The Journal of Infectious Diseases MAJOR ARTICLE

The Journal of Infectious Diseases MAJOR ARTICLE The Female Genital Tract Microbiome Is Associated With Vaginal Antiretroviral Drug Concentrations in Human Immunodeficiency Virus Infected Women on Antiretroviral Therapy Renee Donahue Carlson, 1 Anandi N. Sheth, 1 Timothy D. Read, 1,2 Michael B. Frisch, 1 C. Christina Mehta, 3 Amy Martin, 4 Richard E. Haaland, 4 Anar S. Patel, 1 Chou-Pong Pau, 4 Colleen S. Kraft, 1,5 and Igho Ofotokun 1 1 Department of Medicine, Division of Infectious Diseases, School of Medicine, 2 Department of Human Genetics, School of Medicine, and 3 Department of Biostatistics and Bioinformatics, Rollins School of Public Health, Emory University; 4 Laboratory Branch, Division of HIV/AIDS Prevention, Centers for Disease Control and Prevention; and 5 Department of Pathology and Laboratory Medicine, School of Medicine, Emory University, Atlanta, Georgia (See the perspective by Bavoil et al, on pages 932 5.) Background. The female genital tract (FGT) microbiome may affect vaginal ph and other factors that influence drug movement into the vagina. We examined the relationship between the microbiome and antiretroviral concentrations in the FGT. Methods. Over one menstrual cycle, 20 human immunodeficiency virus (HIV) infected women virologically suppressed on tenofovir (TFV) disoproxil fumarate/emtricitabine and ritonavir-boosted atazanavir (ATV) underwent serial paired cervicovaginal and plasma sampling for antiretroviral concentrations using high-performance liquid chromatography tandem mass spectrometry. Analysis of 16S ribosomal RNA gene sequencing of cervicovaginal lavage clustered each participant visit into a unique microbiome community type (mct). Results. Participants were predominantly African American (95%), with a median age of 38 years. Cervicovaginal lavage sequencing (n = 109) resulted in a low-diversity mct dominated by Lactobacillus (n = 40), and intermediate-diversity (n = 28) and high-diversity (n = 41) mcts with abundance of anaerobic taxa. In multivariable models, geometric mean FGT:plasma ratios varied significantly by mct for all 3 drugs. For both ATV and TFV, FGT:plasma was significantly lower in participant visits with high- and low-diversity mct groups (all P <.02). For emtricitabine, FGT:plasma was significantly lower in participant visits with low- vs intermediate-diversity mct groups (P =.002). Conclusions. Certain FGT mcts are associated with decreased FGT antiretroviral concentrations. These findings are relevant for optimizing antiretrovirals used for biomedical HIV prevention in women. Keywords. female genital tract; HIV prevention; microbiome; pharmacology. Antiretroviral therapy (ART) has averted approximately 7.6 million deaths globally and is now a key component of modern human immunodeficiency virus (HIV) prevention efforts [1], decreasing mother-to-child and sexual HIV transmission and suppressing HIV type 1 viral shedding in blood and genital secretions [2 4]. Optimizing ART to achieve adequate female genital tract (FGT) concentrations is paramount for prevention strategies that utilize ART and important for HIV eradication efforts, which attempt to achieve complete suppression of viral replication at reservoir sites including the FGT [5, 6]. Received 30 May 2017; editorial decision 10 August 2017; accepted 17 August 2017; published online October 5, 2017. Presented in part: 21st International AIDS Conference, Durban, South Africa, 18 22 July 2016 [abstract WEPEB12]; and IDWeek, New Orleans, Louisiana, 26 30 October 2016 [abstract 56501]. Correspondence: I. Ofotokun, MD, MSc, Emory University School of Medicine, 49 Jesse Hill Junior Drive, Atlanta, GA 30303 (iofotok@emory.edu). The Journal of Infectious Diseases 2017;216:990 9 The Author 2017. Published by Oxford University Press for the Infectious Diseases Society of America. All rights reserved. For permissions, e-mail: journals.permissions@oup.com. DOI: 10.1093/infdis/jix420 The FGT microenvironment can be dynamic and is highly variable between women and within the same woman over time, due in part to microbial distribution and diversity. In healthy women, Lactobacillus species typically dominate the FGT microbiome, secreting lactic acid and maintaining a relatively acidic ph. Factors including race and ethnicity, sexual practices, influence of endogenous or exogenous hormones, antimicrobial use, and vaginal hygiene practices [7 10] can alter the FGT microbial flora. Nearly 30% of US women overall [11, 12] and >50% of non-hispanic African American women [12] have bacterial vaginosis (BV), a condition associated with increased bacterial diversity, an abundance of anaerobic species, and increased FGT ph. The extent to which systemically administered drugs distribute into various compartments including the FGT depends on several host-specific (membrane transporters, binding proteins, local ph), and drug-specific factors (protein binding capacity, membrane transporters affinity, lipid solubility, and the dissociation constant [pka]) [13]. Changes in the FGT microbiota toward communities characterized by increased microbial diversity and BV increase vaginal ph and have the potential to 990 JID 2017:216 (15 October) Donahue Carlson et al

alter other local factors that could influence movement of drugs into the FGT. In the current study, we hypothesized that the ability of antiretroviral drugs to concentrate within the FGT would vary according to the FGT microbiome community type (mct) and by antiretroviral drug. More specifically, we hypothesized that FGT concentrations would be increased in Lactobacillus-dominated FGTs. We characterized the FGT mcts in a cohort of virologically suppressed HIV-infected women on a stable ART regimen consisting of tenofovir disoproxil fumarate/emtricitabine (TDF/ FTC) and atazanavir/ritonavir (ATV/r). We additionally examined the relationship between the FGT mcts and the FGT-toplasma trough concentration ratios for these drugs. METHODS Study Design and Population A detailed description of the study population and design has been previously published [14]. In brief, this was a single-center, prospective cohort study of 20 virologically suppressed (viral load <75 copies/ml) HIV-infected women, aged 18 years, with regular menstrual cycles, without symptomatic BV, and taking combination ART 6 months including a regimen consisting of standard doses of TDF/FTC plus ATV/r for 30 days. Women were excluded if they were pregnant; had active sexually transmitted infections, symptomatic BV, or visible genital lesions; or were on other medications that could interact with their antiretroviral drugs. The study was approved by the Emory University Institutional Review Board, and written informed consent was obtained from each participant. Detailed demographic, sexual, medical, and treatment histories were obtained, and pelvic examinations were performed on all participants. Participants underwent 6 twice-weekly study visits over a single menstrual cycle for the collection of antiretroviral trough-timed plasma and genital samples (TearFlo wicks, Hub Pharmaceuticals, Rancho Cucamonga, California), and cervicovaginal lavage (CVL) separated within 2 hours of sample collection into supernatants and cellular pellets via centrifugation. All samples were stored at 80 C until analysis. Laboratory Techniques 16S Ribosomal RNA Gene Sequencing DNA was extracted from CVL pellets using the Qiagen EZ1 DNA Tissue kit (Qiagen, Germantown, Maryland) with the Qiagen bacterial card on the Qiagen EZ1 Advanced XL instrument according to the manufacturer s instructions [15] with modifications (Supplementary Methods). The 16S ribosomal RNA (rrna) gene microbial census sequencing library preparation was carried out using the Illumina MiSeq procedures as previously described [16] with modifications as described in the Supplementary Methods. Universal polymerase chain reaction primers for the V1 V2 hypervariable regions of the 16S rrna gene were used for sequencing: 8F:AGAGTTTGATCCTGGCTCAG, 338R:TGCTGCCTCCCGTAGGAGT. Sequence processing was performed using Mothur software [17]. After generating contigs from reads, sequences with 1 ambiguous bases and a length >385 bases were removed. Sequences starting at position 1046 and ending at position 6424 with maximum homopolymer length of 8 bases were screened and unique sequences underwent a preclustering step using UCHIME [18] for removal of chimeric sequences and classification with a Bayesian classifier and the GreenGenes database [19]. Operational taxonomic unit clustering was performed using 95% sequence homology, and taxonomic assignments (at the genus level) were made using the GreenGenes database [19]. Additional analyses of sequencing data utilized RStudio [20] and the Phyloseq package [21]. Each participant visit was clustered into an mct comprised of similar abundance and type of bacterial taxa using Dirichlet multinomial mixtures with the Dirichlet Multinomial R package [22]. Alpha-diversity was measured for each sample using the Shannon index [23] and results were averaged across mct. Principal coordinates analysis utilized UniFrac distances [24]. Antiretroviral Drug and Endogenous Sex-Hormone Assays High-performance liquid chromatography tandem mass spectrometry was used to measure trough concentrations of plasma and TearFlo genital wick tenofovir (TFV), FTC, and ATV concentrations as described previously [14, 25]. Plasma estradiol and progesterone concentrations were measured using a radioimmunoassay (Siemens Healthcare, Erlangen, Germany) as previously reported [14] and used to characterize each participant visit into follicular (visit occurring before the onset of progesterone rise) or luteal phase (visit occurring at or after the onset of progesterone rise). Participants without a progesterone rise across study visits were classified as having a nonovulatory cycle. Screening for FGT Semen, Blood and Leukocytes, BV, and Herpes Simplex Virus 2 Serology Testing for the presence of semen and semiquantitative measurement of red blood cells (RBCs) and leukocytes were performed on whole CVL using the ABACard p30 antigen detection kit (Abacus Diagnostics, West Hill, California) and Multistix Reagent strip (Siemens Medical Solutions, Malvern, Pennsylvania), respectively. Women were considered to have recent sexual activity (within 7 days of a study visit) if self-reported or if semen contamination was present at the study visit. Gram stain for Nugent score was performed on whole CVL fluid (n = 53) or CVL cell pellet (n = 55), depending on specimen availability. A subset of 15 participant visits had data available for both whole and pellet CVL; for these specimens a correlation analysis of Nugent score was performed, identifying a Pearson correlation coefficient of 0.84267 (P <.0001), suggesting that the source of CVL material (whole vs cell pellet) did not markedly alter the resulting Nugent score. Nugent score was characterized as positive for BV if 7 and negative if <7 [26]. Serology for herpes simplex virus 2 (HSV-2) infection was performed using HerpeSelect 2 (Focus Diagnostics, Cypress, California). Vaginal Microbiome and Genital ARV Levels JID 2017:216 (15 October) 991

Analytic Methods Descriptive statistics (median and first and third quartiles for continuous variables and counts and percentages for categorical variables) were computed at the participant level for baseline information and also at the visit level by mct. All bivariate and multivariate analyses utilized generalized mixed models that included a random intercept to account for repeated measures among participants. Clinical predictors evaluated included age, body mass index (BMI), antibiotic use for vaginal infection in the prior 30 days, recent sexual activity, FGT leukocytes, FGT blood contamination (RBC), plasma antiretroviral concentrations, plasma estradiol and progesterone concentrations, and follicular vs luteal phase for participants with an ovulatory cycle, and ovulatory vs nonovulatory cycles for all participants. FGT leukocytes and FGT RBCs were dichotomized into 2 groups based on median values. To evaluate bivariate associations between FGT antiretroviral concentrations and the independent variables of interest (clinical predictors listed above and mct), separate mixed linear models were carried out for each drug type. Antiretroviral concentrations (for both plasma and FGT) were natural log-transformed to normalize the distributions prior to bivariate and multivariate analyses. The geometric mean ratio (GMR) of FGT to plasma antiretroviral concentration was computed for each drug (termed FGT:plasma henceforth). The association between mct (primary predictor) and the natural log-transformed antiretroviral concentration was assessed using multivariable mixed linear models with indicators for location (FGT or plasma), antiretroviral drug (ATV, FTC, or TFV), and all possible combinations of location, antiretroviral drug, and mct. The model additionally included as covariates the clinical predictors that were found to be significantly associated (P <.1) at the bivariate level with any of the FGT antiretroviral concentrations. Estimates of FGT:plasma and natural log-transformed FGT concentration by mct and pairwise tests of mct differences were produced by the model. SAS version 9.4 software (SAS Institute, Cary, North Carolina) was used for statistical analyses. RESULTS Demographic and Clinical Characteristics Among 119 study visits by 20 women in the parent study, 117 samples had an available CVL pellet specimen to undergo 16S rrna gene sequencing, and 109 samples from 20 women yielded high-quality sequences used for microbiome analyses (7 samples were removed in quality processing steps and one removed postprocessing due to only extremely rare taxa identified). Participants had a median age of 38 years (min 24, max 48), 95% were African American, and approximately half were obese (median BMI, 30 kg/m 2 [min 21 kg/m 2, max 51 kg/m 2 ]). Most (90%) participants had a CD4 count 200 cells/μl and were taking their current ART regimen for a median of 14 (min 3, max 41) months, were not using hormonal contraception (95%), and were sexually active (85%). The majority of participants had prior exposure to HSV-2 by immunoglobulin G (95%), and, at the time of screening, 5 participants had asymptomatic BV by Nugent score and 5 had asymptomatic identification of Candida on Gram stain (Table 1). Most participants were virologically suppressed during the study period (13 of 118 [11.0%] visits in the original cohort had quantifiable viral loads >50 copies/ml [range, 50 395 copies/ml]) as previously reported [14]. Diversity of Microbiome Community Types Unsupervised clustering of similar microbiome communities by identity and distribution of bacterial taxa using Dirichlet Table 1. Baseline Demographic and Clinical Characteristics of Study Participants (N = 20) Variable No. (%) or Median (Range) 5 (25) Age, y 38 (24 48) Weight, kgs 83 (56 122) BMI, kg/m 2 30 (21 51) Race African American 19 (95) White 1 (5) Years since HIV diagnosis 9 (1 17) Nadir CD4 count, cells/μl 110 (2 320) Most recent CD4 count, cells/μl 383 (71 1189) <200 2 (10) 200 500 12 (60) >500 6 (30) ART history Months since first ART regimen 90 (9 115) Months on current ART regimen 14 (3 41) Current hormone contraceptive use a 1 (5) Treatment of vaginal infection within 30 d 6 (30) Antibacterial agent b 5 (20) Antifungal agent c 3 (15) Vaginal product use or douching reported 0 within 7 d of any study visits Sexually active in the past 6 mo 17 (85) 1 sexual partner 16 (94) 2 sexual partners 1 (6) Dysplasia on most recent Pap smear 5 (25) Genital infections at screening visit d Neisseria gonorrhoeae 0 Chlamydia trachomatis 0 Syphilis 2 (10) HSV-2 IgG positive 19 (95) Candida on Gram stain 5 (25) Bacterial vaginosis from vaginal Gram stain e Abbreviations: ART, antiretroviral therapy; BMI, body mass index; HIV, human immunodeficiency virus; HSV-2, herpes simplex virus type 2, IgG, immunoglobulin G. a Depot medroxyprogesterone. b Metronidazole alone or with additional oral antibacterial agents within 30 days of screening. c Fluconazole single oral dose or topical vaginal antifungal agent prescribed within 30 days of screening. d Women were excluded if bacterial vaginosis by Amsel criteria, Trichomonas or vaginal candidiasis by wet mount or potassium-hydroxide staining of wet mount, or if abnormal vaginal discharge or genital ulcers at screening visit. e Bacterial vaginosis diagnosed on vaginal Gram stain by Nugent score 7. 992 JID 2017:216 (15 October) Donahue Carlson et al

A 0.8%, 2.68%, and 0.44%, respectively (Figure 1B). The majority of participants had stable mcts during the course of the menstrual cycle in which they were enrolled in the study; 6 (30%) participants experienced an mct change a total of 9 times during the course of the study (Supplementary Figure 2). Overall, BV was observed in 56 of 98 (57%) of the study visits. The low-diversity mct was rarely associated with BV, in 9.1% (3/33) of the visits, whereas BV was identified among 67.9% (19/28) and 91.9% (34/37) of intermediate- and high-diversity mct visits, respectively (Table 2). Pairwise comparisons of mcts and clinical predictors in univariate analyses identified a positive association with BMI and the intermediate- compared to the high-diversity mct (odds ratio [OR] per unit increase in BMI for high- vs intermediate-diversity mct, 0.68 [95% confidence interval {CI}.5.94]; P =.02). BMI was not significantly Alpha Diveristy Measure Shannon Diversity Index, Mean (SD) by Community Group 0.14 (0.19) 1.27 (0.49) 2.07 (0.34) 2 Microbiome Community Type High Diversity Low Diversity Intermediate Diversity 1 0 Samples B Microbiome Community Type Low Diversity Intermediate Diversity Genus High Diversity Aerococcus Anaerococcus 1.00 Clostridium Dialister Fusobacterium 0.75 Gardnerella Abundance Lactobacillus Megasphaera Mobiluncus 0.50 Mycoplasma Peptoniphilus Porphyromonas 0.25 Prevotella Shuttleworthia Sneathia 0.00 Streptococcus Samples unclassified Ureaplasma Figure 1. A, Alpha-diversity by Shannon diversity index of microbiome community type (mct). Individual cervicovaginal lavage pellet samples from each study visit (horizontal axes) underwent 16S ribosomal RNA gene sequencing, and individual participant visits were classified into mcts using Dirichlet multinomial mixtures [22]. The mcts are defined as high, intermediate, or low diversity (red, blue, and green, respectively) based on mean (standard deviation [SD]) Shannon diversity index. B, Relative abundance of the top 20 most abundant bacterial taxa by genus for each participant visit is depicted in the colored boxes. Vaginal Microbiome and Genital ARV Levels JID 2017:216 (15 October) 993 multinomial mixtures yielded 3 distinct mcts that had average Shannon index scores characterized by low diversity (n = 40 study visits), intermediate diversity (n = 28), and high diversity (n = 41) of bacterial taxa (Figure 1A). Principal coordinates analysis using unifrac distance also demonstrated separation by mct (Supplementary Figure 1). Of classified bacterial taxa, the low-diversity mct was comprised of primarily Lactobacillus species (95.9% relative abundance), whereas the intermediate-diversity mct had a lower proportion of Lactobacillus (49.7%), followed by Prevotella (11.5%), unclassified genus (9.0%), and Megasphaera (8.6%). The high-diversity mct had Megasphaera (23.3%), Prevotella (23.0%), and Shuttleworthia (16.1%), comprising the top 3 identified taxa with a minority of Lactobacillus (1.0%). Among mcts, Gardnerella relative abundance in low-, intermediate-, and high-diversity mcts was

Table 2. Distribution of Clinical Factors and C 24 Antiretroviral Drug Concentrations by Female Genital Tract Microbiome Community Type Over the Menstrual Cycle in 20 Human Immunodeficiency Virus Infected Women a Microbiome Community Type Variable Total Cohort (N = 109 Visits) a associated with other mcts. Age, antibiotic use for vaginal infection within 30 days, recent sexual activity, presence of FGT semen, leukocytes or RBCs, menstrual cycle phase, ovulation status, and plasma estradiol and progesterone levels did not significantly vary by mct (P >.05, data not shown). Predictors of FGT Antiretroviral Concentrations Bivariate analyses of factors associated with FGT antiretroviral trough concentrations for all 3 drugs examined are summarized Low Diversity (n = 40 Visits) Intermediate Diversity (n = 28 Visits) High Diversity (n = 41 Visits) Nugent score, median (Q1, Q3) b 7 (4, 9) 3 (1, 5) 7 (5, 8) 9 (8, 10) Bacterial vaginosis, Nugent score 7, No. (%) c 56 (57) 3 (9.1) 19 (67.9) 34 (91.9) Age, y, median (Q1, Q3) 38 (33, 41) 37 (36, 43) 35 (39, 36) 40 (30, 41) BMI, kg/m 2, median (Q1, Q3)* 27 (25, 37) 26 (26, 37) 37 (35, 41) 27 (23, 32) Antibiotic for vaginal infection within 30 days of screening, 29 (26.6) 14 (35) 3 (10.7) 12 (29.3) No. (%) c Recent sexual activity, No. (%) 47 (43.1) 17 (42.5) 11 (39.3) 20 (48.8) FGT semen contamination, No. (%) 8 (7.3) 0 (0) 5 (17.9) 3 (7.3) FGT leukocytes >125 cells/μl, No. (%) 30 (27.5) 10 (25) 11 (39.3) 9 (22) FGT RBCs >25 cells/μl, No. (%) 62 (57.4) 18 (45) 15 (53.6) 13 (32.5) Menstrual cycle characteristics, No. (%) Nonovulatory phase 31 (28.4) 10 (25) 10 (35.7) 11 (26.8) Ovulatory phase 78 (71.6) 30 (75) 18 (64.3) 30 (73.2) Follicular, No. (%) of ovulatory 39 (50) 15 (50) 8 (44.4) 16 (53.3) Luteal, No. (%) of ovulatory 39 (50) 15 (50) 10 (55.6) 14 (46.7) Plasma hormone concentrations, pg/ml, median (Q1, Q3) Estradiol (n = 108) 27.01 (6.24, 58.5) 27.14 (3.58, 61.15) 35.96 (12.98, 63.15) 24.88 (0.00, 51.10) Progesterone (n = 109) 0.58 (0.31, 4.64) 0.59 (0.29, 4.04) 0.54 (0.3, 4.68) 0.70 (0.38, 5.03) FGT antiretroviral concentrations, ng/ml, median C 24 (Q1, Q3) d ATV 1435 (705, 2655) 1200 (705, 2510) 1818 (1048, 47778) 1435 (541, 2330) TFV 215 (117, 473) 157 (100, 251) 560 (141, 1272) 202 (90, 343) FTC 1252 (495, 1840) 1252 (431, 1780) 1133 (640, 2678) 1284 (455, 1670) Plasma antiretroviral concentration, ng/ml, median C 24 (Q1, Q3) ATV 620 (380, 957) 612 (400, 880) 612 (407, 915) 666 (346, 1390) TFV 74 (44, 104) 89 (63, 120) 55 (40, 88) 76 (35, 101) FTC 69 (46, 131) 98 (54, 152) 55 (40, 88) 64 (43, 135) FGT:plasma antiretroviral concentration, median (Q1, Q3) ATV 2.29 (1.17, 5.30) 2.18 (1.20, 4.65) 4.42 (1.76, 8.16) 1.52 (0.86, 3.67) TFV 3.22 (1.27, 8.62) 2.03 (0.71, 4.42) 10.80 (4.92, 20.59) 2.73 (1.69, 6.14) FTC 13.36 (7.34, 24.75) 11.36 (3.36, 17.84) 20.06 (11.73, 38.26) 13.10 (7.89, 23.26) Abbreviations: ATV, atazanavir; BMI, body mass index; C 24, antiretroviral drug concentration measured 24 hours after last dose; FGT, female genital tract; FTC, emtricitabine; Q1, quartile 1; Q3, quartile 3; RBC, red blood cell; SD, standard deviation; TFV, tenofovir. a Samples collected from 20 participants during 109 study visits unless N for the individual variable is otherwise specified. Study visits were completed in a single menstrual cycle for 16 participants or were completed outside the cycle window because of early menses prior to completion of study visits (4 participants, 5 [4.6%] study visits). Trough FGT antiretroviral concentrations were obtained, with the median time from last antiretroviral drug doses to FGT sampling 24 (Q1, Q3: 23, 25) hours, though 9 (8.3%) sets of samples were collected >4 hours from a true 24-hour trough. Plasma sampling occurred a median of 24 (Q1, Q3: 22, 24) hours from last antiretroviral drug doses and within an hour of FGT sampling for 82 (75.2%) of visits and not more than 1 hour from genital sampling in the remaining visits (27 visits [24.8%]). b Participant visits with Gram stain and Nugent scores available, n = 98 (low-diversity microbiome community type [mct], n = 33; intermediate-diversity mct, n = 28; high-diversity mct, n = 37). c Metronidazole alone or in combination with another oral antibiotic(s). Among the 5 participants with antibiotic use within 30 days of screening, the mct at the time of the first study visit was of low diversity (n = 3), intermediate diversity (n = 0), and high-diversity (n = 2). d FGT antiretroviral concentrations available for 107 visits (n = 2 missing from 2 study visits by 1 participant, both visits with high-diversity mct identified). *Denotes variable with a statistically significant (P <.05) association with mct in pairwise bivariate mixed models identified. Odds ratios per unit increase in BMI (kg/m 2 ) for high- vs intermediate-diversity mct, 0.68 (95% confidence interval [CI],.5.94]; P =.0191), for low- vs intermediate-diversity mct, 0.87 [.72 1.04]; P =.1145), and for low- vs high-diversity mct, 1.16 [.85 1.58]; P =.3370). We were unable to estimate associations between semen contamination and low-diversity mcts as no low-diversity visits with semen contamination were present. Other bivariate associations among those tested each had P >.05. in Table 3. Significant predictors include recent sexual activity for FGT TFV (P =.049) and presence of FGT RBCs >25 cells/ ml for FGT ATV (P =.036). FGT TFV concentrations were positively, though nonsignificantly, associated with BMI (4.5% increased TFV per unit [kg/m 2 ] increase in BMI; P =.090). Age, FGT leukocytes, menstrual cycle phase, ovulation status, and plasma estradiol and progesterone levels were not associated with any of the antiretroviral drug concentrations. Plasma concentrations of ATV were positively associated with FGT ATV 994 JID 2017:216 (15 October) Donahue Carlson et al

Table 3. Bivariate Associations a Between Clinical Predictors and Genital Antiretroviral Drug Concentrations Among 20 Human Immunodeficiency Virus Infected Women Over the Menstrual Cycle (N = 109 Visits) Percentage Change b in FGT Antiretroviral Drug Concentration (95% CI) Variable ATV ATV P Value concentrations (P <.0001); however, plasma TFV and FTC concentrations were not significantly associated with their respective FGT concentrations. FGT:plasma GMRs of ATV, TFV, and FTC varied significantly by mct in bivariate analyses (P <.0001 for all drugs; Table 2). Figure 2 summarizes the results of the multivariate models (controlling for the aforementioned predictors significant on bivariate analyses: BMI, FGT RBCs >25 cells/ml, and recent sexual activity) that examined the relationship between mct groups and FGT:plasma antiretroviral concentration (Figure 2A) and absolute FGT concentration (Figure 2B). In multivariable analysis, estimated ATV FGT:plasma GMRs were lowest for the high-diversity mct group (1.87 [95% CI, 1.32 2.63]), then the low-diversity mct group (2.08 [95% CI, 1.48 2.92]), and highest for the intermediate-diversity mct group (4.11 [95% CI, 2.74 6.17]). Significantly lower ATV FGT:plasma GMRs were observed in the high- and low-diversity mct groups than the intermediate-diversity mct (P =.004 and P =.012, respectively), and there was no statistical difference observed between the high- and low-diversity mcts (P =.657). For FTC, FGT:plasma GMRs were lowest for the low-diversity mct group (8.20 [95% TFV TFV P Value Age, y 0.66 ( 5.26 to 4.15).781 1.88 ( 8.44 to 5.15).587 2.11 ( 7.79 to 3.93).481 BMI, kg/m 2 0.58 ( 3.07 to 4.37).756 4.45 (.69 to 9.85).090 2.56 ( 2.03 to 7.37).276 Recent sexual activity vs none c 24.10 ( 16.61 to 84.66).283 47.66 (.14 117.74).049 21.99 ( 11.38 to 67.92).220 FGT leukocytes >125 cells/μl 29.85 ( 14.38 to 96.92).219 0.82 ( 32.79 to 51.24).968 0.01 ( 28.04 to 38.93).999 FGT RBCs >25 cells/μl 46.34 (2.67 108.58).036 2.23 ( 27.50 to 44.20).899 3.24 ( 26.77 to 27.84).815 Log plasma antiretroviral concentration, 0.44 (.25.63) <.0001 0.02 (.29 to.33).895 0.19 (-0.04 to 0.42).102 ng/ml d Menstrual cycle phase Follicular vs luteal (n = 39 1.74 ( 36.12 to 51.17).936 14.17 ( 40.01 to 22.80).397 6.23 ( 30.39 to 26.32).668 vs 39) Ovulation status Nonovulatory vs ovulatory 4.08 ( 49.97 to 83.89).899 41.01 ( 76.50 to 48.10).258 48.83 ( 76.26 to 10.28).086 cycle (n = 29 vs 78) Plasma hormone concentrations Estradiol 0.14 (.64 to.37).594 0.29 (.22 to.81).264 0.15 (.28 to.57).489 Progesterone 1.06 ( 6.16 to 4.33).691 1.26 ( 6.11 to 3.84).618 0.33 ( 4.32 to 3.82).871 Microbiome community type by level of diversity Low vs intermediate 41.50 ( 69.33 to 11.61).103 53.70 ( 79.26 to 3.36).060 21.57 ( 62.00 to 61.87).507 High vs intermediate 35.20 ( 62.60 to 12.28).120 48.56 ( 70.95 to 8.92).023 29.23 ( 56.18 to 14.31).155 Low vs high 9.72 ( 51.81 to 69.13).747 9.98 ( 60.07 to 102.95).798 10.81 ( 47.23 to 132.68).784 FTC FTC P Value Abbreviations: ATV, atazanavir; BMI, body mass index; CI, confidence interval; FGT, female genital tract; FTC, emtricitabine; RBC, red blood cell; Ref, reference group; TFV, tenofovir. a Bivariate associations between natural log-transformed FGT antiretroviral drug concentrations and each clinical predictor were performed using bivariate mixed linear models accounting for repeated measures using a random intercept for each participant. b Percentage change in FGT antiretroviral concentration is presented for a 1-unit increase in a continuous predictor or for the comparison to the reference outcome of dichotomous predictor variables with 95% confidence interval. c Includes participants with reported sexual activity within 7 days of a study visit (n = 46 visits) or participants with positive semen contamination of vaginal secretions but with no reported sexual activity in the past 7 days (n = 1 visits). d Plasma antiretroviral drug concentrations were log-transformed prior to analyses. The association between the concentration of the same FGT and plasma drug was assessed in each bivariate model. Due to log transformation, FGT concentrations are presented for a 1% change in plasma concentration. CI, 5.84 11.52]), then the high-diversity mct group (13.12 [95% CI, 9.29 18.52]), and highest for the intermediate-diversity mct group (18.77 [95% CI, 12.50 28.18]). A significantly lower FTC GMR was observed in the low- vs intermediate-diversity mcts (P =.002), but the differences between high- vs intermediate-diversity and high- vs low-diversity mct groups were not statistically significant (P =.187 and P =.057, respectively). Last, FGT:plasma GMRs for TFV were lowest for the high-diversity mct group (3.08 [95% CI, 2.18 4.35]), then the low-diversity mct group (1.82 [95% CI, 1.29 2.55]), and highest for the intermediate-diversity mct group (9.49 [95% CI, 6.32 14.24]). The FGT:plasma TFV was significantly higher for high-diversity vs low-diversity mct groups (P =.033) and significantly lower for high- vs intermediate-diversity and low-diversity vs intermediate-diversity mct groups (both P <.0001). Multivariate models evaluating the association between absolute FGT antiretroviral concentrations and mct groups (including the same predictor variables) revealed similar findings (Figure 2A). For FGT ATV concentration, no significant differences were observed by mct group, although there was a trend toward lower ATV concentrations in high-diversity Vaginal Microbiome and Genital ARV Levels JID 2017:216 (15 October) 995

A B FGT:plasma geometric mean ratio (95% CI) Ratio Estimates (95% Cl) 25 20 15 10 5 0 compared with intermediate-diversity mct groups (1111 vs 1944 ng/ml; P =.051). For FTC, similar patterns were noted, but no significant pairwise differences were observed by mct group. Finally, as with FGT:plasma GMRs, significantly lower TFV FGT concentrations were observed for high- vs intermediate-diversity (184 vs 493 ng/ml; P <.001) and low- vs intermediate-diversity mct groups (183 vs 493 ng/ml; P <.0001). DISCUSSION ATV TFV Antiretroviral Drug High Int Low High Int Low High Int Low 1.87 4.11 2.08 3.08 9.49 1.81 13.1 18.81 8.20 (1.33, (2.74, (1.48, (2.19, (6.32, (1.29, (9.29, (12.5, (5.84, 2.64) 6.17) 2.92) 4.35) 14.2) 2.56) 18.5) 28.2) 11.5) We observed diversity in the distribution of FGT bacterial taxa and microbiome communities in this cohort of predominantly African American HIV-infected women who were virologically suppressed on ART. The FGT bacterial taxa clustered into 3 mcts: low diversity, which was Lactobacillus genus dominant; intermediate diversity with lower Lactobacillus proportion; and high diversity with a paucity of Lactobacillus. Most women maintained a stable mct over the menstrual cycle. Certain mcts were associated with lower FGT antiretroviral drug exposure (as measured by FGT:plasma ratio). Specifically, after adjusting for known and potential confounders, FGT:plasma ratios for ATV and TFV were lower in the low- and high-diversity mcts compared with the intermediate-diversity mct and were less than half of those observed for the intermediate-diversity mct group for both ATV and TFV. FGT:plasma ratios for FTC showed similar results for low- vs intermediate-diversity mct but not for the high- vs intermediate-diversity mct. While systemic exposure and effects of certain orally administered drugs have been shown to be modulated by the composition of the gut microbiota [27], our data, to our knowledge, Geometric mean FGT concentration (ng/ml) 2500 2000 1500 1000 500 0 FTC ATV TFV Antiretroviral Drug FTC High Int Low High Int Low High Int Low 1111 1944 1381 184 493 183 894 1163 901 (730, 1690) (1299, 2911) (951, 2006) (121, 281) (329, 738) (126, 266) (588, 1360) (777, 1741) (621, 1309) Microbiome Group by Diversity High Intermediate Low Figure 2. Multivariable model estimates of geometric mean ratio of female genital tract (FGT) to plasma antiretroviral drug concentrations (A) and geometric mean antiretroviral drug concentration (B) by microbiome community type for atazanavir, emtricitabine, and tenofovir using multivariable mixed linear models. Covariates included in each model were body mass index, FGT red blood cells >25 cells/μl, and recent sexual activity. Abbreviations: ATV, atazanavir; CI, confidence interval; FGT, female genital tract; FTC, emtricitabine; Int., intermediate; TFV, tenofovir. are among the first to demonstrate the association between mct and FGT compartmental concentrations of systemically administered drugs. In Centre for the AIDS Program of Research in South Africa - 004 (CAPRISA-004), which examined the efficacy of vaginal 1% TFV gel for preexposure prophylaxis (PrEP) in South African women, FGT TFV concentrations varied markedly among study participants, with lower PrEP efficacy among women with low FGT TFV drug concentrations [28]. Using a metaproteomics approach to vaginal flora characterization, the group demonstrated that the presence of Lactobacillus-dominant vaginal flora was associated with higher PrEP efficacy [29]. In an in vitro study, Gardnerella (a common pathogen identified in BV and in some non-lactobacillus-dominant vaginal flora) was capable of metabolizing TFV [29]. The role of Gardnerella in our study is uncertain and may be negligible given its very low abundance in all mct groups, even in the high-diversity group, which had the highest frequency of asymptomatic BV. However, methodological differences in microbiome quantitation make comparisons across studies challenging. Systematic underestimation of Gardnerella using 16S rrna sequencing of the V1 V3 hypervariable regions has been reported, though complex interactions with other species leading to variability in rates of this bias may occur and may be a limitation of 16S rrna studies [30]. Furthermore, it is well established that systemic drug distribution into body compartments depends upon both host-specific (ie, membrane transporters, drug-binding proteins, local ph), and drug-specific factors (ie, protein-binding capacity, membrane transporters affinity, lipid solubility, and the dissociation 996 JID 2017:216 (15 October) Donahue Carlson et al

constant [pka]) [13]. Therefore, changes in the FGT mct toward communities characterized by increased microbial diversity and BV have the potential to alter local ph and other factors capable of influencing movement of drugs across the FGT compartment. The FGT is naturally acidic with a ph of 3.5 4.5; strongly basic, lipid-soluble drugs achieve higher concentrations in the healthy acidic vagina, whereas the converse is true for acidic drugs, a phenomenon known as ion trapping. Drugs such as ATV and TDF (pka of 4.7 and 3.8, respectively [31, 32]) may therefore trap readily in the Lactobacillus-dominated vagina but be affected by a more basic vaginal ph, whereas drugs such as FTC (pka 2.7, which is far below the ph of the healthy FGT [32]) may not be as affected by a vaginal ph change, consistent with our study s findings. TFV, though not TDF, has been recently found in vitro to have reduced T-cell uptake with increased extracellular ph, lending additional support to this hypothesis [33]. However, differences in antiretroviral drug concentrations across the mcts in our study cannot likely be explained on the basis of FGT ph changes and compartmental drug trapping alone. The low- and high-diversity mcts in our study exhibited markedly different FGT bacterial taxa and, as such, high-diversity mcts could be expected to produce higher local ph values and therefore different patterns of drug trapping. However, FGT antiretroviral concentrations for these 2 microbiota groups were similar, suggesting that additional factors may be at play. One plausible explanation is the presence of microbial strains in certain mcts but not others that are capable of uptake or degradation of antiretroviral drugs (similar to the action of Gardnerella in the CAPRISA -004 study) or altering movement of drugs across the FGT by affecting local drug transporters in a ph-dependent or -independent manner. In fact, preliminary in vitro data suggest that certain FGT microbiota may differentially accumulate or bind TFV, reducing extracellular concentrations. For example, Lactobacillus crispatus significantly increased TFV uptake and reduced extracellular TFV by >75%, whereas other Lactobacillus species (L. jensenii and L. iners) did not [33]. Lactobacillus species level data were not available in this study, but it is possible the low-diversity mct was dominated by species associated with TFV uptake (such as Lactobacillus crispatus), accounting for reduced FGT drug levels. Although the exact mechanism(s) are not delineated, the finding that certain mcts are associated with altered FGT antiretroviral concentrations in vivo may have implications for HIV prevention strategies that utilize antiretroviral drugs including PrEP, prevention of mother-to-child transmission, postexposure prophylaxis, and potentially also treatment for prevention in the setting of suboptimal adherence and/or ongoing plasma viremia. More broadly, suboptimal drug exposure associated with certain FGT flora could underlie poor therapeutic responses to treatment of other genital infections in some women. It should be noted that our study is a proof-of-principle study with a small sample size and was a secondary analysis using stored specimens; variables such as measured FGT ph were not collected in the parent study, thus limiting our ability to directly explore the role of compartmental ph in the observed antiretroviral concentration differences. Additionally, sample size may have been too small to detect differences between mct or FGT concentrations and some variables. Analyses from prospective studies are currently under way to address these limitations. The majority of our study participants were African American women receiving regular HIV medical care in urban Atlanta, and these findings may not be generalizable to other populations. Finally, our participants were treated with the same antiretroviral drug regimen, with consistent reported adherence. Although ART has been speculated to alter the composition of the FGT microbiome [34], the mct diversity observed in out our population is unlikely to be solely ART related as all participants received the same regimen. We were not able to estimate the potential effect of mct on genital antiretroviral concentrations in the setting of inconsistent adherence. These limitations notwithstanding, we identified 3 major mcts in the FGT of HIV-infected women virologically suppressed on ART, characterized by differences in the diversity and distribution of bacterial taxa including varied Lactobacillus abundance. Our data further suggest that certain FGT mcts are associated with decreased FGT antiretroviral concentrations. These findings, if validated in larger studies and diverse populations, may influence antiretroviral medication choice for optimal biomedical HIV prevention in women. Supplementary Data Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author. Notes Acknowledgments. We thank the study participants; the Emory University Center for AIDS Research (CFAR) Clinical Virology Laboratory for gene sequencing; and the Emory CFAR Biomarkers Core Laboratory at the Yerkes National Primate Research Center (2P51RR000165-51) for estradiol and progesterone assays. We also thank Wendy Armstrong, Angela Caliendo, Jeffrey Lennox, and Elizabeth Corwin for their input. Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health (NIH) or the Centers for Disease Control and Prevention. Financial support. This work was supported by the Emory University CFAR, supported by the NIH (grant number P30AI050409) and the Atlanta Clinical and Translational Science Vaginal Microbiome and Genital ARV Levels JID 2017:216 (15 October) 997

Institute, supported by the National Center for Advancing Translational Sciences of the NIH (grant number UL1TR000454). An educational support scholarship supported R. D. C. (Bristol- Myers Squibb Virology Fellows 2014 2015). A. N. S. was supported by the NIH (grant number 1K23AI114407). T. D. R. was partially supported by the NIH (grant number AI121860), and T. D. R. and M. B. F. were supported by the Emerging Infections Program Patient Protection and Affordable Care Act (PPACA): Enhancing Epidemiology and Laboratory Capacity funding from the Emory Public Health Bioinformatics Fellowship (grant number 3U50CK000314-04S1). Potential conflicts of interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed. References 1. Joint United Nations Programme on HIV/AIDS. The Gap Report. http://www.unaids.org/sites/default/files/media_asset/ UNAIDS_Gap_report_en.pdf. Accessed 1 December 2015. 2. Cu Uvin S, Caliendo AM, Reinert SE, Mayer KH, Flanigan TP, Carpenter CC. HIV-1 in the female genital tract and the effect of antiretroviral therapy. AIDS 1998; 12:826 7. 3. Cohen MS, Chen YQ, McCauley M, et al; HPTN 052 Study Team. Prevention of HIV-1 infection with early antiretroviral therapy. N Engl J Med 2011; 365:493 505. 4. Siegfried N, van der Merwe L, Brocklehurst P, Sint TT. Antiretrovirals for reducing the risk of mother-to-child transmission of HIV infection. Cochrane Database Syst Rev 2011: CD003510. 5. Poles MA, Boscardin WJ, Elliott J, et al. Lack of decay of HIV-1 in gut-associated lymphoid tissue reservoirs in maximally suppressed individuals. J Acquir Immune Defic Syndr 2006; 43:65 8. 6. Chun TW, Carruth L, Finzi D, et al. Quantification of latent tissue reservoirs and total body viral load in HIV-1 infection. Nature 1997; 387:183 8. 7. Brotman RM. Vaginal microbiome and sexually transmitted infections: an epidemiologic perspective. J Clin Invest 2011; 121:4610 7. 8. McClelland RS, Richardson BA, Graham SM, et al. A prospective study of risk factors for bacterial vaginosis in HIV-1-seronegative African women. Sex Transm Dis 2008; 35:617 23. 9. Fethers KA, Fairley CK, Hocking JS, Gurrin LC, Bradshaw CS. Sexual risk factors and bacterial vaginosis: a systematic review and meta-analysis. Clin Infect Dis 2008; 47:1426 35. 10. Gajer P, Brotman RM, Bai G, et al. Temporal dynamics of the human vaginal microbiota. Sci Transl Med 2012; 4:132ra52. 11. Ravel J, Gajer P, Abdo Z, et al. Vaginal microbiome of reproductive-age women. Proc Natl Acad Sci U S A 2011; 108(suppl 1):4680 7. 12. Allsworth JE, Peipert JF. Prevalence of bacterial vaginosis: 2001 2004 National Health and Nutrition Examination Survey data. Obstet Gynecol 2007; 109:114 20. 13. Thompson CG, Cohen MS, Kashuba AD. Antiretroviral pharmacology in mucosal tissues. J Acquir Immune Defic Syndr 2013; 63(suppl 2):S240 7. 14. Sheth AN, Evans-Strickfaden T, Haaland R, et al. HIV-1 genital shedding is suppressed in the setting of high genital antiretroviral drug concentrations throughout the menstrual cycle. J Infect Dis 2014; 210:736 44. 15. Qiagen. EZ1 DNA Tissue Handbook. https://www.qiagen. com/it/resources/resourcedetail?id=4c45d35b-10d8-4d5ba3a7-88e29ab1d88f&lang=en. Accessed 7 January 2017. 16. Illumina Inc. Preparing libraries for sequencing on the MiSeq. https://support.illumina.com/content/dam/illuminasupport/documents/documentation/system_documentation/miseq/preparing-libraries-for-sequencing-on-miseq-15039740-d.pdf. Accessed 25 November 2015. 17. Schloss PD, Westcott SL, Ryabin T, et al. Introducing mothur: open-source, platform-independent, community-supported software for describing and comparing microbial communities. Appl Environ Microbiol 2009; 75:7537 41. 18. Edgar RC, Haas BJ, Clemente JC, Quince C, Knight R. UCHIME improves sensitivity and speed of chimera detection. Bioinformatics 2011; 27:2194 200. 19. DeSantis TZ, Hugenholtz P, Larsen N, et al. Greengenes, a chimera-checked 16S rrna gene database and workbench compatible with ARB. Appl Environ Microbiol 2006; 72:5069 72. 20. RStudio Team. RStudio: integrated development for R. Boston, MA: RStudio, Inc, 2015. 21. McMurdie PJ, Holmes S. phyloseq: an R package for reproducible interactive analysis and graphics of microbiome census data. PLoS One 2013; 8:e61217. 22. Holmes I, Harris K, Quince C. Dirichlet multinomial mixtures: generative models for microbial metagenomics. PLoS One 2012; 7:e30126. 23. Shannon CE. A mathematical theory of communication. Bell Syst Tech J 1948; 27:379 423, 623 56. 24. Lozupone C, Knight R. UniFrac: a new phylogenetic method for comparing microbial communities. Appl Environ Microbiol 2005; 71:8228 35. 25. Kuklenyik Z, Martin A, Pau CP, et al. Effect of mobile phase ph and organic content on LC-MS analysis of nucleoside and nucleotide HIV reverse transcriptase inhibitors. J Chromatogr Sci 2009; 47:365 72. 26. Nugent RP, Krohn MA, Hillier SL. Reliability of diagnosing bacterial vaginosis is improved by a standardized method of Gram stain interpretation. J Clin Microbiol 1991; 29:297 301. 998 JID 2017:216 (15 October) Donahue Carlson et al

27. Saad R, Rizkallah MR, Aziz RK. Gut pharmacomicrobiomics: the tip of an iceberg of complex interactions between drugs and gut-associated microbes. Gut Pathog 2012; 4:16. 28. Karim SS, Kashuba AD, Werner L, Karim QA. Drug concentrations after topical and oral antiretroviral pre-exposure prophylaxis: implications for HIV prevention in women. Lancet 2011; 378:279 81. 29. Klatt NR, Cheu R, Birse K, et al. Vaginal bacteria modify HIV tenofovir microbicide efficacy in African women. Science 2017; 356:938 45. 30. Brooks JP, Edwards DJ, Harwich MD Jr, et al; Vaginal Microbiome Consortium. The truth about metagenomics: quantifying and counteracting bias in 16S rrna studies. BMC Microbiol 2015; 15:66. 31. Bristol Myers Squibb Australia Pty Ltd. Evotaz [package insert]. http://www.medicines.org.au/files/bqpevota.pdf. Accessed 7 December 2015. 32. Gilead Sciences. Truvada [package insert]. http://www. gilead.com/~/media/files/pdfs/medicines/hiv/truvada/ truvada_pi.pdf. Accessed 6 February 2016. 33. Taneva E, Cameron S, Cheshenko N, Srinivasan S, Fredricks DN, Herold BC. Modulation of tenofovir (TFV) pharmacokinetics (PK) and antiviral activity by vaginal microbiota: implications for topical preexposure prophylaxis. In: HIV Research for Prevention Conference, Chicago, IL, 2016. 34. Pyles RB, Vincent KL, Baum MM, et al. Cultivated vaginal microbiomes alter HIV-1 infection and antiretroviral efficacy in colonized epithelial multilayer cultures. PLoS One 2014; 9:e93419. Vaginal Microbiome and Genital ARV Levels JID 2017:216 (15 October) 999