Molecular assessment of bacterial vaginosis by Lactobacillus abundance and species diversity

For women with high-risk for STIs, routine screening is offered at the STI outpatient clinic of the Public Health Service Amsterdam (GGD, Amsterdam). The subjects included in this study had either vaginal signs and/or symptoms, were referred by a physician for STI testing or tested because a sexual partner with a proven STI had notified them. Samples were collected in June and July 2012. A standard cervical examination was performed and a cotton swab was used to remove abundant mucus prior to the collection of a sample for Chlamydia trachomatis and Neisseria gonorrhoeae screening. The swab to remove abundant mucus is not normally used for routine testing and is discarded. However, for this study the cotton swab was re-suspended in screw-cap coded tubes with Amies transport media to which additional 15 % glycerol and cysteïne solution had been added. Immediately, the tubes were placed in liquid nitrogen and stored at −80 °C. Samples did not disclose any subject names or other patient identification (date of birth, patient file number etc.), and were sent on dry ice to the Netherlands Organisation for Applied Scientific Research (TNO) for further analysis. We studied the vaginal microbiota of 40 subjects, of which 20 BV-negative and 20 BV-positive, by selection of low (0–3) and high (7–10) Nugent scores, respectively.

STI-screening consisted of HIV status (determined by serology), syphilis (serology and/or nucleic acid amplification test (NAAT), gonorrhea (culture and/or NAAT), Chlamydia (urogenital, anogenital and throat region; NAAT), herpes type 1 and 2 (PCR), Trichomonas (PCR), hepatitis B (serology), Moluscum contagiosum (clinical appearance), scabies (clinical appearance), ulcus (clinical appearance), and pelvic inflammatory disease (PID; clinical appearance).

To determine if women met the criteria for BV or had vulvovaginal candidiasis (VVC), microscopic scoring was performed (potassium hydroxide (KOH) preparation and Gram-stain). A vaginal smear was examined using the Nugent scale [7], which includes the scores 0–3 as Normal; 4–6, Intermediate; and 7–10 for BV. The diagnosis of BV was based on the Nugent Gram stain and the presence of three Amsel criteria [6], characteristic vaginal discharge, clue cells, and positive amine test. Measurement of the pH was not part of the routine screening procedure.

The DNA isolations were performed as described in detail by Zhao and others [15]. Briefly, samples were mixed with 150 μl Agowa lysis buffer BL, 350 μl zirconium beads (0.1 mm; suspended in milli Q-water), and 200 μl phenol and lysed in a BeadBeater (BioSpec Products, Bartlesville) for 2 min. The aqueous phase was collected after spinning (5 min ~9.000 g) and DNA was isolated via Agowa binding beads.

Sequence analysis was performed on a 454 GS-FLX-Titanium Sequencer (Life Sciences (Roche), Branford, CT) as described previously [16]. The amount of bacterial template in the isolated DNA samples was determined with a universal quantitative PCR for 16S rRNA gene [17]. Afterwards, a 16S rRNA gene amplicon library spanning variable regions V5-V7 was generated [16]. FASTA-formatted sequences and corresponding quality scores were extracted from the data file generated by the GS-FLX Titanium sequencer using the GS Amplicon software package (Roche, Branford, CT) and processed using modules from the Mothur v. 1.22.2 software platform [18].

On average 2745 (minimum 96, maximum 5932, standard deviation 1304) sequence reads were generated for the total of 40 amplified DNA fragments. Only one DNA sample (TCMID116) resulted insufficient reads (less than 100) and was excluded from subsequent analyses. Sequences were de-noised using a pseudo-single linkage algorithm with the goal of removing sequences that are likely pyrosequencing errors using the “pre.cluster” command [19]. Potentially chimeric sequences were detected and removed using the “chimera.uchime” command [20]. Fragments were aligned and a consensus was made for identification purposes.

The family, genus and species determination is based on the comparison of the specific probe sequence with sequences derived from type strains in the RDP database [21]. A sequence matched with the RDP database with a homology score of 1 is named after the species, if possible. A homology score <1 is named after the genus or family name. Names of certain species, genera or families which are presented more than once, result from a difference on sequence level. The amplicon sequences, numbers of reads for each sequence, and taxonomic classifications have been presented in Additional file 1.

The Pearson correlation coefficient of counts of all pairs of unique sequences from the 40 samples (585 in total) were calculated (leading to a 585 by 585 matrix of correlation coefficients). Some sequences were highly positively correlated, having correlation coefficients close to or equal to 1, namely those that occurred in only 1 sample, and those that most likely originate from the same species. The complexity of the data was reduced by binning the counts of such highly correlated sequences to one new sequence count variable. The total number of new variables (unique sequences and sets of highly correlated sequences) was 149. The lower boundary used for binning counts of correlated sequences was a Pearson correlation coefficient of 0.95. The resulting data were clustered. Data were normalized by calculating relative numbers per sample. The square root of these values was taken. The Kruskal-Wallis test (a robust variant of one-way ANOVA), was performed on the relation between relative species abundance and the five main sample clusters.

Using random forest analysis, correlations between relative abundances of bacterial sequences and the clinical variables with a moderate variability were examined by calculating a “Node Purity value” (IncNodePurity). The mean increase in node purity is a measure of how each variable, the sequence abundance of a specific species, genus or family, contributes to the classification of the sample by minimizing the residual sum of squares in regression.

Literature research and sequence analysis of a number of vaginal microbiome samples were used for the design of a tailor-made nucleotide-based microarray [8, 22]. Taxonomic selection of vaginal bacterial species for the microarray was expanded based on denaturing gradient gel electrophoresis (DGGE) analysis (data not shown). For each bacterial species represented on the microarray, one or more unique short oligonucleotide sequences from within the 16 s rRNA gene were selected [8, 22]. The microarray data, based on consistent signal to background ratios of fluorescence intensity after hybridization, have been indicated for each selected oligonucleotide probe in Additional file 2.

A PCR-based profiling technique for high-throughput analysis of the microbiome was performed as described by Budding and others [23]. The amplification of the IS-regions was performed with the IS-pro assay (IS-Diagnostics, Amsterdam, the Netherlands). The IS-pro method involves bacterial species differentiation by the length of the 16S-23S rRNA intergenic spacer region with taxonomic classification by phylum-specific fluorescent labeling of PCR primers. The IS-pro procedure includes PCR’s for the phyla Firmicutes, Actinobacteria, Fusobacteria, and Verrucomicrobia (FAFV), Bacteroidetes and Proteobacteria [23, 24]. The IS-profiling was performed on vaginal microbiota samples selected from each of five clusters identified in this study. A total of 15 samples were analyzed by IS-profiling, including representatives of BV-negative cluster I (TCMID 101, 103, 113), and cluster II (TCMID 102, 111, 112), as well as BV-positive cluster III (TCMID 126, 135), cluster IV (TCMID 134, 136), and cluster V (TCMID 117, 118, 120). As a control, IS-pro analyses was performed on a L. crispatus strain isolated from TCMID 103 (BV-negative, Nugent score 0), and on a L. iners strain, isolated from TCMID 134 (BV-positive, Nugent score 10).

Species diversity based on next generation sequencing, microarray and IS-profiling has been calculated by the Gini-Simpson index [25]. The Gini-Simpson index measures the degree a sequence abundance contributes to the classification of the sample. Each microarray probe was scored by counting only positive detection, i.e. having a value above the detection threshold (≥5 signal-to-background ratio). Relative abundance is calculated by selecting all representing probes and calculating the number of percent of a species of a particular kind relative to the total number of species per sample. The read-out of the IS-profiling was analyzed by the mean log2 intensity in Relative Fluorescence Units (RFU) per phylum for BV positive clusters and BV negative clusters. Analysis was conducted using the software environment for statistical computing and graphics R version 3.2.2 [26] and the MeV microarray software suite version 4.9 [27].

A total of 40 cervical swab samples were collected from women older than 18 years of age. The bacterial populations in these samples were analyzed from the first twenty consecutive patients with a BV-positive score (Nugent scale 7–10), and the first 20 consecutive patients with a BV-negative score (Nugent scale 0–3). The latterincluded 18 (45 %) of European or Asian origin and 2 (5 %) of a Hispanic background, with the remainder Caucasian. The BV-positive women comprised 9 (23 %) of European or Asian origin versus 11 (28 %) of a Hispanic background, (Additional file 3). Characteristics of the women and STI-status are shown in Table 1. According to the Amsel criteria, 21 women were positive for BV, with only one discrepancy between the Amsel and Nugent method. Two subjects (5 %) tested positive for VVC. An STI was detected in 10 women (25 %), with C. trachomatis being the most prevalent (18 %). None of the women had HIV, Syphilis, Gonorrhoea, Hepatitis B, Moluscum contagiosum, Scabies, genito-ulcerative disease, or PID.

Analysis of the 16S rRNA amplicon sequencing data (Fig. 1) revealed five distinct types or clusters with two subgroups perhaps within the fifth type. Three clusters were identified in the BV-positive subjects and two in the BV-negative samples (Nugent score ≤ 3). Sample TCMID 121, positive for BV according to the Nugent criteria, was classified as cluster I. Apart from a high abundance of L. iners, there was a high abundance of Prevotella, G. vaginalis, and Sneathia sanguinegens in this sample.

Results of the Kruskal-Wallis test, that determined which species distribution differentiated most significantly between clusters, are shown in Table 2. Clearly, in BV-negative samples, a relatively low species diversity is observed, with the microbiome dominated by either L. crispatus or L. iners. Cluster I was characterized by L. iners at 81 %, while cluster II was dominated by L. crispatus at 79 % with L. iners present at 17 %. Cluster III comprised a group of mainly BV-positive women (5 out of 6), with the vaginal microbiome dominated by G. vaginalis (43 %) and Leptotrichia amnionii (12 %). Sample TCMID 109 scored negative for BV according to the Nugent criteria (score 1), however according to the Amsel score this sample was BV-positive (vaginal discharge clue cells, and positive amine test; in combination with pseudohyphae). The analysis of 16S rRNA amplicon sequences indicated a high abundance of G. vaginalis in this sample, thereby classifying it as cluster III. Sequences of the Family Lachnospiraceae were the most abundant in cluster IV (52 %). Cluster V could be identified as the most diverse cluster, including bacterial species at similar abundance, including Sn. sanguinegens (22 %) and G. vaginalis (15 %).

The correlation between L. crispatus and L. iners abundance across the women is indicated in Fig. 2. In the BV-negative subjects, there was a negative correlation of almost one between L. crispatus and L. iners. These women had a high abundance of L. crispatus or L. iners. All women with BV, and only 37 % of the women without BV were deficient of L. crispatus. The power of various 16S rRNA amplicon sequence based identifiers for the prediction of the Nugent-score is shown using the Node Purity value (Fig. 3). The strongest correlation between sequence abundance and Nugent score was found with sequences belonging to non-lactic acid bacteria, including the Family of the Coriobacteriaceae (including A. vaginae), Leptotrichiaceae (Sn. sanguinegens or L. amnionii), and of Veillonellaceae (including Dialister micraerophilus). The presence of sequences belonging to the Family of the Coriobacteriaceae was the most discriminating, comparing the two BV negative clusters I and II with the three BV positive clusters III, IV and V (see Fig. 3). There was no correlation between the sequence abundances and other clinical variables, including STI status, VVC, age, or symptoms and signs reported by patients. A larger sample size may be needed to detect correlations between these variables and vaginal microbiota composition.

The bacterial species diversity of the various clusters is based on the 16S rRNA amplicon sequences and expressed as the Gini-Simpson index (Fig. 4). In the clusters of BV positive women (cluster III, cluster IV and cluster V) the microbiome showed a higher diversity of species than in the clusters of BV negative women based on the Gini-Simpson index (cluster I and cluster II), although there were a few individual exceptions.

The cluster analysis of microarray results is shown in Fig. 5. The abscissa orders the women on the basis of their cluster defined by the previous cluster analysis of 16S rRNA amplicon sequencing. Firmicutes, mainly Lactobacillaceae, were more present in BV negative than in BV positive clusters. Bacteroidetes abounded in the BV positive clusters. Cluster II, again (Table 2) identified by the presence of L. crispatus, and cluster IV, again identified by the presence of Family Lachnospiraceae, could also be distinguished. A separate cluster III marked by G. vaginalis or Sn. sanguinegens (Table 2) could not be distinguished in the microarray profiles.

In Fig. 6, the mean log2 intensity in RFU is shown per phylum for BV positive clusters and BV negative clusters. Firmicutes, including the Family of Lachnospiraceae, and Lactobacillaceae, was the most diverse in the BV positive clusters. An increased diversity of species in BV positive women as compared to BV negative women was found by the use of IS-profiling. Cluster IV included the samples with the highest diversity index.

We evaluated four molecular indicators to assess BV (Table 3). First, the diversity of the bacterial population, expressed as a number between 0 and 1 by the Gini-Simpson index. Second, the relative abundance of the genus Lactobacillus, and third, the relative abundance of L. crispatus (both expressed as percentage of the total bacterial population). As a fourth additional Lactobacillus marker, we included the relative abundance of L. iners, although this bacterium is known to be associated with the BV-negative as well as the BV-positive vaginal microbiota [28]. Diversity of the bacterial population between BV positive and negative samples was well-discriminated by 16S rRNA amplicon sequencing. The average Gini-Simpson diversity index for the BV-positive samples was 0.70 ± 0.15, while the average of BV-negative samples was 0.06 ± 0.04. Although the microarray and to a lesser extent the IS-pro method were not able to discriminate BV on basis of bacterial diversity, we argue that this result does effect the fidelity of this marker, as these methods easily overestimate species diversity from the inclusion of background signals. The dominance of the genus Lactobacillus could be assessed by the 16S rRNA sequencing method (0.98 ± .15 vs 0.04 ± 0.04 of BV-negative vs BV-positive), but also by both other molecular methods. The presence of L. crispatus can be considered the best indicator for the absence of BV, as this bacterium was completely absent (0.00 %) in all samples, confirmed by all three molecular methods. However, this absence did not provide any indication, as the BV-negative cluster I vaginal microbiota profiles do not contain any L. crispatus. Our observations confirm that although L. iners is in many studies found to be associated with BV, its dominance is also a good indicator for BV-negative samples. The bacterium L. iners occured in BV-associated samples only for relative abundances between 0 % and in an exceptional case 32 % of the bacterial population, depending on the detection method used.

This work confirmed the presence of two clusters of bacterial populations BV negative women, dominated by either Lactobacillus iners or Lactobacillus crispatus and three distinct clusters in the BV positive women. The cluster profiles in BV positive subjects were relatively high in bacterial species diversity and dominated by a variety of anaerobic species. The Gini-Simpson index of species diversity, and the relative abundance of Lactobacillus species appeared consistent indicators for BV. The 16S rRNA amplicon sequencing method was most suitable to assess species diversity, while all three molecular composition profiling methods used in this study were able to indicate Lactobacillus abundance in the vaginal microbiota. An affordable and simple molecular test showing a depletion of the genus Lactobacillus in combination with an increased species diversity of vaginal microbiota could serve as an alternative diagnostic method for the assessment of BV.

The 16S rRNA amplicon sequence analysis of the vaginal microbiome led to the identification of three clusters of microbiome patterns in Dutch women with BV, and two clusters in women without BV. The latter two clusters resemble two of the major community state types (CSTs), found in US subjects who did not have symptomatic BV, which were dominated by L. crispatus and L. iners [28]. However, the present study did not find CSTs of L. gasseri, or L. jensenii, as observed in other studies [29]. All women with BV were deficient in L. crispatus. However, L. iners was detected in various ratios among women with BV, albeit with lower abundance than BV-negative subjects, as previously confirmed [12].

In women without BV, the vaginal microbiome was dominated by L. iners or L. crispatus or a mixture of both. L. crispatus has been regarded as an important marker for health, yet it is not present in all women deemed healthy [30] and in our study particularly not in the women with cluster 1 microbiome (Table 2). It appears that L. crispatus is not solely responsible for maintaining a BV negative state, and it is easily displaced when BV occurs [31]. In BV negative subjects, there was a negative correlation of almost one between L. crispatus and L. iners. Species of the vaginal microbiome associated with BV in this study include the Family of Coriobacteriaceae (including A. vaginae), Leptotrichiaceae (Sn. sanguinegens or L. amnionii) and of Veillonellaceae (including D. micraerophilus). The phyla Bacteroidetes, Actinobacteria, and Fusobacteria, were dominant in BV positive samples. As observed before, unlike the gut [32], the aberrant or disturbed vaginal microbiome is highly diverse, with no systematic occurrence of a single bacterial species shared in the BV-associated microbiota, also evident in our study. Therefore, an overall increased diversity of bacterial species in combination with a depletion of the genus Lactobacillus appear good universal molecular markers for diagnosis of BV, in comparison to Nugent scoring, Amsel test or other commercial methods.

With regard to the amplicon sequencing method, the family, genus and species assignment is based on the comparison of specific V5-V7 16S rRNA sequences with those derived from the type strains present in the RDP database. In some cases the obtained V5-V7 16S rRNA amplicon sequence could not be unambiguously assigned to one species (e.g. the sequence of Sneathia sanguinegens could not be discriminated from that of Leptotrichia amnionii). Although the amplicon sequencing methodology appears most accurate in the read-out of species diversity compared to the other methodologies used in this study, also with this method under- or overrepresentation of certain species can occur.

To study the influence of ethnicity on the vaginal microbiota a larger sample size needs to be assessed. Because of financial restrictions, the sample size could not be expanded in this study. However, the findings on bacterial diversity agree with a recent study of samples from Rwandan women [33]. In the present study only BV positive and BV negative women were selected: none with an intermediate Nugent score. A future study could be undertaken to determine if intermediate scores with high diversity actually fall under a BV diagnosis. All the subjects were at high risk of STIs and for that reason attended the clinic for regular check-up. Equating high risk with exposure to STI pathogens is not easy, so the ability to study samples from these subjects on multiple days would be useful to identify if and when exposure occurs and if having BV influences the outcome. A larger sample size is needed to detect correlations between clinical variables, including STI status, VVC, age, or symptoms and signs reported by patients and vaginal microbiota composition.

For the benefit of clinicians, a simple and inexpensive method is needed to diagnose BV. This could be based on molecular markers identifying women with a vaginal microbiome less resilient to (or at risk for) negative health consequences. A multiplex PCR, which included L. crispatus, L. iners, and G. vaginalis, A. vaginae, and Megasphaera,has been developed for the diagnosis of BV by Kusters and others [34]. Another has used a G. vaginalis, A. vaginae, Lactobacillus genus—qPCR tool and found a sensitivity of 93.4 % and specificity of 83.6 % to diagnose BV [35]. Based on these studies and our data of Dutch women, the bacterial diversity and overall abundance of the genus Lactobacillus could be considered sufficient as molecular markers to determine BV in the majority of subjects. It remains to be seen if the assessment of BV-clusters with a wide variety of species is significant for the clinical practice, given that there is no consensus about their number and composition.

Another major issue in patient management is that currently no treatments have been developed specifically against any BV-cluster. Formation of an epithelial polymicrobial biofilm with G. vaginalis appears to play an important role in BV [36]. Ideally, treatment should be more targeted to destabilize BV biofilms and allow restoration of the subject’s indigenous microbiome associated with health. The failure of industry to develop new treatments leaves patients with sub-optimal care, and with no new therapeutic agents on the horizon, making the best use of current agents, perhaps in combination with (personalized) probiotics, could provide better management of BV for women around the world [37, 38].

The research proposed in this study was evaluated by the ethics review board of the Academic Medical Center (AMC), University of Amsterdam, The Netherlands. According to the review board no additional ethical approval is required for this study, as the samples had been obtained by removal of abundant mucus and are usually discarded as part of the standard procedure for cervical examinations (document reference number W12_086 # 12.17.0104). Clients of the STI clinic are notified that remainders of their samples may be used for scientific research, after anonymization of client clinical data and samples. If clients object, data and samples are discarded. This procedure has been approved by the AMC ethics review board (reference number W15_159 # 15.0193).

Clients of the STI clinic are notified that clinical data based on the remainders of their samples may be used for scientific publication, after anonymization. This procedure has been approved by the AMC ethics review board (reference number W15_159 # 15.0193).

Vaginal sample information and nucleotide-based microarray data (signal to background ratios of fluorescence intensity for each probe) is presented in Additional file 1; the 16S rRNA amplicon sequencing data (DNA sequence, and number of reads) and the family, genus and species identification by the match with the RDP (ribosomal database project) for each 16S rRNA amplicon (V5-V7 region) in Additional file 2 and country of origin of 20 women with BV and 20 women without BV in Additional file 3.