The Vaginal Microbiome: The Influence of Intramuscular Depot-medroxyprogesterone Acetate Initiation on Vaginal Microbiota and A Comparison of PCR Approaches for Use in Predicting HIV Acquisition

The vaginal microbiome is a key factor in women’s reproductive health and hormones may play an important role in the composition of vaginal bacterial communities. The vaginal microbiome is commonly evaluated using broad-range polymerase chain reaction (PCR) with next generation sequencing (NGS) or taxon-specific quantitative PCR (qPCR); both approaches have advantages and disadvantages. We sought to describe how DMPA-IM initiation influences the vaginal microbiome in postpartum women and provide evidence as to whether both PCR approaches provide the same value for predicting important biological outcomes of interest, such as HIV acquisition, to help identify optimal PCR approach(s) for use in future research and clinical applications. To assess the impact of DMPA-IM initiation on the vaginal microbiome, we used data from the Postpartum Family Planning (PFP) cohort, a prospective study of postpartum women in Kenya initiating DMPA-IM or non-hormonal contraception (non-HC). Enrollment occurred at the time of contraception initiation, approximately 6-14 weeks postpartum, and vaginal swabs for Nugent score determination, taxon-specific qPCR, and broad-range 16S rRNA gene PCR coupled with NGS were collected at enrollment and three-months post-initiation. Generalized estimating equations and linear mixed models were used to estimate mean change in Nugent score, total bacterial load, taxa concentrations, and bacterial community alpha diversity in women using DMPA-IM compared to those using non-HC. The effect of DMPA-IM on microbial community composition was assessed by comparing change in community type (CT) membership, created through hierarchical clustering, using a Wilcoxon signed-rank test. The PFP cohort enrolled 54 HIV-negative women, 33 choosing DMPA-IM and 21 choosing non-HC. Women who chose DMPA-IM were more likely to be married (97% vs. 67%) and have resumed intercourse since delivery (52% vs. 29%) compared to women who chose non-HC. Baseline distributions of CTs was similar between contraceptive groups, however bacterial vaginosis by Nugent score was more common among DMPA-IM users. After three months, non-HC users were more likely to have Lactobacillus spp. dominant microbiomes compared to DMPA-IM users (n=11, 61% vs. n=8, 31%; chi-square p=0.046). Nugent score decreased significantly among DMPA-IM users (change=-1.89 points, p=0.02), however there was not a corresponding decrease in alpha diversity (change=0.03, p=0.83). Among non-HC users, Nugent score remained more stable (change=-0.73 points, p=0.33) while alpha diversity decreased (change=-0.34, p=0.05). While there were significant changes in Nugent score and alpha diversity within contraceptive groups, the observed changes were not significantly different between the groups. After three months, significant decreases in the concentrations of Sneathia species, Mycoplasma hominis, and Parvimonas species Type 1 were seen among non-HC users, however concentrations remained stable among DMPA-IM users; contraceptive method was associated with significantly different changes in M. hominis concentration between groups (p=0.010). To assess how abundance measures from qPCR and NGS compare in their relative value for predicting HIV acquisition, we used data from a case-control study evaluating the association between vaginal bacteria and HIV acquisition comprised of women from Eastern and Southern Africa at risk of HIV. A subset of 55 cases and 55 matched controls had both NGS and qPCR sequencing performed on their samples. We generated a series of models using logistic regression to assess the performance of bacterial concentration, from 9 selected taxa, and relative abundance measures as predictors of HIV. We performed principal components analysis (PCA) to assess the combined predictive value of bacterial abundances of taxa of interest. We also performed least absolute shrinkage and selection operator (LASSO) to select the most informative taxa for HIV prediction. Level of discrimination was quantified using the area under the receiver‐operating curve (AUC) and AUCs were compared between models. Within the nested case-control study, median concentrations and relative abundances of bacteria associated with HIV acquisition were low. A reference model containing demographic and risk factor variables had an acceptable discrimination value (AUC=0.706, 95% Confidence Interval (CI): 0.609-0.804). The first three principal components (PCs) from qPCR data resulted in an AUC=0.787 (95%CI: 0.702-0.872) and the first three PCs from NGS data resulted in an AUC=0.761 (95%CI: 0.672-0.850). AUCs from LASSO selection resulted in similar discrimination (AUC=0.771 vs. AUC=0.784, respectively); there was no significant difference in discrimination between PCR approaches. In a cohort of postpartum African women, vaginal bacterial diversity did not change in DMPA-IM users despite a reduction in Nugent-BV, but decreased significantly among women using non-HC. Choice of contraception may influence Lactobacillus recovery in postpartum women. Additionally, our findings suggest that postpartum use of DMPA-IM and non-HC may have differential impacts on the vaginal concentrations of some bacteria that have previously been associated with HIV acquisition. In the nested case-control study, we found the abundance of bacterial taxa associated with HIV acquisition from qPCR, in the form of concentration, and from broad-range PCR with NGS, in the form of relative abundance, provide similar predictive discrimination between women who acquired HIV and women who remained HIV uninfected.