Introduction

Bacterial vaginosis (BV) is an enigmatic condition and despite over 50 years of medical research, its aetiology and pathogenesis remains unknown. It is the commonest cause of vaginal discharge globally [1], [2], is associated with adverse pregnancy outcomes [3], [4], [5], [6] and enhancement of heterosexual HIV transmission [7]. There is debate as to whether BV is caused by a single bacterial agent or whether it has a polymicrobial aetiology. Gardnerella vaginalis has been the bacterium most intensively studied in relation to BV; however it is not specific for BV and is commonly present in women with normal vaginal flora although usually in lower numbers than in those with BV [8]. Recent studies have shown that a biofilm, predominantly comprised of G. vaginalis and Atopobium vaginae is associated with BV [9], and that the presence of these two organisms together in high copy numbers have high sensitivity (95%) and specificity (99%) in predicting BV [10]. However, the advent of molecular screening has also identified a number of fastidious, primarily anaerobic bacteria, in association with BV [11], [12], [13], [14], [15], [16], [17], including the Clostridia-like bacteria vaginosis-associated bacteria (BVAB) 1, 2 and 3, the closely related lactic-acid producing Leptotrichia and Sneathia genera, and the uncultivated Megasphaera-like phylotype (termed type I) [13]. In particular, Fredricks found that while no one bacterial vaginosis-candidate organism (BV-CO) was universally present in BV, the presence of either BVAB2 or Megasphaera type I had a sensitivity of 100% and specificity of 91.3% in predicting BV. The causal role of these BV-COs remains to be established and there are no published reports on their behavioural associations and whether they are sexually transmitted.

Observational evidence shows a strong association between sexual activity and BV [18], [19], [20], [21], [22]. In a recent meta-analysis BV was associated with recognized STI risk factors such as multiple sexual partners and condoms were protective against BV [21]. We have recently reported that BV is absent in women with no history of sexual activity with another individual, importantly this study considered both coital and non-coital sexual activity in their classification of sexual exposure [22], [23]. There are data to also suggest transmission of BV may occur between women [19], [21], [24], [25], [26], [27]. Women-who-have-sex-with-women (WSW) have consistently demonstrated higher BV prevalence rates than exclusive heterosexually active women, and consistent with the concept of WSW transmission, studies of female-female monogamous couples demonstrate high concordance for BV (73–95%) [24], [28]. However, there is some evidence against the sexual transmission of BV. Studies testing the impact of male partner treatment have failed to consistently reduce BV recurrence in women [29], [30], [31], [32], and some studies have identified BV in women who have not engaged in penile-vaginal sex [22], [33], [34].

Using samples from two published studies that collected detailed behavioural data and included women with a broad range of sexual experience including sexually-naive women, we sought to examine the relative strength of associations between the aforementioned key BV-COs and BV status established by Nugent's score (NS), and to establish the behavioural characteristics associated with each BV-CO.

Methods

Results

Overall, 339 specimens were analysed; 193 from FUSS and 146 from the clinic cohort. Among the study population, 233 women had NF and 106 women had BV. Table S1 describes the demographic and behavioural characteristics of the study populations. As expected, more clinic-based women were older, more sexually-experienced, more likely to smoke cigarettes (p<0.001), but they were less likely report sex toy use (p = 0.02) than FUSS participants. The study populations were similar with respect to other reported practices including oral contraception use and douching.

Discussion

Our paper explores the association between eight key BV-COs with BV and examines their sexual risk factors; Table S6 summarizes our major findings. Seven of the eight BV-COs demonstrate a strong unadjusted association with BV, and confirm other studies' findings [8]–[14]. Four of the eight BV-COs (Megasphaera type I, BVAB2, A. vaginae and G. vaginalis) were significantly associated with BV after logistic regression analysis, while BVAB3, Sneathia and Leptotrichia spp. were not, and BVAB1 was too uncommon to meaningfully comment on. Six of the eight BV-COs demonstrated significant associations with increasing sexual exposure (Megasphaera type I, BVAB2, BVAB3, Sneathia spp., Leptotrichia spp. and G. vaginalis), and only two BV-COs, A. vaginae and G. vaginalis, were commonly detected in women with no sexual experience. Our findings suggest that there may be a gradient of importance among the eight BV-COs based on their strength of association with BV and their epidemiological association with increasing sexual exposure. The finding that the majority of BV-COs were absent or rare in sexually-unexposed women and associated with increasing sexual exposure suggest sexual transmission of BV-COs is occurring.

For agents to be causative of BV one would anticipate that they would be absent in NF and universally present in BV; no single BV-CO in our study satisfied both these criteria. We have shown, however, that Megasphaera type I and BVAB2 were both strongly associated with BV and were also rare or absent in sexually-naive women. The suggestion that Megasphaera type I and BVAB2 are key BV-COs is supported by Fredricks et al. who showed that the combination of BVAB2 and/or Megasphaera type I conferred the best BV diagnostic predictability for BV (sensitivity 98.8%, specificity 88.5%) [15].

The other BV-COs which were significantly associated with BV were G. vaginalis and A. vaginae, bacteria known to be integrally related to BV disease; however these BV-COs were also detected at lower loads in women with NF (data not shown). G. vaginalis has long been implicated in the development of BV and recent published findings have suggested that G. vaginalis biofilms may be critical in BV pathogenesis and symptomatology [42], [43]. Our results demonstrating that G. vaginalis was ubiquitous in women with BV, and further analysis indicating a 6–7 log₁₀ greater bacterial load in BV than NF (data not shown), are consistent with historical findings [10], [44]. Our data also show that G. vaginalis was relatively common in truly virginal women (prevalence 28%); however the prevalence of G. vaginalis increased dramatically with numbers of sexual partners, indicating that sexual activity or transmission contributes to increasing detection of G. vaginalis. Studies using enzyme assays to identify G. vaginalis biotypes and genomic comparisons of G. vaginalis isolates suggest particular strains of G. vaginalis may have differing pathogenic properties, such as biofilm formation [45]. Whether differing biotypes of G. vaginalis are present in sexually-inexperienced compared to experienced women is not known and would be of considerable interest.

A. vaginae was independently associated with BV and whilst also commonly seen in those with NF (62%), it was generally at low loads (<4 log₁₀, data not shown). The relatively high prevalence of A. vaginae in NF samples was similar to that of Menard et al [10] at 69%, but significantly higher than a previous publication that detected A. vaginae in 12% of NF [46]. A key difference is that in the current study, our assay used primers described by Fredricks et al. [40] which target a 81-bp region of the 16S rRNA gene, whilst the assay in the earlier publication targeted a 430-bp region [47]. Generally, the longer amplicon assay displays relatively reduced sensitivity, and the difference in sensitivities between these two assays is discussed in detail by Menard [10]. A. vaginae, like G. vaginalis, was detected in sexually-inactive women, but unlike G. vaginalis, A. vaginae showed no significant association with increasing sexual exposure.

Importantly, our data seem to indicate that Leptotrichia and Sneathia spp. are not associated with BV once adjusted for the presence of other BV-COs, and their strong relationship with increasing sexual risk was seen similarly in women with NF and BV. This could be important, because if a BV-CO is sexually transmitted but not associated with disease causation, then one would expect to see increasing prevalence with increasing sexual risk equally in both BV and NF. This was the case for Sneathia and Leptotrichia spp. but not for any of the other BV-COs. Collectively these findings suggest that Sneathia and Leptotrichia spp. could just be epidemiologically associated with BV, or sexually transmitted “passengers” rather than being involved in the development of BV; a similar scenario to that of some Ureaplasma and Mycoplasma spp. in NGU [48]. However, as Sneathia and Leptotrichia spp. were detected in 74 and 73 percent of BV cases in our study respectively, their role in symptom/disease pathogenesis, especially in a polymicrobial disease context, cannot be excluded. Future work involving the exploration of metabolic interactions between BV-COs will assist in further understanding the contribution of these organisms to the development of BV.

Our data show that Megasphaera type I may have distinctive behavioural risk factors. Although we only had 12% of participants with a female partner in the last 12 months, these women demonstrated a strong association with Megasphaera type I detection. It is important to note that WSW have greater BV population prevalence than other women [21], [24], [27], [49], [50] and the high prevalence of BV in WSW has to date remained unexplained. We note with interest that Fredricks et al. [15] found Megasphaera type I detection was highly sensitive for BV (94.5; 95%CI 86.7–97.9) in a population of women with a history of sexual contact with another woman. In our study Megasphaera type I was 91.3% (95%CI 70.4–98.4) sensitive for BV in WSW, but only 72.2% (95%CI 61.2–81.2) sensitive for BV in non-WSW. It is possible that Megasphaera type I may be an important BV-CO in the development of BV in WSW, and that WSW with BV differ in their microbiological BV profile to other women. Clearly larger prospective studies are needed to better understand this relationship, but an association between Megasphaera type I and WSW to our knowledge has not been previously described.

This study has several limitations. We used a study population that combined samples from two previous studies collected at different time points. One group was a symptomatic clinic-based population; the other a lower risk university population. By combining these populations investigators were able to obtain a uniquely diverse population of women who ranged from sexually-inexperienced to highly sexually-experienced, and we considered this necessary to be able to comprehensively address the contribution of BV-COs to sexual activity. Prior published studies investigating the association of BV-COs with BV have not contained women with limited or no sexual exposure. The fact that these populations substantially differ in their risk profile, however, is also limitation of the study. We have adjusted for lifetime sexual partner number in the multivariable analyses to control for the key differences in sexual exposure in the study groups, and although we describe known differences in the two study populations not all differences between the groups may not be accounted for. Questions were asked in a standardized way in both studies to allow these data to be aggregated where necessary, and self-collected paper-based questionnaires were used for both studies with the option of an identical self-completed online questionnaire for FUSS (40% of FUSS participants used online surveys). Self-sampling was employed in FUSS and clinician-collected samples were used in the clinic study; published studies confirm equivalency of self-collected to clinician-collected samples for the diagnosis of BV, and excellent agreement has been reported between self-collected versus clinician-collected swabs for molecular quantification (qPCR) of BV-associated bacteria [37], [38]. Specimens were stored in the same way in both studies as whole samples at −80°C, and we confirmed the integrity of stored samples in both studies.

Due to funding limitations, our use of molecular methods to characterise the vaginal bacterial communities was limited to a representative panel of eight of the most promising and specific BV-COs, acknowledging there is an increasing number of BV-COs being cited in the literature, particular from next-generation sequencing projects [51], [52]. Further research should include a greater number of BV-COs and in particular more Megasphaera spp. including the type II phylotype. Our data was cross-sectional, therefore we have no capacity to comment on temporal relationships between the detection of BV-COs and sexual behaviour, and we can only make inferences regarding BV causality and pathogenesis. Due to the complexity of the current analyses in this manuscript, we also did not include possible metabolic interactions between species in this study, and have limited ourselves to exploring the associations between individual BV-COs and behaviour and BV disease status. Given the polymicrobial nature of BV, assessing for metabolic interactions is the subject of further research by investigators. Another limitation of cross-sectional data is that fluctuations in vaginal flora are known to occur rapidly and to be influenced by various factors including hormonal variation and day-to-day behaviours [53]., clearly longitudinal studies with frequent sample collection are required to further understand the complex relationship between BV-COs and development of BV. As with all BV research, the lack of a gold standard test hampers our ability to explore specific disease associations. We excluded any women with intermediate vaginal flora (NS4-6) in an attempt to minimize the possible misclassification of flora caused by a cross-sectional ‘snapshot’ design in an actively changing environment. It is important to also note that this data cannot be used to determine the population prevalence of any of the BV-COs as the sampling method was not appropriate for this.

The strengths of our study design include the systematic objective assessment using Nugent's method of all vaginal smears by an experienced microbiologist and 10% being re-scored by an alternative microscopist. Not all studies examining the role of BV-COs in BV have applied methods such as the Nugent method to classify flora objectively and some have used more subjective clinical descriptions and/or patient history [12], [14], [16]. Another important strength of our research is that highly detailed behavioural data was collected and the FUSS study was quite unique, including women with a broad-range of sexual experience, in particular a considerable number with no reported sexual exposure to another person or limited non-coital exposure only. Previous BV study populations that have examined BV-COs have recruited women from STD clinics or women from high prevalence populations. The inclusion of sexually-naive women in our study, in addition to comprehensive, confidential sexual behaviour data from participants, has enabled us to explore specific sexual practices and their associations to individual BV-COs in considerable depth.

Supporting Information

Acknowledgments

We would like to acknowledge Leonie Horvath and Irene Petalotis from the MSHC microbiology laboratory. Jun Kit Sze and Afrizal from IT, Lisa Kennedy and Carol Hopkins for helping with FUSS procedures and Marcus Chen and Anna Ralph for reviewing the manuscript.