Association of Vaginal Microbiota With Signs and Symptoms of the Genitourinary Syndrome of Menopause Across Reproductive Stages

Abstract

The genitourinary syndrome of menopause (GSM) describes signs and symptoms resulting from effects of estrogen deficiency on the female genitourinary tract, including the vagina, labia, urethra, and bladder. Signs/symptoms associated with GSM may occur during any reproductive stage from multiple etiologies but are most common during menopause due to low estrogen. Vaginal microbiota, particularly Lactobacillus spp., are beneficial to the female genital tract; however, their abundance declines during menopause. We aimed to longitudinally assess vaginal microbiota characterized by 16S rRNA gene amplicon sequencing and GSM-associated endpoints across reproductive stages. In a 2-year cohort study of 750 women aged 35–60 years at enrollment and 2 111 semiannual person-visits, low-Lactobacillus vaginal microbiota communities were observed at 21.2% (169/798), 22.9% (137/597), and 49.7% (356/716) of person-visits among pre-, peri-, and postmenopausal women, respectively (p < .001). Compared to communities that have high Gardnerella vaginalis relative abundance and diverse anaerobes, the following communities were associated with a lower covariate-adjusted odds of vaginal atrophy: L crispatus-dominated communities among postmenopausal women (odds ratio [OR] = 0.25; 95% confidence interval [CI]: 0.08, 0.81), L gasseri/L jensenii (OR = 0.21; 95% CI: 0.05, 0.94) and L iners (OR = 0.21; 95% CI: 0.05, 0.85) among perimenopausal women, and L iners-dominated communities (OR = 0.18; 95% CI: 0.04, 0.76) among premenopausal women. Postmenopausal women with L gasseri/L jensenii-dominated communities had the lowest odds of vaginal dryness (OR = 0.36; 95% CI: 0.12, 1.06) and low libido (OR = 0.28; 95% CI: 0.10, 0.74). Findings for urinary incontinence were inconsistent. Associations of vaginal microbiota with GSM signs/symptoms are most evident after menopause, suggesting an avenue for treatment and prevention.

The genitourinary syndrome of menopause (GSM) is a term endorsed by the North American Menopause Society (NAMS) to describe signs and symptoms resulting from effects of estrogen deficiency on the female genitourinary tract that women often experience during menopause (1,2). Signs involve physical changes to the vulva, vagina, and lower urinary tract such as vulvovaginal atrophy and thinning and loss of moisture of vaginal mucosal walls (1,2). The NAMS 2020 position statement indicated that GSM symptoms, such as vaginal dryness, urinary incontinence, and sexual dysfunction (eg, low libido), affect 27%–84% postmenopausal women, and that over half of women with vaginal symptoms reported that these symptoms adversely affected their sexual health and quality of life (2).

Signs/symptoms associated with GSM can occur during any reproductive stage, but they most often emerge during perimenopause and worsen if untreated (2). Complicating matters is that these signs/symptoms may reflect multiple etiologies (2). Halting onset and progression of GSM is also complicated by the limitations of available treatments. A first-line treatment for vulvovaginal symptoms is nonhormonal moisturizers and isomolar lubricants. Most commercial varieties have a higher osmolality than vaginal cells, which damages epithelial integrity (3), and may exacerbate loss of beneficial lactic acid-producing vaginal bacteria (4,5). Systemic hormone therapy can ameliorate GSM signs/symptoms but has some contraindications (2). Topical estrogen is preferred for treating mild GSM symptoms but is also contraindicated in some women (2) and may not improve urinary symptoms (6). Newer treatments like laser therapy require more evidence (2).

After menopause, the composition of vaginal microbiota has been shown to be less likely dominated by Lactobacillus spp. and more likely to be composed predominantly of anaerobic and aerobic bacteria (7–9). Lactobacillus spp. contribute to urogenital health among reproductive-aged women (10); however, less is known about peri- and postmenopausal women. A role for Lactobacillus spp. in GSM, especially vulvovaginal signs and symptoms, is plausible, given that lactobacilli support structural integrity of the cervicovaginal epithelium (11), protect epithelial cells against oxidative stress (12), and produce lactic acid, which creates an acidic vaginal environment (pH ~4) that is both antimicrobial and anti-inflammatory (13,14). Some studies have found low-Lactobacillus vaginal microbiota to be associated with vaginal atrophy signs (7,8); however, despite the strong associations of vaginal signs/symptoms with indicators of sexual dysfunction (2), no large longitudinal vaginal microbiota study has focused on indicators of sexual dysfunction included in the definition of GSM endorsed by NAMS. This definition also includes urinary symptoms (1,2); and indeed, bacterial taxa commonly found in the vagina are functionally and phylogenetically similar to strains isolated from the bladder (15,16). Further, a recent study reported estrogen therapy in postmenopausal women to be associated with bladder Lactobacillus spp. relative abundance and modest changes in urinary incontinence symptoms (17).

Studies to date on vaginal microbiota and GSM have been small, cross-sectional, or limited to postmenopausal women (7,8,18,19). We aimed to longitudinally test the association between vaginal microbiota and GSM in a large prospective cohort of women across a broad age range (35–60 years at enrollment). Also, previous work has focused primarily on vulvovaginal aspects of GSM (7,8). To address the multidimensionality of GSM (1,2), we examine a vulvovaginal sign (atrophy), a vaginal symptom (self-reported dryness), a urinary symptom (self-reported urinary incontinence), and a sexual dysfunction symptom (self-reported decreased sexual interest). We hypothesize Lactobacillus-dominant vaginal microbiota to be associated with lower burden of GSM signs/symptoms, especially vaginal atrophy, as compared with a low-Lactobacillus composition.

Method

Study Sample

Study participants were from the Human Papillomavirus in Perimenopause (HIP) study, a cohort that recruited 885 women aged 35–60 years from 4 clinical sites in Baltimore, Maryland, from 2008 to 2010. Eligible women were English speakers with intact uterus and no history of HIV infection or organ transplantation who provided informed consent and primary locator information (20). The Johns Hopkins University School of Public Health and University of Maryland Baltimore Institutional Review Boards approved the study; participants provided written informed consent. Participants attended semiannual clinic visits for up to 24 months of follow-up. Visits included an interview, pelvic exam, and speculum cervical exam and clinician-collected mid-vaginal swab for microbiota analysis (Elution-swab system stored in 1 mL of Amies transport media, Copan) (20). Vaginal microbiota was profiled in samples collected during 2 301 person-visits (PVs; up to 5 visits at baseline, 6, 12, 18, and 24 months) from 812 women. The final analytic data set comprised 2 111 PVs from 750 women with data on microbiota and reproductive stage (defined below). Among these women, 152, 169, 186, 152, and 91 contributed 1, 2, 3, 4, and 5 observations, respectively, to the final analytic data set. There were 373, 407, 433, 437, and 461 observations at baseline, 6, 12, 18, and 24 months, respectively.

Microbiota Assessment and Data Processing

DNA was extracted from vaginal Elution-swabs on a QiaSymphony instrument, where the final extracted DNA is eluted off the column with water, as previously described (21). Negative controls (water) were extracted at the same time as samples using the same protocol as samples. Vaginal microbiota was characterized by sequencing amplicons of the V3–V4 region of the 16S rRNA gene. Libraries were constructed by a 2-step PCR protocol and dual-barcoding strategy. Sequencing was performed on an Illumina HiSeq 2500 instrument modified to generate 300 bp paired-end reads as previously described (21). Raw data were processed using DADA2 (v.1.5.2) (22), and amplicon sequence variants (ASVs) were taxonomically classified using the RDP Naive Bayesian Classifier trained with the SILVA v128 16S rRNA gene database (23). ASVs of major vaginal taxa were assigned species-level annotations using speciateIT (http://ravel-lab.org/speciateit/). Taxa identified as contaminants in negative controls and samples with <1 000 reads were removed.

Microbiota Community State Types

VALENCIA, a nearest centroid-based classifier, was used to classify samples into community state types (CSTs) (24). Four CSTs are dominated by Lactobacillus spp.: CST-I, L crispatus; CST-II, L gasseri; CST-III, L iners; and CST-V, L jensenii. Low-Lactobacillus CST-IV is subclustered: CST-IV-A, diverse anaerobes with Candidatus Lachnocurva vaginae (formerly bacterial vaginosis [BV]-associated bacterium 1 [BVAB1]) and Gardnerella vaginalis; CST-IV-B, diverse anaerobes with G vaginalis but low Ca. L. vaginae; and CST-IV-C, diverse anaerobic and aerobic bacteria.

Study Outcomes and Covariate Assessment

Primary study outcomes were the GSM endpoints vaginal atrophy and vaginal dryness. Secondary outcomes were urinary incontinence and low libido. Together, these endpoints represent the multidimensional GSM definition endorsed by NAMS and address heterogeneity in GSM presentation. They also serve as a sensitivity analysis by determining whether vaginal microbial profiles more consistently relate to vaginal than urinary endpoints (where vaginal microbiota may be proxy for urinary microbiota) or sexual endpoints (which may reflect an indirect association due to its correlation with vaginal endpoints) (2). Presence of vaginal atrophy was determined at enrollment, and 12-, and 24-month visits during pelvic exams and was operationally defined as clinician-indicated presence of the atrophy signs pallor, dryness, diminished rugosity, blanching, friability, or petechiae as a composite measure to reduce sparsity. Pelvic exams were inconsistently performed at 6- and 18-month visits, thus excluded for this endpoint. Participants were interviewed at 6-, 12-, 18-, and 24-month follow-up visits for presence of common menopausal symptoms. Low libido was defined as less sexual interest than usual; vaginal dryness by self-report; and urinary incontinence by experience of uncontrolled urine leakage, especially upon sneezing/laughing.

Women were categorized as postmenopausal if they reported that they did not have a menstrual period within the last 12 months and were not taking exogenous hormones that result in cessation of menstrual periods. Among women who reported their last menstrual period to have been within the last 12 months, those who reported greater cycle length variability without exogenous hormone use were categorized as perimenopausal; otherwise, they were categorized as premenopausal. This operational staging definition was used to align with the 3 broad reproductive stages and they use menstrual cycle as the principal staging criteria recommended by the 2001 Stages of Reproductive Aging Workshop (STRAW) and the 2011 STRAW + 10 (25,26). Other covariate data from interviews included age (years), age at menarche (years), education, race, smoker status, marital status (married, never married, widowed/separated/divorced), number of lifetime male sex partners, recent sexual activity (within 6 months), and exogenous hormone use (contraceptive or hormone replacement pills, injections, creams, or intrauterine devices). Age, age at menarche, education, race, and smoking were time-independent variables in analysis; others were time varying. The covariates were selected because they may influence microbiota (27) or GSM endpoints (28).

Statistical Analysis

We performed descriptive analysis by menopause stage. We computed number (%) of PVs and chi-squared tests to compare categorical variables and median (range) and Kruskal–Wallis tests to compare continuous variables, all for clustered data (see Supplementary Methods).

We measured within-person microbial diversity (alpha diversity) of 168 identified lowest-level taxa using Shannon diversity index (29) and computed medians (interquartile range [IQR]) by CST. To reduce the data dimension, we filtered taxa detected in <5% of samples or in <10^–4 proportion of reads. Retained taxa were individual taxa if they had a relative abundance with a 2% trimmed range ≥5% (ie, 98th percentile minus 2nd percentile ≥0.05). Taxa that did not satisfy this criterion overall but did within a CST were agglomerated. Taxa that did not meet this criterion were agglomerated and labeled “other taxa.” Microbiota data for each sample are a vector of taxa read counts that are compositional (30), meaning that since total read counts differ by sample and are constrained by the DNA sequencer, taxa read counts do not reflect absolute microbial abundance, only relative abundance. Thus, counts were normalized by a sample’s total read count to compute relative abundances constrained to sum to 1. We computed taxa geometric mean relative abundance by menopause stage and CST after filtering and agglomerating. We used weak Dirichlet priors for multiple imputation of zeroes as in our previous work (31). Taxa correlations were quantified by Sparse Correlations for Compositional Data (SparCC) (32) (see Supplementary Methods).

We used weighted generalized estimating equations (WGEE) to fit marginal structural logistic models (33) of each GSM endpoint on natural cubic splines of Shannon diversity (3 knots), menopause stage, their interactions, and study visit. Inverse-probability weighting addressed time-varying confounding of stage and missing data (34); inverse density weighting addressed time-varying confounding of Shannon diversity (see Supplementary Methods). We similarly regressed GSM endpoints on CST, stage, their interactions, and study visit. Due to sparsity, we combined CSTs II and V into CST-II/V, and we combined CSTs IV-A and IV-C into CST-IV-A/C. In secondary analysis, we examined microbiota at 2 adjacent visits to assess low Lactobacillus persistence and GSM endpoints. Due to sparsity, we combined CSTs into Lactobacillus-dominant (CSTs I, II, III, and V) and low-Lactobacillus (CSTs IV-A, IV-B, and IV-C) communities. We defined Lactobacillus dominance at both visits as “persistent Lactobacillus dominance,” and low-Lactobacillus communities at both visits as “persistent low Lactobacillus.” Once again, we used WGEE. We also used WGEE for first-order Markov transition models of Lactobacillus dominance with transitions (onset or remission) of GSM endpoints (Supplementary Methods).

Lastly, we assessed filtered and agglomerated microbial taxa and GSM endpoints using compositional data analysis with microbiota as exposures. Microbial taxa are interrelated, thus isolating the association of a taxon with outcomes requires adjusting for other taxa; but microbiota relative abundance data are high-dimensional and perfectly collinear due to the sum-to-1 constraint. We used compositional Bayesian generalized linear mixed models with random intercepts and weak priors for normalization (35) for within-person correlation and high-dimensional compositional microbiota, extending our previous work (31). We transformed taxa relative abundances into isometric log ratios (ILRs), which measure a taxon’s relative dominance by comparing a log taxon relative abundance to log geometric mean relative abundances of other taxa, adjusting for relative dominances of the other taxa (Supplementary Methods) (36). Models included menopause stage, ILRs, their interactions, and covariates. Transformed exponentiated coefficients of taxon main effects are interpreted as odds ratios per doubling of a taxon’s relative dominance adjusted for relative dominances of other taxa and covariates. Multiple imputation of 50 data sets addressed zeros in logarithms (31). We adjusted individual taxon p values for multiple comparisons by comparing p values to .05 divided by the number of principal components (effective number of independent tests) determined from a scree plot. This approach overcomes overly conservative Bonferroni corrections for correlated tests, controls type I error (37), and unlike false discovery rate, uses the same criterion for all endpoints. We performed all analyses with R Statistical Software version 3.6.3 (packages in Supplementary Methods). p < .05 (p < .05/# of principal components for individual taxa) or 95% confidence intervals (CIs) that excluded the null were considered statistically significant.

Results

The 750 participants had median (IQR) enrollment age 47 (42–52) years, which was highest among postmenopausal women (53 years among 716 PVs) and lowest among premenopausal women (43 years among 270 PVs) (p < .001), but each stage included women from nearly the full study age range (Table 1). Most of the 750 participants were White (75.5%) and were college graduates (62.7%), and few were smokers at enrollment (9.9%). Postmenopausal women were least likely to report recent sexual activity (p < .001). Postmenopausal women had the highest Shannon diversity and highest proportion of PVs within CSTs IV-B and IV-C, and the lowest proportion of PVs within all other CSTs; perimenopausal women had the highest proportion of PVs within CSTs III and V, and premenopausal women had the highest proportion of PVs within CSTs I, II and IV-A (p < .001). More advanced stage was associated with reporting low libido (45.8% [postmenopausal] vs 16.5% [premenopausal] of PVs) and vaginal dryness (58.0% [postmenopausal] vs 13.8% [premenopausal] of PVs) (both p < .001); however, urinary incontinence was most commonly reported by perimenopausal women (27.5% of PVs, p = .067). More advanced stage was also associated with vaginal atrophy (50.9% [postmenopausal] vs 14.9% [premenopausal] of PVs, p < .001). Participant-level longitudinal patterns of CST, vaginal atrophy, and symptoms are shown in Supplementary Figure 1A–C .

Table 1.

Characteristics of 750 Women Across 2 111 Person-Visits (PVs) in Human Papillomavirus in Perimenopause Study by Menopause Stage

	All Womena (N = 750)	Postmenopausal (716 PVs from 375 womenb)	Perimenopausal (597 PV from 310 womenb)	Premenopausal (798 PVs from 270 womenb)
Characteristics	Median (range) or n/N (%)	Median (range) or No./PV (%)	Median (range) or No./PV (%)	Median (range) or No./PV (%)	p Value
Baseline age (y)	47.0 (35.0–60.0)	53.0 (35.0–60.0)	46.0 (35.0–55.0)	43.0 (35.0–57.0)	<.001
Race	N = 750	716 PVs	597 PVs	798 PVs	.66
Black/African-American	128 (17.1)	122 (17.0)	102 (17.1)	146 (18.3)
White	566 (75.5)	551 (77.0)	451 (75.5)	579 (72.6)
Other race	56 (7.5)	43 (6.0)	44 (7.4)	73 (9.1)
Baseline education	N = 750	716 PVs	597 PVs	798 PVs	.073
Secondary school	112 (14.9)	135 (18.9)	68 (11.4)	91 (11.4)
Some postsecondary school	168 (22.4)	146 (20.4)	141 (23.6)	179 (22.4)
College graduate	470 (62.7)	435 (60.8)	388 (65.0)	528 (66.2)
Age at menarche (y)	13.0 (8.0–20.0)	13.0 (9.0–18.0)	13.0 (8.0–20.0)	13.0 (8.0–20.0)	.26
Current marital status		716 PVs	597 PVs	798 PVs	.17
Never married		111 (15.5)	99 (16.6)	178 (22.3)
Widowed, separated, or divorced		136 (19.0)	113 (18.9)	121 (15.2)
Married		469 (65.5)	385 (64.5)	499 (62.5)
Baseline smoker	74/750 (9.9)	80/716 (11.2)	57/597 (9.5)	62/798 (7.8)	.35
Lifetime sex partners		716 PVs	594 PVs	798 PVs	.14
<3 men		184 (25.7)	98 (16.5)	139 (17.4)
3–5 men		199 (27.8)	184 (31.0)	242 (30.3)
6–10 men		199 (27.8)	167 (28.1)	236 (29.6)
>10 men		134 (18.7)	145 (24.4)	181 (22.7)
Recent sex (in past 6 mo)		508/711 (71.4)	473/595 (79.5)	676/796 (84.9)	<.001
Exogenous hormone use		169/716 (23.6)	141/597 (23.6)	221/798 (27.7)	.36
Shannon diversity		0.99 (0.00–3.36)	0.60 (0.00–2.99)	0.60 (0.00–2.93)	<.001
Community state type		716 PVs	597 PVs	798 PVs	<.001
I (L crispatus dominated)		136 (19.0)	197 (33.0)	308 (38.6)
II (L gasseri dominated)		60 (8.4)	59 (9.9)	81 (10.2)
III (L iners dominated)		131 (18.3)	161 (27.0)	189 (23.7)
IV-A (Ca L. vaginae, G vaginalis, diverse anaerobes)		10 (1.4)	12 (2.0)	17 (2.1)
IV-B (G vaginalis, diverse anaerobes, low Ca L. vaginae)		106 (14.8)	67 (11.2)	108 (13.5)
IV-C (diverse anaerobes)		240 (33.5)	58 (9.7)	44 (5.5)
V (L jensenii dominated)		33 (4.6)	43 (7.2)	51 (6.4)
Vaginal atrophy		167/328 (50.9)	46/261 (17.6)	56/375 (14.9)	<.001
Vaginal dryness		349/602 (58.0)	121/495 (24.4)	88/636 (13.8)	<.001
Urinary incontinence		200/601 (33.3)	177/495 (35.8)	175/636 (27.5)	.067
Low libido		276/602 (45.8)	168/495 (33.9)	105/635 (16.5)	<.001

Notes: ^aTime-invariant variables only. ^b205 women changed stage over time and are in multiple stage columns.

The final set of lowest-level 168 classified taxa included 42 species, 111 genera, 10 families, 1 order, 3 classes, and 1 phylum. The final microbiota data set comprised 119 555 234 reads. Figure 1 and Supplementary Figure 2 convey individual and geometric mean relative abundances, respectively, of the most abundant taxa from samples collected at 2 111 PVs. Samples classified as CST-I, II, III, and V had low diversity and high relative abundance of L crispatus, L gasseri, L iners, and L jensenii, respectively, for all stages; CST-I had the lowest Shannon diversity (median = 0.29; IQR: 0.09, 0.68). CST-IV was more diverse than Lactobacillus-dominated CSTs, where CST-IV-A had the highest Shannon diversity (median = 2.17; IQR: 1.99, 2.29). Samples classified as CST-IV-A or CST-IV-B had high relative abundance of G vaginalis; however, CST-IV-A samples also had higher relative abundances of Atopobium vaginae, Megasphera, Sneathia sanguinegens (among perimenopausal women), Ca. L. vaginae, and an agglomeration of low-abundance taxa often found in both CSTs IV-A and IV-B (Supplementary Methods) compared to other CSTs. In contrast, samples classified as CST IV-C had high relative abundances of Streptococcus and an agglomeration of low-abundance taxa (Supplementary Methods), as well as higher relative abundances of Anaerococcus, Peptoniphilus, and Finegoldia in postmenopausal women compared to other CSTs and menopausal stages. Figure 2 shows positive correlations among taxa commonly found in CST-IV-C; and among taxa commonly found in CSTs IV-A and IV-B, although low-abundance agglomerations did not consistently correlate with taxa commonly found in their CST namesakes. Also, Lactobacillus spp. tended to weakly positively correlate with each other and negatively correlate with taxa found in CSTs IV-A and IV-B except for L iners, which weakly positively correlated with these taxa (Figure 2).

Figure 1.

Heat map of relative abundance of 27 individual taxa and agglomerations (Supplementary Methods) of 2 111 person-visits from 750 women sorted by menopause stage and GSM endpoints. CST-1, L crispatus dominated; CST-II, L jensenii dominated; CST-III, L iners dominated; CST-IV-A, Ca L. vaginae, G vaginalis, diverse anaerobes; CST-IV-B, G vaginalis, diverse anaerobes, low Ca L. vaginae; CST-IV-C, diverse anaerobes; CST-V, L gasseri dominated. NA refers to missing GSM endpoint, either due to nonresponse or design. Rows are taxa; columns are person-visits. Taxa with g_ are classified at the genus level. Remaining taxa are species or agglomerations (Supplementary Methods). CST, community state type; GSM, genitourinary syndrome of menopause.

Figure 2.

Correlation and hierarchical clustering of 27 taxa and agglomerations measured at 2 111 person-visits from 750 women. Taxa with g_ are classified at the genus level. Remaining taxa are species or agglomerations (Supplementary Methods).

Among postmenopausal women, higher Shannon diversity was associated with lower odds of vaginal atrophy, vaginal dryness, and low libido, up to a threshold; then higher odds beyond the threshold. However, results were only statistically significant for vaginal dryness (p = .036) (Figure 3). There was little evidence that Shannon diversity related to study outcomes in earlier stages.

Figure 3.

Association of Shannon diversity (alpha diversity) and genitourinary syndrome of menopause (GSM) endpoints by menopausal stage at 2 111 person-visits from 750 women. Vertical axis is the odds ratio of the GSM endpoint relative to a Shannon diversity of 0; the horizontal axis is Shannon diversity.

The highest proportions of women with vaginal atrophy were those with vaginal microbiota classified as CST-IV-A/C (58.6%) and IV-B (54.7%) among postmenopausal women, CST-I (22.4%) among perimenopausal women, and CST-IV-A/C (20.0%) and IV-B (19.6%) among premenopausal women (Supplementary Table 1). Adjusted odds of vaginal atrophy among postmenopausal women were lower in all CSTs than in CST-IV-B, but only CST-I was statistically significant (OR = 0.25; 95% CI: 0.08, 0.81). Among perimenopausal women, significantly lower adjusted odds of vaginal atrophy were found for both CST-II/V (OR = 0.21; 95% CI: 0.05, 0.94) and CST-III (OR = 0.21; 95% CI: 0.05, 0.85). CST-III was also associated with lower adjusted odds of vaginal atrophy (OR = 0.18; 95% CI: 0.04, 0.76) among premenopausal women (Table 2).

Table 2.

Associations of Vaginal Microbiota Community State Types (CSTs) With Genitourinary Syndrome of Menopause (GSM) Endpoints by Menopause Stage Among Women in Human Papillomavirus in Perimenopause Study

		CST-IV-Ba		CST-Ia		CST-II/Va		CST-IIIa		CST-IV-A/Ca
GSM Endpoint	Menopause Stage	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	OR	95% CI	p for Interactionb
Vaginal atrophyc	Post	1.00	Ref.	0.25	0.08, 0.81	0.39	0.12, 1.26	0.42	0.16, 1.11	0.46	0.17, 1.23	Ref.
	Peri	1.00	Ref.	0.44	0.12, 1.61	0.21	0.05, 0.94	0.21	0.05, 0.85	0.23	0.05, 1.12	.32
	Pre	1.00	Ref.	0.45	0.12, 1.66	0.43	0.15, 1.22	0.18	0.04, 0.76	0.33	0.07, 1.54	.34
Vaginal drynessd	Post	1.00	Ref.	0.52	0.17, 1.57	0.36	0.12, 1.06	0.49	0.16, 1.51	0.71	0.26, 1.93	Ref.
	Peri	1.00	Ref.	0.53	0.18, 1.56	1.02	0.34, 3.09	0.70	0.25, 1.95	0.65	0.21, 1.96	.39
	Pre	1.00	Ref.	1.02	0.26, 4.00	1.09	0.27, 4.42	1.60	0.38, 6.67	0.85	0.16, 4.53	.57
Urinary incontinenced	Post	1.00	Ref.	0.67	0.23, 1.94	0.62	0.21, 1.84	1.23	0.40, 3.74	1.54	0.56, 4.28	Ref.
	Peri	1.00	Ref.	0.40	0.16, 0.96	1.19	0.46, 3.11	0.84	0.36, 1.98	1.08	0.38, 3.06	.41
	Pre	1.00	Ref.	1.30	0.49, 3.44	0.85	0.31, 2.30	1.20	0.48, 3.01	0.81	0.27, 2.38	.44
Low libidod	Post	1.00	Ref.	0.42	0.13, 1.35	0.28	0.10, 0.74	0.31	0.11, 0.90	0.70	0.25, 1.97	Ref.
	Peri	1.00	Ref.	0.52	0.21, 1.34	1.29	0.48, 3.48	1.31	0.53, 3.26	1.79	0.56, 5.69	.066
	Pre	1.00	Ref.	0.72	0.23, 2.28	1.24	0.44, 3.51	0.81	0.28, 2.35	0.48	0.12, 1.95	.068

Notes: CI = confidence interval; OR = odds ratio.

^aCST-IV-B, G vaginalis, diverse anaerobes, low Ca L. vaginae; CST-I, L crispatus dominated; CST-II/V, L jensenii/L gasseri dominated; CST-III, L iners dominated; CST-IV-A/C, Ca L. vaginae, G vaginalis, diverse anaerobes/diverse anaerobes. ^bCST-by-menopause stage interaction, with postmenopause as reference. ^cAssessed at enrollment, and 12- and 24-mo visits. ^dAssessed at 6-, 12-, 18-, and 24-mo visits.

Among postmenopausal women, the highest proportions reporting vaginal dryness were those with vaginal microbiota classified as CST-IV-A/C (63.8%) or -IV-B (57.6%), whereas the highest proportions among peri- and premenopausal women were those with microbiota classified as CST-II/V (34.5%) and CST-I (16.5%), respectively (Supplementary Table 1). Adjusted odds of vaginal dryness among postmenopausal women were lower for all CSTs relative to CST-IV-B, with the lowest odds for CST-II/V (OR = 0.36; 95% CI: 0.12, 1.06). CST did not statistically significantly relate to vaginal dryness among peri- and premenopausal women (Table 2).

Among postmenopausal women, those with vaginal microbiota classified as CST-III (42.5%) were most likely to report urinary incontinence. The highest proportions among perimenopausal and premenopausal women were those with microbiota classified as CST-II/V (47.6%) and CST-III (31.3%), respectively (Supplementary Table 1). The only statistically significant association between CST and urinary incontinence relative to CSTI-IV-B was for CST-I among perimenopausal women (OR = 0.40; 95% CI: 0.16, 0.96) (Table 2).

The highest proportions of women reporting low libido were those with vaginal microbiota classified as CST-IV-B (54.1%) among postmenopausal women and CST-II/V among peri- (45.2%) and premenopausal women (22.9%) (Supplementary Table 1). For postmenopausal women, adjusted odds of low libido were lower for all CSTs relative to CST-IV-B, and statistically significant for CST-II/V (OR = 0.28; 95% CI: 0.10, 0.74) and CST-III (OR = 0.31; 95% CI: 0.11, 0.90) (Table 2). No CST was statistically significantly associated with low libido among peri- or premenopausal women (Table 2). Vaginal dryness was significantly associated with low libido (p < .001 for all stages; Supplementary Results)

Assessment of Lactobacillus dominance (CSTs I, II, III, or V) versus low Lactobacillus (CSTs IV-A, IV-B, or IV-C) persistence in 2 adjacent visits revealed patterns that differed between postmenopausal and premenopausal women for vaginal dryness, urinary incontinence, and low libido (Supplementary Tables 2 and 3). Odds of each symptom were lower with persistent Lactobacillus dominance than persistent low Lactobacillus among postmenopausal women but were only statistically significant for urinary incontinence (OR = 0.30; 95% CI: 0.11, 0.79). Results among postmenopausal women significantly differed from those among premenopausal women for vaginal dryness and urinary incontinence (p for interaction < .05 for both). Assessment of Lactobacillus dominance versus low Lactobacillus with GSM endpoint transitions estimated mostly lower adjusted odds of endpoint onset and higher adjusted odds of remission among postmenopausal women, but only remission of low libido was statistically significant (OR = 4.27; 95% CI: 1.51, 12.05) (Supplementary Tables 4 and 5). No consistent patterns were found for peri- or premenopausal women.

Compositional principal component analysis identified 4 components, thus statistical significance for individual taxa was p < .05/4 = .0125. The 5 most significant taxa for each GSM endpoint by stage are shown in Supplementary Figures 3–5. For postmenopausal women, no taxon was significantly related to any endpoint after multiple comparisons adjustment (Supplementary Figure 3; Supplementary Tables 6–9). Perimenopausal women with higher relative dominance of a taxa agglomeration found in CST-IV-C had higher odds of low libido (p = .012) (Supplementary Figure 4; Supplementary Tables 6–9). For premenopausal women, higher L iners relative dominance was associated with lower odds of vaginal atrophy (p = .0056) (Supplementary Figure 5; Supplementary Tables 6–9).

Discussion

Overall, we found strong evidence that vaginal microbiota is associated with GSM endpoints among postmenopausal women. Postmenopausal women with microbiota classified as CST-IV-B (low Lactobacillus and high G vaginalis relative abundance) tended to have the highest covariate-adjusted odds of vaginal atrophy, vaginal dryness, and low libido, and microbiota dominated by Lactobacillus spp. (L crispatus, L iners, L gasseri/L jensenii) tended to have the lowest odds. Alpha diversity and individual taxon results were generally consistent with CST results.

Lactobacillus spp. has been associated with urogenital health among reproductive age women, but the precise mechanisms that may explain why Lactobacillus spp. dominance is inversely associated with GSM endpoints among postmenopausal women are unknown. A role for vaginal microbiota in GSM, especially vaginal signs/symptoms, is plausible, given the lower likelihood of Lactobacillus dominance after menopause (7,9) and that some (7,8), though not all (19), studies found vaginal signs/symptoms to be associated with low Lactobacillus spp. abundance. Lactobacillus spp. has been shown to protect against oxidative stress (12); and they produce lactic acid, which has anti-inflammatory properties (13,14). Notably, Lactobacillus spp. also reduces G vaginalis-induced cell cytotoxicity (38). Herein, CST-IV-B had high relative abundance of G vaginalis and the highest odds of multiple GSM endpoints; but, it is unknown if findings of other CSTs relative to CST IV-B reflect functional benefits from Lactobacillus spp. or lack of harm from absence (or low abundance) of G vaginalis or other CST IV-B taxa, or both. G vaginalis is thought to be a key bacterium in the etiology of BV, a common clinical condition in reproductive age women characterized by low Lactobacillus abundance and vaginal symptoms including odor and discharge (10,39). However, magnitudes of association of individual taxa, such as Lactobacillus spp. or G vaginalis, with GSM endpoints in either direction, were small, which may be due, in part, to taxa networks, such as CSTs, that work in concert, not in isolation.

Not unexpectedly, patterns for urinary incontinence differed somewhat from those for vaginal endpoints. The direction of findings is consistent with Lactobacillus spp. dominance in general as beneficial for urinary incontinence, but findings were weaker than those for vaginal atrophy. Urinary incontinence may relate more strongly to urinary microbiota than to vaginal microbiota; however, evidence suggests that urinary and vaginal microbiota are highly concordant (15,16). Recent work suggests that urogenital microbiota may play a role in urinary symptoms among postmenopausal women (17), although our results do not conclusively prove this.

A novel finding here is the association of Lactobacillus-dominated CSTs with lower odds of low libido among postmenopausal women. There is no underlying hypothesis for a direct association with vaginal microbiota; however, it is plausible that an indirect relation exists due to the associations of vaginal signs and symptoms with sexual symptoms as highlighted by NAMS (2) and with vaginal microbiota shown here and in previous reports (7,8). Although the low-Lactobacillus condition BV has been associated with sexual symptoms in reproductive age women (40), we did not find an association of low libido with vaginal microbiota among premenopausal women here.

Many findings differed, some significantly, between postmenopause and other stages. Multiple possibilities may explain these findings. First, as noted by NAMS (2), signs and symptoms similar to GSM can occur during any stage due to, for example, allergic or inflammatory conditions. Thus, between-stage differences in associations may reflect heterogeneity in the condition leading to the endpoints. Second, microbial profiles between stages may also reflect different etiologies. Although postmenopausal women are less likely to have a vaginal bacterial community dominated by Lactobacillus spp. than premenopausal women, nearly half of postmenopausal women have a high relative abundance of Lactobacillus spp. (8), and presence of Lactobacillus spp. has been associated with exogenous hormone use (41). We addressed this explanation, in part, by adjusting for exogenous hormone use in analysis. In contrast, low-Lactobacillus profiles in premenopausal women, which tend to differ in composition from low-Lactobacillus profiles in postmenopausal women, may reflect BV. The distinction in etiology impacts treatment indication (2). Thus, stratifying on stage here may be a proxy for stratifying on etiology of both vaginal microbiota and GSM-associated endpoints. Third, microbes are metabolically active and co-occur in concert even at low abundance, and may depend on the local immunological environment, which may differ by menopause stage (18,42). Lastly, study outcomes were generally uncommon for premenopausal women; thus, results may be due to low power or chance.

Strengths of this study include a large well-characterized longitudinal cohort of women across menopause stages with repeated vaginal microbiota assessment and rigorous statistical analysis that addressed microbial communities as a whole (CSTs) and individual taxa. Most research on vaginal microbiota and GSM has been cross-sectional, in small samples, or limited to one menopause stage. In contrast, we used a longitudinal design to examine changes in microbial communities and GSM endpoints across menopause stages. Also, we assessed urinary and sexual dimensions as well as the vulvovaginal dimension of GSM using modern methods to address time-varying confounding and missing data. Lastly, findings were consistent overall with a 2014 preliminary cross-sectional report on 88 women from the same cohort in which postmenopausal women with vaginal microbiota dominated by Lactobacillus spp. had the lowest odds of vaginal atrophy and dryness, and those with microbiota consistent with CSTs IV-A/C and IV-B had the highest odds (7). Gardnerella vaginalis could not be assessed in the 2014 report owing to technological limitations and differences in sequencing (V1–V2 pyrosequencing).

Some limitations must be noted. First, microbiota relative abundance data are prone to measurement error. This issue was addressed by analysis that included multiple imputation for measurement error and a robust CST classifier (24). Self-reports, especially menstrual cycle length variability used for staging, are prone to more misclassification error than other methods (eg, menstrual diaries). However, since measurement of microbiota and self-reported cycle length are nondifferential with respect to each other, these issues likely led to conservative results. Second, data sparsity required combining some CSTs; this was overcome in analysis using more granular taxon data. As in any study of microbial relative abundances, individual taxa are highly correlated and compositional; however, we used appropriate CST level and compositional analysis to address this. Lastly, as in any observational study with concurrent exposure and outcome assessment, reasons for association are unknown. It is unknown if the putative exposure (vaginal microbiota) affects the putative outcome (GSM endpoint) or vice versa, or if findings are due to unmeasured genetic, behavioral, relational, community, or societal confounders (27). We performed secondary analysis on clinically relevant changes in microbial composition and on transitions of GSM endpoints to help address temporal ordering.

In summary, vaginal microbiota are associated with GSM endpoints among postmenopausal women. A CST-IV-B microbial profile (high proportion of G vaginalis with low Lactobacillus) related to presence of most GSM endpoints, most strongly for vaginal atrophy, and Lactobacillus-dominated CSTs related to absence of these endpoints. Although findings suggest a role for vaginal microbiota at the intersection of reproductive and aging biology, longitudinal research on microbial metabolic activity and the host immune environment can help identify mechanisms explaining these findings and may inform intervention. Research on potential microbiota-modulating interventions to alleviate GSM symptoms is needed to determine whether vaginal microbiota impact GSM.

Funding

This work was supported by the National Institutes of Health (grant numbers R01 AG048069, R56 AG068673, R03 AG070178, R01 AI116799, R21 AI107224, R01 CA123467). All authors are supported by grants or contracts from the National Institutes of Health.

Conflict of Interest

J.R. is a cofounder of LUCA Biologics, a biotechnology company focusing on translating microbiome research into live biotherapeutic drugs for women’s health. M.S., P.E.G., A.E.B., and R.M.B. have nothing to disclose.

Author Contributions

M.S., P.E.G., and R.M.B. were responsible for conception and study design. P.E.G., J.R., A.E.B., and R.M.B. were responsible for data acquisition and management. M.S. performed statistical analysis and drafted the manuscript. All authors contributed to data interpretation and edited the manuscript for content; all authors reviewed the manuscript and approved the final version.