Factors associated with incidence and spontaneous clearance of molecular-bacterial vaginosis: Results from a longitudinal frequent-sampling observational study

Abstract

Background.

We sought to assess time-independent and time-varying factors associated with incidence and spontaneous clearance of molecular bacterial vaginosis (without treatment).

Methods.

Mid-vaginal samples were self-collected daily by 100 participants recruited at the University of Alabama Birmingham over 10 weeks (4,778 samples). Vaginal microbiota was characterized by 16S rRNA gene amplicon sequencing and clustered into community state types (CSTs). A low-Lactobacillus CST IV defined the molecular-BV outcome in this study. Factors associated with molecular-BV incidence and spontaneous clearance were modeled using Andersen-Gill recurrent event Cox models. Community class identified the predominant CST of a participant during follow-up.

Results.

Menstruations (aHR 2.09 [95% CI: 1.51–2.89] in the prior 24 hours) and CST III (Lactobacillus iners-dominated) at the previous sample (aHR 2.25 [1.48–3.40]) were associated with increased molecular-BV incidence. Participants with a majority of L. iners-dominated samples longitudinally (community class LI) displayed less stable patterns of vaginal microbiota. In LI participants, reduced molecular-BV spontaneous clearance was observed in African-American participants (aHR 0.44 [0.26–0.75]) compared to White participants, older participants [age 40–49 (aHR 0.38 [0.23–0.61]), age 30–39 (aHR 0.48 [0.28–0.83]) compared to participants aged 18–29, and after douching (0.45 [0.28–0.73] within prior 72 hours).

Conclusions.

While it is now well-documented that vaginal microbiota are dynamic, there is little available data on factors associated with spontaneous clearance of molecular-BV. L. iners-dominated vaginal microbiota are more likely to be dynamic and associated with different risk factors for incidence and clearance of BV. Among L. iners-dominated participants, age, race and douching were linked to reduced clearance. Most transitions to molecular-BV during menstruations were short-lived.

Short summary

Vaginal microbiota fluctuate between community states though most transitions are short-lived. L. iners-dominated microbiota are less resilient and associated with specific factors linked to reduced molecular-BV clearance.

INTRODUCTION

Bacterial vaginosis (BV) affects nearly 30% of reproductive-age North-American women (1) and is characterized by a vaginal microbiota with low abundance of Lactobacillus spp. and higher abundances of anaerobic bacteria (2). BV is associated with vaginal symptoms (3) and a significant 2–4 fold increased risk of sexually transmitted infections (4–6), including HIV (7, 8). Community state types (CST) can be defined based on the clustering of vaginal bacterial community compositions determined by 16S rRNA gene amplicon sequencing (9). CST IV is characterized by low relative abundance of Lactobacillus spp and a wide array of strict and facultative anaerobes associated with BV, and was termed molecular-BV by McKinnon et al. (10). Recently, Lactobacillus iners-dominated communities have gained attention as being a risk factor for STIs and other adverse outcomes (11), hypothesized to be due in part to the inability of L. iners to produce D-lactic acid (12).

Previous longitudinal studies focusing on daily or frequent sampling with Gram stain assessment (termed Nugent-BV (10)) have described several patterns of vaginal microbiota dynamics. In 1999, Schwebke et al. (13) demonstrated that there are stable patterns of normal microbiota, intermediate microbiota or BV, or short transitions in and out of BV, and these findings have been observed in other studies (14–18). These fluctuation patterns have been documented in BV treatment studies as well, where molecular characterization of the vaginal microbiota demonstrated increases of BV-associated bacteria during menstruations, while women with BV could face recurrent BV quickly after antibiotic treatment (19, 20). Such data expanded and challenged research practice and pointed to a need for vaginal microbiota to be more frequently assessed to gain a better understanding of its natural history and pathogenesis.

Studies using mostly Nugent-BV assessment and frequent sampling designs have shown that fluctuations in the vaginal microbiota are associated with menstruations, spermicide and lack of condom use, while vaginal intercourse and receptive oral sex have been suspected but with inconsistent results (13–18). Other longitudinal studies with infrequent sampling have demonstrated that sex with a new partner (21), digital sex (22) and vaginal douching (23, 24) were associated with incident BV, while condom use was associated with lower BV risk (25, 26). However, factors associated with spontaneous clearance of BV (without antibiotic treatment) have been seldom studied.

We aimed to assess the association between both time-fixed and daily time-varying factors and incidence and spontaneous clearance of molecular-BV defined by 16S rRNA gene amplicon sequencing (10) in a cohort of 100 participants who collected samples daily for 10 weeks.

METHODS

This is a secondary data analysis utilizing data from an observational cohort study in which factors associated with incidence and spontaneous clearance of molecular-BV are assessed.

Human Microbiome Project – University of Maryland (HMP-UMB)

The HMP-UMB study recruited 135 non-pregnant reproductive-age cisgender women (between 18 and 49 years old) at the University of Alabama at Birmingham between September 2009 and July 2010 to a prospective observational study that has been previously described (27). In brief, participants self-collected mid-vaginal swabs daily for 10 weeks. After exclusion of participants with fewer than five samples with 16S rRNA gene sequence data over the course of the study, this analysis included 100 participants with 4,778 vaginal samples. A clinician performed a pelvic examination and clinical evaluation at baseline, week 5 and week 10, or at an interim visit if a participant returned for a visit due to vulvovaginal symptoms. A questionnaire was administered at all clinical visits and participants also reported time-varying behaviors and menstrual bleeding on daily diaries, which were submitted weekly along with vaginal samples. During the clinical visits, BV was diagnosed according to Amsel’s criteria (at least three of the following four findings): (i) vaginal pH> 4.5; (ii) homogenous white/grey vaginal discharge; (iii) the presence of clue cells, and (iv) a positive whiff test (fishy odor after addition of 10% potassium hydroxide). When symptomatic BV was diagnosed, the participant was prescribed antibiotic treatment following standard practice (metronidazole or clindamycin) (28). Antibiotic use for BV (16 events) from the moment it was taken until the end of the study was controlled for in all analyses as a time-varying covariate. In addition to antibiotic treatment for BV, many participants also reported antibiotic use for other indications over the course the study. This additional antibiotic consumption was not taken into account in the present study (antibiotic use in the last 30 days was reported 32 times throughout the study). The study protocol was approved by the IRBs of the University of Alabama Birmingham and the University of Maryland Baltimore. All participants provided written informed consent.

Vaginal microbiota characterization

DNA was extracted from a Copan ESwab placed in Amies liquid transport, using enzymatic and physical lysis of bacterial cells followed by purification of genomic DNA using a QIAsymphony robotic platform and QIAGEN CellFree 500 kits (QIAGEN, Valencia, CA, USA) according to the manufacturer’s instructions (27). The V3-V4 regions of the 16S rRNA gene were amplified and sequenced on an Illumina HiSeq 2500 instrument (29). Bioinformatic sequence processing and taxonomic assignments also followed the procedures of Holm et al. (29), and resulted in a bacterial phylotypes relative abundance table. Community state types (CSTs) (Figure S1) were assigned using VALENCIA, a nearest centroid-based classification algorithm (git-hub.com/ravel-lab/valencia) (30). VALENCIA identified seven CSTs, four of which were dominated by the indicated Lactobacillus species: CST I: Lactobacillus crispatus, CST II: L. gasseri, CST III: L. iners, and CST V: L. jensenii. CST IV was characterized by low levels of Lactobacillus spp. and was subdivided into three sub-CSTs based on the most abundant organisms detected: CST IV-A: BVAB1 and Gardnerella vaginalis, CST IV-B: G. vaginalis and Atopobium vaginae, and CST IV-C: Streptococcus sp. and Corynebacterium (10). For some analyses, CSTs were also grouped into a binary outcome: Lactobacillus-dominated (CST I, II, III, V) versus molecular-BV (CST IV).

Community classes

The community class variable reflects a profile of longitudinal sampling and was defined by hierarchical clustering of a participant’s CST profile using Euclidian distance and Ward linkage, as described previously (31s). The community classes that we observed were LC (CST I-dominated, mostly L. crispatus), LG (CST II-dominated, mostly L. gasseri), LI (CST III-dominated, mostly L. iners), DA (CST IV-A-dominated, diverse), DB (CST IV-B-dominated, diverse) and DC (CST IV-C-dominated, diverse) (Figure 1).

Figure 1.

Community state types (CST) dynamics per participant over the course of the study, in the HMP-UMB study on 100 participants in Birmingham, AL.

Each of the 4,778 samples of the study is represented by a rectangle and colored according to its CST. Each row represents one participants. Participants were grouped into community class according to their main CST over the course of the study, determined through hierarchical clustering on the proportions of each CST. LC (L. crispatus): mostly CST I; LG (L. gasseri): mostly CST II; LI (L. iners): mostly CST III; DA: mostly CST IV-A; DB: mostly CST IV-B, DC: mostly CST IV-C.

Statistical analyses

Comparison of baseline characteristics across community classes was performed using Fisher’s exact test. Incidence of molecular BV was defined as at least one sample categorized as molecular-BV (CST IV) after a non-molecular-BV sample (any other CST). Clearance of molecular BV was defined as at least one non-molecular-BV sample after a molecular-BV sample, i.e., a spontaneous transition from molecular-BV to Lactobacillus-dominated state. The effects of time-independent and time-dependent covariates on the rate of transition from Lactobacillus-dominated state to molecular-BV (incident molecular-BV versus non-incidence) and molecular-BV to a Lactobacillus-dominated state (molecular-BV clearance versus non-clearance) were assessed with Andersen-Gill recurrent event Cox models, using the “survival” package (version 3.2–11) in R (version 4.1.0). The models included assessment for time-varying factors such as douching and sexual exposure variables in different time windows (within prior 24 hours, within prior 48 hours, within prior 72 hours of the outcome) and CST at the previous sample (approximately 1.5 days prior), and time-independent factors such as age, race and contraception. The effect of condom use was evaluated by comparing condom use versus no condom use during vaginal intercourse. The effect of condomless vaginal intercourse was evaluated by comparing vaginal intercourse without condom use versus no vaginal intercourse. For each exposure assessed, multivariate models were computed using relevant confounders. Transition intensities (rates), daily probabilities of transition and sojourn length were computed from multi-state models with the “msm” package (version 1.6.9) in R.

RESULTS

Participants contributed a median of 50 (IQR 35–65) mid vaginal samples. Individuals with a low-Lactobacillus CST IV as their main CST over the course of the study (community classes DA/DB/DC) were more likely to be African-American (p=0.011) and report history of vaginal douching (p=0.060), and were less likely to use hormonal contraception (p=0.060) than those in other community classes (Table 1).

TABLE 1.

Baseline characteristics of participants in L. crispatus-dominated and L. gasseri-dominated community classes (LC/LG), L. iners-dominated community class (LI) and diverse community classes (DA/DB/DC) in the HMP-UMB study in Birmingham, AL (N=100).

	Total number of observations		Participants in LC/LG community class* (n=33)		Participants in LI community class* (n=23)		Participants in DA/DB/DC community class* (n=44)		p-value**
Age	100								0.163
40+ years	9	9%	4	12%	2	9%	3	7%
30–39 years	37	37%	17	52%	7	30%	13	30%
18–29 years	54	54%	12	36%	14	61%	28	64%
Race	100								0.011
African American	61	61%	13	39%	14	61%	34	77%
Hispanic or Other	6	6%	3	9%	1	4%	2	5%
White	33	33%	17	52%	8	35%	8	18%
Contraception	86								0.060
Hormonal	17	20%	7	26%	7	33%	3	8%
IUD	5	6%	1	4%	0	0%	4	11%
Non-Hormonal	64	74%	19	70%	14	67%	31	82%
Lifetime number of partners	100								0.531
>7	41	41%	15	45%	7	30%	19	43%
1 to 6	59	59%	18	55%	16	70%	25	57%
High school education	100								0.179
graduate	43	43%	18	55%	6	26%	19	43%
more than high school	54	54%	15	45%	16	70%	23	52%
less than highschool	3	3%	0	0%	1	4%	2	5%
Marital status	100								0.488
not married	54	54%	15	45%	12	70%	27	61%
separated	20	20%	6	18%	6	26%	8	18%
married	26	26%	12	36%	5	22%	9	20%
Smoking in last 2 months	100								0.730
yes	16	16%	5	15%	5	22%	6	75%
no	84	84%	28	85%	18	78%	38	86%
Alcohol consumption in last 2 months	100								0.402
yes	77	77%	28	85%	17	74%	32	73%
no	23	23%	5	15%	6	26%	12	27%
Pregnancy in lifetime	100								0.256
yes	67	67%	19	58%	15	65%	33	75%
no	33	33%	14	42%	8	35%	11	25%
Lubricant use in last 2 months	91								0.281
yes	22	24%	5	19%	8	36%	9	22%
no	69	76%	23	85%	14	64%	32	78%
Condom use in last 2 months	86								0.643
often/always	38	44%	9	38%	10	48%	19	48%
rarely/never	48	56%	16	67%	11	52%	21	53%
Douching	100								0.060
ever douched yes	51	51%	12	36%	11	48%	28	64%
ever douched no	49	49%	21	64%	12	52%	16	36%
last 2 months yes	12	12%	3	9%	2	9%	7	16%	0.674
last 2 months no	88	88%	30	91%	21	91%	37	84%

^*Participants were grouped into community class according to their main CST over the course of the study, determined through hierarchical clustering on CSTs. LC: mostly CST I; LG: mostly CST II; LI: mostly CST III; DA: mostly CST IV-A; DB: mostly CST IV-B, DC: mostly CST IV-C.

^**p-value was calculated using Fisher’s exact test.

The temporal dynamics of CST by community class are presented in Figure 1. Eight participants (8%) remained in the same CST for the length of the study, three from the LC community class (L. crispatus-dominated CST I), one from the LG community class (L. gasseri-dominated CST II), one from the DB community class (CST IV-B) and three from the DC community class (CST IV-C). Overall, there were 292 molecular-BV incidence events (for 2,214 non-incidence events) and 295 molecular-BV clearance events (1,877 non-clearance events) in the study.

Factors associated with incident molecular-BV, whole cohort

Menstruations within prior 24 hours was associated with incidence of molecular-BV (aHR 2.09 [1.51–2.89]), as well as menstruations in the prior 48 hours and in the prior 96 hours (Table S2 and Figure 2). Other factors, including time-varying intravaginal practices (douching, lubricant use), type of undergarment, and sexual behaviors (condom use, anal sex, digital, receptive oral sex), and time-fixed factors such as race, age and hormonal contraception were not associated with incident molecular-BV. In a multivariate model using race, contraception and antibiotic use for BV as confounders, an age between 30 and 39 was associated with a reduced rate of transition to molecular-BV compared to age 18–29 (Table S3). Participants with a CST III at the previous sample were more likely to transition to molecular-BV compared to participants with a CST I (aHR 2.25 [1.48–3.40]. When controlling for confounders such as contraception, menstruations in prior 72 hours, douching in prior 72 hours and antibiotic use for BV, participants with a CST III at the previous sample remained more likely to transition to molecular-BV compared to CST I (Table S3).

Figure 2.

Time-varying and time-independent factors associated with incidence and clearance of molecular-BV (CST IV) in bivariate analyses, in 100 participants from the HMP-UMB study in Birmingham, AL.

Analyses were adjusted for antibiotic treatment for BV in participants from the moment they took the antibiotic until the end of the study.

CST: Community State Type; BV: Bacterial Vaginosis; ATB: Antibiotics for BV.

*Reference categories for age, race, contraception and previous CST were “18–29 years old”, “White”, “Non-hormonal”, “CST I” (for incidence) and CST IV-A (for clearance).

Factors associated with molecular-BV spontaneous clearance, whole cohort

Concerning transitions from molecular-BV to a Lactobacillus-dominated state (spontaneous clearance of molecular-BV), only thong undergarment use in the prior 24 hours, in the prior 48 hours and in the prior 72 hours were associated with higher clearance rates of molecular-BV (aHR 1.61 [1.02–2.54] in the prior 24 hours). Of note, this practice was mostly driven by twelve participants who wore thong undergarment repeatedly throughout the study. Other factors including prior CST, hormonal contraception, sexual practices and vaginal douching, were not statistically associated with molecular-BV clearance.

Description of community state type transitions by community class

Participants had a daily probability of 8% to transition to molecular-BV (incidence probability: 0.083 [95% CI 0.075–0.093]) and a daily probability of 11% to clear molecular-BV (clearance probability: 0.112 [95% CI 0.100–0.125]). Because distinct microbiota might exhibit different resilience and stability, we investigated stability patterns by evaluating daily probabilities of transition and sojourn length in and out of molecular-BV by community class, which corresponded to the main CST detected throughout the study for an individual (Table 2). After a molecular-BV incidence event, participants from LC or LG community class persisted on average 2.3 days in a molecular-BV state and participants from LI community class persisted on average 2.1 days in a molecular-BV state. For participants from low-Lactobacillus CST IV-dominated community classes (DA, DB and DC), molecular-BV lasted for 12.2 days on average, with short-lived transitions to Lactobacillus-dominated states lasting 3.2 days on average.

TABLE 2.

Daily probabilities of incidence and clearance of molecular-BV (CST-IV) by community class in the HMP-UMB study in Birmingham, AL (N=100).

	LC/LG*		LI*		DA/DB/DC*
Number of transitions to molecular-BV (incidence)	65		85		142
Number of non-transitions to molecular-BV (non-incidence)	1175		722		317
Number of transitions from molecular-BV (clearance)	65		90		140
Number of non-transitions to molecular-BV (non-clearance)	93		122		1662
Daily probability of incident molecular-BV	0.040	0.032–0.050	0.082	0.067–0.103	0.257	0.222–0.296
Daily probability of clearing molecular-BV	0.347	0.281–0.426	0.357	0.301–0.424	0.068	0.058–0.080
Number of consecutive days in molecular-BV	2.3	1.8–2.9	2.1	1.7–2.7	12.2	10.2–14.4
Number of consecutive days out of molecular-BV	19.6	14.9–25.3	9.2	7.3–11.8	3.2	2.7–3.8

^*Participants were grouped into community class according to their main CST over the course of the study, determined through hierarchical clustering on CSTs. LC: mostly CST I; LG: mostly CST II; LI: mostly CST III; DA: mostly CST IV-A; DB: mostly CST IV-B, DC: mostly CST IV-C.

The daily incidence probability of molecular-BV for LI participants was twice as high as that for LC/LG participants (0.082 [95% CI 0.067–0.103] vs. 0.040 [95% CI 0.032–0.050]). These participants also spent less time in a Lactobacillus-dominated state than participants in the LC/LG class: 9.2 days [95% CI 7.3–11.8] versus 19.6 days [95% CI 14.9–25.3].

Higher fluctuation, and therefore less stable patterns, in the LI community class are evident in Figure 1 and in Table 2. Participants in the LI community class had 71% of their samples corresponding to a L. iners-dominated CST III compared with 83% of CST I, II and V for participants in LC/LG community class and 80% of CST IV for participants in DA/DB/DC community class (in CST IV-A, IV-B and IV-C respectively) (Table 3).

TABLE 3.

Number of observations for each community state type (CST) and each time-dependent variable by community class in the HMP-UMB study in Birmingham, AL (N=100).

	LC/LG/LJ*		LI*		DA/DB/DC*
	n	%	n	%	n	%
CST
I, II, V	1189/1431	83%	86/1042	8%	155/2305	7%
III	81/1431	6%	742/1042	71%	315/2305	14%
IV-A, IV-B, IV-C	161/1431	11%	214/1042	21%	1835/2305	80%
Time-varying covariates
Menstruations	240/1431	17%	184/1042	18%	518/2305	22%
Rectal Sex	3/1420	0%	6/1042	1%	32/2264	1%
Douching	4/1418	0%	4/1042	0%	19/2263	1%
Lubricant Use	2/1400	0%	25/1028	2%	41/2247	2%
Condom Use	17/1420	1%	48/1041	5%	82/2267	4%
Digital Sex	39/1419	3%	86/1040	8%	111/2264	5%
Oral Sex	42/1420	3%	57/1042	5%	86/2265	4%
Thong Use	120/1415	8%	243/1041	23%	172/2260	8%
Condomless Vaginal Sex	89/1420	6%	93/1042	9%	210/2262	9%

^*Participants were grouped into community class according to their main CST over the first 14 days of the study, determined through hierarchical clustering on CSTs. LC: mostly CST I; LG: mostly CST II; LI: mostly CST III; DA: mostly CST IV-A; DB: mostly CST IV-B, DC: mostly CST IV-C.

Restricted analysis: Factors associated with incidence and spontaneous clearance of molecular-BV among participants from the L. iners-dominated community class LI

We hypothesized that participants from the L. iners-dominated (LI) community class may have a microbiota that was more easily disrupted by behavioral factors than participants from other community classes thus explaining why LI community class tends to have less stable patterns. We therefore restricted analysis to LI participants and found there were 85 molecular-BV incidence events (for 722 non-incidence events) and 90 molecular-BV clearance events (for 122 non-clearance events) (Table 2). In LI participants (Table S3 and Figure 3), the only factor significantly associated with incidence of molecular-BV was menstruations in the prior 24 hours and in the prior 48 hours, as was also indicated in the analysis of the whole cohort above.

Figure 3.

Time-varying and time-independent factors associated with incidence and clearance of molecular-BV (CST IV) in bivariate analyses, in 23 participants from community class LI (L. iners-dominated trajectory over time), from the HMP-UMB study in Birmingham, AL.

Analyses were adjusted for antibiotic treatment for BV in participants from the moment they took the antibiotic until the end of the study.

CST: Community State Type; BV: Bacterial Vaginosis; ATB: Antibiotics for BV.

*Reference categories for age, race, contraception and previous CST were “18–29 years old”, “White”, “Non-hormonal”, “CST I” (for incidence) and CST IV-A (for clearance).

Concerning molecular-BV clearance in LI participants, several factors were associated with a lower probability of clearing molecular-BV, including African-American participants (aHR 0.44 [0.26–0.75]) compared to White participants, older participants compared to 18–29 (age 40–49 and age 30–39 (aHR 0.38 [0.23–0.61] and 0.48 [0.28–0.83], respectively), and douching in the prior 24, 48 or 72 hours (aHR 0.45 [0.28–0.73]).

DISCUSSION

Main findings

Epidemiologic studies have indicated that race, older age, recent new sex partner, lifetime number of partners, vaginal douching, and smoking are among the most significant risk factors for incident BV, while hormonal contraception and condom use are largely considered protective (32s-35s). However, most studies have approached this topic by evaluating time-varying factors with weeks or months between sampling and surveys. We sought to assess time-varying exposures on a frequent basis with their immediate effect on vaginal microbiota dynamics with daily diaries and samples collected daily or every two days. We confirmed that menstruation was statistically associated with incident molecular-BV, and found that having a CST III (L. iners-dominated) at the previous sample increased the rate of molecular-BV incidence. In addition, this study revealed that participants most often in a CST III state (LI community class) exhibited less stable patterns than other participants.

A focus of this analysis was also to assess the factors associated with spontaneous BV clearance, i.e., transitioning to a Lactobacillus-dominated state regardless of antibiotic use for BV, a topic that has been understudied. In a study that followed participants every 4 to 8 weeks, Taha et al. reported Trichomonas vaginalis detection and multiple sex partners were associated with reduced BV clearance while a pH less than 4.5 was associated with BV clearance, in a non-treated cohort (36s). Others have approached the topic in the context of BV chronicity and patterns of recurrence (37s, 38s). In our analysis, among all participants (whatever the vaginal microbiota profiles), no factors were associated with molecular-BV spontaneous clearance, except for thong undergarment. However, in participants with a L. iners-dominated CST III longitudinal profile (community class LI, which demonstrate more unstable patterns), we revealed that specific risk factors were associated with lower molecular-BV clearance rates, such as African-American race, age over 30 and douching. No factor was associated with improved clearance rates in L. iners-dominated individuals.

Finally, we used molecular characterization of the vaginal microbiota to demonstrate that molecular-BV is usually a short-lived state (lasting between 2 and 3 days for participants who present a vaginal microbiota usually dominated by Lactobacillus spp.), or an enduring pattern (12.2 days on average) in which participants had short-lived transitions to other Lactobacillus-dominated CSTs.

Interpretation

Menstruation is an important factor for incident molecular-BV in the whole cohort and also in the analysis restricted to participants from the LI community class. This result confirms previous studies describing a decrease of L. crispatus and L. gasseri during menstruations and a concomitant increase of G. vaginalis or other bacterial anaerobes (19, 31s, 39s). General antibiotic use and antibiotic use for BV treatment was reported throughout the study, likely affecting the vaginal microbiota composition and dynamics. We tested the association between antibiotic use for BV from the day it was taken until the end of the study and incidence/clearance of molecular-BV in a univariate analysis and controlled for this exposure when modelling the impact of other factors (bivariate analyses). Previous studies showed that among patients treated for BV with antibiotics, there are initial declines in the abundances and proportions of BV-associated bacteria and a rapid expansion of lactobacilli, usually L. iners (19, 27, 40s); however, patients often quickly relapse to molecular-BV (27). In our study, having been treated with an antibiotic during the study was not associated with molecular-BV incidence or clearance, therefore the effect of the antibiotics taken by participants, if any, does not seem to affect transition rates in and out of molecular-BV. Interestingly, sexual practices were not associated with any type of transition, suggesting a limited contribution to vaginal microbiota dynamics. Additional studies on factors associated with BV incidence and clearance are needed to inform how sexual practices, which are intertwined with issues such as sexual transmission of bacteria, partner concordance or behavior substitution, affect the vaginal microbiota.

Interestingly, among the whole cohort, the CST of the previous sample was an important risk factor, with a CST III (L. iners-dominated) increasing the likelihood of a transition to molecular-BV. As L. iners-dominated communities exhibited highly fluctuating patterns suggesting low resiliency (41s), we evaluated separately the factors associated with transitions in LI individuals. This restricted analysis is consistent with prior work demonstrating that L. iners-dominated CST III is associated with C. trachomatis infection (11, 42s) and that L. iners displays different properties than other Lactobacillus spp., in particular an inability to produce D-lactic acid (12). With this analysis, it is not yet possible to know if L. iners-dominated vaginal microbiota is detrimental in itself or if it is associated with negative outcomes because it fluctuates more easily in and out of CST IV. In any case, our analysis suggested that individuals with L. iners-dominated communities had specific risk factors, in particular regarding the persistence of a molecular-BV state (defined as an absence of spontaneous clearance of molecular-BV in our time-to-event analysis). Among these specific risk factors, douching was associated with lower rates of molecular-BV clearance events in participants from the LI community class. There is little data available on the effect of douching on BV clearance, although both Onderdonk et al. and Pavlova et al. demonstrated with in-vivo and in-vitro studies, respectively, reductions in the abundance of bacteria following douching (43s, 44s). Sabo et al. demonstrated in a U.S. cohort a higher likelihood of detection of BV-associated bacteria among participants who reported vaginal washing (7 participants out of 26) (45s). However, Brown et al. recently found douching cessation was not associated with major changes in vaginal microbiota in a pilot study of 34 participants (46s).

Studying factors related to the temporal dynamics of the vaginal microbiota is challenging because rapid fluctuation between Lactobacillus-dominated CSTs and molecular-BV CST IV are common (31s). The approach of assessing factors associated with molecular-BV incidence and clearance in a particular community class may serve as a starting point for future research to better understand the temporal profiles and resilience of certain communities of vaginal microbiota, because these factors may be different depending on a woman’s community class. To our knowledge, this analysis is the first to report on factors associated with both incidence and spontaneous clearance of molecular-BV.

Conclusions

Most community state type transitions identified in our study were short-lived. We highlighted distinct factors associated with molecular-BV incidence such as menstruations and a L. iners-dominated CST III at the previous sample. Further, participants’ vaginal microbiota dynamics can be classified into community classes based on the most common CST over time. L. iners-dominated participants had specific factors associated with a reduced clearance of molecular-BV, such as African-American race, age 30–49 and douching, indicating that these individuals may have less resilient vaginal microbiota and may experience acute transitions in and out of molecular-BV. Defining vaginal microbiota profiles may aid future studies in identifying risk factors and responses to treatment, as resilience of stable versus fluctuating patterns may vary.