Introduction Worldwide, an estimated 33.3 million people are infected with HIV-1, 60% in sub-Saharan Africa, where women account for the majority of those infected . Antiretroviral therapy (ART), through reducing HIV-1 plasma  and genital HIV-1 RNA concentrations , has been associated with >90% reduction in HIV-1 transmission in observational studies  and a recent trial of earlier ART initiation . However, only about half of HIV-1–infected adults qualify for ART initiation per current country guidelines, and only 37% of those qualifying for ART in Africa received treatment . Thus, new HIV-1 prevention strategies that will reduce HIV-1 risk for those not on ART remain an urgent need. Bacterial vaginosis (BV) is a common disorder characterized by changes in vaginal flora in which normally predominant Lactobacillus species are replaced by potential pathogens including Gardnerella vaginalis, genital Mycoplasma, and fastidious anaerobic bacteria ,. For unknown reasons, BV is considerably more common among women in sub-Saharan Africa and other resource-poor countries than in developed countries, affecting up to 55% of women in some studies –. BV has been associated with a 60% increased risk of HIV-1 acquisition in women , and, among women with HIV-1, with higher HIV-1 concentrations in cervicovaginal fluids –. Bacteria associated with BV can induce viral replication and shedding in the genital tract ,, which may lead to increased HIV-1 infectiousness for women with BV ,. However, to date, no study has examined whether BV increases the risk of female-to-male HIV-1 transmission. We hypothesized that HIV-1–infected women with BV have an increased risk of female-to-male HIV-1 transmission than women with normal vaginal flora. To answer this question, we prospectively studied a cohort of African heterosexual couples in which the female was HIV-1 seropositive and the male was HIV-1–seronegative who were enrolled in a randomized placebo-controlled trial of dually HIV-1 and herpes simplex virus (HSV) type 2–seropositive heterosexual African adults, and their HIV-1–seronegative partners. Methods Ethics Statement The University of Washington Human Subjects Review Committee, University of California San Francisco Committee on Human Research, the Kenya Medical Research Institute (KEMRI) National Ethics Review Committee, and ethics review boards at each study site reviewed and approved the study protocol and consent documents. Population and Procedures We used data from a cohort of southern and East African HIV-1 serodiscordant heterosexual couples enrolled in a clinical trial (the Partners in Prevention HSV/HIV Transmission Study) evaluating HSV-2–suppressive therapy with acyclovir 400 mg bid provided to the HIV-1–infected partner to prevent HIV-1 transmission to their HIV-1–seronegative partners. As previously reported, acyclovir decreased plasma HIV-1 levels in the HIV-1–infected partners, but did not reduce HIV-1 transmission risk . The present report is a secondary analysis of data from the subset of 2,236 couples from this prospective cohort in which the HIV-1–infected partner was female . HIV-1–infected partners were required to be seropositive for HSV-2, with a CD4 count ≥250 cells/mm3, and without history of AIDS-defining conditions; couples were followed for up to 24 mo. HIV-1–infected women were seen monthly and underwent a pelvic examination at enrollment and every 3 mo to collect a vaginal swab for Gram stain for evaluation of BV. Enrollment vaginal swabs were collected on all participants. Quarterly vaginal swab collection was performed as part of a protocol modification implemented at each site once approved by the site institutional review board; vaginal Gram stain results were not obtained prior to site-specific approval of the protocol modification. Plasma for HIV-1 RNA quantification was collected at baseline, 3-, 6-, and 12-mo visits, and at study exit; CD4 counts were performed at baseline and every 6 mo. HIV-1–infected partners who met national guidelines for initiation of ART during follow-up were referred to local HIV-1 care clinics, and those who became pregnant were referred to antenatal clinics for prevention of mother-to-child transmission services. HIV-1–infected women underwent a speculum pelvic examination at a visit 6 mo after enrollment, during which an endocervical Dacron swab for HIV-1 RNA quantification was obtained; swabs were not collected at a defined time in the menstrual cycle, although women usually deferred sampling during menstruation. HIV-1–uninfected men were seen quarterly for HIV-1 serologic testing. Participants received comprehensive HIV-1 prevention including HIV-1 risk-reduction counseling (both individual and as a couple), quarterly sexually transmitted infection (STI) symptom assessment with syndromic treatment of STIs, and provision of free condoms. All participants provided written informed consent. Laboratory Methods for Diagnosis of BV Vaginal swabs collected at enrollment and quarterly follow-up visits were rolled onto glass slides, air dried, and methanol fixed at the study site and subsequently Gram stained at the Center for Microbiology Research Laboratory at KEMRI. Vaginal flora was evaluated using Nugent's criteria : normal vaginal flora, intermediate flora, and BV categories were defined by Nugent's scores of 0–3, 4–6, and 7–10, respectively. Each slide was double-read by two technologists. A digital image of approximately every tenth slide was sent electronically to one of the investigators (CAS) for external quality control (EQC). Our target for concordant results between the laboratory and EQC was ≥90%. A discordant result was defined as a difference in the Nugent's score ≥1, which also caused a change in flora category (e.g., a score difference of 3 to 4, which changes the diagnosis from normal vaginal flora to intermediate flora). External quality control performed on 1,722 (8.7%) of 19,882 slides (the total number of slides included HIV-1–positive women included in this study, and HIV-1–negative women evaluated for a separate analysis) demonstrated an overall concordance of 92.2% (Κ = 0.84); the concordance surpassed our predefined value of ≥90% based on expected inter-observer agreement in other studies ,. Other Laboratory Procedures HIV-1 serologic testing was by dual rapid HIV-1 antibody tests performed locally, with positive results confirmed by HIV-1 Western blot at the University of Washington . For couples in which the initially HIV-1–uninfected male partner seroconverted to HIV-1 seropositive, analysis of HIV-1 env and gag gene sequences from both members of the couple were used to evaluate transmission linkage within the partnership . Serologic testing for HSV-2 and nucleic acid amplification testing for STIs (specifically Chlamydia trachomatis, Neisseria gonorrhoeae, and Trichomonas vaginalis) was done at study enrollment . CD4 quantification was performed using standard flow cytometry. All laboratory procedures followed Good Laboratory Practices, and laboratories were enrolled in External Quality Assessment programs. HIV-1 RNA was quantified from plasma at baseline, at months 3, 6, and 12, and at study exit; and from the 6-mo endocervical swab specimen (collected at the same visit as the 6-mo plasma specimen) with the COBAS AmpliPrep/COBAS TaqMan real-time HIV-1 RNA assay version 1.0 (Roche Diagnostics). Endocervical swabs were eluted in 1,000 µl of GUSCN lysis buffer, eluted for 15 min, vortexed briefly, and microfuged for 5 s at 14,000g to pellet debris before removal of fluid for testing. A final dilution step with 10× PBS was used to achieve sufficient volume for the COBAS AP/TM assay, with a lower limit of quantification of 240 copies (per milliliter for blood plasma and per swab for endocervical samples). Plasma and genital HIV-1 RNA concentrations were log10-transformed to approximate normality. Samples below the limit of quantification were assigned values at half that limit. Statistical Analysis The primary outcome was female-to-male HIV-1 transmission, defined as those HIV-1 seroconversion events that were genetically linked within the partnership. Male partners who acquired HIV-1 from an outside partner contributed follow-up time up to HIV-1 seroconversion and were censored thereafter. Follow-up for men was also censored after their HIV-infected partner initiated ART. The primary exposure was vaginal flora status, as measured at the quarterly study visit prior to each HIV-1 test, in order to represent vaginal flora status during the time of potential HIV-1 exposure to the male partner. If the result at the visit 3 mo prior to HIV-1 testing was expected but missing, the result 6 mo prior was used; if the results at both the 3 and 6 mo prior to HIV-1 testing were expected but missing the period was excluded from analysis. We analyzed vaginal flora in three categories: BV (Nugent score ≥7) and intermediate flora (Nugent score 4–6), each compared with normal flora (Nugent score ≤3). We performed two sensitivity analyses to assess the robustness of our vaginal flora exposure: first, we analyzed vaginal flora at the visit concurrent with HIV-1 serologic testing, and second, we analyzed vaginal flora based on the most severe exposure (highest Nugent category) occurring at either the prior or current visit. Association between vaginal flora and time-varying covariates was assessed using logistic regression for each of intermediate and BV compared to normal flora, with GEE methods to account for correlation between visits. HIV incidence rates and confidence intervals were computed using Poisson rates; absolute rate differences were calculated . To assess the risk of HIV-infection we performed multivariable Cox proportional hazards analysis to adjust for potential confounding factors, including demographic, medical, and behavioral characteristics. Variables were selected a priori for inclusion based on previously published association with HIV transmission, and included: (1) characteristics from the time of study enrollment: age (of both partners), region (East versus southern Africa), HSV-2 status of the HIV-1 uninfected male partner, male partner circumcision status, trial randomization assignment (acyclovir versus placebo), and laboratory confirmed STIs at enrollment (i.e., N. gonorrhoeae, C. trachomatis, and T. vaginalis) of both partners; and (2) time-dependent variables, including: pregnancy, hormonal contraceptive use, plasma HIV-1 levels, and CD4 count in the female HIV-1–infected partner, genital ulcer disease in both partners, and sexual behavior during the month prior to each visit, as reported by the male HIV-1–uninfected partner (analyzed as any unprotected sexual intercourse with the study partner, any report of outside partners, and total number of sex acts with the study partner). Robust standard errors were used to account for multiple observations from each person in the time-dependent analyses. Differences in plasma and cervical viral load were assessed using linear regression methods, adjusted for repeated observations. Data were analyzed using SAS version 9.2 (SAS Institute Inc.). Results Study Population A total of 2,236 couples were included in this analysis (Table 1). The median age of HIV-1–infected female partners was 30 y and the median age of HIV-1–uninfected male partners was 35 y. Most couples were married and cohabitating. Couples engaged in sex a median of four times per month, and 30.5% of couples reported sex that was unprotected by condom use during the month prior to enrollment. Among the HIV-1–infected female participants, the median CD4 count was 481 cells/mm3 (interquartile range [IQR] 354–663) and the median plasma HIV-1 RNA concentration was 3.95 log10 copies/ml (3.24–4.53). 10.1371/journal.pmed.1001251.t001 Table 1 Enrollment characteristics, prospective study of 2,236 African HIV-1–seropositive women and their HIV-1 uninfected male partners. Enrollment Characteristics Median (IQR) or n (%) HIV-1–Infected Female HIV-1–Uninfected Male Couple Individual characteristics Age, years 30 (25–35) 35 (30–42) — Education, years 8 (6–10) 9 (7–12) — Hormonal contraceptive use 430 (19.2%) N/A — Male circumcised N/A 1,228 (54.9%) — Couple characteristics a East Africa (versus southern Africa) — — 1,452 (64.9%) Married — — 1,647 (73.7%) Living together — — 1,997 (89.3%) Years lived together — — 5 (2–9) n children together — — 1 (0–2) Sexual behavior (prior month) a n sex acts with study partner — — 4 (2–8) Any unprotected sex with study partner — — 682 (30.5%) Any sex with outside partner — — 96 (4.3%) HIV-1–seropositive female partner characteristics Plasma HIV-1 RNA, log10 copies/ml 3.95 (3.24–4.53) N/A — CD4 count, cells/mm3 481 (354–663) N/A — Randomized to acyclovir (versus placebo) 1,108 (49.6%) N/A — a Couple demographic and behavior characteristics as reported by the HIV-1–uninfected man. N/A, not applicable. Follow-up and HIV-1 Incidence Median follow-up for the HIV-1–seropositive female and HIV-1–seronegative male partners was 20.8 (IQR 15.3–24.1) and 19.3 mo (IQR 13.5–24.0), respectively. Over 3,318 person-years of follow-up, 90 incident HIV-1 infections among men were identified, of which 57 (63.3%) were determined by viral sequencing to be genetically linked within the partnership, for an incidence of linked transmission of 1.72 cases per 100 person-years (95% CI 1.30–2.23). Seven HIV-1 infections occurred in men whose HIV-1–seropositive female partner had no vaginal flora result during the interval when HIV-1 seroconversion occurred. In four of these seven cases, the vaginal swab collection was not expected, while in the remaining three, the result was missing. Thus, 50 HIV-1 incident infections among men with virologically linked HIV-1 transmissions with their female HIV-1–infected partners for whom BV data were available were included in this analysis. BV at Baseline and during Follow-up Of 12,126 visits expected to have vaginal swabs collected during the study, 10,232 (84.4%) had vaginal Gram stain data available. At enrollment, 869 women (41.1%) had BV, 487 (23.0%) had intermediate flora, and 757 (35.8%) had normal vaginal flora. Across all quarterly follow-up visits, the median proportion of women with BV and intermediate vaginal flora was 34.9% (IQR 34.2%–36.3%) and 22.8% (IQR 22.0%–23.9%), respectively, while the median proportion of women with normal vaginal flora was 42.8% (IQR 40.1%–44.1%). Of the 2,221 women with at least one Gram stain result available from the prior 3-mo visit (our main exposure), 337 (15.2%), 113 (5.1%), and 340 (15.3%) had BV, intermediate vaginal flora and normal vaginal flora, respectively, throughout follow-up. An additional 1,151 (51.8%) women had at least a single episode of BV during follow-up. During follow-up, HIV-1–infected women who had one or more intervals with BV were slightly younger than women who had normal vaginal flora and more likely to have an uncircumcised male partner (Table 2). While periods where unprotected sex was reported did not differ by vaginal flora, HIV-1–infected women with BV were more likely to report an outside sexual partner in the last 30 d than HIV-1–infected women with normal vaginal flora. Plasma HIV-1 RNA concentration was slightly elevated, and mean CD4 count was slightly lower in HIV-1–infected women during intervals with BV in comparison to intervals with normal vaginal flora (Table 2). 10.1371/journal.pmed.1001251.t002 Table 2 Participant characteristics during quarterly follow-up intervals with BV and intermediate vaginal flora versus normal vaginal flora. Participant Characteristics Follow-up Intervals for Analysis of HIV-1 Transmission from Women to Mena (n = 2,236 HIV-1–seropositive Women), n (%) or Median (IQR) BV Intervals (n = 4,025) p-Valueb Intermediate Vaginal Flora Intervals (n = 2,566) p-Valuec Normal Vaginal Flora Intervals (n = 4,474) Demographic characteristics Age of HIV-1–seronegative partner, years 35 (29–42) 0.02 36 (30–43) 0.53 36 (30–42) Children within the partnership Having at least one child within the partnership 2,405 (59.8%) 80% among HIV-1–infected persons in sub-Saharan Africa  and thus is unlikely to limit the generalizability of our findings. In addition, the relatively small number of female-to-male HIV-1 transmissions (nine among women with normal vaginal flora versus 31 among women with BV) requires mention. Finally, residual or unmeasured confounding, which cannot be completely excluded, could affect the significance of our findings. This study clearly demonstrates that BV is associated with an increased risk of female-to-male HIV-1 transmission. BV is a highly prevalent condition among HIV-1–infected women. The association of BV with increased infectiousness of HIV-1–infected women requires additional research to understand potential pathogenic mechanisms as well as the etiology, treatment, and prevention of BV. While a large community randomized controlled trial that provided presumptive treatment of STIs including metronidazole for BV failed to reduce HIV-1 incidence , ongoing studies are evaluating more frequent presumptive BV therapy , while others are studying naturally occurring and genetically enhanced probiotics to reduce recurrent BV ,,,. A lactobacillus-predominant vaginal flora might not only reduce the risk of HIV-1 acquisition in women ,, but also HIV-1 transmission to male partners, and points to the potential benefits of using the human microbiota to prevent disease.