Samples included in this study were collected as part of the ECHO (clinicaltrials.gov/NCT02550067) and UChoose (clinicaltrials.gov/NCT02404038) Trials. Both trials were implemented in accordance with Good Clinical Practice. The ECHO Trial was approved by the Research Ethics Committees at the University of Cape Town (HREC 371/2015), University of Witwatersrand (HREC PRC 141112), KEMRI (SERU/CMR/P0014/3109), University of Washington (STUDY00000261), and FHI360 (523201). The UChoose Trial was approved by the Research Ethics Committee at the University of Cape Town (HREC 801/2014). All participants 18 years or older provided informed consent, while informed assent from the participant and informed consent from a parent or legal guardian were obtained for participants younger than 18 years old.

The ECHO Trial was a randomized, multicentre, open-label trial carried out at 12 research sites in SSA to evaluate HIV incidence among non-pregnant, HIV-seronegative women aged 16–35 years who self-reported not using injectable, intrauterine, or implantable contraception for the previous 6 months. Women were randomized to copper-IUD, the LNG-implant or DMPA-IM (150 mg of depot medroxyprogesterone acetate), provided on site at enrolment and then every three months for 18 months. Inclusion criteria for the parent trial were previously described (9). At three sites, the Emavundleni Clinical Research Site, Cape Town, South Africa, the Wits Reproductive Health and HIV Institute (WRHI), Johannesburg, South Africa and the Kenya Medical Research Institute (KEMRI), Kisumu, Kenya, participants were sequentially recruited into a nested sub-study of mucosal immunology. From this subset of participants, women assigned to DMPA-IM were included in the current analysis. Samples at baseline and after two injections of DMPA-IM (months zero and six) were included in this analysis.

The UChoose Trial was an open-label, randomized crossover trial at the Masiphumelele Clinical Research Site, Cape Town, South Africa, aiming to evaluate the feasibility of hormonal contraceptive options as proxy for HIV prevention methods among non-pregnant, HIV-seronegative adolescents aged 15-19 years who were either contraception-naïve, or on a short-term contraceptive method and willing to change to another method. Detailed eligibility criteria have been previously described (20). Briefly, adolescents were randomized to a combined contraceptive intravaginal ring, combined oral contraceptive pills or the bi-monthly injectable hormonal contraception NET-EN (200mg of norethisterone enantate). From these participants, only women assigned to NET-EN were included in the current analysis. Samples at baseline and after two injections of NET-EN (months zero and four) were included in this analysis.

In both trials, detailed interviewer-assisted questionnaires assessing demographics, medical and obstetric history, and sexual behaviour were completed. At baseline and after two contraceptive injections, cervicovaginal secretions were collected using menstrual cups inserted for at least 30 minutes. Clinicians removed the menstrual cup prior to speculum insertion and immediately placed the cup into a sterile tube. Menstrual cup secretions were extracted by centrifuging at 453g at 4°C for 10min upon reaching the laboratory, and diluted 1:4 in sterile phosphate buffered saline (PBS). Using a speculum, vaginal swabs were collected from the lateral vaginal wall for microbiota analysis. The swab was placed into a sterile vial with Digene^® transport media for DNA extraction. All samples were immediately placed at 4°C, transported to the laboratory within 4hrs of collection, and stored at -80°C until analysis. Other samples collected include genital swabs for STI and BV testing, and blood for herpes simplex virus 2 (HSV-2) serology.

In both trials, blood was collected at baseline and serology was performed to screen for IgG antibodies to herpes simplex virus 2 (HSV-2) gG2 using an enzyme-linked immunosorbent assay (HerpeSelect, Focus Diagnostics, USA). Genital swabs were collected to test for STIs (Chlamydia trachomatis and Neisseria gonorrhoeae) at baseline (pre-contraceptive initiation), and for Nugent-BV and candidiasis at baseline and after two injections of either NET-EN or DMPA-IM. In the UChoose Trial, vulvovaginal swabs were tested for C. trachomatis and N. gonorrhoeae by multiplex PCR, as previously described (21). In the ECHO Trial, the GeneXpert Instrument Systems platform (Cepheid Inc., US) with the Abbott Real Time PCR assay (Abbott Molecular, US) was used to test for C. trachomatis and N. gonorrhoeae in endocervical swabs at the South African sites, while the Panther System (Hologic Inc., US) was used in Kenya. For both studies, microscopic slides with vaginal smears were Gram-stained for Nugent scoring (to diagnose Nugent-BV, where participants with a Nugent score 0-3 were BV-negative, those with a Nugent score of 7-10 were BV-positive and those with an intermediate microbiota had a Nugent score of 4-6) and candidiasis screening (Candida hyphae and spores). Participants were provided treatment for curable STIs, using both syndromic and aetiological diagnoses. BV was treated syndromically in accordance with the local syndromic management guidelines.

For this analysis, specimens from the UChoose and ECHO Trials included here were assayed as part of the same Luminex runs, using identical assay lot numbers. After filtering (0.2μM cellulose acetate filters) cervicovaginal fluid, the concentrations of IL-1β, IL-6, TNF-α and IL12(p70) (inflammatory cytokines), CXCL8/IL-8, Exotaxin, CXCL10/IP-10, CCL2/MCP-1, CCL3/MIP-1α, CCL4/MIP-1β and CCL5/RANTES (chemokines), IL-2, IL-4, IL-5, IL-13, IL-15, IL-17 and IFN-γ (adaptive cytokines), IL-7, IL-9, FGF-basic, G-CSF, GM-CSF, PDFG-BB, and VEGF (growth factors), and IL-1RA and IL-10 (regulatory cytokines) were measured by Luminex using the Bio-Plex™ Pro Human cytokine 27-plex assay (Bio-Rad Laboratories Inc^®, USA) according to the manufacturer’s instructions, alongside intra- and inter-plate controls, and acquired using a Bio-Plex Suspension Array Reader (Bio-Rad Laboratories Inc^®, USA). The Bio-plex manager software (version 4) was used for data analysis and a 5 Parameter Logistic (5 PL) regression formula was used to calculate cytokine concentrations from the standard curves. Cytokine concentrations below the lower limit of detection of the assay were reported as the mid-point between zero and the lowest concentration measured for each cytokine.

For this analysis, specimens from the UChoose and ECHO Trials included here were assayed as part of the same library preparation and sequencing runs. Lateral vaginal wall swabs (Digene^®) were slowly thawed on ice, shaken at 100rpm for 3 minutes, and 250μl of suspension was used for DNA extraction. Bacterial gDNA extraction was carried out using the Qiagen DNeasy Powersoil HTP 96 kit according to the manufacturer’s instructions. If there was insufficient volume in the sample vial, 550μl of sterile PBS was added to the vial, which was then shaken for another 3 minutes at 100rpm, and 250μl of suspension transferred for DNA extraction. After DNA extraction, sequencing of the V3-V4 hypervariable region of the bacterial 16S rRNA gene was carried out as described previously (22) except using 357F/806R primers ( Supplementary Table 1 ). Samples were amplified in triplicates. Amplicons were purified using AMPure XP beads (Beckman Coulter) and pooled in equal mass amounts to be sequenced using the Illumina MiSeq platform (300bp paired-end, V3). Following demultiplexing, raw reads were processed and classified using DADA2 (23). Samples with <2000 reads were excluded from further analyses. Taxonomic annotation was carried out using the Ribosomal Database Project’s (RDP) Naïve Bayesian classifier (24). Sequence classification was trained against an updated version of the Silva training set version 132 (25), available at https://github.com/itsmisterbrown/updated_16S_dbs. Run-specific contamination filtering was performed using microfiltR (https://github.com/itsmisterbrown/microfiltR). Sequencing runs were merged using custom scripts (https://github.com/itsmisterbrown/marker_gene_dataset_merging_functions).

With the sample size (n=44 NET-EN, n=53 DMPA-IM) of this study, the minimal difference detectable with 80% power at α=0.05 would have been a 2.45-fold increase in the proportion of women with genital inflammation (when binarizing genital inflammation as absent/present based on the upregulation of ≥5/8 cytokines to above their 75^th percentile) (8, 26) assigned to DMPA-IM versus NET-EN.

All statistical analyses were performed in RStudio Version 3.6 and STATA version 11.0 (StataCorp, Texas, US). Cross-sectional differences in baseline group characteristics were tested using Pearson´s Chi-squared test or Fisher’s exact test (when the expected value was <5) for count data. The unpaired Mann-Whitney U-test was applied for differences in medians. Multivariate logistic regression analyses were performed using the cytokine and microbiota data, and confounders were identified using a step-up model-building strategy (including site, age, sexual risk behaviour, bacterial STIs, HSV-2 serology, Nugent-BV and candidiasis). Adjusted R² and Akaike information criterion (AIC) values were compared as a measure of model fit. For our population, the most robust model was adjusted for baseline HSV-2 serology, C. trachomatis, N. gonorrhoeae and candidiasis. A false discovery rate step down procedure (27) was used to adjust for multiple comparisons. 95% confidence intervals and p values of ≤0.05 were used to assess statistical significance. The presence of genital inflammation in women was defined by their relative concentrations of eight proinflammatory cytokines and chemokines (IL-1β, IL-6, CXCL8/IL-8, CXCL10/IP-10, CCL2/MCP-1, CCL3/MIP-1α, CCL4/MIP-1β and TNF-α), with genital inflammation present if five or more of these eight markers were above the 75th percentile, as previously described (8, 26). Vaginal microbial community state types (CSTs) were generated using partitioning around medoids (PAM) clustering using the cluster package (28). Downstream microbiota data analysis, including ecological diversity metrics, ordination and fold change differences, was performed in R (v3.6.0) using the packages phyloseq (29), NMF (30), metagenomeSeq (31), vegan (32), and DESeq2 (33). The permutational multivariate analysis of variance (PERMANOVA) test was used to assess the heterogeneity of dispersion between groups of samples. The Data Integration Analysis for Biomarker discovery using Latent cOmponents (DIABLO) framework, as part of the mixOmics R Bioconductor package (34), was used for multi-omics analyses integrating the microbial and inflammation data. Cytokines and bacterial taxa that accounted for the highest degree of variance between the two study arms were selected via least absolute shrinkage and selection operator (LASSO) followed by sparse partial least-squares discriminant analysis (PLS-DA).

Among the 97 HIV-seronegative SSA women included in this analysis, 44 were initiating NET-EN (Cape Town, South Africa) and 53 were initiating DMPA-IM (Cape Town [n=23] and Johannesburg [n=15], South Africa and Kisumu, Kenya [n=15]). In accordance with the design of the parent trials described earlier, women initiating DMPA-IM were older than women initiating NET-EN (median 24 [interquartile range (IQR) 22-29] vs. 17 [IQR 16-18] years, respectively; p<0.0001). Likely due to these age differences, more women initiating DMPA-IM had previously been pregnant (81.1% vs. 11.4%, p<0.0001), had less alcoholic drinks per week (median 0 [IQR 0-3] vs. 3 [IQR 0-5], p=0.011) and a higher HSV-2 seroprevalence (64.2% vs. 25.0%, p<0.0001), indicating previous infection with HSV-2. Possibly also age-related, prevalence of active C. trachomatis and N. gonorrhoeae infections tended to be higher in women initiating NET-EN (14/44; 31.8%) compared those initiating DMPA-IM (9/53; 17.0%; p=0.141), as did Nugent-BV prevalence (40.9% vs. 25.0%; p=0.085), while sexual risk behaviour was comparable between groups ( Table 1 ). All but one participant per group continued to use their assigned intervention through follow-up.

At baseline, cervicovaginal cytokine concentrations were largely similar between women initiating DMPA-IM or NET-EN. Possibly due to differences in lived experiences of participants from different sites, some cytokines differed by recruitment site, age group, STI or Nugent-BV status, using univariate analyses prior to adjustment for multiple comparisons ( Supplementary Table 2 ). However, IFN-γ was the only cytokine that remained significantly higher in women initiating DMPA-IM (120.14 [IQR 98.95 - 149.03] pg/ml) compared to those initiating NET-EN (98.78 [IQR 89.52 - 112.95] pg/ml; adjusted p=0.039) after adjusting for multiple comparisons ( Figure 1A ). We thus proceeded to only conduct paired longitudinal analyses that assessed the change in cervicovaginal cytokine concentrations from baseline to follow-up after two injections of the respective contraceptive.

The fold-changes in individual cervicovaginal cytokine concentrations from baseline to after two consecutive contraceptive injections were largely similar between women using DMPA-IM and those using NET-EN ( Figure 1B ). Further, the absolute differences in cytokine concentrations from baseline to follow-up did not differ between women using DMPA-IM and those using NET-EN ( Supplementary Figure 1B ). Albeit not significantly, cervicovaginal fluid from women using DMPA-IM had decreased levels of 20/27 cytokines compared to matched baseline samples, with CXCL10/IP-10 (1.8-fold), CCL2/MCP-1 (1.8-fold), IL-6 (1.6-fold) and GCSF (1.6-fold) being the most notable examples ( Figure 1B ). NET-EN use was also associated with a decrease in 21/27cytokines, including CXCL10/IP-10 (2.2-fold), IL-1β (1.8-fold), CXCL8/IL-8 (1.5-fold) and CCL3/MIP-1α (1.5-fold). Both contraceptives increased VEGF 1.1-fold ( Figure 1B ).

No clustering by contraceptive type, age category or site was observed when assessing change in log₁₀ cytokine concentrations from baseline to follow-up using unsupervised clustering ( Figure 1C ). When binarizing genital inflammation as absent/present (based on the upregulation of ≥5/8 cytokines to above their 75^th percentile) (8, 26), the proportion of women with genital inflammation decreased in both groups (by 8.5% in DMPA-IM and by 2.2% in NET-EN users) from baseline to follow-up, albeit non-significantly. Overall, these data indicate that neither DMPA-IM nor NET-EN use changed cervicovaginal cytokine profiles notably.

The influence of possible confounders, such as site, age, marital status, number of partners, number of sexual acts per week, and condom use on these findings were assessed in multivariate linear regressions. We found that these variables did not influence the observed associations between contraceptive type and change in cytokine concentrations. After adjusting for baseline HSV-2 serology, active bacterial STIs and candidiasis, DMPA-IM use resulted in a general modest decrease in absolute cytokine concentrations from baseline to follow-up compared to NET-EN use (reference category), albeit not significantly ( Figure 1D ).

Due to trial design differences, the age of women initiating DMPA-IM and NET-EN differed, and since participant age may influence responses to hormonal contraceptive initiation, we conducted a sensitivity analysis that included only participants ≤24 years old (DMPA-IM: n=27; NET-EN: n=43). In agreement with observations from the complete dataset, participants generally experienced a non-significant decrease in absolute cytokine concentrations from baseline to after two contraceptive injections and this did not differ between DMPA-IM and NET-EN users ( Supplementary Figure 2A ). Fold-decreases from baseline to follow-up were of similar magnitude in the same cytokines as in the full dataset ( Supplementary Figure 2B ). As in the complete data set, change in absolute cytokine concentrations from baseline to after two contraceptive of women aged ≤24 years using DMPA-IM differed slightly to that of women using NET-EN (reference) after adjusting for confounding factors, albeit not significantly ( Supplementary Figure 2C ).

To account for differences in geography, a sensitivity analysis including women from Cape Town, South Africa only (DMPA-IM: n=17; NET-EN: n=43) was carried out. Similar to findings from all sites, cytokine concentrations were not significantly elevated after two contraceptive injections in DMPA-IM or NET-EN users compared to baseline, and changes in absolute cytokine concentrations from baseline to follow-up did not differ significantly by contraceptive type ( Supplementary Figure 3A ). Similar to when including all sites, a fold-decrease in cytokine concentrations from baseline to follow-up was observed for most cytokines in DMPA-IM and NET-EN users ( Supplementary Figure 3B ), although DMPA-IM use was associated with a non-significant increase in CXCL8/IL-8 (1.4-fold), VEGF (1.4-fold), PDGFbb (1.3-fold), and IL12p70, CCL5/RANTES, IL-2, FGFbasic, GM-CSF and IL1RA (all 1.1-fold), while NET-EN initiation was associated with a 1.1-fold increase in VEGF only. However, after adjusting for confounding factors, changes in absolute cytokine concentrations from baseline to follow-up were comparable among DMPA-IM and NET-EN users ( Supplementary Figure 3C ) when restricting this analysis to isiXhosa women from Cape Town, South Africa.

Together, these data suggest that NET-EN and DMPA-IM use do not have an adverse impact on cytokine profiles in the FGT, also when accounting for differences in participants’ age and geography.

Women were assessed for both Nugent-BV and candidiasis at baseline and after two injections of their respective contraceptive. While baseline Nugent-BV rates tended to be lower among women who were subsequently assigned to DMPA-IM (25.0%) compared to those who were subsequently assigned to NET-EN (40.9%, p=0.085), Nugent-BV prevalence was more comparable between NET-EN (45.5%) and DMPA-IM (37.0%, p=0.177) users after two consecutive injections of the respective contraceptive. Candidiasis was generally less prevalent but increased 2-3-fold in both groups compared to baseline (1.9% at baseline vs. 6.7% at follow-up for DMPA-IM users, p=0.308; and 6.8% at baseline vs. 15.9% at follow-up for NET-EN users, p=0.179).

At baseline, participant age did not influence the genital microbiota distribution (Bray-Curtis distances); however, the overall microbial community composition differed significantly between the NET-EN and DMPA-IM arms ( Supplementary Figure 4A ; PERMANOVA p=0.004) and between sites within the DMPA-IM arm ( Supplementary Figure 4C ; PERMANOVA p=0.040). Similarly, within-participant α-diversity (assessed using Shannon indices) in the DMPA-IM arm differed significantly to the NET-EN arm ( Supplementary Figure 4B ; p=0.001). Due to these baseline differences in genital microbiota composition, only paired longitudinal changes were analysed further (DMPA-IM: n=48; NET-EN: n=39).

Based on these paired pre (baseline)-post (follow-up) analyses, we found no longitudinal change in α-diversity (within-participant Shannon diversity; Figure 2A ) or β-diversity (between-participant Bray-Curtis distances, Figure 2B ) from baseline to after two injections of either contraceptive, suggesting that there were no notable shifts in the vaginal microbiota composition with neither DMPA-IM nor NET-EN use. To confirm that participant age, and particularly site, did not influence our findings, we carried out sensitivity analyses only including participants aged ≤ 24 years or only including participants from Cape Town. We found no difference in α- ( Supplementary Figure 5A ) or β-diversity ( Supplementary Figure 5B ) in paired analyses from baseline to follow-up with the initiation of DMPA-IM or NET-EN among women aged ≤ 24 years (DMPA-IM: n=30; NET-EN: n=39) or among those from Cape Town only (DMPA-IM: n=20; NET-EN: n=39; Supplementary Figures 5C, D ) (all p>0.05).

To assess whether DMPA-IM use influenced community composition differently to NET-EN use, PAM clustering of Bray-Curtis distances was carried out to categorize women into one of four community state types (CSTs), based on bacterial species proportions ( Supplementary Figure 4D ). CST-I and -III were Lactobacillus-dominant CSTs, with CST-I being dominated by L. crispatus and CST-III by L. iners. CST-IVA had a higher diversity of BV-associated bacteria and CST-IVB had a higher abundance of L. iners and Gardnerella vaginalis. CST-I was associated with the least, and CST-IVB with the highest overall α-diversity ( Supplementary Figure 4E ). At baseline, there were no differences in the distribution of CSTs between women in the NET-EN or DMPA-IM groups (p=0.106; Figure 3 ). Overall, 35.4% of DMPA-IM and 38.5% of NET-EN users remained in the same CST category as at baseline after two injections ( Figures 3A, C ). About 15% of women moved from a more diverse to a lactobacilli-dominant state in both arms, while 25% (12/48) women moved from a lactobacilli-dominant CST to a more diverse CST in the DMPA-IM arm compared to 13% (5/39) in the NET-EN arm (p=0.154) ( Figures 3B, D ).

To assess whether specific bacterial taxa differed by contraceptive type, fold change differences in abundance of individual bacterial taxa were determined for both groups, adjusting for geographic site. DMPA-IM use was associated with a markedly larger number of bacterial shifts compared to NET-EN use ( Figure 4A ; Supplementary Table 3 ; all p ≤ 0.01) from baseline to follow-up. With NET-EN use, there was a statistically significant decrease in Mycoplasma hominis (log₂ 2.66-fold), Clostridiales (log₂ 1.76-fold), and Bacteroidia spp. (log₂ 1.52-fold) abundances at follow-up compared to matched baseline samples. In contrast, DMPA-IM use was accompanied by an increase in several BV-associated bacteria, including Sneathia spp. (log₂ 0.66-fold), Atopobium spp. (log₂ 0.80-fold), Fastidiosipila spp. (log₂1.19-fold), BVAB1/Lachnocurva vaginae (log₂1.45-fold), Prevotella amnii (log₂ 1.47-fold), Sneathia sanguinegens (log₂ 1.81-fold) and Sneathia amnii (log₂ 3.04-fold). On the other hand, there was a decrease in bacteria associated with both BV and genital health: Ureaplasma urealyticum (log₂ 1.57-fold), Peptoniphilus asaccharolyticus/grosssensis/harei (log₂ 1.29-fold), Staphylococcus spp. (log₂ 1.25-fold), Ezakiella spp. (log₂ 1.25-fold), Peptostreptococcus anaerobius (log₂ 0.90-fold), Prevotella bivia/denticola (log₂ 0.84-fold) and Lactobacillus spp. (log₂ 0.31-fold) abundances.

To confirm that these associations between contraceptive type and genital microbiota were not confounded by other variables, multivariate linear regressions comparing standardised changes in abundances of the most prevalent bacterial taxa in DMPA-IM users compared to NET-EN users (with the latter being considered the reference category), adjusting for active bacterial STIs, candidiasis, and HSV-2 seroprevalence were conducted. We found that DMPA-IM use was associated with a decrease in Lactobacillus spp. (non-L. iners/L. crispatus; β=-0.12; p=0.035) and an increase in Saccharimonadales spp. (β=0.02; p=0.02) relative to NET-EN use, although not after adjusting for multiple comparisons ( Figure 4B ; adjusted p=0.26 and p=0.30, respectively).

Together, these data suggest that NET-EN and DMPA-IM use do not have significant non-optimal effects on the vaginal microbiota, even when accounting for differences in participants’ age and geography.

Finally, we integrated cervicovaginal cytokine and microbiota data to identify genital biomarkers that were distinct between DMPA-IM and NET-EN users ( Figure 5 ). LASSO and sparse-PLS-DA were used to identify and classify specific cytokines and bacterial taxa that explained the most variance between NET-EN and DMPA-IM use. The identified biomarkers included eight cytokines (VEGF, IL-12(p70), IL-1RA, IL-13, IL-7, IL-6, CCL2/MCP-1 and PDGF-BB) and 12 bacterial taxa (M. hominis, Saccharimonadales, Parvimonas, Muribaculaceae, Peptostreptococcaceae, Clostridiales, Fusobacterium nucleatum, Eggerthella T1, Bacteroidia, Megaspheara, Prevotella timonensis, and Pyramidobacter) ( Figures 5B, D ). DMPA-IM use appeared to mostly be influenced by changes in the microbiota while cytokine changes were more influential among NET-EN users, although none were significant ( Figures 5B, D ), and there was no distinct separation between the two groups using sPLS-DA ( Figure 5C ). Based on these biomarkers, we found no clustering between DMPA-IM and NET-EN users, indicating that neither contraceptive had a marked influence on vaginal cytokines or bacteria ( Figure 5A ).