Introduction

The healthy human vagina is host to a Lactobacillus-dominated microbiota, which appears to have a critical role in thwarting colonization or infection by bacteria, viruses, and parasites that have been linked with negative health outcomes [1]. Approximately one in three women in the U.S. have bacterial vaginosis (BV) [2], a condition characterized by few lactobacilli, and an overgrowth of a diverse mixture of fastidious anaerobes. Many epidemiological studies have shown BV to be associated with increased risk of sexually transmitted infections (e.g., gonorrhea, chlamydia, HIV, trichomoniasis), and pelvic inflammatory disease [3–7]. In pregnancy, BV has been associated with infections of the placenta and amniotic fluid and with preterm birth [8–12]. However, despite the clinical implications of BV, little is known about its etiology.

BV is an imbalance of the vaginal microbiota and was defined in the 1980s as a diagnostic entity. In the clinical setting, BV may present asymptomatically, but it is often accompanied by increased pH, thin vaginal discharge, fishy odor, and the presence of “clue” cells, which are vaginal epithelial cells covered in bacteria [13]. Presently, the gold standard for laboratory-based BV diagnosis is the Nugent scoring method [14]. By this method, Gram stained smears of vaginal fluid are quantified for the presence of Gram positive ‘beneficial’ Lactobacillus morphotypes as compared to Gram negative organisms and Gram variable Actinobacteria such as Gardnerella, Atopobium and Mobiluncus. This approach results in an overall score in which 0–3 indicates a “normal” lactobacilli-dominated microbiota, 7–10 indicates BV, and the score of 4–6 has been referred to as an “intermediate microbiota” whose significance is incompletely characterized. Collectively, samples yielding a Nugent score greater than 3, i.e. both intermediate and BV together, have been referred to as “abnormal” vaginal flora or microbiota [7, 15, 16].

It has long been thought that clue cells, one of the clinical features of BV, are the result of an exfoliation of the vaginal epithelium in the presence of BV-associated bacteria [17, 18]. However, until recently there was little quantitative evidence to support this common assertion. Recently, we examined the quantity of epithelial cells present in human clinical vaginal smears and reported that specimens from women with Nugent-defined BV (score 7–10) exhibited significantly more exfoliated epithelial cells than women with a Lactobacillus-dominant microbiota (Nugent scores 0–3)[19].

This previous study had some potential limitations. First, there was no matching performed in the study. Given the higher risk of BV among African-American women, it is possible that race could be a confounder. The previous study also did not examine exfoliation across the entire range of Nugent scores (Nugent-defined intermediate scores of 4 to 6 were not studied). Interestingly, women with intermediate Nugent scores, like those with high Nugent scores, have been shown to have higher risks of sexually transmitted infections [7, 20–22], suggesting that shared features of the microbiology in women with Nugent scores greater than 3 may be responsible for increased susceptibility to secondary infection. Here, we sought to confirm the earlier finding of increased epithelial exfoliation among women with BV and to expand on the previous findings to determine if increased epithelial exfoliation is a shared feature of vaginal microbiomes with Nugent score greater than 3.

We used a matched cross-sectional design in which women defined as BV-positive (Nugent 7–10), BV-negative (Nugent 0–3), and BV intermediate (Nugent 4–6) were matched on the basis of age, race, and lifetime number of sexual partners. The aim of the present study was to quantify the extent of epithelial exfoliation in clinical specimens representing all three Nugent-defined categories.

Materials and methods

Statistical analysis

Categorical variables were compared using either chi-square statistics or Fisher’s exact tests, as appropriate. Potential confounders (listed in Table 1) were evaluated for association with outcome (epithelial exfoliation) and exposure (BV status/Nugent score). Criteria for confounding variables were: significance with both outcome and exposure at the alpha < 0.05 level for inclusion in adjusted regression models; and change in beta for the exposure variable of interest by 10% or more for inclusion in final models. No variables met these criteria, so unadjusted models are presented. Poisson regression models of epithelial counts were used to estimate incidence rate ratios (IRR) for BV status and Nugent score. Linear and Poisson regression models adjusted for clustering were used to factor-in the clustering introduced by matching BV positive specimens to negative and intermediate specimens using cluster-robust standard errors using the vce(cluster) option [24, 25]. These analyses were completed using Stata (version 13.1, Stata Corp, College Station, TX). GraphPad Prism 7.0 software was used to generate Fig 1. In addition to the analyses above, epithelial counts were compared between Nugent defined groups using the Friedman test (D’Agostino & Pearson normality test indicated that epithelial count data from all three groups were not normally distributed), followed by post hoc analysis using Dunn’s multiple comparisons test (analyses were performed in Prism).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Relationship between Nugent score and epithelial exfoliation.

Number of exfoliated epithelial cells by Nugent-defined BV group (negative, intermediate, and positive) is plotted as the harmonic mean of six counts from two observers. “Int.” = intermediate; “ns” = not significant. The central tendencies and error bars shown are medians and interquartile ranges respectively. The Friedman test (P < 0.01) was used to compare differences between groups, followed by Dunn’s multiple comparison post hoc test (** P < 0.01, * P < 0.05).

https://doi.org/10.1371/journal.pone.0177797.g001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Characteristics of women from the Contraceptive CHOICE Project included in matched^* analysis, N = 150.

https://doi.org/10.1371/journal.pone.0177797.t001

Results

Discussion

Many studies of bacterial vaginosis describe the presence of clue cells as evidence of a vaginal epithelial exfoliation response. Here, we tested whether the total number of epithelial cells in vaginal specimens varied as a function of the vaginal microbiota (independent of the presence or numbers of clue cells). In this study, we show a correlation between total epithelial exfoliation (cell counts in vaginal specimens) and BV (women with Nugent scores of 7–10 had higher levels of epithelial cells compared to women with scores of 0–3). In fact, epithelial exfoliation was not specific to BV. Rather, women with Nugent scores greater than or equal to 4 had higher levels of epithelial cells in vaginal specimens compared to BV-negative women. We emphasize that our measures of exfoliation were not limited to epithelial “clue” cells; rather, all epithelial cells were counted to estimate the extent of exfoliation in relation to Nugent criteria.

The fact that BV-positive and BV-intermediate were phenotypically and statistically indistinguishable with respect to epithelial exfoliation is consistent with a role for Gram negative bacteria and Actinobacteria in BV pathophysiology. Typically, the difference between women who score as intermediate (4–6), versus those who have Nugent-defined BV (7–10) lies in the amount of lactobacilli present. Whereas intermediate women have some lactobacilli, women with BV have many fewer or absent lactobacilli. On the other hand, BV and intermediate women typically share the common feature of higher levels of bacteria that are phenotypically Gram negative or that vary morphologically from the distinct large rod-shaped lactobacilli. Thus, the presence of the exfoliation in Nugent intermediate women argues against the absence of lactobacilli as a driver of this phenotype. The finding that Nugent-defined BV and intermediate samples share an exfoliation response is consistent with a causal role for these organisms, whose overgrowth has been associated with BV.

Indeed, this interpretation parallels our previous findings in a mouse model, showing that G. vaginalis can elicit an exfoliation response of the vaginal epithelium [19]. In this previous model, we showed that infected animals exhibited a significantly higher number of epithelial cells in vaginal washes compared to mock infected animals. The animal model also exhibited several other clinical features of BV, including vaginal bacterial sialidase activity and vaginal mucus degradation.

Our analysis is based on stored specimens. We did not assess the association of exfoliation and Amsel’s criteria as these data were not readily available for analysis. Since clue cells have been defined according to their presence at the point of care in wet mount microscopy, which must be performed on fresh vaginal secretions, we were unable to perform a comparison of the amounts of clue cells in relation to the total number of exfoliated cells.

In the current study, epithelial counts and Nugent score were derived from the same Gram-stained slide examined at a different magnification. We acknowledge the possibility that BV could result in inherent differences in the type and/or quantity of biological material collected by vaginal swabbing, or that procedure of Gram-staining could influence epithelial cells differently because of inherent biological differences in samples from women with and without BV. However, as noted in the Materials and Methods, swabs take up a limited volume far below the total volume of vaginal fluid available. Studies by previous researchers estimate that women produce vaginal fluid at volumes between 4.5 to 7.9 mL per day [26], whereas swabs have a limited capacity at least an order of magnitude below this. Previous studies have shown a very good agreement between Nugent scores and Nugent-based BV diagnosis when comparing self-collected versus provider-collected samples, including studies involving pregnant [27] and nonpregnant [28, 29] women. Self-collected vaginal swabs also result in nearly identical 16S-rDNA-based bacterial community profiles as physician collected swabs [30]. Taken together, these findings argue that (1) uniform use of the same swabs (Starplex) across all women in the study should result in the collection of similar amounts of biological material, and (2) that the biological material sampled during self-collection of vaginal swabs is indistinguishable from clinician-collected material in terms of Nugent score and microbial community profiles. We also acknowledge that self-collection of vaginal swabs may limit the study by introducing more variation between samples, for example if some participants swab more rigorously than others. However, this is a concern that could also apply to contexts where multiple clinicians collect material from participants.

In the present study, possible confounding factors might include variables such as smoking status, which was previously shown to be associated with increased cervicovaginal epithelial exfoliation [31, 32]. Other studies have identified smoking as a risk factor for BV [33, 34]. Our study sample contained relatively few (20%) current smokers. Nonetheless, tobacco use and other potential confounders (listed in Table 1) were evaluated for association with outcome (epithelial exfoliation) and exposure (BV status/Nugent score). None of these variables achieved significance for both outcome and exposure at the alpha < 0.05 level (see S1 Table).

Another potential weakness of our study is a relatively low racial diversity of our participants, with 76% being black. As noted in the Materials and methods section, the demographic characteristics of the CHOICE study were representative of the St. Louis region (approximately half black and half white). Considering the fact that black women are differentially impacted by BV, they were over-represented in this study. Future studies would need to assess potential differences in exfoliation among different demographic groups from diverse geographic locations.

The relationship between exfoliation and intermediate Nugent score is a new finding with potential clinical implications. Recent studies using modern sequencing technology have measured the levels of different bacteria implicated in BV and compared these data across Nugent scores. These studies demonstrate that women with intermediate Nugent scores also tend to have high levels of BV bacteria measured by quantitative molecular methods [35, 36]. The finding of greater exfoliation in women with intermediate Nugent scores may explain previous findings that these women, like their BV-positive counterparts, are at increased risk of acquiring sexually transmitted infections (trichomoniasis, HIV, and gonorrhea) compared to women with Nugent scores 0–3[7, 20–22]. Our data suggest that epithelial exfoliation may be a host response to elevated numbers of BV bacteria, or the results of the action of bacterial virulence factors on epithelial cells, even if women do not meet the full Nugent criteria for BV (i.e. very low numbers of lactobacilli). Results of several previous studies suggest that certain detergents used in personal lubricants (e.g. nonoxynol-9) stimulate rapid epithelial exfoliation, a finding that may underlie the failure of such products to protect their users from HIV infection, despite their ability to disrupt viral membranes [37–40]. Further studies should explore whether epithelial exfoliation exposes niches in which sexually transmitted pathogens more easily gain or maintain residence in the reproductive tract.

Supporting information

Acknowledgments

We thank Deborah Frank for editorial advice (science writer employed by Washington University Department of Obstetrics and Gynecology) and Jennifer Bick for technical assistance with Nugent scoring (employed by CHOICE). We also gratefully acknowledge the support staff of the Contraceptive CHOICE project and the participants for generously sharing the samples used in this study.