Genital microbiota of women using a 90 day tenofovir or tenofovir and levonorgestrel intravaginal ring in a placebo controlled randomized safety trial in Kenya

In a phase-IIa trial, we investigated the influence of 90 days continuous-delivery tenofovir (TFV) intravaginal rings (IVRs) with/without levonorgestrel (LNG) on the genital microbiota of Kenyan women. Eligible women (n = 27; 18–34 years; negative for HIV, sexually transmitted infections, and Amsel-bacterial vaginosis) were randomized 2:2:1 to use of IVRs containing TFV, TFV/LNG, or placebo. Using vaginal wall and IVR swabs at IVR insertion and removal, the genital microbial composition was determined using 16S rRNA gene sequencing. The presence of Candida spp. was determined using qPCR. The vaginal total bacterial burden appeared to decrease with TFV and TFV/LNG IVR use (log100.57 and log100.27 decrease respectively; p > 0.05). The TFV/LNG IVR was more ‘stabilizing’: 50% of the participants’ microbiota community state types remained unchanged and 50% shifted towards higher Lactobacillus abundance. Specifically, TFV/LNG IVR use was accompanied by increased abundances of Lactobacillus gasseri/hominis/johnsonii/taiwanensis (16.3-fold) and L. fermentum/reuteri/vaginalis (7.0-fold; all p < 0.01). A significant shift in the overall microbial α-diversity or β-diversity was not observed for either IVR, and IVR use did not influence Candida spp. prevalence. TFV/LNG and TFV IVRs did not adversely affect the genital microbiota and are safe to use. Our findings support further studies assessing their efficacy in preventing HIV/HSV-2 and unintended pregnancies.


In a phase-IIa trial, we investigated the influence of 90 days continuous-delivery tenofovir (TFV) intravaginal rings (IVRs) with/without levonorgestrel (LNG) on the genital microbiota of Kenyan women. Eligible women (n = 27; 18-34 years; negative for HIV, sexually transmitted infections, and
Amsel-bacterial vaginosis) were randomized 2:2:1 to use of IVRs containing TFV, TFV/LNG, or placebo. Using vaginal wall and IVR swabs at IVR insertion and removal, the genital microbial composition was determined using 16S rRNA gene sequencing. The presence of Candida spp. was determined using qPCR. The vaginal total bacterial burden appeared to decrease with TFV and TFV/LNG IVR use (log 10 0.57 and log 10 0.27 decrease respectively; p > 0.05). The TFV/LNG IVR was more 'stabilizing': 50% of the participants' microbiota community state types remained unchanged and 50% shifted towards higher Lactobacillus abundance. Specifically, TFV/LNG IVR use was accompanied by increased abundances of Lactobacillus gasseri/hominis/johnsonii/taiwanensis (16.3-fold) and L. fermentum/ reuteri/vaginalis (7.0-fold; all p < 0.01). A significant shift in the overall microbial α-diversity or β-diversity was not observed for either IVR, and IVR use did not influence Candida spp. prevalence. TFV/LNG and TFV IVRs did not adversely affect the genital microbiota and are safe to use. Our findings support further studies assessing their efficacy in preventing HIV/HSV-2 and unintended pregnancies.
With global estimates of 1.7 million incident HIV infections 1 and 121 million unintended pregnancies annually 2 , development of a multipurpose, user-controlled product that prevents heterosexual transmission of HIV and provides contraception for women is invaluable. Daily oral emtricitabine (FTC)/tenofovir disoproxil fumarate (TDF) has revolutionized HIV prevention as the first pre-exposure prophylaxis (PrEP) option. However, a high level of adherence with the dosing regimen is essential for protection and is often difficult to achieve 3,4 . Longacting contraception options not requiring daily adherence have been pivotal in the prevention of unintended pregnancies. Innovative multipurpose technologies (MPT) could be developed by leveraging current long-acting contraceptive strategies with HIV PrEP co-delivery or co-formulation to improve protective coverage.

Results
Participant characteristics. Twenty-six of the twenty-seven randomized women had bacterial absolute abundance data available: TFV/LNG IVR (n = 11), TFV IVR (n = 10), and placebo IVR (n = 5). The median age was 22 years [interquartile range (IQR) [21][22][23][24][25][26], median body mass index was 22.4 , and demographic characteristics were similar across study arms (Table 1). Most women used male condoms during the study (74.1%), 7.4% previously used a copper-intrauterine device (≥ 6 months before enrolling in the study), and 18.5% used no contraceptive method. Despite being asymptomatic and BV-negative by Amsel criteria at screening, 4/27 (14.8%) women were Nugent-BV-positive at the time of IVR insertion and 11/27 (41.0%) at the IVR removal visit. With the exception of an estimated increase in the prevalence of BV-positive Nugent score, differences in clinicial characteristics at IVR removal were small. Total bacterial load did not change with IVR use. The total FGT 16S bacterial load did not change between the baseline and IVR removal visits for women in any arm ( Fig. 1A; TFV/LNG: log 10 8.29 vs log 10 8.14, p = 0.56; TFV: log 10 7.84 vs log 10 7.94, p = 0.36; placebo: log 10 8.14 vs log 10 8.45, p = 0.25) and the difference in the degree of change across arms relative to the placebo arm was small (Fig. 1B). At IVR removal, no participantlevel differences were observed in any arms between the lateral vaginal wall bacterial load and that of the IVR surface ( Fig. S1; TFV/LNG: log 10 8.14 vs log 10 8.56, p = 0.11; TFV: log 10 7.94 vs log 10  www.nature.com/scientificreports/ log 10 8.45 vs log 10 8.18, p = 0.25). Compared to the placebo arm, a higher estimated bacterial load was found on the TFV/LNG IVR surface (median difference of log 10 0.33, p = 0.05) than on the TFV IVR (median difference of log 10 0.04, p = 0.20 vs placebo median difference of log 10 − 0.34, Fig. 1C) relative to the vaginal wall bacterial load. Overall, genital bacterial load was positively correlated with the total abundance of the Lactobacillus genus, with a log 10 0.95 increase in total bacterial load with every log 10 increase in Lactobacillus abundance (Linear regression p = 0.03 after adjusting for multiple comparisons).  www.nature.com/scientificreports/ imprecise since only 4 women were in this arm. Similarly, the correlation between IVR surface bacterial load and length of time of IVR use was low for either arm (Fig. S2).

Change in bacterial load
The overall microbiota diversity did not change with IVR use. To measure the effect of IVR use on the genital microbial composition of women, bacterial diversity was assessed using the Shannon diversity index and Bray-Curtis distances. We did not find a large change in overall microbiota diversity between baseline and IVR removal with use of any IVR (

TFV/LNG IVR use was associated with a shift towards a less diverse CST. Based on hierarchical
clustering of their dominant vaginal microbiota taxa, women were categorized as having community state type (CST) I if they had a genital microbiota dominated by L. crispatus, or CST III if their microbiota was dominated by L. iners. CST IVA had a higher abundance of the BV-associated bacterium G. vaginalis and CST IVB was composed of a broader range of BV-associated bacteria (Fig. S4). We found some discordance between Nugent-BV status and CST IV categorization. At baseline, of the 11/26 (42.3%) of women with a microbiota typically associated with microbial dysbiosis (CST IV A/B), 3/11 were Nugent-BV-positive, 3/11 had an intermediate Nugent score and 5/11 were Nugent-BV-negative. Further one participant with Nugent-BV was categorized as having a CST III microbiota. During follow up, 50.0% (5/10) of women using the TFV/LNG IVR transitioned to a CST less associated with BV (2/5 transitioned from CST IVB to CST IVA) or a higher abundance of Lactobacillus species (1/5 transitioned from CST IVB to CST III, 1/5 from CST IVA to CST I, and 1/5 from CST III to CST I) while there were no CST shifts in the other 50.0% of women (Fig. 4A, Table S1). Among women using the TFV IVR, 22.2% (2/9) of women did not experience CST shifts, while the majority transitioned to a more diverse (44.4%; 4/9) or a less diverse CST (33.0%; 3/9). With the placebo IVR, 75.0% of women (3/4) did not transition to a different CST and one participant transitioned to a more diverse CST (CST IVA to CST IVB).
Overall, 41.6% (10/24) women had differences in assigned CST between the vaginal wall and the IVR surface at IVR removal (Fig. 4B), with 6/10 having a more diverse and 4/10 having a less diverse IVR CST, although none of these broad transitions were statistically significant (TFV/LNG p = 0.58; TFV p = 0.57; placebo p = 0.85).
Neither TFV/LNG nor TFV IVR use was associated with major increases in pathogenic bacterial taxa. Longitudinal fold-changes in relative abundances of each individual bacterial taxa measured was determined for each study arm, accounting for individual variations in time between IVR insertion and removal. Using a threshold cutoff of log 2 0.05, the placebo IVR was associated with the most fluctuations, with a significant fold-change in 23 bacterial taxa (Fig. 5, Table S2): including decreases in Finegoldia spp. and Dialister spp. (~ log 2 twofold), and increases in numerous taxa such as a > log 2 twofold increase in Proteobacteria phylum, Atopobium vaginae, Corynebacterium coyleae/mucifaciens, Corynebacterium genitalium, Prevotella buccalis, and Porphyromonadaceae as well as a log 2 9.17-fold (standard error (SE) = 2.5) increase in L. crispatus/acidophilus. Women randomized to the TFV/LNG and TFV IVRs appeared to experience a more 'stabilizing' effect, with less change in vaginal bacterial taxa. Women using the TFV IVR were observed to have an average 5.42-fold (SE = 1.63) increase in Dialister micraerophilus abundance, a BV-associated bacterium. Women using the TFV/ LNG IVR were observed to have a 3.79-fold (SE = 1.55) decrease in the pathogenic bacterium Streptococcus Candidiasis was similar across IVR types. The prevalence of Candida spp. were also measured at both visits. No C. parapsilosis was detected in any sample. There was an overall 30.8% prevalence of Candida at baseline and 44% at the IVR removal visit (p = 0.54; Table 2). C. krusei (23%) and C. albicans (17%) were more common ( Table 2, Fig. 6). The distribution of each Candida spp. did not differ across study arms at baseline (C. albicans p > 0.99; C. glabrata p = 0.57; C. krusei p > 0.99) or at IVR removal (C. albicans p > 0.99; C. krusei p = 0.64). C. glabrata was not detected at the IVR removal visit. There were no differences in the overall microbial diversity (PERMANOVA p = 0.07) or differences in abundances of individual bacterial taxa between women who remained Candida-negative and those who acquired a Candida infection by the IVR removal visit.
Higher microbial diversity was associated with lower TFV levels. At the IVR removal visit, we found a higher estimated concentration (84.6-fold) of vaginal TFV in women with lower microbial diversity (median 53550 ng/swab with CST I vs 632.5 ng/swab with CST IVB; p = 0.16; Fig. 7), although sample sizes were www.nature.com/scientificreports/ small. In addition, after adjusting for the number of days of IVR use, a log 10 increase in the absolute abundance of Lactobacillus genus was associated with a log 10 0.41 ng/swab increase in TFV concentrations (p > 0.99).

Discussion
Overall, we found no evidence that either the TFV/LNG or TFV IVR significantly affected a woman's genital bacterial or candidal composition. Our results are consistent with previous findings, including a recent safety study of the same IVR among women in the US and Dominican Republic 16 .
We mostly found small longitudinal changes in total bacterial load or in the overall microbial diversities in all study arms, suggesting that these IVRs do not induce major shifts in the genital microbiota. At the IVR removal visit, CSTs between the lateral vaginal wall and the IVR did not differ, similar to previous findings 16 . Despite no CST changes in most women using the placebo IVR, its use led to fluctuations in multiple bacterial taxa, with statistically significant fold changes in both Lactobacillus spp. and BV-associated bacteria. The TFV IVR was only associated with an increase in D. micraerophilus abundance, a BV-associated bacterium strongly associated with microbial diversity and high-inflammation states [17][18][19] . However, we did not see a shift to a more BV-associated microbial composition, likely because the increase in D. micraerophilus was insufficient to drive this change. Interestingly, there was a statistically significant decrease in S. anginosus/milleri/sanguinis abundance with the TFV/LNG IVR and a markedly larger fold-increase in Lactobacillus abundance, suggesting a positive effect of the TFV/LNG combination on microbial composition. Previous studies of vaginal rings found no significant destabilization of the genital microbiota, and an increase in Lactobacillus spp. in some cases 16,20,21 , which was attributed to ethinyl estradiol [22][23][24] . Here we found a similar stabilizing effect, with a decrease in microbial diversity with the TFV/LNG IVR despite containing LNG, a progestin, indicating that the decrease in microbial diversity www.nature.com/scientificreports/ and increase in Lactobacillus spp. could be associated with progestin use, particularly LNG, in line with a previous study reporting an increase in genital Lactobacillus spp. concentration with LNG implant use 25 .
Although not statistically significantly so, we found that genital TFV concentrations at the IVR removal visit were positively associated with Lactobacillus spp. abundances. This is similar to other studies showing that TFVbased HIV PrEP products were less effective in Lactobacillus-deficient states 14 . There was a decreasing trend in microbial diversity with increased TFV concentration, with a 84.6-fold higher TFV concentration found with an L. crispatus-dominant microbiota (CST I) compared to a more diverse BV-associated microbiota (CST IVB). Based on our differential abundance analyses, this was not driven by G. vaginalis or Prevotella spp as previously described 14,26 . More work is needed to determine if this decrease is clinically relevant to impact the IVR capacity to prevent HIV or HSV-2. In particular, it would be important to assess the relationship between CSTs and concentrations of the TFV active metabolite, TFV-diphosphate, in mucosal tissues.
A primary limitation was the small sample size, due to the study being a phase-IIa safety trial. Thus, we presented substantial descriptive data that examine actual numbers and trends, rather than just statistical comparisons. More in-depth comparisons, in larger cohorts, that also include women with BV at IVR insertion, might better help identify potential genital microbial changes induced by these IVRs.
In this study measuring the safety of TFV/LNG and TFV polyurethane IVRs on the genital microbiota among women living in Kenya, we found no evidence of adverse changes to genital microbial health with up to 90 days of use. We found a trend towards the establishment of high Lactobacillus states with the TFV/LNG IVR, which could positively impact genital health in a population with a high BV prevalence [27][28][29] . These results indicate that  www.nature.com/scientificreports/

Methods
Participant recruitment. CONRAD B17-144 was a phase-IIa randomised, placebo-controlled, investigational new-drug-enabling trial assessing the safety, pharmacokinetics, pharmacodynamics, tolerability, acceptability of, and adherence to two 90 days IVRs. Women were randomised in a 2:2:1 ratio to receive a TFV-only IVR 30 to prevent HIV/HSV, a combination TFV/LNG IVR that additionally may prevent pregnancy 30-32 and a placebo IVR 31 . All rings were composed of polyurethane tubing 31 . This study recruited women at the Jaramogi Oginga Odinga Referral Hospital, Kisumu, Kenya, who were generally healthy, non-pregnant, at lower risk for HIV (based on a validated risk score 33 ), seronegative for HIV and hepatitis B surface antigen, and bacterial vaginosis (BV)-negative by Amsel criteria. Women diagnosed with BV by Amsel criteria during the screening procedures were treated and permitted to be re-assessed for eligibility > 2 months after treatment, and enrolled if BV was not detected at that time. Eligible women were not using hormonal contraception at enrolment and had not been diagnosed or treated for STIs in the last three months (Supplementary Methods). The study was approved by the Scientific and Ethics Review Unit at the Kenya Medical Research Institute and the Human Subjects Division at University of Washington, and was registered with ClinicalTrials.gov (Identifier: NCT03762382; registered 03/12/2018). All participants provided written informed consent and all methods were carried out in accordance with relevant guidelines and regulations. Study procedures. Results from the parent study will be published in a separate manuscript (Mugo et al. 32 ,in preparation). Paired data from the baseline/IVR insertion and IVR removal visits were included for this analysis, focused on microbiota changes. IVR removal was scheduled for 90 days after IVR insertion or prior to the IVR expiry date, whichever date came first (At 90 days: 12/27; before expiry date: 9/27; other reasons including pregnancy or STIs: 6/27). In cases where matched clinical data from the IVR removal visit were not available, samples collected during the next visit, which was scheduled to occur 24 h later, were used for analyses.

Sample collection at IVR insertion and removal.
After randomization, the IVR was inserted deeply into the vagina by a study clinician at the baseline visit and removed at the IVR removal visit. At both baseline and the IVR removal visits, vaginal swabs were collected for (1) TFV PK measurements, (2) Nugent score and vaginal pH and (3) microbiota 16S rRNA sequencing. A swab of the IVR surface was also collected at the IVR removal visit. Nugent-BV was defined as having a Nugent score ≥ 7; women with a Nugent score 4-6 were cat- Microbial DNA extraction. DNA extraction from lateral vaginal wall swabs was carried out (Qiagen All-Prep PowerViral ® DNA/RNA kit; Supplementary Methods) and was stored at − 20 °C until used for 16S rRNA sequencing, total 16S bacterial load using real-time polymerase chain reaction (qPCR), and Candida spp. quantitation (qPCR).
16S rRNA gene sequencing and analysis. The V3-V4 hypervariable region of the bacterial 16S rRNA was amplified using modified universal primers 35 . Samples were sequenced using the Illumina MiSeq platform and (300 bp paired-end). DADA2 v1.12.1 36 was used to process, merge and filter raw reads and samples with < 2000 reads were excluded from further analyses. Taxonomic annotation was carried out using the RDP database (v11.5) and a BLAST search was carried out for further taxonomic classification of unannotated Comparing the difference in log 10 vaginal TFV concentrations at the IVR removal visit among women using the TFV and tenofovir/levonorgestrel arms categorized by their CST. The p value was determined using the Mann-Whitney U test. www.nature.com/scientificreports/ amplicon sequence variants (ASV). Taxonomic annotation was available for 993 ASVs, with 277 (27.9%) having species-level annotation. The ASVs were merged at the lowest available taxonomic level to generate relative abundances for a total of 105 bacterial taxa 17 (Supplementary Methods). The total bacterial load per swab was determined using a qPCR of the 16S rRNA gene 37 . The BactQuant assay targets the V3-V4 region, and gives an estimate of the total 16S rRNA copies per swab. Based on these values and the relative abundances of taxa generated by 16S rRNA sequencing, absolute bacterial abundances (copies/ swab) were calculated for each taxon. One sample (out of five), in the placebo arm, without sufficient 16S DNA to pass quality control checks was excluded from analysis (final n = 4). All further analyses were carried out using absolute abundances for bacterial taxa.
Candida spp. quantification. Concentrations of C. albicans, C. glabrata, C. krusei and C. parapsilosis, were measured using species-specific qPCRs at the baseline and IVR removal visits (Supplementary Methods). The final concentrations were reported as number of log 10 copies per swab.
Statistical analysis. All microbiota data analysis was carried out in R (v3.6.0). Paired comparisons between the ring insertion and ring removal visits were considered primary endpoints of this analysis. Participant microbiota were categorised into four CSTs using Ward clustering of weighted Unifrac dissimilarity matrices derived from bacterial absolute abundances. TFV and TFV/LNG study arms were compared to the placebo arm using Mann-Whitney U-tests for independent samples and permutational multivariate analysis of variance (PER-MANOVA) for microbial diversity. Paired analyses comparing microbiota at IVR insertion and removal visits were carried out using Wilcoxon Signed Rank tests and generalized estimating equation models. Correlations between bacterial absolute abundance and variables such as time of IVR use and genital TFV concentrations were estimated using linear regression models. The number of days of IVR use was categorized as ≤ 47 days (< 2 menstrual cycles), 48-71 days (~ 2 menstrual cycles) and 72-95 days (~ 3 menstrual cycles), and was adjusted for when measuring fold changes in individual bacterial taxa within study arms, using generalized linear models as part of the DeSeq2 package. Log 2 fold-changes were determined only for bacterial taxa found in at least 15% of samples to minimize the influence of minority taxa. Due to the small sample size, the study was not powered to estimate differences between the study arms with a high degree of precision and analyses were descriptive in nature. As a result, p-values and confidence intervals (CI) were not adjusted for multiple comparisons unless explicitly stated, in which case a false discovery rate step down procedure was used 38 (Supplementary Methods).

Data availability
The dataset generated and analyzed in this study has been deposited at NCBI SRA under Accession ID PRJNA834833 (https:// www. ncbi. nlm. nih. gov/ biopr oject/ PRJNA 834833). Linked data are available from the corresponding author on reasonable request. www.nature.com/scientificreports/