Association Between Polycystic Ovary Syndrome and the Vaginal Microbiome

A Case-Control Study

Xiang Hong; Pengfei Qin; Kaiping Huang; Xiaoling Ding; Jun Ma; Yan Xuan; Xiaoyue Zhu; Danhong Peng; Bei Wang

Disclosures

Clin Endocrinol. 2020;93(1):52-60. 

In This Article

Results

The Study Population

In total, 39 PCOS and 40 healthy control women were included. Therein, there were no statistical differences (P > .05) in age and educational level between the two groups. Because most of the women who took part in the prepregnancy health examination programme were attempting their first pregnancy, so the odds of parous women in the PCOS group seemed to be higher than control group, but without statistical significance (P > .05). The proportion of overweight/obese women in the PCOS group was higher than the control group (P = .012). Although the history of vaginitis in PCOS women was more common than the control group (43.6% vs. 15.4%, P = .006), the distribution of vaginal cleanliness grading at the time of sampling between the groups was not statistically significant (P = .102). Equally, we noted no vaginal grade IV cleanliness in both groups. Grade I just existed in control women (Table 1).

The α and β Diversity Between PCOS and Healthy Control Women

All samples were successfully sequenced. Rarefaction curves (Figure S1) showed an obvious slow-rising platform, meaning that sequencing depths were qualified. Microbiome diversity between the two groups was disparate. From Figure 1, the Simpson index for the PCOS group was less than controls (median 0.49 vs. 0.80, P = .008), but the Shannon and Chao1 indices of the PCOS group were greater than controls (Shannon: median 1.07 vs. 0.44, P = .003; Chao1: median 85.12 vs. 66.13, P < .001). After adjusting for the age, BMI and vaginal cleanliness grading, the Chao1 index of PCOS group was still greater than controls (P = .01). This result suggests that the vaginal bacterial species in the PCOS group would be more diverse than the control group. From PCoAs, microbiome communities in the PCOS group were visibly different from the controls, in terms of binary Jaccard and unweighted UniFrac distance (P = .001 for both analyses, Figure 2). The cluster heatmap, based on binary Jaccard distances, revealed distinctions between the PCOS and the control groups (Figure S2). Most control samples were clustered in the middle, but cases were divided into two parts on both sides of the graph. The R 2 was 0.162 for UniFrac distance PCoA, which meant a 16.2% variation in vaginal microbiome could be explained by PCOS in our samples. For weighted UniFrac distances, the β diversity between the two groups was not statistically significant (P = .154) (Figure 2).

Figure 1.

Violin plot for α diversity indices between PCOS and healthy control women. Differences between groups were tested by Mann-Whitney analysis. Thick dotted lines represent medians, and thin dotted lines represent quartiles [Colour figure can be viewed at wileyonlinelibrary.com]

Figure 2.

Principal coordinate analysis for β diversity indices between PCOS and healthy control women. R 2 and P values were derived from PERANOVA tests. (A) Binary Jaccard distance; (B) unweighted UniFrac distance; (C) weighted UniFrac distance

Specific Genus Comparisons Between PCOS and Control Samples

LEfSe analysis showed the different bacteria between groups with LDA threshold score of > 4.0 at different levels (Figure S3). At the genus level, three genera reached the LDA threshold score of > 4.0 (Figure 3A). The average relative abundance of Mycoplasma and Prevotella in PCOS samples was significantly higher than the control group (P < .001 and P = .002, respectively, Figure 3B,C). The prevalence rate of Mycoplasma was 100% in PCOS samples, but 27.5% in control samples.

Figure 3.

LEfSe analyses and a scatter diagram for key species. (A) LEfSe analysis; the LDA threshold was > 4, (B) scatter diagram for Mycoplasma; (C) scatter diagram for Prevotella; the long horizontal lines in the middle represent the medians, and short lines represent quartiles. Differences between groups were tested by Mann-Whitney analysis [Colour figure can be viewed at wileyonlinelibrary.com]

The Relative Abundance of Lactobacillus spp. in PCOS and Control Samples

At the genus level, 72% of vaginal microbiome communities in participants were dominated by the Lactobacillus genus (57/79). Although the relative abundance of Lactobacillus in PCOS samples was lower than the control group (LDA > 4, Figure 3A), the difference was very small in the scatter diagram (Figure 4A). At the species level, the average relative abundance of L. crispatus in PCOS samples was statistically lower than the control group (P = .001, Figure 4C). After adjusting for the age, BMI and vaginal cleanliness grading, the difference was still significant (P = .03). Although the relative abundance of L. iners in PCOS group seemed to be higher than control group, no statistical differences were observed (P = .842, Figure 4B). The relative abundance of L. gasseri was not significantly different between groups (P = .797, Figure 4D).

Figure 4.

Scatter diagram for Lactobacillus genus and specific species. Differences between groups were tested by Mann-Whitney analysis. (A) Lactobacillus genus; (B) L iners; (C) L crispatus; (D) L gasseri [Colour figure can be viewed at wileyonlinelibrary.com]

43% of vaginal microbiome was CST III (34/79), then type V (21/79) and finally type I (17/79). In total, the distribution of CST between groups was not statistically significant (P = .159, Table 2), but the proportion of CST type I in PCOS samples was significantly much lower than control samples (χ 2 = 5.79, P = .016).

Exploration of Diagnostic Bacterial Biomarkers for PCOS

Mycoplasma was the most distinguishing genus in terms of PCOS distinction. If the positive carrier of Mycoplasma was used to judge the PCOS, sensitivity was 100% (39/39), but specificity was 72.5% (29/40), and the positive predictive value was 78% (39/50). Using ROC analysis, the area under the curve (AUC) for Mycoplasma relative abundance was 0.958 (95% CI: 0.901–0.999), and the optimal cut-off value was 0.02% (Figure S4A). This meant that women, with more than 0.02% Mycoplasma relative abundance in the vaginal microbiome, were very likely to have PCOS.

Similarly, we analysed the distinguishing effects for PCOS of L. crispatus and Prevotella. The AUC was 0.706 (95% CI: 0.588–0.825) and 0.698 (95% CI: 0.582–0.814), respectively, and the cut-off values were 13% and 0.25%, respectively (Figure S4B and S4C). These data suggested that if the relative abundance of L. crispatus was < 13%, or the relative abundance of Prevotella was > 0.25%, the women would be very likely to have PCOS.

Sensitivity Analysis

We did four subgroup analyses based on the BMI, vaginal cleanliness and hyperandrogenism status at the baseline. Among the women with normal BMI (19 PCOS and 20 controls) and the women with vaginal cleanliness grade II (30 PCOS and 22 controls), the Chao1 indices of PCOS group were all greater than control (P < .001), and binary Jaccard distances between groups were significantly different (P < .001). The PCOS women with hyperandrogenism had a less Simpson index and greater Shannon and Chao1 indices than healthy controls (P values were all < .05). Meanwhile, PCOS women without hyperandrogenism had a greater Chao1 index than controls (P = .002); details can be seen at Table S5 and Table S6. These results suggested that the BMI, vaginal cleanliness and hyperandrogenism could not explain the variance between PCOS and controls. In other words, if we control these factors, the vaginal microbiome was still different between groups. According to the recommendations of International PCOS Network in 2018,[28] we changed the polycystic ovary morphology definition as there were 20 or more follicles in each ovary, then 2 females could not meet the diagnostic criteria. After excluding them, the analysis for the residual samples (37 PCOS vs. 40 controls) had similar results.

Comments

3090D553-9492-4563-8681-AD288FA52ACE

processing....