsis* \[[@CR53]\]. *AQP* genes have been identified to play important roles in ABA responses in different plant species including *Arabidopsis* \[[@CR12]\], rice \[[@CR54]\], *Brassica napus* \[[@CR55]\], and radish \[[@CR1]\]. All of the *BrPIP* genes except *BrPIP1;2b; 2;3b* and *2;5a* were up-regulated in response to exogenous ABA application (Fig. [6d](#Fig6){ref-type="fig"}). Most of the *BrPIP* genes showed moderate up-regulation (below 3 fold). However, the *BrPIP1;5, 2;4a, 2;4b, 2;6,* and *2;7a* exhibited 9-, 16-, 5-, 4- and 4- fold higher expression, respectively, in response to ABA treatment. These results indicate that responsiveness of *BrPIP* genes to ABA treatment varied greatly. Therefore, it could be deduced that *BrPIP* gene expression responses are complex, likely due to involvement in both ABA-dependent and ABA-independent signaling pathways. Under water logging stress, all of the *BrPIP* genes were significantly down-regulated. A very few cases showed up-regulation at the end of the time courses, although their expression pattern remained below the control (Fig. [6e](#Fig6){ref-type="fig"}). The hydraulic conductivity of tissues is regulated by three different pathways of water flow in plants, the symplastic, transcellular and apoplastic pathways \[[@CR56]\]. In the symplastic pathway, water and solutes are transported from cytoplasm of one cell to that of a neighboring cell via plasmodesmata. In the transcellular pathway, water and dissolved nutrients pass across through plasma membrane and vacuolar membrane. The apoplastic pathway facilitates the transport of water and solutes across cell wall. Apoplastic water movement is faster than symplastic water movement. Under water logging conditions, apoplastic water movement may be more active and the symplastic water movement system may be stop or inactive. *AQPs* are mostly involved in symplastic water transport in plants \[[@CR57], [@CR58]\], consistent with our findings that all *BrPIP* genes showed down-regulation under water logging, when symplastic water movement would be expected to be down-regulated. The cold-upregulated *AQP* genes such as *BrPIP1;4* could be candidates for introgression or overexpression to develop cold stress tolerant genotypes, whereas *BrPIP1;5* genes might candidates for cold as well as ABA-responsive *B. rapa*. The *BrPIP* gene *BrPIP2;6* was cold- and *Fusarium*-stress responsive; *Br.PIP*2*;7c* was cold- and salt-stress responsive; *BrPIP2;4a* was drought- and ABA-responsive. In addition, to obtain drought and salt stress-tolerant genotypes, breeders might focus attention on *BrPIP*2;*4b. BrPIP1;3b* could be useful for salt and *Fusarium* fungus tolerance. Additionally, to develop *Fusarium* fungus tolerance, introgression of *BrPIP2;1* and *BrPIP2;2a* might be useful (Fig. [6a-f](#Fig6){ref-type="fig"}). Our findings are also supported by the review of Afzal et al. \[[@CR59]\] the argues that *AQP* genes play an important role in plant defense responses against biotic and abiotic stressors and the report of Reddy et al. \[[@CR60]\] of the functions of this gene family in abiotic stress tolerance in Sorghum. There have been no previous reports on responses of *AQP* to biotic stress. From our analysis, we have identified 4 *BrPIP* genes that showed responsiveness to biotic stress in the form of *Fusarium oxysporum* f.sp*. conglutinans* fungus. Three *BrPIP* genes showed the highest expression at 4 dai, and one showed the highest expression at 11 dai (Fig. [6f](#Fig6){ref-type="fig"}). This soil pathogenic fungus specifically attacks *Brassica* species, causing wilting, yellowing, necrosis of various plant parts and finally plant death \[[@CR61]\]. The highly responsive *BrPIP* genes reported here might play an important role against the fungus *F. oxysporum* f.sp. *conglutinans*. Conclusions {#Sec14} =========== In this study, we demonstrated that *BrPIP* genes showed organ-specific expression in *B. rapa* plants and might be related to different cellular controls of water flow. In addition, four out of 22 *BrPIP* genes showed responses to *F. oxysporum* f.sp. *conglutinans* fungal infection in *B. rapa* plants. Our expression analysis illustrates the possible involvement of *BrPIP* genes in different abiotic and biotic stress-related physiological processes. Several *BrPIP* genes seem to participate in multiple processes; for instance, *BrPIP1;3b, 1;4,2;4a, 2;6, 2;7a* showed responsiveness to cold and drought stresses. *BrPIP1;3b, 1;4, 2;4a, 2;4b, 2;6* and *2;7a* showed higher expression under salt and drought stresses and might be useful for developing salt and drought tolerance cultivars through conventional, molecular or transgenic breeding approaches. By contrast, *BrPIP1;4; 1;5, 2;3b,2;4a,2;5b,2;6, 2;7a, 2;7b* and *2;7c* genes exhibited several-fold higher expression compared to the control during cold and salt stresses. Remarkably, *BrPIP1;3a, 1;4, 2;4a,2;6* and *2;7a* exhibited responses to three abiotic stress (cold, salt and drought) and could be good sources for breeding targeted abiotic stress-tolerant cultivars. It is interesting to note that all *BrPIP* genes were significantly down-regulated by water logging stress, while *BrPIP1;5* and *2;4a* showed the highest expression to ABA treatment. The highly induced *BrPIP* genes reported here might be involved in maintaining water homeostasis in plant responses to abiotic stresses and ABA, and several of these genes might be functional against multiple stresses. The comprehensive expression analysis under different stress stimuli supplies novel information to assign putative stress-related physiological functions of *BrPIP* genes and facilitates selection of potential genes for further functional genomics studies in different *Brassica* crops. Methods {#Sec15} ======= Identification and sequence analysis of aquaporins in *B. rapa* {#Sec16} --------------------------------------------------------------- *B. rapa* AQP members were identified using the key word "aquaporin" for the SWISSPROT tool of the *B. rapa* database (<http://brassicadb.org/brad/index.php>; \[[@CR22]\]. We also investigated the microarray annotated database for two cold-treated *B. rapa* inbred lines, Chiifu and Kenshin, using the keyword "aquaporin". The CDS (coding DNA sequence) and protein sequences of the identified AQPs were processed or deduced using the *B. rapa* genomic database, after which the AQP protein sequences were further examined to confirm the presence of the characteristic MIP and trans-membrane helical domains using the SMART program (<http://smart.embl-heidelberg.de/>; \[[@CR62]\] and TMHMM Server v.2.0 (<http://www.cbs.dtu.dk/services/TMHMM/>) \[[@CR63]\]. Prediction of subcellular localization of identified *B. rapa* AQPs was carried out using Plant-mPLoc (<http://www.csbio.sjtu.edu.cn/bioinf/plant/>). Additionally, the primary gene structure (protein length, molecular weight and iso-electric point) was analyzed using ExPasy (<http://au.expasy.org/tools/pi_tool.html>). Open reading Frame Finder (ORF) was obtained using ORF finder at NCBI (<http://www.ncbi.nlm.nih.gov/gorf/gorf.html>). Multiple sequence alignments using the identified protein sequences were made by CLUSTAL Omega (<http://www.ebi.ac.uk/Tools/msa/clustalo/>). The protein homology study was done using the Basic Local Alignment Search Tool (BLASTp) (<http://www.ncbi.nlm.nih.gov/BLAST/>) to confirm the identified *AQP* genes. The exon--intron organization of *BrAQP* genes was identified by comparing predicted coding sequences (CDS) with the corresponding genomic sequences using the GSDS 2.0 software ([http://gsds.cbi.pku.edu.cn](http://gsds.cbi.pku.edu.cn/)). The conserved motifs in the encoded proteins were identified using Multiple Expectation Maximization for Motif Elicitation (MEME; <http://meme-suite.org/tools/meme>) with the following parameters: maximum number of motifs 10; width of optimum motif ≥15 and ≤50. Phylogenetic analysis {#Sec17} --------------------- The predicted protein sequences of the 59 *BrAQP* genes were downloaded from the *B. rapa* genomic database (<http://brassicadb.org/brad/>). *Arabidopsis* and tomato AQP protein sequences were collected from TAIR (<http://www.arabidopsis.org/>) and the Sol Genomics network (<http://solgenomics.net/>), respectively. All sequences were then aligned using Clustal X \[[@CR64]\]. A phylogenic tree was constructed with MEGA6.0 software ([http://www.megasoftware.net](http://www.megasoftware.net/)) \[[@CR65], [@CR66]\] using the neighbor-joining method and 1,000 bootstrap replicates. The different domains might contribute to the topology of the phylogenetic tree with pairwise gap deletion option. Chromosomal location and gene duplication analysis {#Sec18} -------------------------------------------------- Sub-genome fractionation, and positional information of all candidate *AQP* genes along through the ten (10) chromosomes of *B. rapa* were retrieved from *B. rapa* database and the locations of the *AQP* genes were drafted using Map Chart version 2.2 (<http://www.wageningenur.nl/en/show/Mapchart.htm>). The *AQP* genes were BLAST searched (<http://blast.ncbi.nlm.nih.gov/Blast.cgi>) against each other to identify duplicate genes, in which the similarity of the aligned regions covered \>80% and the aligned region had identity \>80% \[[@CR67]\]. Tandem duplicated genes were defined as an array of two or more homologous genes within a range of 100-kb distance. We calculated the non-synonymous substitution (*Ka*), synonymous rate (*Ks*), and evolutionary constriction (*Ka/Ks*) between the duplicated *AQP* gene pairs of *B. rapa* based on their coding sequence alignments, using the Nei and Gojobori model \[[@CR68]\] as employed in MEGA 6.0 software (66). The nonsynonymous to synonymous ratio (*Ka/Ks*) between duplicated genes was analyzed to identify the mode of selection. *Ka/Ks* ratio \>1, \<1 and =1 indicate positive selection, purifying selection and neutral selection, respectively. We calculated the divergence time of duplicated gene pairs using *T* = *Ks/2R* Mya (Millions of years), where *T* refers to divergence time, *Ks* refers to the synonymous substitutions per site, and *R* is the rate of divergence of plant's nuclear genes. For dicotyledonous plants *R* = 1.5 × 10^−8^ synonymous substitutions per site per year (38). Microarray expression analysis {#Sec19} ------------------------------ Temperature-treated microarray data for *AQP* genes were collected from the data of Jung et al. (30). For that data, two inbred lines of *B. rapa* ssp. *pekinensis*, namely cold-tolerant Chiifu and cold-sensitive Kenshin, were treated with different temperatures viz*. 22,* 4, 0, −2, and −4 °C for 2 h. A heat map was generated based on transcript abundance value of 59 AQP genes using Cluster 3.0 and tree view software (<http://bonsai.hgc.jp/~mdehoon/software/cluster/software.htm#ctv>). Microsynteny analysis of the AQP gene family {#Sec20} -------------------------------------------- The microsyntenic relationship of AQP genes among *B. rapa, B. oleracea* and *A. thaliana* were detected using Blast against whole genomes of such crop species. *AQP* gene positions on chromosomes were collected from databases and the relationship among the three crop species were plotted using Circos software (<http://circos.ca/>) \[[@CR69]\]. Plant materials, growth and treatments {#Sec21} -------------------------------------- Chinese cabbage (*B. rapa* ssp. *pekinensis*) inbred lines cold-tolerant Chiifu and cold-sensitive Kenshin were used for cold-stress experiments, and Kenshin was used for other abiotic stress treatments. Seed sterilization, culture, seedling management were conducted according to the methods described by Ahmed et al. \[[@CR70]\]. Plants were culture on semisolid media for 2 weeks, after which those plants were transferred into liquid media to minimize stress during the treatment time. The 3-week-old plants were used for abiotic stress treatments (cold, drought, salt, ABA and water logging) and treatments were applied over a continuous time course (with samples taken at 0, 1, 4, 12, 24 and 48 h). Plants were transferred to the incubator at 4 °C to induce cold stress. Drought stress was simulated by drying the plants on Whatmann 3 mm filter papers. To induce salt ABA and waterlogging stress, plants were placed on petri dishes with medium containing 200 mM NaCl, 100 mM abscisic acid (ABA) and abundant of water respectively, for the recommended time courses. Fresh roots and leaves (third and fourth leaves) of *B. rapa* plants were harvested, immediately frozen in liquid nitrogen, and then stored at −80 ° C for RNA extraction. *B. rapa* (SUN-3061) was used for analysis of organ-specific expression and for biotic stress treatment (with *F. oxysporum* f.sp. *conglutinans*). The plants were grown for 3 weeks under culture room conditions with 16 h light and 8 h dark maintaining 25 °C temperature prior to fungus treatment. The fungal spore concentration 1x10^6^ spores per ml solution was used for inoculation using the method described by Ahmed et al. \[[@CR71]\]. Samples were collected from infected and mock-infected plants at 0 h, 3 h, 6 h, 4 d, 8 d and 11 d after inoculation (dai). The local (fourth) and systemic (fifth) leaves were harvested and immediately frozen in liquid nitrogen. Samples were then stored at −80 ° C until RNA extraction. RNA extraction and cDNA synthesis {#Sec22} --------------------------------- Total RNA was extracted from the samples (roots and leaves) using the RNeasy mini kit (Qiagen, USA) following the manufacturer's protocol. The concentration of RNA from each sample was determined by UV spectrophotometry at A260 using a NanoDropND-1000 (Nano Drop Technologies, USA). DNA contamination was removed using RNase-free DNase (Promega, USA) following the manufacturer's protocol. A 6 μl sample of total RNA was converted to cDNA using the First-Strand cDNA synthesis kit (Invitrogen, Japan) following the manufacturer's instructions. qPCR expression analysis {#Sec23} ------------------------ For each treatment, qRT-PCR was performed on three biological replicates. The 10 μl reaction volume consisted of the following: 5 μl 2x Quanti speed SYBR mix, 1 μL (10 pmol) each forward (F) and reverse (R) gene-specific primers, 1 μl template cDNA (50 ng) and 2 μl distilled, deionized water (ddH~2~O). The conditions for real-time PCR were as follows: initial denaturation at 95 °C for 5 min, followed by 40 cycles of denaturation at 95 °C for 10 s, annealing at 58 °C for 10 s, and extension at 72 °C for 15 s. The qRT-PCR reactions were normalized using the *B. rapa Actin* gene as reference for all comparisons \[[@CR72]\]. The fluorescence was measured following the last step of each cycle, and three replications were used for each sample. Amplification detection and data were processed using the Light cycler® 96 SW 1.1 software and the cq value was calculated using the 2^-ΔΔ^C~T~ method to determine the relative expression. The relative expression data was statistically analyzed (Tukey HSD test) and lettering was done using Minitab 17 software ([https://www.minitab](https://www.minitab/). com/products/minitab/). Additional files {#Sec24} ================ Additional file 1: Table S1.*In silico* analysis of aquaporin genes identified in *B. rapa* with their closest *Arabidopsis* homologs and sequence characteristics. (DOCX 21 kb) Additional file 2: Table S2.Homology analysis of *AQP* genes of *B. rapa*. (DOCX 66 kb) Additional file 3: Figure S1.Alignment of amino acid sequences of *Arabidopsis* and (1a) BrNIP (1b) BrSIP (1c) BrPIP and (1d) BrTIP subfamily members. Upper red line indicates predicted MIP domain and the blue portion of the alignment denotes predicted transmembrane domains. The two conserved NPA motifs are shown in bold pink letters. Residues comprising the ar/R filter are marked in yellow and labelled H2, H5, LE1 and LE2. Residues occupying the conserved Froger's positions one to five (from N- to C-terminus P1 to P5) are marked in green. (PPTX 437 kb) Additional file 4: Table S3.Pair-wise sequences similarity (%) of aquaporin proteins in *B. rapa. (XLSX 32 kb)* Additional file 5: Figure S2.Depiction of *BrAQP* duplicated gene pairs on *10* chromosomes of *B. rapa*. (PPTX 400 kb) Additional file 6: Figure S3.Schematic representation of motif compositions in the BrAQP protein sequences. Different motifs, numbered 1--10, are displayed in different colored boxes. The names of all members are displayed on the left, while the length of the motif is shown in the scale at the bottom of the figure. (PPTX 269 kb) Additional file 7: Table S4.Best possible match sequences of motifs (1--15) presented in Additional file Figure S3. (DOCX 11 kb) Additional file 8: Figure S4.The intron-exon structures of *BrAQP* genes. Names of the genes are on the left. The thick blue lines, exons; fine red lines, introns. (PPTX 346 kb) Additional file 9: Figure S5.Expression profiles of *BrPIP* genes in various tissues as determined by RT-PCR analyses. Four amplified bands from left to right for each gene represent amplified products from R, roots; S, stems; L, leaves; Fb, flower buds (PPTX 103 kb) Additional file 10: Table S5.Primer sequence used for real time and RT-PCR amplification of *Aquaporin* genes of *B. rapa. (DOCX 12 kb)* ABA : Abscisic acid AK : Karyotype AQP : Aquaporin Ar/R : Aromatic/ Arginine At : *Arabidopsis thaliana* B. oleracea : *Brassica oleracea* Br : *Brassica rapa* BRAD : Brassica database C : Chiifu CDS : Coding DNA sequence Cl : Cluster dai : Days after infection *F.oxysporum* : *Fusarium oxysporum* f.sp : Fungal species Gm : *Glycine max* K : Kenshin Ka : Nonsynonymous substitutions per nonsynonymous site kb : Kilo basepair Ks : Synonymous substitutions per synonymous site LF : Least fractionated MF1 : Medium fractionated MF2 : Most fractionated MIP : Major intrinsic protein MYA : Million year NIP : NOD26-like intrinsic protein Nt : *Nicotiana tabacum* PIP : Plasma membrane intrinsic proteins qPCR : Quantitative polymerase chain reaction ROS : Reactive oxygen species RT-PCR : Reverse transcription polymerase chain reaction SIP : Small basic intrinsic protein Sl : *Solanum lycopersicum* Ta : *Triticum aestrivum* TIP : Tonoplast intrinsic protein XIP : X intrinsic protein Special thanks to Professor YoonKang Hur, department of Biology, College of Biological Sciences and Biotechnology, Chungnam National University, Daejeon, Korea Republic for providing seeds of *B. rapa* inbred lines 'Chiifu' and 'Kenshin'. Funding {#FPar1} ======= This research was supported by the Golden Seed Project (Center for Horticultural Seed Development), the Ministry of Agriculture, the Food and Rural Affairs (MAFRA), the Ministry of Oceans and Fisheries (MOF), the Rural Development Administration (RDA) and the Korea Forest Service (KFS). Availability of data and materials {#FPar2} ================================== We declare that the dataset(s) supporting the conclusions of this article are included within the article (and its additional file(s)). Authors' contributions {#FPar4} ====================== MAK performed research, analyzing real time data and drafted the manuscript, JIP and ISN formulate concept and designed research, UKN acquisition and analyse data, write part of the manuscript and critical review for intellectual content of it, MKB and HTK perform *in silico* analysis, designed the stress experiments and cultured the plants. All authors read and approved the final manuscript. Competing interests {#FPar5} =================== The authors declare that they have no competing interests. Consent for publication {#FPar6} ======================= Not applicable. Ethics approval and consent to participate {#FPar7} ========================================== Not applicable. Deposition of data {#FPar3} ================== The complete raw microarray data have been deposited in the Omics database of NABIC ([http://nabic.rda.go.kr](http://nabic.rda.go.kr/)) as enrolled numbers NC-0024-000001 and NC-0024-000002.

Introduction {#Sec1} ============ Tuberculosis (TB) is becoming a major worldwide health problem in recent years. The World Health Organization estimates that in 2016 alone 10 million people contracted TB and about 1.7 million people died from the disease, mostly in low and mid-income countries^[@CR1]^. TB is caused by *Mycobacterium tuberculosis (Mtb)* and for curing TB, a prolonged treatment with a combination of multiple agents is required^[@CR1]^. The multiple drugs needed for a cure may be out of reach for people in parts of the world. Furthermore, the lengthy regimen leads to adherence problems and undesired interruptions that subsequently drive the emergence and spread of MDR *Mtb* strains. Novel drugs that could target MDR *Mtb* are urgently needed to combat TB. One of the fundamental aspects of mycobacteria that confers resistance to antimicrobial agents and promotes the virulence of *Mtb* is a thick layer of mycolic acids that constitute their cell wall^[@CR2],[@CR3]^. Mycolic acids are C~74~--C~90~ fatty acids that are synthesized by a unique combination of two canonical enzymatic systems that *Mtb* has: a fatty acid synthase-II (FAS-II) that is comprised of a group of discrete enzymes, a system that is typically found in plants and prokaryotes. The second system is the mycobacterial FAS-I, which is a multi-domain, multi-functional enzyme complex that is typically found in fungi and higher eukaryotes^[@CR4],[@CR5]^ but not in prokaryotes. The FAS-I complex of mycobacteria is an essential enzymatic complex^[@CR6]^, making it an attractive drug target. The mycobacterial FAS-I has a distinct capacity to elongate fatty acids beyond C~16~, to C~24/26~^[@CR7],[@CR8]^, which are further elongated to meromycolate (C~56~) by the FAS-II system and then condensed with C~26~ to form mycolic acids^[@CR2],[@CR5],[@CR9]^. A group of compounds that interfere with fatty acid synthesis by *Mtb* have become invaluable drugs for fighting TB. Specifically, isoniazid and ethionamide, first and second line drugs against TB, inhibit mycolic acid synthesis through inhibition of an enoyl reductase (ER) of the FAS-II system^[@CR10]^. Pyrazinamide (PZA) is a first line and an indispensable drug for shortening the regimen required for curing TB^[@CR11]^. Analogs of PZA were shown to bind and inhibit mycobacterial FAS-I^[@CR12],[@CR13]^ by a mechanism that is still unknown. Here we report the 3.3 Å resolution structure of the \~2 MDa FAS-I from *Mtb*, determined by single particle cryo-EM. The *Mtb* FAS-I structure reveals unique adaptations at the catalytic modules that allow production of long fatty acids. The structural information for *Mtb* FAS-I we have obtained makes an important milestone in the quest for novel drugs as it now enables computer-based drug discovery approaches. Importantly, the resolution achieved by cryo-EM is sufficient for elucidating inhibition mechanisms by putative small molecular weight inhibitors. Differences in the catalytic domains of *Mtb* FAS-I compared to a structurally similar fungal FAS-I homolog suggest a possibility to target *Mtb* FAS-I with high specificity that would be needed to minimize undesired toxicity toward other systems like the human FAS-I. Results {#Sec2} ======= Determining the *Mtb* FAS-I structure to near atomic resolution {#Sec3} --------------------------------------------------------------- *Mtb* FAS-I is an α6 subtype complex, composed of six long polypeptide chains with an overall molecular weight of \~2 MDa. Each of the α chains has seven catalytic domains and is 3069 amino acids long. Structural information at near atomic resolution is available for FAS-I from fungi^[@CR14]--[@CR16]^. The FAS-I system from the non-pathogenic mycobacterium *M. smegmatis* was investigated using cryo-electron microscopy (cryo-EM), yielding a 7.5 Å resolution structure^[@CR17]^. The FAS-I system from *Mtb* itself was also studied using cryo-EM, but a structure was obtained to only 20 Å resolution due to conformational flexibility of the purified FAS-I complex^[@CR18]^. The lack of structural information at near atomic resolution for the *Mtb* FAS-I system hinders the ability to investigate the mechanism by which PZA or its analogs inhibit FAS-I, and precludes rational design approaches for developing novel inhibitors as potential TB therapeutics. To enable structural studies of *Mtb* FAS-I we developed a platform for producing the activated form of the protein complex^[@CR19]^. This platform relies on co-expressing *Mtb* FAS-I in *E. coli* together with 4′-phosphopantetheinyl transferase (AcpS) protein, which activates the acyl carrier protein (ACP) module of FAS-I. Following gentle purification, homogeneous FAS-I complex can be clearly seen using negative-stain EM (Supplementary Fig. [1](#MOESM1){ref-type="media"}), as well as using cryo-EM, which yields well-defined 2D classes (Fig. [1a](#Fig1){ref-type="fig"}). A 3D map with D3 symmetry imposed was reconstructed from a final data set of 40,160 particles at 3.3 Å overall resolution with good angular coverage (Supplementary Fig. [2a](#MOESM1){ref-type="media"}, [b](#MOESM5){ref-type="media"}). The map reveals a barrel shaped complex (Fig. [1b](#Fig1){ref-type="fig"}) as previously observed for fungal and mycobacterial FAS-I^[@CR14],[@CR15],[@CR17],[@CR18],[@CR20],[@CR21]^. 3D classification indicated that the core barrel domains had a uniform conformation, and conformational heterogeneity was observed only for the ACP modules (see below). The local resolution of the map extends beyond 3.0 Å at core regions of the catalytic subunits and decreases to \~6 Å at the more peripheral and flexible regions (Fig. [1b](#Fig1){ref-type="fig"}, Supplementary Fig. [3](#MOESM1){ref-type="media"}). This map allowed us to model the structure of FAS-I including unambiguous assignment of side chain rotamers for most amino acids at the better-resolved regions (Fig. [1c](#Fig1){ref-type="fig"}, Supplementary Fig. [4](#MOESM1){ref-type="media"}).Fig. 1Determining a 3.3 Å resolution map using single particle cryo-EM. **a** Cryo-EM micrograph of *Mtb* FAS-I particles. FAS-I particles are clearly visible in the micrographs. Scale bar represents 50 nm. Red and yellow arrows highlight some particles with 'top' or 'side' views, respectively. Representative 2D-class averages of 'top', 'side', and 'tilted' views are shown on the right side of the micrograph from top to bottom, respectively. **b** Electron density map of FAS-I colored by the local resolution estimate. Resolution lower than 3.0 Å is colored purple/blue, between 3.0 and 5.0 Å is colored cyan/green and higher than 5.0 Å is colored in yellow/red. 'Side' views (upper images) and 'top' views (lower images) of the density map are shown. The panels on the right are cross-sections of the map, illustrating the quality of density in the central regions of the map. **c** A segment spanning residues 2765--2773 of FAS-I illustrating the map quality at inner regions at better than 3.0 Å resolution ACP adopts variable orientations at the vicinity of KS {#Sec4} ------------------------------------------------------ Out of seven functional domains of *Mtb* FAS-I, we modeled the six domains that make the barrel-like cage (Fig. [2a](#Fig2){ref-type="fig"}, Supplementary Table [1](#MOESM1){ref-type="media"}). The density for the ACP was weak and thus we did not model it (see Methods section). The function of ACP is to carry the growing fatty acid at the inner chamber between the catalytic domains^[@CR21],[@CR22]^ (Supplementary Fig. [5](#MOESM1){ref-type="media"}). The weak density for the ACP implied that FAS-I particles are heterogeneous with respect to the locations of the six ACP domains at the inner chamber. To gain insights for the putative location of ACP we classified the data set into five distinct 3D classes and refined the maps separately with imposed D3 symmetry. Clear densities for the ACP domains are visible in all the five different classes at resolutions of about 15 Å (Fig. [2b](#Fig2){ref-type="fig"}, Supplementary Fig. [6](#MOESM1){ref-type="media"}). In all classes, the ACP domains are near the ketoacyl synthase (KS) modules in the central wheel as was previously observed for the yeast FAS-I^[@CR15],[@CR21]^. However, the exact orientations (and likely conformations) of the ACP domains vary between the different classes (Supplementary Fig. [6](#MOESM1){ref-type="media"}), explaining the weak density in the original reconstruction that included all particles. Imposing D3 symmetry for this analysis may obscured low populated states in which some of the ACP domains are located at different positions in the catalytic chamber, as was previously observed in a fungal FAS-I^[@CR22]^. We therefore used the localized reconstruction method^[@CR23]^ to extract individual asymmetric units (referred to as sub-particles) from the 2D particle images, each focusing on a single ACP domain. After excluding sub-particles that overlap in the 2D images, we obtained a data set of 166,918 sub-particles. We then classified the sub-particle data set in 3D into 3, 5, or 10 classes (Supplementary Fig. [7](#MOESM1){ref-type="media"}). The localized reconstruction revealed orientations of the ACP which were not observed in the context of the whole complex, however the resolution of the ACP within the sub-particle classes was similar to that obtained using 3D classification of the whole complex. This indicates that the ACP domains are not rigidly held in a single position within the chamber but rather occupy a continuum of closely related orientations and possibly conformations. Interestingly, in all classes the ACP domain remains localized at the vicinity of the KS module.Fig. 2Structure of the *Mtb* FAS-I. **a** The final model of Mtb FAS-I is shown using ribbon representation. A 'side' view (upper image) reveals the barrel-like shape of FAS-I. One of the six chains that compose the D3-symmetric barrel is colored. Each of the six different catalytic domains that make the barrel is shown using a distinct color: acetyltransferase (AT) in blue, enoyl reductase (ER) in cyan, dehydratase (DH) in green, malonyl transacylase (MPT) in brown, ketoacyl reductase (KR) in light green, and ketoacyl synthase (KS) in red. A 'top' view (lower image) shows the wheel-like organization of the KR and KS domains. **b** The *Mtb* FAS-I model as red ribbon is shown inside a low-pass filtered map from one specific 3D class as 'tilted' and 'top' views (upper and lower images, respectively). A density for the acyl carrier protein (ACP) is visible. **c** A schematic diagram showing the domain organization of *Mtb* FAS-I. Color coding and abbreviations are the same as in **a** Inverted surface electrostatic potentials of the *Mtb* FAS-I {#Sec5} ------------------------------------------------------------ Putative docking of ACP domains to the KS modules is partially mediated by a long loop (residues 2210--2226) that projects from the neighboring ketoacyl reductase (KR) module (KR-loop) (Supplementary Fig. [8](#MOESM1){ref-type="media"}). An equivalent such loop is not seen in the case of FAS-I from fungi. It was previously noted that electrostatic interactions govern the interaction of the negatively charged ACP with the positively charged catalytic modules in *S. cerevisiae* FAS-I^[@CR15]^. Examining the surface electrostatic potential near the active site of the KS module from *Mtb* FAS-I reveals mixed charged residues (Fig. [3a](#Fig3){ref-type="fig"}) in contrast to the positively charged surface seen in FAS-I from the thermophilic *T. lanuginosus* fungus (PDB: [4v59](https://www.rcsb.org/structure/4v59))^[@CR14]^ (Fig. [3b](#Fig3){ref-type="fig"}). However, the KR-loop that mediates the interaction of ACP with KS (Supplementary Fig. [8](#MOESM1){ref-type="media"}) donates negatively charged residues (Fig. [3c](#Fig3){ref-type="fig"}) and ultimately makes a negatively charged surface near the active site of KS (Fig. [3d](#Fig3){ref-type="fig"}). Negatively charged surfaces are also found near the catalytic sites of the acetyltransferase (AT), ER, dehydratase (DH) (Supplementary Fig. [9](#MOESM1){ref-type="media"}), and KR modules (Supplementary Fig. [10](#MOESM1){ref-type="media"}), suggesting that in the case of *Mtb* FAS-I electrostatic interactions may also mediate attachment of ACP to the catalytic modules but with an opposite charge as in the fungal FAS-I system. Compared to the other catalytic modules, MPT has a more positively charged surface near its catalytic site, which might lower the probability of ACP to interact with it during the transport of the growing fatty acid between the catalytic modules. Since MPT has a special role in terminating the synthesis cycle (Supplementary Fig. [5](#MOESM1){ref-type="media"}), its positively charged surface may function as a safety mechanism that prevents premature termination of synthesis. Since ACP needs to shuttle malonyl from MPT to KS during synthesis, it could be that the electrostatic potential of ACP changes when it binds the growing aliphatic chain. Such a change might be induced by the burial of the aliphatic chain of the fatty acid at the core of the ACP as was previously suggested to happen during substrate shuttling^[@CR21]^. Such electrostatic modulations of the ACP are likely to be transient as the aliphatic chain is only transiently concealed by the ACP^[@CR24]^.Fig. 3A KR-derived loop provides negatively charged surface to KS. **a** Electrostatic potential at the solvent accessible surface mapped on the molecular surface of the KS module from *Mtb* FAS-I. Negatively charged areas are in red (−5.0* kT*/*e*) and positively charged areas are in blue (+5.0 *kT*/*e*). Electrostatic potentials were calculated using APBS tools^[@CR36]^. The catalytic site is indicated with a black arrow. **b** Electrostatic potential of the KS module from fungal FAS-I (PDB: [4v59](https://www.rcsb.org/structure/4v59)). **c** Electrostatic potential of the loop donated from the *Mtb*-KR module on top of a surface representation of the KS module in gray. **d** Combined electrostatic potentials of the KS module with the KR-originating loop. The surface near the active site of KS is dominated by negative charge in the case of *Mtb* and by positive charge in the case of the fungal FAS-I Large inner cavities at the catalytic modules of the *Mtb* FAS-I {#Sec6} ---------------------------------------------------------------- In addition to differences in electrostatic potentials, comparing *Mtb* FAS-I to the fungal FAS-I (PDB: [4v59](https://www.rcsb.org/structure/4v59))^[@CR14]^, reveals a marked structural difference in the catalytic clefts of some of the active modules (Fig. [4](#Fig4){ref-type="fig"}). The fungal FAS-I is an homolog that retains a barrel shape complex and that structural information at near-atomic resolution is available for. The catalytic cleft of the ER from *Mtb* FAS-I is wider compared to the fungal-ER, and the flavin mononucleotide (FMN) co-factor in *Mtb* FAS-I is more exposed compared to the FMN in the fungal FAS-I. This difference is mostly due to the local amino acid composition rather than changes in backbone conformation (Fig. [4](#Fig4){ref-type="fig"}, Supplementary Movies [1](#MOESM4){ref-type="media"} and [2](#MOESM5){ref-type="media"}). In this regard, the quality of density in the inner region of the protein (Fig. [1b, c](#Fig1){ref-type="fig"}) is sufficient to accurately model side-chain rotamers (Supplementary Fig. [11](#MOESM1){ref-type="media"}) that ultimately determine the shape of the cleft. The catalytic cleft of the KR module is also much wider in *Mtb* FAS-I compared to fungal FAS-I (Fig. [4](#Fig4){ref-type="fig"}, Supplementary Movies [3](#MOESM6){ref-type="media"} and [4](#MOESM7){ref-type="media"}). A notable difference that contributes to this change is a distinct backbone conformation of a loop that partially conceals the active site (Fig. [4](#Fig4){ref-type="fig"}). A similar difference in width is also observed between the DH modules of *Mtb* and the fungal FAS-I (Fig. [4](#Fig4){ref-type="fig"}, Supplementary Movies [5](#MOESM8){ref-type="media"} and [6](#MOESM9){ref-type="media"}). More striking is the larger inner volume of the *Mtb* DH catalytic chamber compared to the fungal protein (Supplementary Fig. [12](#MOESM1){ref-type="media"}). In contrast to these observations, the catalytic pockets of the KS, MPT, and AT modules (Fig. [4](#Fig4){ref-type="fig"}, Supplementary Movies [7](#MOESM10){ref-type="media"} and [8](#MOESM11){ref-type="media"}) have similar sizes in *Mtb* compared to the fungal FAS-I (Supplementary Fig. [13](#MOESM1){ref-type="media"}). The extended catalytic clefts of the KR, DH, and ER modules may reflect evolutionary adaptation of *Mtb* FAS-I for producing the unusually long C~26~ fatty acids; larger clefts are probably required for accommodating the longer aliphatic chains, while the ACP-attached pantothenic acid reaches into the active sites. As for the other catalytic modules, the KS and MPT from both species seem to have long channels or clefts that effectively do not restrict the length of the growing aliphatic chain (Supplementary Fig. [14](#MOESM1){ref-type="media"}), and the AT module transfers similarly sized acetyl groups (Supplementary Fig. [5](#MOESM1){ref-type="media"}) in both species.Fig. 4Catalytic clefts of *Mtb* FAS-I structurally differ from fungal FAS-I. The four catalytic domains that take part in extension and modification of the growing aliphatic chain during fatty acid synthesis are shown using ribbon representation (left column). The polypeptide chains of *Mtb* FAS-I (gray) and of the fungal FAS-I (pink, PDB code: [4V59](https://www.rcsb.org/structure/4v59)) are superimposed. Surface representations of *Mtb* FAS-I (middle column) and of fungal FAS-I (right column) illustrating the differences in shapes of the catalytic clefts between *Mtb* and fungal FAS-I. The enoyl reductase domains (ER, top row) are shown with the bound FMN co-factors in cyan. The fungal FAS-I structure was determined with a bound NADPH that is colored green. In the ketoacyl reductase domain (KR, second row from top) of *Mtb* FAS-I, two short loops were not visible in the density maps (blue and orange dots). NADPH from the structure of the fungal FAS-I is shown in green space-filling spheres. A loop that partially conceals the catalytic cleft in the fungal FAS-I (hot pink, residues 750--761) adopts a different structure in *Mtb* FAS-I (black, residues 2237--2248). The NADPH from the fungal FAS-I is shown in the surface representation of *Mtb*-KR to help visualize the catalytic cleft. The catalytic histidine residues of the dehydratase domains (DH, third row from top) are colored blue. The catalytic lysine and histidine residues of the ketoacyl synthase domains (KS, lower row) are colored blue. A long loop that projects from *Mtb*-KR toward the active site of *Mtb*-KS is shown in black. The corresponding loop in the fungal FAS-I is colored in hot pink. For better 3D perception, movies of the various catalytic modules are included (Supplementary Movies [1](#MOESM4){ref-type="media"}, [2](#MOESM5){ref-type="media"}, [3](#MOESM6){ref-type="media"}, [4](#MOESM7){ref-type="media"}, [5](#MOESM8){ref-type="media"}, [6](#MOESM9){ref-type="media"}, [7](#MOESM10){ref-type="media"} and [8](#MOESM11){ref-type="media"}) Discussion {#Sec7} ========== Despite overall similar architecture, *Mtb* FAS-I has distinct features that differentiate it from previously determined fungal FAS-I systems. The transition from positively charged surfaces of the fungal FAS-I catalytic modules to negatively charged surfaces in *Mtb* FAS-I and the likely complementary change in the charge of the ACP reflect the large evolutionary distance between mycobacteria and fungi. The unique ability of *Mtb* FAS-I to produce C~26~ fatty acids can explain its larger catalytic clefts in the KR, DH, and ER modifying modules as compared with the fungal FAS-I. In spite of the evolutionarily conserved biochemical process of fatty acid synthesis, the combination of large catalytic clefts together with the altered electrostatic potentials near the active sites, makes the catalytic modules of *Mtb* FAS-I unique. Considering that targeting mycolate synthesis is a proven approach for fighting TB^[@CR9]^, inhibiting *Mtb* FAS-I is a promising strategy for fighting MDR-*Mtb*. The structural data that we provide here can now be used for virtual screening approaches^[@CR25]^ for identifying novel drug candidates. Moreover, the near atomic resolution that we achieved in this study using only a modest number of particles suggests that it would be possible to further use structure-guided approaches to improve putative hits from such screening efforts. Along these lines, elucidating the inhibition mechanism of PZA analogs and rationally improving them using structural data should also be possible on the base of single particle cryo-EM studies. Methods {#Sec8} ======= Protein expression and purification {#Sec9} ----------------------------------- To produce FAS-I for cryo-EM analysis, we followed a protocol that we previously developed^[@CR19]^. Briefly, FAS-I expression plasmid (pMT100 StrepFlag-FAS-I) was transformed into *E. coli* BL21 (DE3) (Novagen) cells together with AcpS expression plasmid (pACYCDuetAra-Acps). Expression of AcpS was initiated by adding 0.2% Arabinose followed by the induction of FAS-I expression by adding 200 μM Isopropyl β-D-1-thiogalactopyranoside at 15 °C for 16--18 h. Cells were lysed using French press in buffer 'A' (100 mM potassium phosphate buffer (KPB) pH 7.2, 150 mM potassium chloride, 1 mM tris(2-carboxyethyl)phosphine, 1 mM ethylenediaminetetraacetic acid) supplemented by protease inhibitor cocktail (cOmplete, Roche) and 20% sucrose. After clarifying the supernatant by 1 h centrifugation at 150,000×*g*, protein debris were pelleted by precipitation at 20% (w/v) ammonium sulfate. Next, FAS-I was precipitated by adjusting the ammonium sulfate concentration to 50% (w/v) and applying 150,000×*g* for 1 h. Precipitate was dissolved in buffer 'A' supplemented by protease inhibitor cocktail (cOmplete, Roche) without sucrose, and FAS-I was captured using Strep-Tactin Sepharose column (GE Healthcare) and separated by FPLC system. Purified FAS-I was eluted using buffer 'A' supplemented with 2.5 mM d-desthiobiotin. Negative staining {#Sec10} ----------------- 3.5 μl of the purified FAS1 sample at 0.4 mg/ml was applied to glow-discharged, homemade 300 mesh carbon-coated copper TEM grids for 30 s. Excess liquid was blotted and, after a wash with distilled water, the grids were stained with 2% uranyl acetate. Samples were visualized in an FEI Tecnai T12 TEM operated at 120 kV, equipped with a Gatan OneView camera. Cryo-EM image acquisition {#Sec11} ------------------------- 3.5 μl of FAS-I solution at 10 mg/ml was applied to glow-discharged C-flat 2/2, 200 mesh holey carbon grids and plunge frozen in liquid ethane cooled by liquid nitrogen, using a Leica EM-GP plunger (3 s blotting time, 80% humidity). Grids were screened on an FEI F20 microscope equipped with a K2 Summit direct detector (Gatan). Cryo-EM data were collected on a Titan Krios electron microscope (FEI) operated at 300 kV. Coma-free alignment was performed with AutoCTF (FEI) and beam size was 645 nm. Movies were recorded on a K2 Summit direct detector mounted at the end of a GIF Quantum energy filter (Gatan) using a slit of 20 eV. 5248 Movies were collected in super-resolution counting mode at a nominal magnification of 130,000× corresponding to a physical pixel size of 1.054 Å. The dose rate was set to 4.49 electrons per physical pixel per second and the total exposure time was 20 s, resulting in an accumulated dose of 80.1 electrons per Å^2^. Each Movie was fractionated into 40 frames of 0.5 s. Nominal defocus range was −0.5 to −2.5 μm. All dose-fractionated images were recorded using an automated low dose procedure implemented in SerialEM^[@CR26]^. Stage navigation was used to navigate to hole centers and image shift used to target 5 distinct imaging locations within each hole. Single particle cryo-EM image processing {#Sec12} ---------------------------------------- Image frames were Fourier-binned 2 × 2, aligned (5 × 5 tiles, B-factor 100) and dose-weighted using MotionCor2^[@CR27]^. Contrast transfer function (CTF) parameters were estimated using Gctf^[@CR28]^ on the non-dose-weighted image sums. 1912 images showing resolution better than 6.0 Å were selected for further processing. All subsequent image processing was done using RELION 2.0^[@CR29]^. About 500 particles were manually picked and subjected to reference-free 2D classification, followed by automated picking using the newly generated 2D class averages. 91,589 automatically picked particles, extracted into 400^[@CR2]^ pixel boxes and subjected to multiple rounds of 2D classification in order to clean the data set. 2D classes showing secondary structure features were retained at each round, resulting in a total of 40,160 particles. 3D classification with D3 symmetry imposed did not result in maps with significant difference at the core elements. Additionally, separating the data set into different numbers of 3D classes resulted in lower resolution in the classes. Therefore all 40,160 were subjected to 3D refinement with a sphere mask of 290 Å and D3 symmetry imposed, using the published cryo-EM structure of FAS-I from *M. smegmatis*^[@CR17]^ (PDB: [4V8L](https://www.rcsb.org/structure/4V8L)) filtered to 60 Å as an initial model. Only movie frames 2--15 were used at this stage. The global resolution in the refined structure, estimated by the gold-standard FSC = 0.143 criteria, was 3.3 Å (Supplementary Fig. [2a](#MOESM1){ref-type="media"}). Local filtering for subsequent model building was done using RELION, and local resolution estimate was calculated using the program blocres in BSOFT^[@CR30]^. Localized reconstruction was performed as described in Ilca S.L. et al.^[@CR23]^ using scripts (version 1.2) embedded in Scipion^[@CR31]^ and using RELION 2.1 for 3D classification and reconstruction. The position of a single ACP domain and the center of the complex were marked in the refined, D3 symmetrized map using UCSF Chimera^[@CR32]^. These coordinates were then used to locate all symmetry-related positions of the ACP domains in the 2D particle images. A minimum distance of 25 pixels was set between extracted ACP coordinates to avoid overlaps. A total of 166,918 ACP domains ('sub-particles') were extracted (160^2^ pixel boxes). The sub-particle data set was 3D classified into 3, 5 and 10 classes using the alignment and orientation parameters obtained for the whole complex and a spherical mask of 120 Å diameter. Local resolution estimate of some sub-particle 3D maps indicated regions where the ACP was resolved to better than 10 Å, however other ACP regions remained noisy at that resolution. Therefore, to avoid overfitting we low-pass filtered all classes to 15 Å. Model building and refinement {#Sec13} ----------------------------- The [4V8V](https://www.rcsb.org/structure/4v8v) PDB entry was used as an initial model. First, it was docked into the electron density map using UCSF Chimera^[@CR32]^, and then it was real-space refined using Phenix^[@CR33]^. The model was then manually adjusted to fit the electron density map while converting the amino acids to correspond to the *Mtb* FAS-I sequence using Coot^[@CR34]^. Subsequently, we used iterative real-space refinements and manual model improvements using Phenix and Coot, respectively. The final model spans residues 32--3061. We did not observe interpretable density for the first 31 residues, the last 34 residues, or for residues 2289--2296 and 2350--2354 and hence did not model them. Also, density for the ACP domain (residues 1742--1976) was weak, and hence it was not modeled. Structural analysis and visualization {#Sec14} ------------------------------------- To visualize structures, electron density maps and to perform structural analysis we mainly used PyMol^[@CR35]^ and UCSF Chimera^[@CR32]^. Electrostatic potentials were calculated using APBS tools^[@CR36]^. To detect and visualize internal cavities we used the cavity detection module of PyMol with cavity detection cutoff of 2 solvent radii and cavity detection radius of 10 solvent radii. Electronic supplementary material ================================= {#Sec15} Supplementary Information Peer Review File Description of Additional Supplementary Files Supplementary Movie 1 Supplementary Movie 2 Supplementary Movie 3 Supplementary Movie 4 Supplementary Movie 5 Supplementary Movie 6 Supplementary Movie 7 Supplementary Movie 8 Supplementary Movie 9 **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Nadav Elad, Szilvia Baron. Electronic supplementary material ================================= **Supplementary Information** accompanies this paper at 10.1038/s41467-018-06440-6. We thank Wim Hagen from EMBL Heidelberg for his invaluable assistance in collecting cryo-EM data. We thank Prof. Deborah Fass for critical comments and suggestions. This project was continuously supported by the Kimmelman Center for Biomolecular Structure and Assembly. O.Z. was supported by the Legacy Heritage Clinical Research Initiative of the Israel Science Foundation (Grant no. 1629/10) and by the Israeli Ministry of Health Chief Scientist award (Grant no. 6223). S.B. was supported by the Israeli Minister of Absorption for Post Doctorate Fellowship no. 17733. Electron microscopy studies were supported in part by the Irving and Cherna Moskowitz Center for Nano and Bio-Nano Imaging at the Weizmann Institute of Science. High-resolution cryo-EM data were acquired at the cryo-EM facility of the European Molecular Biology Laboratory (EMBL) in Heidelberg supported by iNEXT (Project no. 1975) and funded by the Horizon 2020 program of the European Union. Research in the Diskin lab is supported by a research grant from the Enoch Foundation, a research grant from the Abramson Family Center for Young Scientists, a research grant from Ms. Rudolfine Steindling, by the Minerva Foundation with funding from the Federal German Ministry for Education and Research, and by a grant from the Israel Science Foundation (Grant no. 682/16). R.D. is an incumbent of the Tauro career development chair in biomedical research. O.Z., Z.S. and R.D. designed the project. O.Z. oversaw the project. Y.P. developed the expression vector. S.A., S.B. and J.G. developed protein production, purification, and validation methods. S.B. and G.R. produced and purified protein. N.E. and S.B. performed initial EM characterization of the protein. N.E., S.B. and R.D. collected cryo-EM data. N.E. and R.D. processed cryo-EM data, calculated density maps, and modeled the structure. R.D. performed structural analysis and wrote the manuscript with the help of all other authors. Atomic model for the *Mtb* FAS-I as well as EM density maps are available at the protein data bank under accession code 6GJC and at the electron microscopy data bank under accession code EMD-0011. Other data are available from the corresponding authors upon reasonable request. Competing interests {#FPar1} =================== The authors declare no competing interests.

Background {#Sec1} ========== Rheumatoid arthritis (RA) is a crippling disease associated with significant morbidity and mortality. Disease-Modifying Anti-Rheumatic Drugs (DMARDs), are used to control the disease as well as to slow the disease progression. Despite the use of these agents, some patients continue to have persistently active disease \[[@CR1]\]. The introduction of biologic DMARDs has helped those not responding to non-biologic DMARDs. Biologic DMARDs recommended by current guidelines and are in use include, Tumor Necrosis Factor inhibitors (TNFi), and non TNF alpha agents, rituximab (anti CD-20 monoclonal antibodies), abatacept (inhibitor of Cytotoxic T lymphocyte Antigen-4 (CTLA4) co-stimulation) and tociluzimab (IL-6 inhibitor) \[[@CR2]\]. Current treatment guidelines recommend monotherapy, usually with methotrexate as the first line DMARD, but recommend combinations of DMARDs in those with high disease activity, in the presence poor prognostic factors and in non-responding patients. Due to fears of hepatotoxicity combination therapy with methotrexate and leflunomide has not been used frequently \[[@CR3]\]. Recommended next line of treatment in DMARD combination failures are the TNFi. Some data also suggest that adding a TNFi early in active disease is superior to adding conventional DMARDs to methotrexate therapy \[[@CR4]\]. TNFi are the first line biologic agents recommended as they give long term efficacy and safety data from clinical trials and clinical registries \[[@CR5]\]. However risk of reactivation of tuberculosis (TB) especially in populations from TB endemic regions such as South Asia remains a concern when using a TNFi in this region \[[@CR6]--[@CR9]\]. Moreover TNFi in recommended dosages are expensive and are not available routinely through the government health services. The use of rituximab as a first line biologic agent has recently gained acceptance \[[@CR10]\] and it is now a licensed indication for rheumatoid arthritis \[[@CR10], [@CR11]\]. Rituximab is effective in both methotrexate non-responders \[[@CR12]\] as well TNFi non-responders \[[@CR13]\] and has the advantage of needing only two infusions for a long lasting remission. Although initially recommended for use in TNFi non-responders, current evidence from clinical trials, and changes in recommendations \[[@CR2]\] have promoted it as second line therapy after failure of non-biologic DMARDs. Moreover, not reactivating latent TB makes it a suitable biologic to use in TB endemic regions of the world \[[@CR14]\]. The usually recommended dose of Rituximab is 1000 mg but some clinical trials have shown that the dose of 500 mg to be efficacious as well \[[@CR15], [@CR16]\]. Data from the MIRROR study \[[@CR15]\] and SERENE study \[[@CR17]\] suggested that overall efficacy of two infusions of rituximab 500 mg and two infusions of 1000 mg could not be clearly differentiated, although some of the efficacy end points suggested improved outcomes in the rituximab 1000 mg group \[[@CR15]\]. More recently, a systematic review of 6 clinical trials and 2 cohort studies \[[@CR18]\] have shown similar efficacy of low dose rituximab compared to high dose regimen. Another study on international cohort data of a large number of patients who have received rituximab has also given similar findings \[[@CR19]\]. Although available trial data favors leflunomide-methotrexate combinations in non-responders to conventional DMARD combinations \[[@CR20], [@CR21]\], only a few clinical trials have addressed adequately the efficacy and safety of this cheaper alternative. The recently published SMILE study has given evidence on the safety of combining lefluonomide with methotrexate as this combination was well tolerated with adverse effect profile comparable to monotherapy with either methotrexate or leflunomide \[[@CR22]\]. Efficacy data from South Asian patients using the cheaper low dose rituximab-methotrexate regimen and the even more cheaper leflunomide-methotrexate combination is not available. To our knowledge no previous trial has compared the leflunomide-methotrexate combination versus a biologic agent given with methotrexate in refractory RA. To address these questions we carried out a randomized double blind controlled clinical trial to study the efficacy and safety of low dose rituximab-methotrexate combination compared to leflunomide-methotrexate combination in patients with rheumatoid arthritis not responding to initial treatment with non-biologic DMARDs. Methods {#Sec2} ======= Patients {#Sec3} -------- Patients were recruited from the two-rheumatology clinics at the National Hospital of Sri Lanka (NHSL). The patients were included if they were older than 18 years, fulfilled the revised 1987 ACR criteria for diagnosis of rheumatoid arthritis \[[@CR23]\], and had active disease despite treatment with non biologic DMARDs containing at least 10 mg of methotrexate per week for more than 6 months. Active disease was defined by the presence of at least four swollen and four tender joints and a raised serum C-reactive protein (CRP) level of ≥0.6 mg/dl, and erythrocyte sedimentation rate (ESR) of \>28 mm per hour. Patients were on stable doses of non-steroidal anti-inflammatory drugs (NSAIDS) or corticosteroids not exceeding 10 mg per day of prednisolone (or the equivalent) for at least 3 months. Patients previously given biologic DMARDS or leflunomide were excluded from the study. Patients were also excluded if they had an autoimmune disease other than rheumatoid arthritis, ACR functional class IV disease, an active infection and history of recurrent clinically significant infections or recurrent bacterial infections. Screening for tuberculosis was not done, as rituximab has not been shown to reactivate tuberculosis. Study protocol {#Sec4} -------------- The ethics committee of the Faculty of Medicine, University of Colombo, Sri Lanka, approved the study. The trial was registered prior to commencement with the Sri Lanka Clinical Trials Registry (SLCTR), a publicly accessible primary registry ([www.slctr.lk](http://www.slctr.lk)) recognized by the World Health Organization, and the protocol can be accessed through this registry. Written informed consent was obtained from all study participants. At study entry, all patients were taking methotrexate at a stable dose of 10-20 mg for at least 12 weeks. Forty patients were randomly assigned to the two groups using block randomization. One group received two infusions of rituximab 500 mg reconstituted to 500 ml 0.9% sodium chloride intravenous infusions on days 0 and days 14, together with placebo tablets manufactured matching to yellow colored leflunomide tablets. The second group received leflunomide10mg as starting dose increased to 20 mg, in the absence of elevation of liver enzymes with a placebo infusion of 500 ml of 0.9% sodium chloride. Loading dose of leflunomide was not given since it has not shown any advantage over standard maintenance dose \[[@CR24]\]. Methotrexate, steroids and NSAID's were continued in stable doses in both groups. Any other DMARDs that the patient was using (eg. sulphasalazine, hydroxychloroquine) were stopped prior to randomization. Both groups received a 14-day course of steroids, 100 mg of methylprednisolone 2 h prior to the infusion and prednisolone 60 mg daily (on days 2 to 7) and 30 mg daily (on days 8 to 14), similar to steroids regimens used in the initial rituximab trials \[[@CR12]\]. Patients were subsequently continued on their maintenance steroid doses. Premedication with IV chlorpheniramine 4 mg and oral paracetamol 1 g were given prior to each infusion to both groups. Patients were screened and enrolled into the study based on inclusion and exclusion criteria and were given a study number in the order that they entered the study by HW. Random allocation sequence was generated by PG using excel random sequence generator version 1.0 using blocks of 10 (block randomisation) to allocate patients in 1:1 ratio to the two groups ensuring equal numbers are allocated to the two groups. Forty opaque sealed envelopes were prepared, indicating the allocation of each randomised patient. The envelopes and the allocation sequence were kept under lock and key with one investigator (PG) who gave envelopes to study coordinators when each patient was enrolled into the study, for preparation of rituximab or placebo infusion. Leflunomide, which came as foil packed identifiable tablets were removed from foil and were put into air tight containers containing one months supply, with label of the bottle containing the patient ID. Identical looking yellow coloured placebo tablets manufactured by the same company manufacturing leflunomide were put into airtight containers similarly labelled with one month supply, which were given to patients randomised to rituximab. Study nurses and pharmacists prepared, supplied and administered the study drugs to the patients, supervised by unblinded study coordinators ensuring accurate study drug administration. Patient assessment was done by investigators (LSW and LW) who were blinded to treatment. Blinding of patients was maintained throughout the study with administration of matching placebo tablets and 0.9% saline to leflunomide and rituximab arms respectively as described. Outcome assessments {#Sec5} ------------------- Patients were assessed at weeks 4, 8, 12, 16, 20, and 24. Patients and those assessing clinical response (LSW and LW) remained blinded to the treatment. At each visit, history, physical examination and the ACR core set of disease-activity measures and DAS 28 scores were assessed. This included swollen joints count (66 joints), tender joint count (68 joints), patient's assessment of pain on a scale from 0 (no pain) to100 (unbearable pain), patient's global assessment of disease activity on a scale from 0 (disease inactive) to 100 (maximal disease activity), physician's assessment of disease activity and patient's assessment of physical function by health-assessment questionnaire (HAQ). Disease-activity score, DAS 28 (with physician's assessment of 28 joints and patient's self-assessment of disease activity), and the response according to the criteria of the European League Against Rheumatism (EULAR response) were also measured. Laboratory monitoring included measurement of inflammatory markers, blood counts and routine biochemistry. Baseline and 24 week B cell counts, memory B cell (CD19 + 27+) percentages and T cell counts were measured using a Coulter Epics XL flow cytometer. Pneumococcal and tetanus antibody titers were measured at baseline and at 24 weeks by Enzyme linked immunosorbent assay (ELISA). Patients were followed for up to 48 weeks for adverse effects. End points {#Sec6} ---------- The primary endpoint was 20% improvement in the American College of Rheumatology core set of disease-activity measures (ACR 20 response) response at 24 weeks. The secondary outcomes included changes in 50% and 70% improvement (ACR50 and ACR70 scores respectively), DAS, EULAR disease score and changes in different immune markers. Statistical tests and analysis {#Sec7} ------------------------------ Sample size and power of the study was calculated based on the proportion of patients achieving ACR 20 in either group. Assuming that the proportion of patients receiving methotrexate-leflunamide group achieving an ACR 20 response at week 24 would be 25% and the proportion of patients reaching ACR 20 in methotrexate-rituximab group would be 55% (observed with 500 mg rituximab in study by Emery et al.) \[[@CR16]\], total of 38 patients was required to detect a 30% difference between the two groups, giving 70% power for a significance level of 0.05. Therefore 40 patients with DAS of more than 3.2 were randomly assigned to receive leflunomide (10-20 mg/day) or rituximab (500 mg on days 1 and 15) added to methotrexate(10-20 mg/week). The primary analyses were based on intention- to-treat analysis. Statistical analyses were done using non-parametric tests, Mann-Whitney U and Wilcoxan Rank Sum test for paired comparisons and Chi square test for comparison of categorical variables using SPSS 15. An investigator with expertise in statistics did the statistical analyses (US). The local agents of rituximab provided the study drugs, some laboratory reagents and a study coordinator, and no payment was made to the patients or the investigators. The investigators independently developed the trial protocol, collected data, analysed the results and wrote the manuscript. Results {#Sec8} ======= Flow diagram of participants at each stage of the trial is given in Fig. [1](#Fig1){ref-type="fig"}. Baseline characteristic were comparable among the 40 patients in the two groups with no significant differences, except for the mean disease duration, which was longer in the leflunomide group \[Table [1](#Tab1){ref-type="table"}\]. The study population was primarily female and had RA for a mean duration of 67 months. All patients had similar disease activity with a DAS score of \>5.1 at enrolment.Fig. 1Consort flow diagram of participants at each stage Table 1Baseline Characteristics of the Patients in the two groupsRituximab Group (*n* = 20)Leflunomide Group (*n* = 19)*P* valueMean Age in Years(+/−SD)44.15(±12)48.2(±10)0.318Female sex(%)16(80%)18(95%)0.169^a^Mean Disease duration in months(+/−SD)67.9 (±80)102.8 (±63)0.034Mean number of swollen joints (+/−SD)9.35(±4)7.58(±4)0.074Disease activity Score (DAS) (+/−SD)6.88(±1)6.43(±0.5)0.067Mean methotrexate dose in mg17160.67Prednisolone dose8.0 mg7.76 mg0.730Patients who were on sulphasalazine and or hydroxychroloquine2019NTPatients with erosions on plain X Ray4 (20%)5 (26%)0.640HAQ score8.23 (±3.36)8.07 (± 3.37)0.843Tender joint count16.35(± 6.31)12.89(5.62)0.955Pain score (out of 100)68.50 (±22.64)72.53 (±15.70)0.723Patient assessment score (out of 100)67.4 (±19.02)64.37 (± 15.18)0.743Physician assessment score (out of 100)68.85(±16.96)64.05 (± 15.09)0.3539ESR\*71.90 (±36.41)62.05 (±28.92)0.473CRP\*\*30 (0-600)18 (0-140)0.20RF\*\*\*256 (0-1024)128 (0-512)0.524*n* number of patients, *SD* Standard Deviation, *RF* Rheumatoid Factor, *NA* Not applicable. *P* values for difference between means were compared using Mann Whitney U test ^a^For comparison of gender, Chi-square test was used for categorical variable. \*ESR was measured in mm/h. \*\*Assay cut off for CRP = 6 mg/L and tested using particle agglutination test, \*\*\*Assay cut off for RF = 20 IU/ml and tested using particle agglutination test. NT- Not tested (significance cannot be tested since one value is 100%) The primary endpoint ACR20, at 24 weeks in the rituximab and leflunomide groups were 85% and 84% respectively, and this difference was not statistically significant \[Table [2](#Tab2){ref-type="table"}\].Table 2Clinical Responses at Weeks 24 and mean DAS at baselineACR ResponseRituximab Group (*n* = 20)Leflunomide Group(*n* = 19)*P* valueNo ACR response3 (15%)3 (16%)0.93ACR 2017 (85%)16(84%)0.93ACR 5012 (60%)12 (64%)0.79ACR 707 (35%)6(32%)0.84Mean DAS at base line6.88 (±0.93)^a^6.43(±0.46)^b^.067\*Mean DAS at 24 weeks3.26 (±0.74)^a^3.25 (±1.02)^b^0.84\*DAS remission \<2.64 (20%)5 (26%)0.640DAS low activity \<3.28 (40%)8 (42%)0.894DAS Moderate activity(3.2-5.1)12 (60%)11 (58%)0.894DAS High activity (\>5.1)0 (0%)0 (0%)NTEULAR response moderate12 (60%)11(58%)0.894EULAR good response8 (40%)8 (42%)0.894HAQ score2.872 (±2.087)2.132 (±1.240)0.388Tender joint count1.80 (± 2.26)1.16 (1.74)0.564Pain score23.70 (±22.64)22.7 (±18.20)0.499Physician global assessment score11.75 (±10.51)15.74 (±17.62)0.983Patient global assessment score20.25(±16.12)20.0 (±15.81)0.927Anti-Pneumococcal antibody (units/ml)132.05 (±81.3)116.7 (±82.1)0.429Anti tetanus antibody0.760 (±0.43)0.842(±0.32)0.7414^\#^CD31624.4 (±847.1)1811.1 (±837.5)0.50^\#^CD1942.38 (±62.6)209 (±164.6)0.0002^\#^CD2710.1 (±8.7)49.38 (±37.9)0.0061^\#^ESR\*28.05(±16.55)30.42(±18.42)0.535CRP\*\*6 (0-84)3(0-54)0.7RF\*\*\*84(0- 372)60 (0-720)0.92*ACR* American College of Rheumatology, *DAS* Disease activity Scores, *EULAR* European League Against Rheumatism good response criteria**,** *NT* Not Tested. \*ESR was measured in mm/h. \*\*Assay cut off for CRP = 6 mg/L and tested using particle agglutination test, \*\*\*Assay cut off for RF = 20 IU/ml and tested using particle agglutination test. Anti tetanus antibody \<0.01 IU/ml - Non protective, 0.01-- 0.09 IU/ml - No reliable protection Anti pneumococcal anti body - Minimum accepted level 20 U/ml Binary outcomes were compared using chi-square test \#Numeric outcomes were compared using Mann Witney U test ^a^DAS at baseline and at 24 weeks in rituximab group, *p* \< 0.001 based on paired t-test ^b^DAS between baseline and at 24 weeks, in leflunomide group, *p* \< 0.001 based on paired t-test The EULAR responses were also high in both groups. At baseline, both groups had a high disease activity (DAS28 \> 5.1) in 95% of the rituximab and 100% of the leflunomide group. At 24 weeks low disease activity or DAS \< 3.2 as well as EULAR good response was seen in 40% of the rituximab and 42% of the leflunomide group respectively with none of the patients having DAS high disease activity (DAS \>5.1) \[Table [2](#Tab2){ref-type="table"}\]. The addition of either medication produced significant changes in disease activity scores from baseline level. None of the differences in clinical responses in any of the outcome criteria assessed in the two groups were statistically significant \[Table [2](#Tab2){ref-type="table"}\]. There were no significant differences in B cell, T cell or B cell memory percentages between the two groups at the start of the study. Compared to baseline, 24 week post-treatment levels showed the rituximab group having significant reduction in B cells (*p* \< 0.001) and memory B cells (*p* \< 0.001), \[Fig. [2a and b](#Fig2){ref-type="fig"}\] and pneumococcal antibody levels (*p* \< 0.05) \[Fig. [3b](#Fig3){ref-type="fig"}\] with no significant change in T cells (*p* \> 0.05) \[Fig. [2c](#Fig2){ref-type="fig"}\] or tetanus antibody levels (*p* \> 0.05) \[Fig. [3a](#Fig3){ref-type="fig"}\]. The leflunomide group also showed significant change in memory B cells (*p* \< 0.05) but T cells \[Fig. [2](#Fig2){ref-type="fig"}\] and tetanus antibody levels \[Fig. [3b](#Fig3){ref-type="fig"}\] did not show significant difference from baseline (*P* \> 0.05). Both groups showed a significant reduction in pneumococcal antibody levels (*P* \< 0.05) \[Fig. [3a](#Fig3){ref-type="fig"}\] and B memory cells (*P* \< 0.01) \[Fig. [2a](#Fig2){ref-type="fig"}\]. There were no significant differences in other laboratory measurements in either group during the study period (rheumatoid factor, ESR, CRP, IgG, IgM levels and liver function tests).Fig. 2**a** **b** and **c** Measured B and T lymphocyte counts at baseline and at 24 weeks. **a** Memory B lymphocyte counts (CD 19+ 27+) at baseline and 24 weeks. Leflunomide shows significant difference in Memory B lymphocytes (CD19 + 27+) at 24 weeks compared to baseline (*P* \< 0.01). Rituximab shows significant difference of Memory B lymphocytes (CD19 + 27+) at 24 weeks compared to baseline (*p* \< 0.001). **b** Mean B lymphocyte counts (CD 19) at baseline and 24 weeks. Leflunomide shows no significant difference in B lymphocyte counts (CD19) at 24 weeks compared to baseline (*P* \> 0.05). Rituximab shows significant difference in B lymphocyte counts (CD 19) at 24 weeks compared to baseline (*p* = \<0.001). **c** T lymphocyte counts (CD 3) in patients at baseline and 24 weeks. Leflunomide shows no significant difference in T lymphocyte counts (CD 3) at 24 weeks compared to baseline (*P* \> 0.05). Rituximab shows no significant difference in T lymphocyte counts (CD 3) at 24 weeks compared to baseline (*P* \> 0.05). Data Analysed using non-parametric Wilcoxan Rank Sum test as paired samples Fig. 3**a** and **b** Anti pneumococcal and anti tetanus antibody status at baseline and at 24 weeks. **a** Anti pneumococcal antibody titres at baseline and 24 weeks. Leflunomide shows significant difference of anti pneumococcal antibodies at 24 weeks compared to baseline (*P* \< 0.05). Rituximab shows significant difference of anti pneumococcal antibodies at 24 weeks compared to baseline (*P* \< 0.05) Anti pneumococcal anti body - Minimum accepted level - 20 U/ml. **b** Anti-tetanus antibody titres at baseline and 24 weeks. Leflunomide shows no significant difference of anti-tetanus antibodies at 24 weeks compared to baseline (*P* \> 0.05). Rituximab shows no significant difference of anti tetanus antibodies at 24 weeks compared to baseline (*P* \> 0.05). Data Analysed using non parametric Wilcoxan Rank Sum test as paired samples. Anti tetanus antibody \<0.01 IU/ml Non protective, 0.01-- 0.09 IU/ml No reliable protection There were five serious adverse events (SAE) in the rituximab group and three in the leflunomide group (including one death). In the rituximab group, four SAEs were due to infections, two patients developing lower respiratory tract infections, one having cellulitis in leg and the other having an infected toe, and all requiring IV antibiotics. In patients who had respiratory infections, the organism could not be identified but tuberculosis was excluded using the relevant investigations. In the leflunomide group, one patient developed an infective diarrhea (no organism isolated) and the other two patients had cardiac events. There were total of 3 cardiac events in both groups. In the rituximab group, one patient developed unstable angina at 16 weeks and was treated medically. In the leflunomide group, there were two cardiac events, one patient was admitted complaining of non-specific chest pain with elevated blood pressure and the other patient died of a myocardial infarction (MI). This patient with MI was admitted to hospital with a febrile illness and whilst in hospital developed an extensive anterior MI. The patient had multiple cardiovascular risk factors including long standing RA, prolonged NSAID use intermittently (more than 5 years), hypertension for 10 years and dyslipidemia. In the causality assessment the death was not considered to be due to study drug. Discussion {#Sec9} ========== In this randomized double-blind controlled clinical trial among patients with rheumatoid arthritis refractory to conventional DMARDs, leflunomide-methotrexate combination was as efficacious as two infusions of low-dose rituximab (500 mg)-methotrexate combination, indicated by similar primary endpoints (ACR20) at 24 weeks. Although the steroids used in the protocol could have partly contributed to the responses seen in both groups, the previous studies using similar steroid regimens have shown that this is unlikely \[[@CR16]\]. The response rates observed in this study are also higher than the rates reported from European trials using similar steroid regimens. The ACR 20, 50 and 70 achieved was 73%, 43% and 23% in the trial done by Edward et al. with 1000 mg rituximab \[[@CR12]\] where similar steroid regimen was used. The study by Emery et al. using similar steroid regimen (100 mg methylprednisolone administered IV prior to rituximab infusions on days 1 and 15, and prednisone administered orally at 60 mg on days 2--7 and at 30 mg on days 8--14) has shown that glucocorticoid treatment had no significant effect on ACR20 response, through detailed statistical analysis on their data \[[@CR16]\]. However they have shown that patients who received this regimen of glucocorticoids were more likely to achieve an ACR20 response early, (by week 4) than those who received no glucocorticoids. Therefore it is unlikely that the higher effects we observed were due to the use of high dose steroids in both arms. The study by Emery et al. \[[@CR16]\] used 2 doses of 500-mg infusions of rituximab with MTX in one arm and at week 24, ACR20, 50 and 70 was achieved in 55%, 33% and 13% respectively in that arm. The SERENE study \[[@CR17]\] achieved 54.5% 26.3% and 9% rates of ACR 20, 50 and 70 with two doses of rituximab 500 mg. Thus achieving ACR 20, 50, 70 in 85%, 60% and 35% respectively in the low dose rituximab group in our study suggest a higher response in our population. Higher responses rates were also achieved for EULAR good response and DAS low activity in our study, (Table [2](#Tab2){ref-type="table"}) compared to response rates reported in previous studies. In the SERENE study \[[@CR17]\] the EULAR good response rates were seen only in 17.4% and 11.8% patients receiving two doses of 500 mg and 1 g of rituximab respectively. In the study by Emery et al. also, EULAR good response was achieved in only 28% and 14% patients receiving rituximab 500 mg and 1 g doses respectively \[[@CR16]\]. Therefore our study showing DAS low activity and EULAR good response in 40% and 42% in the two arms are much higher compared to the response rates observed in other similar studies. This makes our results even more significant. The trial comparing methotrexate-leflunomide therapy in non-responders to methotrexate \[[@CR21]\] showed an ACR 20, 50,70 of 46%, 26% and 10% for the leflunomide-methotrexate group. Thus the responses seen in our methotrexate and leflunomide group of 85%, 64% and 32% for ACR 20, 50,70 responses observed in the leflunomide group are also comparatively higher in our population. These data raise the question whether South Asian populations respond differently to these drug combinations than their European counterparts giving these higher response rates. These results are also important as the response rate of low dose rituximab in our study is comparable to response rates seen with high dose standard rituximab therapy, reported in European trials \[[@CR12], [@CR13], [@CR16], [@CR17]\]. Some of the recent evidence from other studies also suggest that low dose rituximab (500 mg) has almost similar efficacy as the high dose of 1000 mg rituximab \[[@CR17]--[@CR19]\]. Rituximab causes B lymphopenia and a reduction in memory B-lymphocytes. Although rituximab produced B cell lymphopenia as expected, no measurable reduction from baseline levels were seen in tetanus antibody levels. However there was a significant reduction in pneumococcal antibody levels with both drugs \[Fig. [3a](#Fig3){ref-type="fig"}\]. The leflunomide group also showed a significant reduction in memory B cells \[Fig. [2a](#Fig2){ref-type="fig"}\] although B cells, T cells \[Fig. [2b and c](#Fig2){ref-type="fig"}\] and tetanus antibody levels \[Fig. [2b](#Fig2){ref-type="fig"}\] did not show a significant reduction from baseline. The observed changes in immune parameters with leflunomide are novel and need further characterization. Leflunomide showing comparable efficacy to rituximab could be explained to some extent by the significant reduction of memory B cells seen with leflunomide, similar to the effect produced by rituximab. We observed a trend towards increased incidence of infections in the rituximab group with four patients requiring hospitalizations due to infections. Increased incidence of lower respiratory tract infections with rituximab has been reported from the European trials \[[@CR15]\]. Although pneumococcal infections were not increased in our population, since pneumococcal antibody levels significantly reduced in both groups, possibility of reduced protection against pneumococcal infections with both treatments is a concern. However studies of rituximab, including those with long term follow up, in combination with other biologics and biologic registry data have shown that the incidence of serious infections with rituximab is low \[[@CR25]--[@CR28]\]. Our study also supports these findings, as there were no serious infections in our treatment population, which is reassuring as infections are more common in our setting. No patients in our study developed tuberculosis during the study period. Like many countries in South and South-East Asia, TB is endemic in Sri Lanka and the use of a medication that does not increase the baseline risk of tuberculosis is advantageous. Few patients from both groups had some cardiac events, which is well recognized among patients with RA. Significant increase in liver toxicity observed in some other reports \[[@CR21]\] was not seen in the methotrexate-leflunomide group. One of the limitations of our study is the small sample size, due to the limited availability of only 40 vials of rituximab for this study. Though the trial was small, it had 70% power to detect a 30% difference in the treatment groups. It is possible that with a larger sample size, smaller differences between the two groups could have been detected. Further observations from larger clinical trials may help to confirm the findings from this study. This study suggests that comparatively cheaper therapies such as methotrexate-leflunomide combination or low dose rituximab-methotrexate combination may be beneficial in South Asian patients who are not responding to standard first line treatment in RA. The trial data has significant beneficial cost implications for treatment of patients with refractory rheumatoid arthritis, especially in resource limited settings in South Asian regions. The cost of two infusions of 1000 mg rituximab was Sri Lankan Rupees (SLRs) 744,000(US\$ 5700), according to the cost of rituximab when this trial was done. Annual costs for TNFi are also similar to this figure or even more expensive. Two infusions of rituximab of 500 mg halves this cost, to SLRs 372,000(US \$ 2850). Leflunomide at a dose of 20 mg per day drastically reduces this cost to only SLRs 25,200 (US \$ 192) for one year of treatment. Thus our finding that the response rates were similar to those reported in clinical trials using high-dose rituximab or the TNFi in European patients has highly favorable cost implications for patients living in a developing country. We highlight that with the lower dose of rituximab used, the biologic, which is relatively affordable in South East Asia is low dose rituximab. As this trial shows that Leflunomide is as efficacious as low dose rituximab given in combination with methotrexate and since recent studies \[[@CR22]\] have shown the safety of Leflunomide-methotrexate combination, this combination could be recommended prior to trying the biologics in patients with refractory RA in South East Asia. Conclusion {#Sec10} ========== This study showed that both leflunomide and low dose rituximab were equally efficacious in controlling disease activity when added to methotrexate in Sri Lankan patients with refractory RA. This has significant cost implications as leflunomide-methotrexate combination is much cheaper compared to low-dose rituximab-methotrexate combination. The lower costs of both these two treatment options should enable more patients with refractory rheumatoid arthritis to be treated successfully in resource-limited settings. ACR : American college of rheumatology CTLA4 : Cytotoxic T lymphocyte Antigen-4 DAS : Disease activity score DMARDs : Disease modifying anti rheumatic drugs ELISA : Enzyme linked immunosorbent assay ESR : Erythrocyte sedimentation rate EULAR : European league against rheumatism HAQ : Health-assessment questionnaire MI : Myocardial infarction NHSL : National Hospital of Sri Lanka NSAIDS : Non-steroidal anti-inflammatory drugs RA : rheumatoid arthritis SAE : Serious adverse events SERENE : Study Evaluating Rituximab\'s Efficacy in MTX iNadequate rEsponders SLCTR : Sri Lanka clinical trials registry SLRs : Sri Lankan Rupees SMILE : Safety of Methotrexate In combination with LEflunomide in rheumatoid arthritis TB : Tuberculosis TNFi : Tumor necrosis factor inhibitors US\$ : United States Dollars We wish to thank the local agents of rituximab for supplying the study medications free of charge for the clinical trial and providing test kits for some laboratory tests. We also acknowledge local agents of leflunomide for supplying the study medications and placebo tablets. Dr. T Pathirajamudali, Dr. D Rathnayake, Dr. G D Wickramasinghe, Ms. G Ratnayaka Ms. Chandrani, Mr. A Lionel and Mr. Nanayakkara are acknowledged for support given in performing the laboratory tests. Funding {#FPar1} ======= The local agents of rituximab provided the study medications, some laboratory reagents and a study coordinator. The local agents of leflunomide provided study medications and matching placebo tablets. There was no payment for the patients or the investigators. The investigators independently developed the trial protocol, collected data, analysed the results and wrote the manuscript. Availability of data and materials {#FPar2} ================================== The datasets used and/or analysed during the current study are available from the first author or corresponding author on reasonable request. HW, PG, RS, SLS, RF, LPW and LSW contributed in designing the study. HW, PG and LSW submitted documents and obtained ethics and administrative approvals and clinical trials registration. Patients were screened and enrolled into the study by HW. Random allocation sequence was generated and kept under lock and key by PG. Patient assessment was done by LSW and LPW. Investigations were done and analysed by RS, MH, PK, PU and US. PG and RF reviewed the adverse events. The sample size calculation and other statistical analyses were done by US, PG and RS. HW, PG and SLS prepared the manuscript and it was reviewed and modified by other co-authors. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== The ethics review committee of Faculty of Medicine, University of Colombo approved the study which was conducted in accordance with Good Clinical Practice guidelines and principles stated in Declaration of Helsinki. All patients provided written informed consent to participate in the study. Consent for publication {#FPar4} ======================= Not applicable. Competing interests {#FPar5} =================== The sponsor paid salary of research assistants who were recruited independently by the principal investigators for the conduct of the study. The principle investigator and other co-investigators were not given any payment in cash or kind for the conduct of the study and have no competing interest with the sponsors. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction {#Sec1} ============ Epigenome-wide association studies (EWAS) have shown that a substantial amount of variation in DNA methylation (DNAme) exists between human populations \[[@CR1]--[@CR7]\]. Therefore, if left unaccounted for, population-associated variation can interfere with the discovery of DNAme alterations associated with disease or environment. This type of confounding, often referred to as population stratification, can be addressed by inferring population-associated variation directly from DNAme data itself \[[@CR8]--[@CR10]\], as is done in genome-wide association studies (GWAS) \[[@CR11]\]. However, unlike genetic markers, epigenetic markers are tissue-specific and, therefore, a DNAme-based method developed in a specific tissue or population may not generalize well to other tissues with unique DNAme profiles. In EWAS, confounding from population stratification is most often addressed using self-reported ethnicity/race to stratify study samples across the phenotype of interest. But, defining ethnicity/race is a complex task requiring the interpretation of a combination of biological and social factors leading to several complications: (i) inconsistent definition of ethnicity/race categories between individuals/organizations \[[@CR12], [@CR13]\]; (ii) self-reporting more than one ethnicity/race \[[@CR14]\]; and (iii) missing ethnicity information altogether. To overcome the limitations of ethnicity/race categories, genetically defined ancestry can be used \[[@CR15]\] as an alternative measure of population-specific variation. In contrast to the discrete nature of ethnicity/race categories, genetic ancestry can be expressed as several continuous variables that reflect ancestry composition \[[@CR16]\]. Though the use of genetic ancestry could help to better design EWAS, genotyping markers might not be collected in DNAme studies. In cases where self-reported ethnicity and genetic ancestry information are unavailable, methods have been developed to infer this information directly from DNAme (Table [1](#Tab1){ref-type="table"}) measured on the popular Infinium Human Methylation 450k Beadchip array (HM450K) \[[@CR8]--[@CR10], [@CR17]\]. Barfield et al. \[[@CR8]\] and EPISTRUCTURE \[[@CR9]\] methods both utilize principal components analysis (PCA) on select DNAme sites to infer genetic ancestry. Since only DNAme sites that are associated with nearby genetic variation are used, these methods produce principal components (PCs) that are often highly correlated with genome-wide genetic variation \[[@CR8], [@CR9]\] and, therefore, can be used as a measurement of genetic ancestry. Zhou et al. \[[@CR10]\] explored using the set of 65 SNPs measured on HM450K to produce ethnicity/race classifications. However, it has not been investigated whether these methods perform well in populations and tissues other than the ones they were developed and tested in (Table [1](#Tab1){ref-type="table"}).Table 1Description of methods to infer self-reported ethnicity or genetic ancestry using HM450K dataName of methodStatistical approachInput HM450K sitesOutputSample characteristicsTissuePopulations^a^Cohort locationBarfield et al. \[[@CR8]\]PCA7703 DNAme sites with a 1000 genomes project SNP at the CpG siteGenetic ancestry as PC scoresBloodCaucasian-Americans, African--AmericansUSAEPISTRUCTURE \[[@CR9]\]PCA4913 DNAme sites associated with local genetic variation (mQTLs)Genetic ancestry as PC scoresBloodEuropeans, Puerto Ricans, MexicansSouthern Germany; USAZhou et al. \[[@CR10]\]Predictive-modeling59/65 SNP sitesEthnicityMultipleWhite, Black or African American, AsianManyPlaNET; this studyPredictive-modeling15 SNPs; 1845 DNAme sitesEthnicity and genetic ancestryPlacentaCaucasians, Asians, AfricansCanada, USA^a^Ethnicity/ancestry as defined in associated study DNAme studies using placental tissue are of particular interest because the functioning of the placenta is essential to a healthy pregnancy \[[@CR18], [@CR19]\]. Although many DNAme alterations associated with placental-mediated diseases have been identified \[[@CR20]--[@CR23]\], the incidence of many of these conditions vary by population \[[@CR24]--[@CR26]\]. In this study we developed PlaNET (Placental DNAme Elastic Net Ethnicity Tool), an ethnicity classifier, using DNAme and genotyping data measured on the HM450K array in multiple cohorts of placentas from North America. PlaNET was developed on overlapping sites from HM450K and the newer Illumina MethylationEPIC BeadChip array (EPIC) to ensure compatibility with future studies. We show that PlaNET out-performs existing methods in predicting ethnicity in placental tissue and can produce accurate measures of genetic ancestry. Importantly, our method can be used to classify individuals into discrete ancestral populations (i.e., African, Asian, and Caucasian) or to describe individuals on an ancestral continuum that may more accurately reflect the nature of modern human populations. In studies where ethnicity information is unavailable, PlaNET can be applied to predict ethnicity after obtaining DNAme data, and used to investigate population-specific differences or to minimize confounding by population stratification in statistical analyses. Results {#Sec2} ======= Datasets {#Sec3} -------- Our goal was to develop a placental DNAme-based ethnicity classifier, which could learn ethnicity-specific DNAme patterns from one set of samples to assign ethnicity labels to a new set of samples. We searched for placental HM450K data on the Gene Expression Omnibus \[[@CR27]\] that contained more than one ethnicity group and made sample-specific ethnicity information available (Table [2](#Tab2){ref-type="table"}). Five distinct cohorts met these criteria (labeled C1--C5), with three major North American ethnicities represented by sufficiently large numbers across more than one dataset: African (*n* = 58), Asian (*n* = 53), and Caucasian (*n* = 389). We opted to include samples from both healthy and abnormal pregnancies (preeclampsia, gestational diabetes mellitus, fetal growth restriction or overgrowth) (Table [2](#Tab2){ref-type="table"}) \[[@CR21], [@CR28]--[@CR33]\]. Though there were significant cohort-specific effects on DNAme that may reflect batch/technical variation (Additional file [2](#MOESM2){ref-type="media"}: Figure S1), we included these multiple datasets and phenotypes to enable the development of a robust classifier that would generalize well in future studies \[[@CR34]\].Table 2Description of HM450K DNAme datasets used to develop and test PlaNETCohort (*n*)GEO accessionDataset summaryLocationSelf-reported ethnicityNon-HM450K genetic data (*n*)AFR (*n* = 57)ASI (*n* = 53)CAU (*n* = 389)C1 (72)GSE7045336 controls, 36 gestational diabetes mellitusBoston, MA, USA131346N/AC2 (24)GSE7337513 controls, 11 preeclampsia (PE)Chapel Hill, NC, USA13110N/AC3 (289)GSE75248289 samples from infants with variable newborn neurobehaviorRI, USA; MA, USA239257N/AC4 (44)GSE10019717 controls, 27 PEToronto, CAN7122550 AIMs (41)C5 (70)GSE100197, GSE108567, GSE74738, unpublished35 controls, 35 fetal growth restriction, PE, and/or preterm birthVancouver, CAN1185150 AIMs (67); Omni2.5 (27)*AFR* African, *ASI* Asian, *CAU* Caucasian Development of a placental DNA methylation ethnicity classifier {#Sec4} --------------------------------------------------------------- To determine the best machine learning classification algorithm that could learn ethnicity-specific patterns from DNAme microarray data, we compared four algorithms previously shown to be well-suited for prediction using high-dimensional genomics data \[[@CR34]--[@CR36]\]: generalized logistic regression with an elastic net penalty (GLMNET) \[[@CR37], [@CR38]\], nearest shrunken centroids (NSC) \[[@CR35]\], k-nearest neighbours (KNN) \[[@CR39]\], and support vector machines (SVM) \[[@CR40]\]. For each algorithm, hyperparameter(s) were selected (e.g. *k* number of neighbors for KNN) that resulted in the highest performance estimated by repeated fivefold cross validation (three repeats). All algorithms performed favorably (logLoss = 0.170--0.276; Additional file [2](#MOESM2){ref-type="media"}: Figure S2a), except KNN (logLoss = 1.82). However, all algorithms showed a bias for high predictability of Caucasians (average accuracy = 0.980), and low predictability of Asians (average accuracy = 0.448) (Additional file [2](#MOESM2){ref-type="media"}: Figure S2b). Considering overall- and ethnicity-specific performance, the GLMNET algorithm was used for the remainder of the study (accuracy = 0.866, 0.625, 0.998 for Africans, Asians, and Caucasians, respectively), and we refer to this classifier as PlaNET (Placental DNAme Elastic Net Ethnicity Tool). For each sample, PlaNET returns a probability that the sample is African, Asian or Caucasian, and the final classification is defined by the ethnicity class with the highest of these probabilities. We reason that these probabilities have the potential to identify samples with mixed ancestry or ethnicity. Therefore, we implemented a threshold function on PlaNET's probability outputs that classifies samples as 'Ambiguous' if the highest of the three class-specific probabilities is below 0.75 ("[Methods](#Sec11){ref-type="sec"}", Additional file [2](#MOESM2){ref-type="media"}: Figure S3). This resulted in 7 self-reported African, 12 Asian, and 13 Caucasian samples as being classified as ambiguous, which led to a slight decrease in performance (Fig. [1](#Fig1){ref-type="fig"}a). However, we note that because genetic ancestry is on a continuum and due to the limitations of self-reported ethnicity, there are likely to be individuals of mixed ancestry/ethnicity in our sample set and, therefore, hypothesize that a model that includes an ambiguous class is more realistic and accurate than one without. Cross validation, where training/validation subsets were created based on cohort-identity, yielded an overall accuracy of 0.900, a Kappa of 0.738, and a positive predictive value of 0.944 (Fig. [1](#Fig1){ref-type="fig"}a), which was consistent when examining performance by dataset (Additional file [2](#MOESM2){ref-type="media"}: Figure S4).Fig. 1Evaluating PlaNET's performance and characterizing ethnicity-predictive HM450K sites. We developed PlaNET (Placental elastic net ethnicity classifier), using placental HM450K data and evaluated its classification performance using leave-one-dataset-out cross validation. **a** Each sample's ethnicity classification from PlaNET is shown with respect to their self-reported ethnicity. Samples were called 'ambiguous' if their predicted probability fell below a 'confidence' threshold of 75%. **b** PlaNET utilizes a subset of ethnicity-predictive sites from the HM450K. To investigate whether genetic signal is present in the measurement for these sites, we cross-referenced ethnicity-predictive sites to an existing placental mQTL database \[[@CR42]\] and determined whether any sites had SNPs present in either the probe body, CpG site of interrogation, or single base extension sites, based on dbSNP137 Ethnicity-predictive sites on the HM450K array are largely linked to genetic variation {#Sec5} -------------------------------------------------------------------------------------- To better understand the basis of PlaNET's ethnicity prediction, we examined the 1860 sites (Additional file [1](#MOESM1){ref-type="media"}: Table S1) automatically selected by the GLMNET model. These sites were enriched for SNP probes, containing 15 of the 59 SNPs explicitly measured on both HM450K and EPIC DNAme arrays (*p* \< 1e−16). Of the remaining 1845 DNAme sites, we found significant enrichment for sites linked to genetic variation: 802 sites (43.1%) have a documented SNP in either the probe body, CpG site of interrogation, or the single base extension site (*p* \< 1e−16) \[[@CR41]\], and 220 sites (11.8%) corresponded to previously identified placental-specific methylation quantitative trait loci (mQTLs) \[[@CR42]\] (*p* \< 1e−16, Fig. [1](#Fig1){ref-type="fig"}b). With respect to chromosomal location, we found significant enrichment for ethnicity-predictive sites on chromosomes 2 (*p* \< 0.01), 15 (*p* \< 0.05), and 17 (*p* \< 0.05) (Additional file [2](#MOESM2){ref-type="media"}: Figure S5a). With respect to CpG density, we found significant enrichment for ethnicity-predictive sites in OpenSea (*p* \< 0.001) and South Shore (*p* \< 0.05) regions (Additional file [2](#MOESM2){ref-type="media"}: Figure S5b), where relatively neutral (unselected) genetic variation is more likely to be located \[[@CR43]\]. Pathway analysis for GO and KEGG terms for genes associated with the 1860 sites, found only one significant (*p* \< 0.05) GO term (homophilic cell adhesion via plasma membrane adhesion molecules). DNAme -inferred ethnicity and genetic ancestry {#Sec6} ---------------------------------------------- To test the ability of PlaNET to identify individuals of mixed ancestry, we examined whether samples classified as 'ambiguous' were also intermediate with respect to genetically defined ancestry. Genetic ancestry was inferred from 50 ancestry informative genotyping markers (AIMs) in samples from cohorts C4 and C5 (*n* = 109), using 1000 Genomes Project samples as reference populations \[[@CR44], [@CR45]\]. These 50 markers were previously selected based on their ability to differentiate between African, European, East Asian, and South Asian populations \[[@CR45]\]. Plotting the first two multi-dimensional scaling coordinates calculated on the 50 AIMs in (Fig. [2](#Fig2){ref-type="fig"}), shows a handful of samples intermediate to three more distinct ancestry clusters. The samples with less extreme genetic ancestry coordinates based on AIMs tended to have lower PlaNET-calculated probabilities associated with the ethnicity classification matching the individual's self-reported ethnicity (Fig. [2](#Fig2){ref-type="fig"}), confirming that PlaNET provides some information on the genetic ancestry composition.Fig. 2Probabilities associated with PlaNET ethnicity predictions and genetic ancestry inferred from AIMs. Ethnicity classifications from PlaNET and associated confidence/probability scores were compared to genetic ancestry inferred from 50 AIMs (*n* = 109, cohorts C4, C5), represented by the first three coordinates from multidimensional scaling using 1000 genomes project samples as reference populations Although genetic ancestry can be adequately inferred from a small set of AIMs, it is best obtained from a large number of unlinked markers \[[@CR46]\]. Therefore, we also inferred genetic ancestry in a smaller number of samples from C5 (*n* = 37) with high-density genotyping array data (Omni 2.5, \> 2.5 million SNPs), again using 1000 Genomes Project samples as reference populations \[[@CR44], [@CR47], [@CR48]\], and compared this to PlaNET's predicted membership probabilities for each ethnicity (Fig. [3](#Fig3){ref-type="fig"}a--c). 10 of these 37 samples were not initially used for previous analyses due to a lack of available self-reported ethnicity information (Fig. [3](#Fig3){ref-type="fig"}a). We found that genetic ancestry coefficients reflected the probabilities associated with ethnicity classification to a high degree (Fig. [3](#Fig3){ref-type="fig"}b, c, *R*^2^ = 0.95--0.96, *p* \< 0.001).Fig. 3Probabilities associated with PlaNET ethnicity predictions and genetic ancestry inferred from high-density genotyping data. PlaNET was tested in a subset of cohort C5 (*n* = 37). **a** PlaNET's ethnicity classifications were compared with self-reported ethnicity. **b** Ethnicity probabilities generated by PlaNET were compared to **c** genetic ancestry coefficients determined from high-density genotyping data (Omni 2.5, \> 2 million SNPs), using the function snmf() from the R package *LEA*, and found to be highly correlated (*R*^2^ = 0.95--0.96, *p* \< 0.001) determined by linear regression Characterizing existing methods to infer population structure in placental DNA methylation data {#Sec7} ----------------------------------------------------------------------------------------------- To evaluate our hypothesis that a placental-specific approach to population inference would outperform existing methods developed in other tissues, we compared the performance of PlaNET to three previously published HM450K methods: Barfield's SNP-based filtering approach \[[@CR8]\], EPISTRUCTURE \[[@CR9]\], and Zhou's SNP-based classifier \[[@CR10]\]. To address the differences in the type of outcomes produced by each method (e.g. PCs or ethnicity classifications), we used PCA to generate metrics that could be compared between methods. PCA was performed on the set of HM450K sites corresponding to each method (Table [1](#Tab1){ref-type="table"}), which were then included in a series of simple linear models, where each PC was either a function of self-reported ethnicity (Fig. [4](#Fig4){ref-type="fig"}a; *n* = 499, cohorts C1--C5), genetic ancestry (Fig. [4](#Fig4){ref-type="fig"}b; *n* = 109, cohorts C4 and C5 only), or cohort-specific patient variables (e.g. microarray batch, sex, gestational age; Additional file [2](#MOESM2){ref-type="media"}: Figure S6). Linear models were constructed for each of the top ten PCs and *R*^2^ for each linear model was compared between each method. For computation of PCs on PlaNET's sites, we used a cohort-specific cross validation framework to account for bias that could be introduced using the same samples for development and testing. Specifically, PlaNET's PCs were computed separately for each cohort using ethnicity-predictive sites selected in all other cohorts ("[Methods](#Sec11){ref-type="sec"}").Fig. 4Comparing PlaNET to existing methods to account for population stratification using HM450K data. For each cohort, principal components analysis was conducted on PlaNET using a model trained on all other cohorts. PlaNET's principal components (PCs) were then compared to the PCs computed on sites from EPISTRUCTURE \[[@CR9]\], Barfield's method \[[@CR8]\], and the 59 SNPs. **a** Amount of variance explained from a series of linear models where principal component "*i*" is a function of self reported ethnicity encoded as a dummy variable. **b** This was then repeated using AIMs coordinates 1 and 2 instead of ethnicity as the independent variable (*n* = 109) We found that for all cohorts, the first two PCs computed on PlaNET's sites and the 59 SNPs were highly correlated with self-reported ethnicity (Fig. [4](#Fig4){ref-type="fig"}a, *R*^2^ = 0.649 ± 0.087, 0.697 ± 0.110, respectively) and genetic ancestry (Fig. [4](#Fig4){ref-type="fig"}b, *R*^2^ = 0.555 ± 0.246, 0.487 ± 0.335). In contrast, the first PC computed on Barfield's and EPISTRUCTURE's sites showed almost no correlation with self-reported ethnicity (Fig. [4](#Fig4){ref-type="fig"}a, *R*^2^ = 0.0452 ± 0.060, 0.066 ± 0.082, respectively), or genetic ancestry (Fig. [4](#Fig4){ref-type="fig"}b, *R*^2^ = 0.0435 ± 0.0548, 0.104 ± 0.0653). Instead, for Barfield and EPISTRUCTURE, the PCs that correlated with ethnicity/ancestry were confined to PCs 3--6 (Fig. [4](#Fig4){ref-type="fig"}a, b), while often the top PCs (e.g., 1--4) for these two methods were associated with variables other than ethnicity/ancestry (Additional file [2](#MOESM2){ref-type="media"}: Figure S6). For example, in cohort C4, EPISTRUCTURE PC1 was most correlated with row position on the HM450K array (*R*^2^ = 0.482), PC2 with gestational age (*R*^2^ = 0.315), PC3 with genetic ancestry coordinate 1 (*R*^2^ = 0.450), and PC5 with ethnicity (*R*^2^ = 0.579; Additional file [2](#MOESM2){ref-type="media"}: Figure S6). Limiting to methods that predict ethnicity classes, we compared the performance of PlaNET to Zhou et al. \[[@CR73]\] SNP-based classifier (Additional file [2](#MOESM2){ref-type="media"}: Figure S7). Both classifiers demonstrated similar accuracy in classifying self-reported Africans (*p* = 0.68, 87.1% for PlaNET; 90.3% for Zhou) and Caucasians (*p* = 0.062, 96.7% vs. 97.9%), but PlaNET was more accurate in classifying self-reported Asians (*p* = 0.00052, 74.4% vs. 41.0%). Application of PlaNET in an EWAS setting {#Sec8} ---------------------------------------- Lastly, to demonstrate the utility of applying PlaNET to placental DNAme data, we applied PlaNET to obtain ethnicity classifications across two previously published EWAS studies using three datasets (Table [3](#Tab3){ref-type="table"}, Additional file [2](#MOESM2){ref-type="media"}: Figure S9). We note that this includes samples from cohorts C4 and C5 that were used to develop PlaNET.Table 3Distribution of PlaNET ethnicity predictions across previously published placental EWAS datasetsGEO accessionPrimary groupsAfricanAsianCaucasianAmbiguousGSE98224EOPET54100Preterm controls1350LOPET1181Term controls0450GSE100197EOPET15151Preterm controls14190LOPET06120Term controls02170IUGR0380GSE71678NA^a^003421*EOPET* early onset preeclampsia, *LOPET* late onset preeclampsia^a^Phenotype of interest is a continuous variable (arsenic concentration) One study used two distinct cohorts from Vancouver, Canada (GSE100197, *n* = 102) and Toronto, Canada (GSE98224, *n* = 48) to investigate placental DNAme alterations associated with preeclampsia status \[[@CR21]\]. We reasoned that correction for ethnicity should decrease false positives in the EWAS and, therefore, increase concordance between hits identified in the two data sets. In the original EWAS, with no adjustment for ethnicity, our group reported that 599 out of the 1703 (35.1%) significant associations found in the Vancouver cohort were also significant in the Toronto cohort, and the correlation of the difference in mean DNAme between controls and preeclampsia-affected samples (i.e. delta betas) at FDR significant sites between discovery and validation was 0.62 \[[@CR21]\]. When we repeated the analysis while adjusting for ethnicity determined by PlaNET, the number of preeclampsia-associated sites that overlapped between cohorts increased to 651/1614 (40.3%) (Additional file [1](#MOESM1){ref-type="media"}: Table S5), and the correlation between delta betas increased to 0.66. We also found that repeating gene set enrichment analysis, which originally found nothing significant \[[@CR21]\], yielded several significantly enriched (FDR \< 0.05) GO terms such as developmental process, inflammatory response, and cell adhesion (Additional file [1](#MOESM1){ref-type="media"}: Table S6). Lastly, we also adjusted for ethnicity determined by Zhou et al.'s SNP classifier, which resulted in a smaller increase in overlapping associations and correlation between delta betas (607/1662 = 36.52%, correlation = 0.65). However, no GO terms were found significant at an FDR \< 0.05. In summary, any adjustment for ethnicity improved the replicability of our preeclampsia EWAS results, with PlaNET performing best when used in placental samples. Next, because adjustment for population stratification can not only be done via correction in linear modeling, but can also be done by stratifying an analysis by population identity, we performed a secondary EWAS confined to samples predicted as Caucasians (*n* = 71/102 for discovery, *n* = 28/48 for validation). This resulted in a decrease in overlap in preeclampsia-associated sites between cohorts: 359/1488 (17%) (Additional file [1](#MOESM1){ref-type="media"}: Table S7), although the correlation between delta betas remained high (*r* = 0.67), indicating the observed decrease in overlap between significantly differentially methylated sites was likely due to a decrease in power from smaller sample size (particularly in the validation group) rather than a decrease in concordance between cohorts. PlaNET can be useful for checking for discrepancies in self-reported ethnicity information. We tested whether PlaNET could identify the ethnicity of samples from an all-Caucasian population. GSE71678 (*n* = 343), a cohort not used in the development of PlaNET, consisted of DNAme data from placental samples collected from a New Hampshire, USA birth cohort that investigated the effects of arsenic exposure on placental DNAme \[[@CR49]\]. PlaNET-determined 342 samples were classified as Caucasian, and 1 sample had a high probability of belonging to the Caucasian group (Probability = 0.73) but was below our confidence threshold and was, therefore, classified as 'ambiguous', confirming ethnic homogeneity was high in this cohort and adjustment for population stratification was not needed in this study. Discussion {#Sec9} ========== In this study, we developed PlaNET, a method to predict Asian, African, and Caucasian ethnicity using placental HM450K array data. To enable compatibility with future studies, PlaNET was developed on sites (452,453 CpGs and 59 SNPs) overlapping between the older HM450K and the newer EPIC Illumina DNAme arrays. Although all samples in this study were reported as a single ethnicity/race, we expected that there would be significant population substructure that might limit our ability to develop predictive models of ethnicity and to assess their performance. Despite this limitation, ethnicity could be predicted with high accuracy as assessed by cross validation. PlaNET's DNAme-based ethnicity classification relies on HM450K sites with large amounts of genetic signal, which supported our initial efforts to filter our data to enrich for genetic-informative sites prior to classifier development ("[Methods](#Sec11){ref-type="sec"}") \[[@CR41], [@CR50], [@CR51]\]. When examining PlaNET's 1860 sites used to predict ethnicity, more than half could be linked to a nearby genetic polymorphism. Of these, 802 CpG sites have documented SNPs in their probe body, single base extension or CpG site of interrogation, which previously have been identified to differ between European and East Asian populations \[[@CR41]\]. Several studies have suggested the genetic influence on DNAme at these sites is primarily technical in nature \[[@CR41], [@CR50], [@CR51]\], suggesting the patterns in DNAme at these sites are likely tissue-agnostic, warranting further investigation in their utility in predicting ethnicity and/or genetic ancestry in tissues other than the placenta. A significant proportion of other ethnicity-predictive CpG sites (*n* = 220) were previously found associated with placental mQTLs in a population with similar demographics to the ones studied here \[[@CR42]\]. This finding, together with EPISTRUCTURE---a method that also relies on mQTLs \[[@CR9]\]---suggests that leveraging the tissue- and population-specificity of mQTLs can produce highly effective DNAme-based population structure inference methods. Of the existing methods to assess population stratification from DNAme data, we note that Barfield's method and EPISTRUCTURE infer continuous measures of genetic ancestry, while Zhou's SNP-based classifier returns discrete ethnicity classifications; however, ours produce both \[[@CR8]--[@CR10]\] (Table [1](#Tab1){ref-type="table"}). EPISTRUCTURE and Barfield's method are unsupervised PCA-based approaches, which rely on the empirical observation that specific DNAme sites can be highly correlated with PCs computed on genome-wide genotype data in adult blood samples \[[@CR8], [@CR9]\]. However, we found that DNAme at these sites did not produce PCs that are highly associated with genotype data in placental samples. Instead, top PCs were more often associated with non-ancestry related variables in the placental samples included in this study, such as gestational age, preeclampsia, and technical variables. Ethnicity- and genetic ancestry-associated PCs were confined to the third to sixth component of variation, suggesting that application of these methods may require identifying which PCs are ethnicity/ancestry-specific, which is impossible when self-reported ethnicity and genetic ancestry information is unavailable (i.e. when these methods are needed most). Future improvements to these types of methods can aim at improving the amount of ethnicity and genetic ancestry-associated signal in the sites used to ensure the top two--three PCs are always associated with ethnicity and ancestry. This aim could also be supported in identifying ethnicity- and ancestry-associated sites that are also robust to changes in non-genetic drivers of DNAme such as cell type, gestational age, and severe pathology. Supervised population inference approaches such as ethnicity classifiers can return an explicit assignment of samples into distinct ancestral groups. In comparison to self-reported ethnicity, an assessment based on DNAme/genetic data is more objectively defined, which allows for more robust investigation of ethnicity-specific effects. An important goal of any population structure inference method would be to identify samples of mixed ancestry, a capability not well supported by Zhou's ethnicity classifier \[[@CR10]\]. In contrast, PlaNET produced membership probabilities corresponding to each ethnicity group that were highly correlated with genetic ancestry estimated from genotyping data. This was consistent whether we used principal components analysis on AIMs data, or model-based estimation of ancestry on high-density genotyping array data \[[@CR47], [@CR52]--[@CR54]\]. In this study, we defined samples of potential mixed ancestry as those with a maximum membership probability of less than 0.75, but we note that this threshold can be manually adjusted by the user and that the probabilities themselves can be used to adjust for population structure in study populations including significant numbers of samples with mixed ancestry. Results of DNAme studies on genetic ancestry and ethnicity, such as this one, depend on the number and proportion of different populations sampled from, as well as the tissue studied. Due to limitations in sample availability, only African, Asian, and Caucasian ethnicities were included in our study. However, we note that these ethnicities are among the most common in North American populations---but future developments should consider inclusion of additional ethnicities. Furthermore, due to limited number of samples with high-density genetic data, we were unable to address the extent of finer population structure that likely exists within the major ancestral groups studied. Differences in ethnic composition in samples from our study and samples used to develop Barfield's method and EPISTRUCTURE may also explain why Barfield's method or EPISTRUCTURE performed poorly in our study \[[@CR8], [@CR9]\]. A lack of generalizability of these methods to our placental samples was likely further compounded by the use of different tissues to develop each method---Barfield and EPISTRUCTURE were both developed and tested in blood tissue only. This is especially important to consider when applying these techniques to tissues with unique DNAme profiles, such as placenta \[[@CR18]\]. It is possible that application of these approaches to other tissues that are more similar to blood (e.g. other somatically-derived tissues) may result in better performance compared to when applied to placenta as seen in this study. However, any DNAme-based test needs to be validated before application to new tissues, which has not yet been done for these methods. A major goal of EWAS is to uncover signal truly associated with the phenotype/environment of interest that might generalize to other relevant populations. This is challenging given the wide host of technical variables that can affect DNAme measurements and the common finding that many phenotypes are associated with relatively small effect sizes \[[@CR33], [@CR55]\]. To this end, adjustment for major confounders such as genetic ancestry or ethnicity can significantly improve EWAS. We demonstrated, in a reanalysis of our previously published PE placentas, that adjustment for ethnicity, determined by PlaNET, improved the replicability of significant associations between independent cohorts. Conversely, overadjustment can occur when populations are relatively homogeneous, resulting in bias and/or loss of precision. We showed that PlaNET can indicate minimal population stratification when applied to a homogenous Caucasian population. Thus, PlaNET will be useful in assessing population stratification in future placental EWAS, as well as conducting ethnicity-stratified analyses, which may lead to important insights into the disparities between populations of pregnancy-related outcomes \[[@CR24]--[@CR26]\]. Conclusions {#Sec10} =========== We demonstrated that ethnicity and genetic ancestry can be accurately predicted using placental HM40K DNAme microarray data with respect to three major ethnicity/ancestral populations. Although samples that were used to develop PlaNET were reported to come from single ethnic populations, our classifier was able to capture mixed ancestry, and outperformed existing prediction methods. PlaNET will be valuable in assessing and accounting for population stratification, which can confound associations between DNAme with disease or environment, in future studies using HM450K or EPIC arrays. The machine-learning approach used to develop PlaNET can easily be applied for other tissues and populations for use in future DNAme studies. Methods {#Sec11} ======= Collection of previously published placental HM450K DNA methylation data {#Sec12} ------------------------------------------------------------------------ Placental DNAme data from liveborn deliveries of healthy and mixed pregnancy complications (*n* = 585), were combined from seven GEO HM450K datasets corresponding to five North American cohorts (summarized in Table [2](#Tab2){ref-type="table"}; sample-specific information in Additional file [1](#MOESM1){ref-type="media"}: Table S4) \[[@CR21], [@CR27], [@CR29]--[@CR32]\]. Five unpublished samples from the C5 cohort were included and are available at GSE128827. Gestational ages of these pregnancies at delivery ranged from 26 to 42 weeks and 50.30% of samples were male. Samples were excluded (*n* = 67) if their self-reported ethnicity was missing or did not fall into one of three major race/ethnicity groups: Asian/East Asian (*n* = 53), Caucasian/White (non-hispanic) (*n* = 389), or African/African American/Black (*n* = 57). Based on census data \[[@CR56]\], we note that self-reported Caucasian/White (non-hispanic) samples are typically of European ancestry, self-reported Asians are typically of East Asian ancestry and self-reported Africans represent diverse ancestries from Africa with a significant potential of admixture from other ancestries \[[@CR57]\]. When possible, data was downloaded as raw IDAT files (GSE75248, GSE100197, GSE100197, GSE108567, GSE74738), otherwise methylated and unmethylated intensities were utilized (GSE70453, GSE73375). DNA methylation data processing {#Sec13} ------------------------------- All samples were analyzed using the Illumina Infinium HumanMethylation450 BeadChip array (HM450K), the most popular measure of DNAme for EWAS. Array data analysis was performed using R version 3.5.0. To allow compatibility of PlaNET with the newest Infinium MethylationEPIC BeadChip array (EPIC), the raw HM450K data (485,512 CpGs, 65 SNPs) was filtered to the 452,453 CpGs and 59 SNPs common between both platforms prior to classifier development \[[@CR10]\]. Because genetic variability can capture ancestry information, we omitted the common filtering step that would remove sites with probes that overlap SNPs (*n* = 52,116 at a minor allele frequency \> 0.05). CpGs were removed if greater than 1% of samples had poor quality signal (bead count \< 3, or a detection *p*-value \> 0.01; *n* = 14,858). The remaining poor quality measurements were replaced with imputed values using K-nearest neighbours from the R package *impute* \[[@CR58]\]. Cross-hybridizing (*n* = 41,937) \[[@CR50], [@CR51]\] and placental-specific non-variable sites (*n* = 86,502) \[[@CR59]\] were also removed, leaving 319,233 sites for classifier development. Biological sex was determined by hierarchical clustering on DNAme measured from sites on the sex chromosomes and then compared to reported sex. Samples with discordant reported and inferred sex were removed (*n* = 3). Samples were also removed if they had a low mean inter-array correlation (\< 0.95, *n* = 5). Intra-array normalization methods, normal-exponential out-of-band (NOOB) \[[@CR60]\] and beta mixture quantile normalization (BMIQ) \[[@CR61]\] were used from R packages *minfi* (version 1.26.2) \[[@CR62]\] and *wateRmelon* (version 1.24.0) \[[@CR63]\] to normalize data. Genotyping data collection and genetic ancestry assessment {#Sec14} ---------------------------------------------------------- In a subset of C5 (*n* = 27) and 10 additional samples, high-density SNP array genotypes were collected. DNA samples from one site from the fetal side of each placenta were collected as previously described \[[@CR45]\] and quality was checked using a NanoDrop ND-1000 (Thermo Scientific) as well as by electrophoresis on a 1% agarose gel. Genotyping at \~ 2.3 million SNPs was done on the Illumina Infinium Omni2.5-8 (Omni2.5) array at the Centre for Applied Genomics, Hospital for Sick Kids, Toronto, Canada. For inferring genetic ancestry, the data for these 37 samples was combined with a previously processed 1000 Genomes Project Omni2.5 dataset (*n* = 1756) to use as reference populations \[[@CR44], [@CR48]\]. Genotypes in this combined dataset were filtered for quality (missing call rate \> 0.05, n removed = 31,604), minor allele frequency (MAF \> 0.05, *n* removed = 114,628), and linkage disequilibrium pruning was performed to select representative SNPs (*R*^2^ \< 0.25, n removed = 919,824) for a final dataset of 218,732 SNPs and *n* = 1793 samples. Genetic ancestry coefficients were estimated using the R package *LEA*, which utilizes sparse non-negative matrix factorization to produce similar results to model-based algorithms ADMIXTURE and STRUCTURE \[[@CR47], [@CR54]\]. Cross-entropy criterion was used to determine the number of ancestral populations (Additional file [2](#MOESM2){ref-type="media"}: Figure S8) \[[@CR64]\]. A smaller panel of 50 ancestry-informative genotyping markers (AIMs) was collected in a subset of samples from cohorts C4 (*n* = 41) and C5 (*n* = 68). AIMs were selected based on their ability to differentiate between African, European, East Asian, and South Asian populations \[[@CR65]--[@CR67]\]. Results from cohort C5 have been published elsewhere \[[@CR45]\], and genotyping data was collected for cohort C4 in the same manner. Briefly, these markers were measured in placental villus DNA using the Sequenom iPlex Gold platform (Génome Québec Innovation Centre, Montréal, Canada). Genetic ancestry inferred from 50 AIMs markers was computed using multi-dimensional scaling after combining with the same 50 AIMs from the 1000 Genomes Project samples, as previously described \[[@CR45]\]. Developing the ethnicity classifier and assessing its performance {#Sec15} ----------------------------------------------------------------- To develop and assess the performance of PlaNET we used a 'leave-one-dataset-out cross-validation' (LODOCV) approach. This approach uses four out of five datasets to develop a predictive model (training), which is then used to generate ethnicity classifications on the samples in the remaining dataset (testing). This differs from the traditional cross validation approach of randomly splitting the full dataset into training and testing. LODOCV produces more accurate estimates of classifier performance for future studies, and has been previously used for evaluating age-predictive models \[[@CR34]\]. Each iteration of LODOCV generates dataset-specific estimates of performance (accuracy, Kappa). After all iterations, overall performance was assessed by aggregating classifications across all datasets. For fitting predictive models within LODOCV-generated training sets, we used the R package *caret* \[[@CR68]\]. Several algorithms were compared: logistic regression with an elastic net penalty (GLMNET) \[[@CR37], [@CR38]\], nearest shrunken centroids (NSC) \[[@CR35], [@CR69]\], K-nearest neighbours (KNN) \[[@CR39]\], and support vector machines (SVM) \[[@CR40]\]. To determine optimum tuning parameters for each algorithm (e.g., '*k*' number of neighbours for KNN, alpha and lambda for GLMNET), we built several models while varying the tuning parameter(s) and compared the performance of these models within each training set using repeated (*n* = 3) fivefold cross validation. Hyperparameter values were left as default settings in *caret* \[[@CR68]\], or a grid of values for GLMNET (alpha = 0.025--0.500, lambda = 0.0025--0.2500). We compared the performance of these models using accuracy, positive predictive value, cohen's Kappa \[[@CR70]\], and logLoss (a measure of classification accuracy that heavily penalizes over-confident misclassifications). The results from this analysis can be found in Additional file [1](#MOESM1){ref-type="media"}: Tables S2, S3. After assessing the classifier performance using LODOCV, a final GLMNET model was fit to the entire dataset (cohorts C1--C5) using the same model fitting procedure described above and is available for use in future datasets (<https://github.com/wvictor14/planet>). Enrichment analysis {#Sec16} ------------------- The DNAme sites (see Additional file [1](#MOESM1){ref-type="media"}: Table S1) and SNPs selected to predict ethnicity in this final model (*n* = 1860) were used for enrichment analysis. For DNAme sites, we looked for enrichment for SNPs in the probe body, CpG site, and single base extension sites based on Illumina's HM450K annotation version 1.2 \[[@CR71]\]. We looked for enrichment for placental mQTLs \[[@CR42]\], chromosomes and CpG islands (HG19; Additional file [2](#MOESM2){ref-type="media"}: Figure S5). Fisher's exact test was used for all enrichment tests using a *p*-value threshold of \< 0.05, and was carried out in R using the function fisher.test(). GO and KEGG pathway analysis was done using the R package *missMethyl* version 3.8 \[[@CR72]\]. Threshold analysis {#Sec17} ------------------ We explored the use of a 'threshold function' to identify samples that are difficult to classify into discrete ethnicity groupings because of mixed ancestry. Because PlaNET's ethnicity classifications are associated with varying degrees of confidence (i.e., probabilities), we reasoned that a sample's most probable ethnicity classification (i.e., max(P(Asian), P(African), P(Caucasian)) would be lower with a higher degree of mixed ancestry. Therefore, we implemented a threshold function on PlaNET's probability outputs that classifies samples as 'Ambiguous' if the highest of the three class-specific probabilities is below a certain threshold. We explored several thresholds and decided on 0.75, which minimized the resulting decrease in predictive performance (Additional file [2](#MOESM2){ref-type="media"}: Figure S3). Comparison of methods for inferring genetic ancestry/ethnicity from HM450K data {#Sec18} ------------------------------------------------------------------------------- Because existing population inference methods and PlaNET use different statistical approaches to infer genetic ancestry/ethnicity (PCA-based vs. predictive modeling), we compared each method based on the amount of population-associated signal in DNAme from each method-specific subset of sites. This was done by applying principal component analysis (PCA) to standardized beta values for HM450k sites associated with each method (Table [1](#Tab1){ref-type="table"}) \[[@CR8]--[@CR10]\] within each cohort. To avoid bias, the PCs associated with PlaNET were calculated for each cohort using a classifier trained on all other cohorts (generated from LODOCV). Several simple linear regression models were applied to estimate the amount of variance explained in PC~*i*~ (*i* = 1, 2, 3,..., 10) by self-reported ethnicity and genetic ancestry when available. Self-reported ethnicity was encoded with indicator variables when testing group-specific associations (Additional file [2](#MOESM2){ref-type="media"}: Figure S6) and also overall association with ethnicity (Fig. [4](#Fig4){ref-type="fig"}a). Genetic ancestry was tested using coordinates one, two, and three, in a total of four different models: three models of each coordinate tested separately (Additional file [2](#MOESM2){ref-type="media"}: Figure S6), and then one model including both coordinates one and two to gain an overall estimate of the association with genetic ancestry (Fig. [4](#Fig4){ref-type="fig"}). To determine other factors that might affect signal in these sites, we also tested for the association between PC~*i*~ and each covariate available for each cohort. All simple regression tests were done in R using the function lm(). For Barfield's approach, we compared the various sets of sites that differ by the distance of a given CpG site to the nearest genetic variant (0, 1, 2, 5, 10, 50 bp) (Additional file [2](#MOESM2){ref-type="media"}: Figure S10). We used the set "0 bp from a genetic variant", following two observations: (1) the sets (0, 1, 2, 5 bp) were not significantly different in their association with ethnicity or genetic ancestry (*p* value \> 0.05), and (2) the sets (10, 50 bp) were significantly less associated with ethnicity and genetic ancestry for the first PC (*p* value \< 0.05). In summary, the closer the genetic variant was to the CpG, the stronger the signal associated with genetic ancestry and ethnicity. To compare PlaNET to Zhou et al.'s SNP-based classifier \[[@CR10]\], we used the package R package *sesame* (version 1.1.0) \[[@CR73]\] to obtain SNP-based ethnicity classifications for samples with idats available (cohorts C3, C4, and C5). To compare class-specific performance, McNemar's Chi-squared Test for Count Data was calculated using the *stats* R package. Application of PlaNET to previous EWAS {#Sec19} -------------------------------------- To demonstrate application of PlaNET, we downloaded placental HM450K DNAme datasets GSE98224, GSE100197, and GSE71678. We note that GSE100197 and GSE98224 overlap cohorts C4 and C5, respectively. To apply PlaNET to obtain ethnicity information, raw data were downloaded from GEO in the form of IDATs and loaded into R using *minfi* (version 1.26.2). Both NOOB and BMIQ normalization were applied before applying PlaNET. The R package *limma* (version 3.36.2) was used to test for differentially methylated sites. For GSE98224 and GSE100197, the processed DNAme data were used, and statistical thresholds were chosen the same as the published analysis \[[@CR21]\]. For enrichment analysis, differentially methylated CpGs were inputted into the *gometh* function from the R package *missMethyl* (version 1.16.0) using all filtered sites as background, and default settings. Additional files ================ {#Sec20} **Additional file 1: Table S1.** Ethnicity-predictive HM450K sites. A list of all PlaNET's 1860 automatically-selected ethnicity-predictive HM450K sites. **Table S2.** PlaNET's cohort-specific ethnicity classification performance. PlaNET's classification performance described across cohorts, assessed using LODOCV. **Table S3.** PlaNET's ethnicity classification performance described by class. PlaNET's classification performance described across ethnicity groups, assessed using LODOCV. **Table S4.** Sample metadata. A table of sample-specific information. **Table S5.** PE linear modeling results while adjusting for predicted ethnicity. Results containing 651 PE-associated sites from linear modeling while adjusting for predicted ethnicity. **Table S6.** PE linear modeling results in Caucasian cohort. Results containing 359 PE-associated sites from linear modeling in Caucasian-only samples. **Table S7.** GO enrichment analysis results. GO terms specifically enriched based on the 651 PE-associated sites obtained from linear modeling. **Additional file 2: Figure S1.** Dataset-specific effects. PC1 by PC2 scatterplot from PCA computed on scaled and centered DNAme beta values from 499 samples and 319,233 sites. **Figure S2.** Performance between machine learning algorithms in training. Resampling results for each machine learning algorithm. **a** performance (LogLoss) between machine learning algorithms in predicting ethnicity, and **b** class-specific accuracy. **Figure S3.** Threshold analysis for determining "ambiguous" samples. Various cutoffs for predicted membership probabilities were compared with respect to changes in predictive performance. **Figure S4.** Dataset-specific performance. PlaNET's classification performance was calculated for each dataset using a model trained to all other datasets. **Figure S5.** Enrichment analysis on ethnicity-predictive HM450K sites. PlaNET's CpG sites used to predict ethnicity was tested for enrichment with respect to **a** chromosomal location, and **b** relation to CpG islands. **Figure S6.** Association of population structure PCs with technical and biological variables. PCs were computed on ethnicity-predictive sites, EPISTRUCTURE, Barfield's method and the 59 SNP probes. Each PC was tested for their association with various cohort-specific technical and biological variables. For a given cohort (e.g. C1), ethnicity predictive sites from a classifier trained on all other cohorts (e.g. C2--C5) was used to avoid bias. **Figure S7.** PlaNET vs. Zhou et al. \[[@CR10]\] snp-based classifier. PlaNET's ethnicity classification performance was compared to Zhou et al. \[[@CR76]\] SNP-based ethnicity classifier in cohorts C3, C4, and C5. **Figure S8.** Estimating k number of ancestral populations using in genetic admixture inference program LEA. The cross-entropy criterion was used to determine the number of ancestral populations for estimating genetic ancestry coefficients. The number of ancestral populations was chosen at the point *k* = 3, when the cross-entropy criterion decreases significantly less with each integer-increase in *k*. **Figure S9.** Application of PlaNET to placental EWAS. Samples from three independent cohorts are plotted along three axes by their probability of belonging to each ethnicity class and colored by their final ethnicity classification determined by PlaNET. **Figure S10.** Evaluation of Barfield's alternative location-based filtering approaches. The signal associated with ethnicity and genetic ancestry was measured in relation to the distance of which a genetic variant lies to a CpG site (0, 1, 2, 5, 10, 50 bp). **a** Amount of variance explained in PC~*i*~ (*i* = 1, 2, 3, ..., 10) by either ethnicity or genetic ancestry. **b** Whether there was difference in the amount of ethnicity or genetic ancestry-associated variation in PC~*i*~, depending on distance to a genetic variant. Direction of association is indicated, where the reference group is the 0 bp set. PlaNET : placental DNAme elastic net ethnicity tool DNAme : DNA methylation CpG : cytosine--phosphate--guanine SNP : single-nucleotide polymorphism AIMs : ancestry informative genotyping markers mQTL : methylation quantitative trait loci PCA : principal component analysis PC : principal component HM450K : Infinium HumanMethylation450 BeadChip EPIC : Infinium MethylationEPIC BeadChip LODOCV : leave-one-dataset-out cross validation GLMNET : generalized logistic regression with an elastic net penalty SVM : support vector machines KNN : K-nearest neighbors NSC : nearest shrunken centroids PlaNET : placental elastic net ethnicity classifier USA : United States of America AFR : African ASI : Asian CAU : Caucasian BMIQ : beta-mixture interquantile normalization NOOB : normal exponential out-of-band normalization **Publisher\'s Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Advice and support from Louie Dinh, Dr. Lisa McEwan, Dr. Sam Wilson, Chaini Konwar, Amy Inkster, and Dr. Maria Penaherrera are greatly appreciated. Special thanks to Dr. Catherine Fry for correspondance regarding GEO data. VY, EMP, GDG, SM, and WR contributed to the design of the study. VY, EMP, and GDG performed all data analysis. EMP, GDG, SM, BC, AMB, KBM, CM, WR helped generate and contribute the data. EMP and WR conceived of the study. All authors approved and contributed to the writing of the final manuscript. All authors read and approved the final manuscript. This work was supported by NIH Grant to WR \[1R01HD089713-01\]. PlaNET can be downloaded as an R package from (<https://github.com/wvictor14/planet>). For reproducibility, all source code for development of software and analysis in this study is available at (<https://github.com/wvictor14/Ethnicity_Inference_450k>) and deposited on Zenodo (10.5281/zenodo.2641272 for source code, 10.5281/zenodo.2641266 for software) \[[@CR74], [@CR75]\]. The processed DNAme data supporting the conclusions of this article are available in the Gene Expression Omnibus repository (<https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE128827>). Raw DNAme data for previously published samples is available in the repositories listed in Table [2](#Tab2){ref-type="table"}. Raw DNAme data for samples published for the first time used in this study are available also at GSE128827. Genetic ancestry coordinates are available in supporting documents along with other sample-specific information (Additional file [1](#MOESM1){ref-type="media"}: Table S4). The genotyping array data used to generate genetic ancestry coordinates are not publicly available due to privacy concerns. The bulk of the data used in this study is from public repositories. Additional molecular investigations of the Vancouver cohort is covered under Ethics approval obtained from the University of British Columbia and BC Women's and Children's Hospital research ethics board in Vancouver, BC, Canada (certificate numbers H16-02280, H04-704488). Not applicable. The authors declare that they have no competing interests.