Influence of Hormonal Contraceptive Use and HIV on Cervicovaginal Cytokines and Microbiota in Malawi

ABSTRACT Important questions remain on how hormonal contraceptives alter the local immune environment and the microbiota in the female genital tract and how such effects may impact susceptibility to HIV infection. We leveraged samples from a previously conducted clinical trial of Malawian women with (n = 73) and without (n = 24) HIV infection randomized to depot medroxyprogesterone acetate (DMPA) or the levonogestrel implant in equal numbers within each group and determined the effects of these hormonal contraceptives (HCs) on the vaginal immune milieu and the composition of the vaginal microbiota. Longitudinal data for soluble immune mediators, measured by multiplex bead arrays and enzyme-linked immunosorbent assays (ELISAs), and vaginal microbiota, assessed by 16S rRNA gene amplicon, were collected prior to and over a period of 180 days post-HC initiation. DMPA and levonogestrel had only minimal effects on the vaginal immune milieu and microbiota. In women with HIV, with the caveat of a small sample size, there was an association between the median log10 change in the interleukin-12 (IL-12)/IL-10 ratio in vaginal fluid at day 180 post-HC compared to baseline when these women were classified as having a community state type (CST) IV vaginal microbiota and were randomized to DMPA. Long-lasting alterations in soluble immune markers or shifts in microbiota composition were not observed. Furthermore, women with HIV did not exhibit increased viral shedding in the genital tract after HC initiation. Consistent with the results of the ECHO (Evidence for Contraceptive Options and HIV Outcomes) trial, our data imply that the progestin-based HC DMPA and levonorgestrel are associated with minimal risk for women with HIV. (This study has been registered at ClinicalTrials.gov under registration no. NCT02103660). IMPORTANCE The results of the Evidence for Contraceptive Options and HIV Outcomes (ECHO) trial, the first large randomized controlled clinical trial comparing the HIV acquisition risk of women receiving DMPA, the levonorgestrel (LNG) implant, or the copper intrauterine device (IUD), did not reveal an increased risk of HIV acquisition for women on any of these three contraceptives. Our study results confirm that the two different progestin-based hormonal contraceptives DMPA and levonogestrel will not increase the risk for HIV infection. Furthermore, DMPA and levonogestrel have only minimal effects on the immune milieu and the microbiota in the vaginal tract, attesting to the safety of these hormonal contraceptives.

region, race, and ethnicity (36)(37)(38)(39)(40). The mechanisms by which genetic and environmental factors impact the cervicovaginal microbiota remain unknown. It is critical though to consider such differences in the normal cervicovaginal microbiota of African women in study outcome and data interpretation.
CST IV, while common in women who are asymptomatic and otherwise healthy, is reminiscent of the syndrome known as bacterial vaginosis (BV), which is also characterized by a lack of Lactobacillus spp. and high pH (41)(42)(43)(44)(45)(46)(47). BV can be associated with malodorous vaginal discharge and irritation but is often present in the absence of symptoms. CST IV microbiota evaluated with molecular tools, with or without symptoms, is now referred to as molecular BV (48). Up to 40% of African women present with symptomatic or asymptomatic BV (34,(49)(50)(51)(52). While, symptomatic BV has been shown to be a major risk factor for the acquisition of HIV and other sexually transmitted infections (STIs), these risks are also increased in asymptomatic women with BV or CST IV microbiota (53)(54)(55)(56)(57). The enhanced risk of HIV acquisition in these women is due, at least in part, to the more inflammatory milieu that promotes recruitment of HIV target cells to the vaginal mucosa (58). In contrast, when the cervicovaginal microbiota is dominated by Lactobacillus species, particularly L. crispatus, HIV incidence is decreased (24) compared to that in women with CST IV cervicovaginal microbiota (24).
The effect of hormonal contraceptives (HCs) on BV or molecular BV remains controversial. Some studies implied that increased HIV susceptibility in women with HC is related to a decrease in Lactobaccillus species and an increase in CST IV cervicovaginal microbiota (59)(60)(61)(62). Interestingly, other studies have suggested that specific HC methods might be protective against BV (20,63,64). A meta-analysis including 55 studies reported a ;25% reduction in both incident and prevalent BV (symptomatic and asymptomatic combined) with HC use (progestin only or combined estrogen-progestin methods) (65). Many of these studies are limited by their observational, and often cross-sectional study, designs as well as the use of different diagnoses of BV using clinical signs and symptoms (Amsel criteria) (41) or a Gram stain often used in research settings (Nugent score) (42). Molecular methods relying on the amplification and sequencing of the 16S rRNA gene can describe the bacterial composition and relative abundance with genus and species resolution, thus affording the opportunity to study the effect of contraceptives on specific members of the microbiota. Few studies have applied these molecular methods to evaluate the impact of contraceptives on the vaginal microenvironment, especially using comparative prospective study designs.
Given these gaps in our understanding of a critical public, and specifically maternal, health issue, we aimed to clarify the effects of DMPA and the levonorgestrel (LNG) implant on cervicovaginal inflammation and microbiota. We leveraged a previously conducted randomized clinical trial of DMPA and LNG implants (66) in Malawi to compare the vaginal microbiota compositions and immune mediator concentrations in the genital tract before and after contraceptive initiation among women living with HIV (WHIV) and women without HIV.

RESULTS
Characteristics of study participants. Between May 2014 and April 2015, we randomized 24 women without HIV (12 to DMPA and 12 to the LNG implant) ( Table 1 and Fig. 1) and 73 WHIV (37 to DMPA and 36 to the LNG implant) ( Table 2). Among women without HIV, the median ages were 27 years among DMPA users and 26 years among implant users. A quarter (n = 3) of implant users, but no DMPA users, had a genital tract infection diagnosed during the study.
Among WHIV, the median ages were 36 and 34 years and CD4 1 T cell counts were 426 cells/mL and 308 cells/mL for DMPA and implant users, respectively (Table 2). Most WHIV (n = 68) were on antiretroviral therapy (ART) at enrollment (89% and 97% in the DMPA and implant arms, respectively). Of the 33 DMPA users on ART, 28 (85%) were on efavirenz-containing therapy, and 15% (n = 5) were on nevirapine-containing therapy. Of the 35 LNG implant users on ART, 30 (86%) were on efavirenz-containing therapy, 3 (8%) were on nevirapine-containing therapy, and 2 (6%) were on atazanavirbased therapy. During the study, genital tract infections (mostly trichomoniasis) ( Table 2 and see Table S1 in the supplemental material) were diagnosed in 49% (n = 17) and 39% (n = 14) of WHIV in the DMPA and implant arms, respectively, and these women received the Malawi standard-of-care treatment. Treatment reduced the incidence of detectable infections in the female genital tract, but as visit intervals increased (and medical observations and treatment access decreased), the infection appeared to increase again in WHIV (Table S1).
Plasma HIV viremia was detectable at any study visit in 30% (n = 11) and 42% (n = 15) of the DMPA and implant arm participants, respectively. Cervicovaginal lavage (CVL) fluid HIV RNA was detectable at any study visit in 27% (n = 10) and 19% (n = 7) of WHIV in the DMPA and LNG implant arms, respectively, but among these women, few women had detectable CVL HIV RNA at multiple study visits (67,68). The ranges of plasma HIV viral load during the study were 42 to 223,582 copies/mL for DMPA users and 41 to 86,766 copies/mL for LNG implant users, whereas the cervicovaginal HIV viral load at any visit ranged from 40 to 3,756 copies/mL for DMPA users and 40 to 1,544 copies/mL for LNG implant users (67,68).
When comparing the levels of each immune mediator at visit 4 (day 3) to visit 2 to assess acute changes with HC use, and when comparing visit 7 to visit 2 to determine a potential long-term effect of HC use, few changes were notable. WHIV on DMPA experienced a decrease in vaginal gamma interferon (IFN-g ) and BD-3 levels and an increase in IL-1a at visit 4 compared to visit 2 ( Fig. 3A), but these changes were transient and no longer detectable at visit 7 ( Fig. 3B). At visit 7, median IL-8 concentrations were lower in WHIV on DMPA than at baseline. However, none of these changes in cytokine levels were statistically significant when correcting for multiple comparisons. These results also imply that any infections other than HIV in the female genital tract during the study period (see Table S1) did not further impact the vaginal immune milieu beyond that of the HIV-associated immune activation that was noted at visit 2 ( Fig. 2). LNG users among WHIV did not experience any changes in vaginal immune mediators ( Fig. 3C and D). In women without HIV, independent of the type of contraceptive, no differences were observed in the vaginal immune milieu (Fig. 4).
When comparing the concentrations of each immune marker at specific time points (visits 2, 4, and 7) between DMPA and LNG implant users among women without HIV (Fig. 5A) and among WHIV (Fig. 5B), there were no differences in immune mediator concentrations noted, with the exception of slightly reduced BD3 levels at visit 7 in WHIV on LNG versus DMPA. In addition, these comparisons were repeated without stratifying by HIV status, with similar findings of no changes in overall cytokine measurements between contraceptive arms or with time within each arm with contraceptive use (data not shown).
Vaginal microbiota characteristics at baseline prior to HC initiation (visit 2). Data were available for 20 of 24 (83.3%) women without HIV and for 62 of 72 (84.9%) WHIV. Among the 20 women without HIV, L. crispatus-dominated vaginal microbiota (CST I) was found in 4 women (20.0%), and 1 woman had a microbiome classified as CST II (L. gasseri dominated). About equal numbers of women without HIV had a vaginal microbiota that was characteristic of CST III (L. iners dominated; n = 7; 35%) or CST IV (n = 8; 40%). Among WHIV, consistent with the more inflammatory vaginal milieu observed at baseline, only 8.1% (n = 5) of women had a CST I vaginal microbiota.   (Table 3). Thus, comparing the number of women with CST I, CST II, CST III, or CST IV at baseline (visit 2), the vaginal microbiota of women with HIV was not distinct from that of women without HIV (P = 0.095) by Fisher's exact test. The vaginal microbiota composition was unknown at visit 3 when women were randomly assigned to DMPA or LNG implant. Among the women without HIV who initiated DMPA (n = 11), women with a CST I (n = 1) or CST II (n = 1) vaginal microbiota at visit 2 were less represented than women with CST III (n = 4) and CST IV (n = 5) ( Table 3). In the LNG group of women without HIV, an equal number of women (n = 3) had a baseline vaginal microbiota characterized as CST I, CST III, or CST IV (Table 3). Among WHIV, in both the DMPA (n = 22 of 32) and LNG (n = 15 of 30) groups, the majority of women, about two-thirds (68.8%) and one-half (50%), respectively, had a CST IV-dominated vaginal microbiota. In contrast, less than 10% of WHIV had CST I vaginal microbiota at the time of DMPA or LNG initiation (Table 3).
Effects of hormonal contraceptive use on vaginal microbiota composition. After the start of HC use, only minor changes in the CST classifications were observed for each individual stratified by HC method and HIV status (Table 4 and Fig. 6); the most common were transitions between CST III and CST IV, occurring in both directions over time. No correlations were observed between HC type and changes in the abundance of a specific bacterial species. However, the assessment of the change in the median log 10 abundance of all Lactobacillus (Lb) species combined at visit 7 versus visit 2 [log 10 (Lb @ V7/V2)] revealed that the distribution of subjects with smaller compared to larger changes in log 10 (Lb @ V7/V2) was different between women who had initiated DMPA or LNG (Fig. S1). Thus, the probability that a larger change in log 10 (Lb @ V7/V2) -or an increase in overall Lactobacillus spp.-after HC initiation occurred was significantly higher for women who were assigned to DMPA (P = 0.024), and this effect was more pronounced when the analysis was limited to WHIV (P = 0.004). In this context it FIG 3 Changes in immune markers of WHIV after contraceptive initiation. Shown are the median concentrations 6 95% CI of various immune markers in CVL samples of WHIV assigned to DMPA (top panels) or the LNG implant (bottom panels) at study visit 4 (gray triangles, left panels) and visit 7 (right panels, gray triangles) compared to visit 2 (black triangles). Statistical comparisons were performed by paired Wilcoxon rank test.
Hormonal Contraceptives, Vaginal Milieu, and HIV mSphere should be noted that although there was no difference in the median log 10 ratio of the abundance of Sneathia sanguinegens (a bacterial species commonly found in CST IV) at visit 7 versus visit 2 [log 10 (Ss @ V7/V2)] between participants in the DMPA and LNG groups, the probability of women experiencing an increase in the abundance of at visit 7 compared to visit 2, was lower (P = 0.08) in the DMPA group. WHIV had a significantly higher probability (P = 0.0098) of having a change in the median log 10 (Ss @ V7/V2) abundance (Fig. S1). Shannon diversity (Fig. S2) and total bacterial load (Fig. S3) did not differ significantly between the DMPA and LNG implant groups together and within group medians after the start of HC use. Potential interactions between HIV status, local immune milieu, vaginal microbiota, and contraceptive use. Despite the lack of major changes in the vaginal immune milieu or microbiota after HC initiation, there were some notable signals of interactions between HIV status, vaginal immune milieu, vaginal microbiota, and HC use. Consistent with the fact that bacterial species characteristic for CST IV promote a more inflammatory milieu, higher levels of some inflammatory cytokines in WHIV compared to women without HIV prior to HC use at visit 2 ( Fig. 2) were only detectable in comparative analyses of women with CST I or CST III vaginal microbiota (Fig. S4). Post-HC initiation, we did not observe any changes in immune markers based on HC method at visits 4 and 7, when we stratified women without HIV or WHIV by CST ( Fig. S5 and data not shown). There was, however, a change in the median log 10 ratio of IL-12p40 to IL-10 concentrations at visit 7 compared to visit 2 (20.137) across all women after initiation of HC use with P = 0.004. The change in the log 10 (IL-12/IL-10 @ V7/V2) after HC initiation was strongest in WHIV (median difference, 20.1550; P = 0.0011) and appeared to be driven primarily by women in the DMPA group (median difference, 20.2110; P = 0.0017) compared to the LNG group (median difference,  (Table 5). Thus, women with CST IV at visit 2 were more likely to have a significant change in the median ratio of log 10 (IL-12/IL-10 @ V7/V2) than women with vaginal microbiota classified as CST I or CST III at baseline (Table 5). A similar effect was observed for the change in the median log 10 IFN-g /IL-10 ratio at visit 7 compared to visit 2 (Table 5). When considering single cytokines, women with vaginal microbiota classified as CST IV at visit 2 had a decrease in the median concentrations of log 10 (IL-12 @ V7/V2) after HC initiation. Across all women, the probability of a change in the median concentration of log 10 (IL-12 @ V7/ V2) was inversely related to the likelihood that a woman had received DMPA (Fig. 7). There was also an inverse relationship between the change of the median concentration of log 10 (IL-12 @ V7/V2) and the median log 10 abundance of L. iners at visit 7 compared to visit 2, but only in WHIV (odds ratio [OR], 20.1; P = 0.0053) (Fig. 7). Among WHIV, this significant correlation was not observed for women using the LNG implant (P = 0.1350), but it was evident in DMPA-treated women (OR, 20.11; P = 0.0224). Such a relationship was not observed for L. crispatus or L. gasseri (data not shown and with no data for L. jensenii due to the limited number of women with L. jensenii). Similar interactions between vaginal microbiota, HC use, and local immune milieu were not observed when other immune markers were considered.

DISCUSSION
Our results of a randomized trial of DMPA or LNG implant among women without HIV and WHIV demonstrate minimal impact of the two different progestin-based hormonal contraceptives on vaginal immune markers or microbiota composition. Furthermore, within each contraceptive arm, we also note no or only minor changes in the composition of the vaginal microbiota at 3 days (visit 4) and 6 months (visit 7) after contraceptive initiation, suggesting that within individuals, there is no altered risk profile. These findings support those from the ECHO trial, add to the reassuring data on DMPA use, and are consistent with recent WHO and CDC medical eligibility criteria (69).
We found no long-lasting changes in local immune mediators among women without HIV or WHIV after contraceptive initiation. Although we observed a transient  decrease in a few inflammatory cytokines in WHIV at day 3 after DMPA start, none of these changes were significant after Benjamini-Hochberg correction or by month 6. As cytokines do not act in isolation, we had also evaluated changes in the ratio of proinflammatory to anti-inflammatory cytokines (IL-12p40/IL-10 and IFN-g /IL-10 ratios), and this analysis confirmed that neither DMPA nor LNG impacted the immune milieu in women without HIV or WHIV. We previously reported that some of the WHIV had genital HIV shedding (67). HC use did not alter the frequency or the amount of HIV  (67), and consistent with the findings of no significant changes in the local immune milieu, we also found no correlation between changes in immune markers and viral shedding. In contrast to our results, some previous studies reported that DMPA alters the levels of certain immune markers in the female genital tract. A case-control study from South Africa evaluated 42 CVL markers and found that women without HIV (n = 64) using either injectable DMPA or noresthisterone-enanthate had lower IL-12p40, IL-15, monocyte chemotactic protein 1 (MCP-1), eotaxin, macrophage-derived chemokine (MDC), and platelet-derived growth factor-AA (PDGF-AA) concentrations than 64 matched non-HC users at baseline (70). In a cross-sectional study of Kenyan and South African women without HIV enrolled in the FEM-PrEP trial, DMPA users (n = 75) had significantly higher macrophage inflammatory protein 1a (MIP-1a), MIP-1b, IL-6, IL-8, CXCL-10, and RANTES concentrations than the non-HC users (n = 99) (71). Our study could not confirm these findings, but our number of women without HIV assigned to DMPA was relatively small. Differences in study design, population, specimen type, specimen collection timing, and analysis methods may explain and contribute to these  different results. We did find, however, an association between the decrease in the median log 10 (IL-12/IL-10 @ V7/V2) ratio in WHIV when these WHIV were classified with a CST IV-dominated vaginal microbiota at baseline and randomized to initiate DMPA. A decrease in the median ratios of IL-12p40 or IFN-g to IL-10 from V7/V2 could be due to an increase in IL-10 levels, a decrease in IL-12p40, or both, but the relative increase in IL-10 exceeded the relative decrease of IL-12p40. Biologically that translates into a less inflammatory milieu. This observation was most prominent in HIV women of the DMPA group, but it was only significant when these women presented with CST IV at baseline, implying that a potentially immunosuppressive effect of DMPA would only have a biological impact in specific subpopulations. These findings underline the complexity of the interaction between HC, local immune milieu, and cervicovaginal microbiota. The caveat of our study is the rather small sample size when we focused our analysis on a subgroup of women depending on HIV status, specific HC use, and CST classification prior to HC initiation. Our vaginal microbiota analysis highlights that most women in our Malawian cohort are not L. crispatus dominant. This finding is consistent with several other microbiome studies of sub-Saharan African women (23,24,30). Although several women in our cohort underwent occasional shifts in the microbiota composition over the course of the study, the changes were generally transient, with the vast majority maintaining a CST III or CST IV state of their vaginal microbiota. Thus, overall, the contraceptive exposure did not lead to any consistent and significant changes. DMPA use resulted in a slight decrease in bacterial load, suggesting that qualitative and quantitative changes in microbiota composition need to be assessed to determine the impact of contraceptives on the vaginal microbiota.
In addition, underlying conditions, such as BV or other infections in the FGT, may impact the outcome of contraceptive use on the local milieu and HIV risk. Indeed, at study entry, the incidence of genital tract infections was higher in WHIV than in women without HIV. A recent study of a Zambian discordant cohort found a significant modification of the risk of HIV acquisition with DMPA use when women presented with BV. Women with BV had more than a 6-fold increased risk of HIV in the setting of BV, whereas there was no significantly increased risk in women without BV (72). Although this finding was not confirmed in some other studies (73), other studies also demonstrated that vaginal inflammation is attributed to vaginal dysbiosis (33,50,74). Despite slightly higher numbers of genital tract infections at visit 7 than at visit 4 in WHIV, we did not observe an increase in local inflammation. On the contrary, in WHIV on DMPA, as discussed above, inflammation appeared to lessen during the study course. Our study, however, was underpowered to evaluate the interactions between microbiota composition, local immune milieu, contraceptive use, and HIV status. Larger cohorts are needed to identify these interactions and should be conducted to conclusively define the individual-level risk factors, including specific vaginal microbiota within CST IV, that may influence one's immunologic response and subsequent risk profile following contraceptive method use.
Our study has several notable strengths, including its randomized prospective study design and before-after comparisons. We had no loss to follow-up during the 6 months after contraceptive initiation. Our study was limited by its small sample size, particularly in the group of women without HIV. Therefore, we were restricted in the evaluations of distinct stratifications (e.g., CST) and interactions between groups. Additionally, we have limited generalizability as the participants were recruited from a single site; however, Malawi represents a country in the heart of the HIV epidemic, reassuring that our results have translational value.

MATERIALS AND METHODS
Study design and enrollment. We previously conducted an open-label randomized trial (ClinicalTrials .gov registration no. NCT02103660) among women desiring to initiate contraception at the Bwaila Hospital, a large district hospital in Lilongwe, Malawi. The study procedures have been previously described (66,67). Briefly, women interested in study participation underwent informed consent for study screening (Fig. 1).
Inclusion criteria included (i) age 18 to 45 years, (ii) known HIV status (as documented during screening), (iii) self-report of at least two regular, monthly cycles (;21 to 35 days) preceding study enrollment, (iv) selfreport of not being on hormonal or intrauterine contraception for at least 6 months, (v) at least 6 months postpartum; (vi) interested in initiating DMPA or the LNG implant, (vii) accepting to be randomized to receive either DMPA or the LNG implant, and (viii) willing to wait 4 to 6 weeks after enrollment to receive the contraceptive method (with condom use or abstinence encouraged during this time frame). We excluded women who were pregnant at screening or desiring pregnancy within the next 12 months or who had any medical contraindications to DMPA or LNG implant per the WHO medical eligibility criteria (75). We also excluded women who were newly diagnosed with HIV at screening but had a known negative HIV test within the past 6 months, because acute HIV infection is associated with high viremia, virus shedding in the FGT, and drastic changes in immune parameters that are not representative of virologic and immune status observed once the viral set point has been reached.
Eligible women underwent consent for study participation and completed an enrollment interview that included questions on demographic, medical, and reproductive health information. We determined cycle phase by patients' self-report of the first day of their last menstrual period and defined the follicular phase as within the first 14 days from their last menses, and the luteal phase from day 15 from the onset of their last menses until the start of the next menses. Participants who were determined to be in the follicular phase on the day of enrollment, also completed visit 1 on that day. Otherwise, they were asked to return for visit 1 during the first 14 days of their next anticipated menstrual cycle. At visit 1, an interval history assessment and physical exam were completed. Participants with untreated visible genital ulcers or lesions at the initial pelvic examination were terminated from the study. Visit 2 was scheduled to occur during the luteal phase of the same cycle visit 1. In a few cases, when the participant missed visit 2 in the corresponding visit 1 cycle, visit 2 was rescheduled to the luteal phase of the subsequent cycle. Although cycle length can vary, for consistency, visits 1 and 2 were scheduled assuming a 28-day cycle. We excluded days of menses to avoid blood contamination of cervicovaginal samples.
At visit 3, within the first 7 days after the start of their next menses, we randomized women to DMPA or the LNG implant using permuted-block randomization schemes that were generated independently for WHIV and women without HIV. Return visits post-HC initiation were scheduled for day 3 (visit 4), day 30 (visit 5), day 90 (visit 6), and day 180 (visit 7). At all visits, patients completed an interval history assessment, a physical and pelvic examination, and specimen collection. Contraceptive use by women on LNG implant was confirmed by palpation at visits 4 to 7. Women who were randomized to DMPA received repeat DMPA injections at visits 6 and 7. Baseline characteristics for women without HIV and WLHIV are listed in Tables 1 and 2.
Specimen collection. Cervicovaginal samples were collected at visits 1, 2, 4, 5, 6, and 7. A catch-all swab was run along the posterior vaginal wall and placed immediately on ice with transfer within 2 h for storage at 280°C until processing for microbiota analysis. Cervicovaginal lavage (CVL) fluid was collected by washing the cervix, vaginal walls, and posterior fornix with 10 mL of phosphate-buffered saline (PBS) and then aspirating the pooled fluid (66,68). The CVL was centrifuged, and the supernatant fluid was stored at 280°C for immune marker analysis. Collection of cervicovaginal samples was rescheduled if the woman had vaginal bleeding at the time of the study visit; if there were no days without bleeding during the study visit window, the sample was missed.
Measurement of soluble immune mediators. The immune mediators were selected based on review of the relevant literature to ensure comparison of our results with prior and ongoing studies (4,18,19). , which were measured by enzyme-linked immunosorbent assay (ELISA) (R&D Systems), all markers were measured by a custom multiparameter bead array (Millipore, Billerica, MA) according to the manufacturer's protocols. CVL samples were run undiluted, except for CXCL-10 and BD3, which were measured at a dilution of 1:10, and SLPI, which was measured at 1:200. Sample dilutions were determined by randomly selecting 3 samples that were tested in the various assays undiluted or at dilutions of 1:10, 1:100, 1:200, and 1:500. Concentrations that were below the lower limit of detection of the assay were reported as the value for the lower limit of detection for that immune mediator. IL-1Ra and IFN-a concentrations were not included in the statistical analysis because the vast majority of values fell above or below the standard curve, respectively, and as limited CVL volumes did not allow for testing a different dilution, concentrations could not be accurately quantified.
Microbiota composition assessment. Total DNA from vaginal samples was isolated by the UNC Microbiome Core Facility using the Qiagen blood and tissue and QIAamp DNA stool protocols (Qiagen). PCR amplification of the V3-V4 region of the 16S rRNA gene was conducted using a protocol previously described and validated (76). Amplicons were pooled in equimolar concentrations and purified prior to loading on an Illumina HiSeq 2500 modified to generate 300-bp paired-end reads (76). Negative and positive-control samples were also as previously described (76). The sequences were demultiplexed using the dual-barcode strategy, with a mapping file linking barcode to samples and split_libraries.py, a QIIME-dependent script (77). The resulting forward and reverse fastq files were split by sample using the Hormonal Contraceptives, Vaginal Milieu, and HIV mSphere QIIME-dependent script split_sequence_file_on_sample_ids.py, and primer sequences were removed using TagCleaner (v.0.16) (78). Further processing followed the DADA2 Workflow for Big Data and dada2 (v.1.5.2) (https://benjjneb.github.io/dada2/bigdata.html) (79). Taxonomy was assigned to each amplicon sequence variant (ASV) generated by dada2 using the RDP Naïve Bayesian Classifier (80) trained with the SILVA 16S rRNA gene database (81). SpeciateIT (version 1.0; https://github.com/ravel-lab/speciateIT), a rapid per-sequence classifier, was used to assign species to the major genera. Sequence counts for ASVs assigned to the same taxonomy were summed for each sample. A table of sequence count assigned to each bacterial taxon for each sample was generated and used in statistical analyses. Bacterial taxa (n = 254) were filtered before analysis if observed in fewer than two samples or present at a frequency of less than 10 25 study-wide (n = 169). CST assignments were performed using VALENCIA, a novel classifier that affords consistent CST assignment without clustering (82).
Shannon diversity was calculated as previously reported (26). Quantification of 16S rRNA gene copy number and estimation of bacterial absolute abundance were performed as follows. The total number of 16S rRNA gene copies in each DNA sample was measured using the TaqMan BactQuant assay targeting the V3-V4 regions of the gene (83). The total number of 16S rRNA gene copies was expressed as number of copies per swab and used as an estimate of bacterial load (total count of bacterial cells present in a sample). An estimate of absolute abundance of each bacterial taxon was calculated for each sample by multiplying the total 16S rRNA gene copies obtained by quantitative PCR (qPCR) and the relative abundance of that taxa obtained by 16S rRNA gene sequencing.
Statistical analysis. Women were grouped by HIV status stratified by HC type. To avoid potential contamination of microbiota or immune marker data by blood, we selected visit 2 as baseline for all analyses as these visits were scheduled to occur in the luteal phase and less likely to be acutely influenced by recent menses. Notably cycle timing was based on self-report only and not confirmed by hormone testing.
For immune marker analysis, we present the median concentration 6 95% confidence interval (CI) for each marker. Data between women of different HIV status and/or HC use at the same visit were compared by Mann-Whitney test. Paired data (visit 4 versus visit 2 and visit 7 versus visit 2) of women in the same group based on HIV and HC status were analyzed by Wilcoxon matched-pair signed-rank test. Differences in immune mediator levels were statistically significant at an a of 0.05; to account for multiple comparisons, we applied the Benjamini-Hochberg method to calculate adjusted P values (P , 0.0035) due to multiple comparisons increasing the false-discovery rate (FDR). All graphs were generated and statistical analyses performed with GraphPad Prism version 7 (La Jolla, CA).
Fisher's exact test was performed to test for differences in the vaginal microbiota between women with and without HIV at baseline (visit 2).
The analysis of the dependence between the probability of DMPA and the change in the log 10 relative abundance of IL-12p40 at visit 7 versus visit 2 within the WHIV group was performed using the Baysian logistic regression adaptive spline model in R. The same method was applied to the analyses of the dependence between changes in the log 10 abundance of IL-12p40 and changes in the log 10

ACKNOWLEDGMENTS
We thank our study staff, the Lilongwe District Management Team, for their support of our study, and the Lighthouse Trust and Area 25 Health Centre for allowing us to inform potential participants about the study at their clinics.
This study was funded by the following grants: CDC no. U48DP001944, CDC no. 200-2015-M-63021, NIH no. 1K01-TW009657-01, NIH no. P30-AI50410, Bill & Melinda Gates Foundation no. OPP1090837, and USAID no. AID-OAA-A-15-00045. Analysis performed by K.D.P. was supported by the UNC Flow Cytometry Core, which is supported in part by Cancer Center Core support grant P30 CA016086 to the UNC Lineberger Comprehensive Cancer Center. Technical assistance was provided by the CFAR HIV/STD Laboratory Core, which is funded by NIH P30 AI050410. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the U.S. Centers for Disease Control and Prevention.
We declare no conflict of interest.