Bacterial vaginosis (BV) is a highly prevalent condition associated with adverse health outcomes. Gram stain analysis of vaginal fluid is the standard for confirming the diagnosis of BV, wherein abundances of key bacterial morphotypes are assessed. These Lactobacillus, Gardnerella, Bacteroides, and Mobiluncus morphotypes were originally linked to particular bacterial species through cultivation studies, but no studies have systematically investigated associations between uncultivated bacteria detected by molecular methods and Gram stain findings. In this study, 16S-rRNA PCR/pyrosequencing was used to examine associations between vaginal bacteria and bacterial morphotypes in 220 women with and without BV. Species-specific quantitative PCR (qPCR) and fluorescence in Situ hybridization (FISH) methods were used to document concentrations of two bacteria with curved rod morphologies: Mobiluncus and the fastidious BV-associated bacterium-1 (BVAB1). Rank abundance of vaginal bacteria in samples with evidence of curved gram-negative rods showed that BVAB1 was dominant (26.1%), while Mobiluncus was rare (0.2% of sequence reads). BVAB1 sequence reads were associated with Mobiluncus morphotypes (p<0.001). Among women with curved rods, mean concentration of BVAB1 DNA was 2 log units greater than Mobiluncus (p<0.001) using species-specific quantitative PCR. FISH analyses revealed that mean number of BVAB1 cells was 2 log units greater than Mobiluncus cells in women with highest Nugent score (p<0.001). Prevotella and Porphyromonas spp. were significantly associated with the “Bacteroides morphotype,” whereas Bacteroides species were rare. Gram-negative rods designated Mobiluncus morphotypes on Gram stain are more likely BVAB1. These findings provide a clearer picture of the bacteria associated with morphotypes on vaginal Gram stain.
Citation: Srinivasan S, Morgan MT, Liu C, Matsen FA, Hoffman NG, et al. (2013) More Than Meets the Eye: Associations of Vaginal Bacteria with Gram Stain Morphotypes Using Molecular Phylogenetic Analysis. PLoS ONE 8(10): e78633. doi:10.1371/journal.pone.0078633
Editor: Adam J. Ratner, Columbia University, United States of America
Received: August 29, 2013; Accepted: September 4, 2013; Published: October 24, 2013
Copyright: © 2013 Srinivasan et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The project was funded with grants R01 AI061628 to Dr. DNF from National Institute of Allergy and Infectious Diseases (NIAID) (www3.niaid.nih.gov), division of National Institutes of Health (NIH) and R01 HG005966-01 to Dr. DNF and Dr. FAM (co-PIs) from National Human Genome Research Institute (NHGRI) (www.genome.gov), division of NIH. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
Bacterial vaginosis (BV) is a dysbiotic condition found in up to 29% of women in the United States when using Gram stain to define BV . The type of vaginal bacterial community a woman harbors can have implications for her health and her newborn infant. BV has been consistently associated with serious reproductive and health sequelae such as increased risk for preterm birth , and sexually transmitted diseases including HIV acquisition  and transmission . Although BV responds to antibiotics, recurrence rates are high . Clinically, BV is often diagnosed using Amsel’s criteria, a set of four signs or laboratory observations . The gold standard for diagnosis of BV in research settings is Gram staining of vaginal fluid with microscopic evaluation of bacterial morphologies and abundances using well validated scoring systems [7,8]. No studies have systematically examined associations between vaginal bacteria described using high-resolution molecular methods and bacterial morphologies seen on Gram stain.
The Gram stain approach for BV diagnosis defined by Nugent et al. evaluates bacterial morphologies and abundances using a standardized weighted scoring system resulting in scores from 0-10 . Scores of 0-3 denote healthy microbiota, with the presence of Gram-positive rods designated Lactobacillus morphotypes. Scores of 4-6 denote intermediate microbiota, and 7-10 BV-like microbiota. Scores of 7-8 indicate increased abundance of Gram-negative or Gram-variable rods or coccobacilli, designated Gardnerella and/or Bacteroides morphotypes. Women with Nugent scores 9-10 have detection of curved Gram-negative rods designated Mobiluncus morphotypes. In this scheme, bacterial morphotypes serve as surrogates for the representation of putative bacterial species present in vaginal fluid. Typically, bacteria need to be present at 105 CFU/mL for detection by Gram stain . Although diagnosis of BV by Nugent score is not a point-of-care approach due to the need for a highly trained microscopist with experience reading vaginal smears, it has the advantages of simple sample collection, efficient storage and transport, a standardized scale for interpretation, and reliability . Recent molecular methods have revealed novel, uncultivated bacteria in women with BV [10-13]. We ask whether bacteria other than the ones previously described (Lactobacillus, Gardnerella, Bacteroides and Mobiluncus) contribute to bacterial morphotypes observed by Gram stain.
Fluorescence in situ hybridization (FISH) of vaginal fluid has demonstrated at least two bacteria have curved rod morphologies; one designated “BV-associated bacterium-1” (BVAB1), and the other Mobiluncus . This observation prompted us to ask whether some curved rod morphotypes seen in Gram stains of vaginal fluid are BVAB1 rather than Mobiluncus species. Given the widespread use of the Gram stain technique to describe the vaginal microbiota and to diagnose BV [14-16], it is imperative to understand bacteria associated with each morphotype, so that misattribution does not lead to false conclusions about the role of particular bacterial species in BV. For example, Gram stain results have been used in several studies to assess BV treatment outcomes and relapse [14,15,17]. Nyirjesy et al. found that a clindamycin-based regimen was more effective in reducing the abundance of Mobiluncus morphotypes, which also correlated with a higher BV cure rate . In another study of bacterial interactions in the vagina, in vitro susceptibility tests were conducted on Bacteroides fragilis and this bacterium was found not susceptible to the antibacterial activity of the lactobacilli tested . Cultivation-independent methods have detected few Bacteroides or Mobiluncus in women with BV, both key Gram stain morphotypes [10,19]. In this study, we systematically investigated associations between vaginal bacteria described by PCR/high-resolution phylogenetic analysis, and bacterial morphotypes observed by Gram stain.
Vaginal samples were collected using Protocol #1789 which was approved by the Institutional Review Board (IRB) at Fred Hutchinson Cancer Research Center (IR# 5485). All study participants provided written informed consent prior to enrollment in the study. Consent forms were also approved by the IRB as part of Protocol #1789.
Study population and sample collection
The study population comprised 220 women seen at the Public Health, Seattle and King County Sexually Transmitted Diseases Clinic (STD Clinic) between September 2006 and June 2010 . Women were eligible if they were of reproductive age, not pregnant and could provide informed consent. Vaginal fluid samples were collected for Gram stain, microscopy with saline and potassium hydroxide preparations, pH, and testing for STDs and other vaginal infections. Vaginal samples for molecular studies were collected using polyurethane foam swabs (Epicentre Biotechnologies, Madison,WI) brushed against the lateral vaginal wall, and stored at -80°C. BV was diagnosed for immediate management using Amsel’s criteria , and confirmed by Gram stain using the Nugent method . Ninety-eight women (43%) had BV by Amsel’s criteria and 117 by Gram stain (53%).
DNA extraction, quantification and qPCR
DNA from vaginal swabs was extracted using the Ultra-Clean Soil DNA Kit or the Bacteremia Kit (Mobio, Carlsbad,CA) which gave similar results. Sham swabs without human contact were included to assess contamination from extraction reagents or collection swabs. Total bacterial load (16S-rRNA gene copies/sample) was evaluated by broad-range quantitative PCR (qPCR) with Escherichia coli plasmid standards ranging from 107 to 10 gene copies for each reaction . Concentrations of BVAB1 and Mobiluncus species (Mobiluncus curtisii and Mobiluncus mulieris) were assessed by bacterium-specific qPCR assays [20,21]
Broad-range PCR and pyrosequencing of 16S rRNA gene amplicons
We performed broad-range 16S-rRNA gene PCR with pyrosequencing using 454 Life Sciences FLX technology (Roche, Branford,CT) targeting the V3-V4 region of the 16S-rRNA gene . Sequence reads (426,602 reads) were classified using a phylogenetic placement tool pplacer  and a curated reference set of key vaginal bacteria . Species level classification was achieved for 98·5% of reads. All sequence reads were deposited in the NCBI Short Read Archive (SRA051298) .
Fluorescence in situ hybridization
Vaginal fluid smears on glass slides were fixed in 95% ethanol. Nonspecific binding was blocked using 2% sheared salmon sperm DNA in hybridization buffer without probe for 2h at 45°C. Hybridization buffer contained 5XSET (0·75M NaCl, 5mM EDTA, 0·1M Tris, pH 7·8), 10% dextran, 0·2% BSA, 0·1mg/mL polyadenosine, and 0·02% SDS. After blocking, slides were immersed in hybridization buffer with 1% sheared salmon sperm DNA and bacterium-specific probes (200ng/100µL hybridization buffer) and incubated overnight at 45°C. Probes included: broad-range-Eub-338-Cy5 (5'-GCTGCCTCCCGTAGGAGT-Cy5-3'), BVAB1-132-Fl (5'-CTGCTATCCCCCCGGTACAGG-Fl-3'), and Mobil-126-Cy3 (5'-TCCCAAAGAAAAGGACAGGTTACTC-Cy3-3'). Cells were also stained with 4',6-diamidino-2-phenylindole (DAPI) which binds to DNA. Post-hybridization, slides were washed, air dried in the dark. Bacteria were visualized using epifluorescence microscopy with a 100X oil immersion objective. Cells from 3-6 representative high-powered fields per smear were enumerated (ImageJ software) . M. curtisii and M. mulieris cultures were used as control bacteria.
Heat-fixed smears on slides were flooded with crystal violet for 60s, washed with tap water; flooded with iodine mordant for 60s, washed with tap water. Decolorizer comprised equal volumes of 100% ethanol and acetone. Cells were counterstained with safranin for 60s, washed with tap water. Bacteria were visualized under 100X oil immersion using bright field microscopy and enumerated using the Nugent method for BV diagnosis .
The frequency matrix of sequence reads in each sample were clustered using the Dirichlet Multinomial Mixture (DMM) model  to evaluate overall associations between bacterial morphotypes and sequence reads. The model accommodates differences in the total number of reads per sample and the sparse taxonomic distribution, where few taxa are very abundant and many taxa are rare. DMM models were fit to the frequency matrix, with the number of distinct mixture components chosen to minimize the Laplace approximation of the model evidence. Results are displayed as a heat map, showing samples assigned to their maximum mixture component. Taxa are ordered by contribution to variation between assigned components, then grouped by genus .
Relationship between bacterial taxa and morphotypes was modeled using a zero-inflated generalized linear model . This approach models samples with zero counts separately from samples with non-zero counts, and is appropriate for our data because number of samples with zero counts is inflated relative to the expectation obtained from samples with non-zero counts.
Association of Gram stain morphotypes with bacterial taxa
Classified 16S-rRNA gene sequence reads from vaginal samples clustered into four groups using the Dirichlet Multinomial Mixture model (DMM) (Figure 1). Cluster-I included vaginal samples from women whose bacterial communities were similar to those observed in BV. Morphotypes by Gram stain in Cluster-I included Gardnerella and Bacteroides, with Lactobacillus morphotypes absent. Cluster-II was dominated by L. iners reads and a high abundance of Lactobacillus morphotypes. Cluster-III contained mixed microbiota by broad-range PCR/pyrosequencing, with L. iners and low abundance of BV-associated bacteria reads including G. vaginalis and Prevotella bivia; most of these samples contained Gardnerella and Bacteroides morphotypes by Gram stain (49/55). G. vaginalis reads were present in 53/55 samples (Rank abundance median, 6·6%) and Bacteroides reads were either absent or rare. Although L. iners reads were present in 47/55 samples in Cluster-III (Rank abundance median, 17·7%), only 16/55 samples were noted to have Lactobacillus morphotypes. Cluster-IV was dominated with L. crispatus by broad-range PCR/pyrosequencing; all samples had Lactobacillus morphotypes.
Classification and clustering of bacterial sequence reads from 220 women resulted in four clusters. Transformed frequencies of sequence reads are shown in the legend with darker colors representing more abundant sequence reads. The abundance of bacterial morphotypes, assessed by Gram stain, are reflected as numbers ranging from 0 to 4 with higher numbers indicating increased abundance. Nugent scores used for diagnosis of BV are presented as follows: BV negative (0-3), Intermediate score (4-6) and BV positive (7-10). White bars denote that data are not available. The model was not informed of bacterial morphotype data. Taxa are ordered based on contribution to variance between clusters, and then are grouped by genus-level classifications. The heat map includes taxa whose total abundance across samples exceeds the 80th quantile. Note absence of Mobiluncus species in the list of taxa presented due to low abundance of sequence reads.
Sequence reads classified as BVAB1 were associated with Mobiluncus morphotypes on Gram stain (Figure 2, Table 1). Most strikingly, Mobiluncus curtisii sequence reads were not associated with presence of Mobiluncus morphotypes. Although there was a significant association between presence of Mobiluncus mulieris sequences and Mobiluncus morphotypes, this species was sparsely represented; only 9/220 samples contained M. mulieris reads at 0·18% median relative abundance, and median values were not different between high morphotype abundance and low abundance groups (Figure 2). In contrast, BVAB1 reads were present in 47 samples (21·4%) with a relative abundance of 0·02%-94% (Median, 8·8%). Rank abundance plots of vaginal bacteria in women with curved rods by Gram stain (Nugent scores 9-10) showed that BVAB1 was the dominant bacterium (26·1%) while relative abundance of M. curtisii and M. mulieris was 0·12% and 0·09% respectively (Figure 3). Presence of Lactobacillus reads was significantly associated with Lactobacillus morphotypes, as expected (Figure 2, Table 1). L. crispatus and L. jensenii reads were not detected in 70% and 82% of women whose samples had low abundance of Lactobacillus morphotypes and this was statistically significant (Table 1). In contrast, L. iners reads were absent in only 13% of women whose samples had low abundance of Lactobacillus morphotypes, suggesting that while L. iners can contribute to Lactobacillus morphotypes, it may be associated with other Gram stain morphotypes or present at lower concentrations. Samples with Gardnerella and Bacteroides morphotypes (scored separately in our analysis) were highly concordant (99·5%) (Figure 1). Although presence of Bacteroides reads was significantly associated with presence of Bacteroides morphotypes, Bacteroides reads were infrequent with a relative abundance of 0·009% in women with BV, hence unlikely to contribute to Bacteroides morphotypes seen in Gram stains. In contrast, Prevotella and Porphyromonas sequence reads were highly associated with Bacteroides morphotypes, and were more abundant (Figure 2, Table 1).
Based on a priori hypothesis and abundance of taxa represented in DMM clustering, bacterial sequence reads (asinh-transformed) obtained using broad-range PCR and pyrosequencing (Y-axis) were correlated with abundance of Gram stain morphotypes (X-axis). The inverse hyperbolic sin (asinh) is a log-like transformation with better (linear) behavior near zero. This makes it applicable to count data with many zeros, as is found in our data. Data are grouped as <=2 (0, 1+, 2+) and >2 (3+, 4+) (X-axis) which indicate average number of bacterial morphotypes observed under oil-immersion per high-powered field. 0, no morphotypes present; 1+, <1 morphotype present; 2+, 1 to 4 morphotypes present; 3+, 5 to 30 morphotypes present; 4+, 30 or more morphotypes present. Open blue circles denote abundance of sequence reads from a single woman; closed black circles indicate median values. Numbers below box plots denote numbers of samples in each group.
|Proportion of zero counts1||<=22||>22||P-value3|
|Average of non-zero counts||<=2||>2||P-value|
|Proportion of zero counts||<=2||>2||P-value|
|Average of non-zero counts||<=2||>2||P-value|
|Proportion of zero counts||<=2||>2||P-value|
|Average of non-zero counts||<=2||>2||P-value|
|Proportion of zero counts||<=2||>2||P-value|
|Average of non-zero counts||<=2||>2||P-value|
Comparison of rank abundance plots in women with Nugent scores of 7-8 (A) and 9-10 (B) show percentage of sequence reads in each group. Women with Nugent scores 9-10 are dominated with BVAB1. Quantitative PCR targeting BVAB1 and Mobiluncus (C) was performed for vaginal samples obtained from women with BV demonstrating higher concentrations of BVAB1 compared to Mobiluncus in each group.
Quantitative PCR targeting BVAB1 and Mobiluncus
To ensure that we could reliably detect and measure concentrations of BVAB1 and Mobiluncus, we performed taxon-directed qPCR in all women with BV. Among women with Nugent scores 9-10, mean concentration of BVAB1 DNA was 2-log units greater than Mobiluncus (Figure 3C) (p<0·0001). Among women with Nugent scores 7-8, Mobiluncus and BVAB1 were detected by qPCR in 41% and 19%, with mean concentrations of 9·2×105 and 5·6×107 copies respectively, although no curved rods were detected by Gram staining. Comparison of results from broad-range PCR/pyrosequencing and qPCR showed that there was 94% concordance for BVAB1, but only 52% for Mobiluncus. Quantitative PCR can detect low levels of Mobiluncus DNA (1·9×103 copies/swab, lowest level detected), which falls below the detection limit of the broad-range PCR/pyrosequencing approach.
FISH targeting BVAB1 and Mobiluncus
We visualized BVAB1 and Mobiluncus by FISH in all samples with Nugent scores of 10 to assess relative quantities using a third approach. In 16/17 samples with swabs available, mean numbers of BVAB1 cells from representative images from each sample were significantly higher than Mobiluncus cells (p<0·0001, Figures 4, 5). In 1/17 samples, bacterial cells were visible using DAPI staining, but no fluorescence was obtained with BVAB1 or Mobiluncus probes. However, there was also a lack of fluorescence with Eub338 (broad-range probe), suggesting there was degradation of RNA and resulting loss of fluorescence in this sample.
Vaginal fluid smears from all women with Nugent 10 were assessed for relative quantities of BVAB1 and Mobiluncus cells. The lines in the box plot represent the mean and whiskers denote 95% confidence intervals.
Vaginal fluid smears from two representative study participants are shown. 4+ curved rods were documented by Gram stain (A & C) for both participants who had BV (Nugent score 10). Panels A (Gram stain) and B (FISH) are vaginal fluid smears from a representative participant with low concentrations of Mobiluncus DNA (2·5×105 copies 16S-rRNA gene/swab) and high concentrations of BVAB1 DNA (2·4×109 16S rRNA gene copies/swab). Panel B shows a field of bacteria hybridizing with probes for BVAB1 (green) while no hybridization was observed with Mobiluncus probe (red). Mean quantity of BVAB1 cells was 661 versus <1 Mobiluncus cell. Panels C (Gram stain) and D (FISH) are vaginal fluid smears from a representative participant with high concentrations of Mobiluncus DNA (1·3×107 16S-rRNA gene copies/swab) and BVAB1 DNA (5·1×108 16S-rRNA gene copies/swab). Panel D shows a field of bacteria hybridizing with both Mobiluncus (red) and BVAB1 (green) DNA. Mean quantity of BVAB1 cells was 908 versus 145 Mobiluncus cells.
The gold standard for BV diagnosis or confirmation is Gram staining of vaginal fluid and interpretation with a weighted scoring system based on abundance of bacterial morphotypes [7,8]. Identity of morphotypes has largely been inferred from cultivation-based studies [27,28]. Since the publication of the Nugent scheme in 1991 , several uncultivated bacteria have been shown to be highly specific for BV such as BVAB1 . Here, we systematically examined associations of bacteria detected using high-throughput sequencing methods with individual components of the Nugent score including Lactobacillus, Gardnerella, Bacteroides, and Mobiluncus morphotypes. Our findings suggest that Bacteroides and Mobiluncus morphotypes do not accurately reflect the presence of these bacteria in the human vagina.
DMM clustering resulted in identification of four community clusters based on distribution of 16S rRNA gene sequence reads (Figure 1). High abundance of reads classified as L. crispatus and L. iners were concordant with Lactobacillus morphotypes. The correlation between Lactobacillus morphotypes and lower abundance of L. iners was not apparent, and it may be that cell numbers of L. iners are sufficiently infrequent such that Lactobacillus morphotypes are not reported in the Nugent score. Many women in Cluster-III are diagnosed as having BV by Gram stain despite presence of L. iners, highlighting the fact that BV is not synonymous with absence of all lactobacilli. To address differences in Lactobacillus morphotypes, Verhelst et al. presented a refined Ison and Hay scoring method for Gram stain to distinguish additional grades within the Lactobacillus-dominated microbiota .
A critical distinction between women with Nugent scores of 7-8 and those with scores of 9-10 is presence of curved Gram-negative rods in the latter group. Mobiluncus has been cultivated from vaginal samples with Nugent scores 9-10, leading to the association between Mobiluncus and curved Gram-negative rod morphotypes [27,30]. A striking observation from our broad-range PCR and pyrosequencing data was the high relative abundance of BVAB1 reads in women with Nugent scores 9-10 (26%) compared to 100-fold lower abundance of Mobiluncus reads (0·2%) (Figure 3B). Statistical analysis revealed that curved rod morphotypes were associated with BVAB1 (p<0.001), but not with Mobiluncus. These data suggested the hypothesis that curved Gram-negative rods observed on Gram stains of vaginal fluid and designated “Mobiluncus morphotypes” are likely to be BVAB1.
We tested this hypothesis using independent methods. It is possible that the V3-V4 region of the 16S rRNA gene selected in our study is not optimal for the amplification of Mobiluncus despite complete homology in our primer target sequence. Hummelen et al. characterized vaginal bacterial communities by targeting the V6 region of the 16S rRNA gene . They showed that an uncultured Lachnospiraceae bacterium (100% identical to BVAB1) was present at >1% relative abundance in 73% of women with BV, and Mobiluncus was below the 1% relative abundance threshold. Given that the Hummelen study targeted a different region of the 16S rRNA gene, it is unlikely that amplification bias accounts for low numbers of Mobiluncus sequences in these studies.
We determined if BVAB1 DNA concentrations were higher than Mobiluncus DNA by using species-specific qPCR in all women with BV (Figure 3C). BVAB1 was detected in all women with Nugent scores 9-10, and median concentration of BVAB1 DNA was significantly greater than concentration of Mobiluncus. In contrast, Mobiluncus DNA was detected in 76% of women with Nugent scores of 9-10, and median concentration was 3-log units less compared to BVAB1 DNA. The qPCR data are concordant with the broad-range PCR/pyrosequencing data showing that BVAB1 is more prevalent and found in higher abundance than Mobiluncus in women with Nugent scores 9-10. Our results are substantiated by observations made by Zozaya-Hinchliffe et al. using qPCR, wherein the investigators showed elevated quantities of BVAB1 (>107 copies) in 89% of women with Nugent scores 9-10, while concentrations of Mobiluncus were much lower (101-102 copies); this observation was not emphasized in that study .
Bacteria can have different numbers of rRNA gene operons/genome, and the exact number is unknown for many uncultivated organisms . Mobiluncus curtisii ATCC43063 has two 16S rRNA gene copies per genome . The exact number of 16S rRNA gene copies per genome for BVAB1 is currently not available. One challenge for correlating 16S rRNA gene copies to absolute quantities of bacteria is that bacteria with high copy numbers/genome may be over-represented using PCR-based approaches. To overcome copy number bias, we performed FISH to quantify BVAB1 and Mobiluncus cells in vaginal fluid from women with Nugent scores 10 (Figures 4, 5). Greater numbers of BVAB1 were detected by FISH (p<0·0001) which confirmed our observations from qPCR and broad-range PCR with pyrosequencing. FISH provided direct microscopic evidence that BVAB1 is more abundant than Mobiluncus, and therefore is more likely contributing to curved Gram-negative rods. Furthermore, the abundance of BVAB1 cells on FISH closely matches the abundance of curved Gram-negative rods seen with Gram stain itself (Figure 5). Together, these separate lines of investigation suggest that curved Gram-negative rods seen on Gram stain of vaginal fluid in women with BV are more likely to be BVAB1 than Mobiluncus.
Bacteroides morphotypes were previously linked to Bacteroides species based on cultivation studies . Since then, changes in taxonomy of the Bacteroides fragilis group based on phylogeny and biochemical properties have led to re-classification of many of these bacteria as Prevotella and Porphyromonas. The genus Bacteroides now contains only bile-resistant Gram-negative rods, while the saccharolytic, bile-sensitive pigmented and non-pigmented species were re-classified to the genus Prevotella and the pigmented asaccharolytic species to the genus Porphyromonas [35-38]. Although current evidence suggests that there is no association of the Bacteroides fragilis group with BV , there continue to be studies investigating this link . In our analysis as well, Bacteroides reads were uncommon in women with high abundance of Bacteroides morphotypes (Figure 2). Instead, we found that Prevotella and Porphyromonas were significantly associated with the Bacteroides morphotype (Table 1). Hummelen et al. also did not report high abundance of Bacteroides, but noted high abundance of Prevotella and Porphyromonas in women with BV . Cultivation studies may detect Bacteroides despite being present at low abundance because these bacteria are readily cultivable. Although we scored for Gardnerella and Bacteroides morphotypes separately in our study, it can be difficult to consistently differentiate between them by Gram stain . A limitation in our analysis of associations between bacteria detected by PCR and Gardnerella or Bacteroides morphotypes is this high level of correlation for these two bacterial types, restricting our ability to link sequence reads to one of these morphotypes.
In conclusion, we systematically examined correlations between vaginal bacterial species described by broad-range PCR and high-throughput sequencing with individual components of the Gram stain Nugent score. Curved Gram-negative rods observed by Gram stain are more likely to be the uncultivated bacterium, BVAB1, rather than the widely accepted Mobiluncus species. Prevotella and Porphyromonas are likely major contributors to Bacteroides morphotypes, whereas Bacteroides species are infrequent. Different bacterial communities with similar Gram stain characteristics may impart different risks for complications associated with BV, highlighting the importance of distinguishing between these bacterial communities using modern molecular approaches. In the future, it would be more accurate to list the bacterial morphotypes in the Nugent score as Lactobacillus, Gardnerella/Prevotella/Porphyromonas, and BVAB1/Mobiluncus morphotypes.
We thank Dwyn Dithmer for her clinical contributions. Kathy Thomas and Linda Drolette provided help with clinical data management.
Conceived and designed the experiments: SS JMM DNF. Performed the experiments: CL TLF KJA. Analyzed the data: SS MTM FAM NGH. Contributed reagents/materials/analysis tools: MTM FAM NGH DNF. Wrote the manuscript: SS DNF.
- 1. Koumans EH, Sternberg M, Bruce C, McQuillan G, Kendrick J et al. (2007) The prevalence of bacterial vaginosis in the United States, 2001-2004; associations with symptoms, sexual behaviors, and reproductive health. Sex Transm Dis 34: 864-869. doi:10.1097/OLQ.0b013e318074e565. PubMed: 17621244.
- 2. Hillier SL, Nugent RP, Eschenbach DA, Krohn MA, Gibbs RS et al. (1995) Association between bacterial vaginosis and preterm delivery of a low-birth-weight infant. N Engl J Med 333: 1737-1742. doi:10.1056/NEJM199512283332604. PubMed: 7491137.
- 3. Taha TE, Hoover DR, Dallabetta GA, Kumwenda NI, Mtimavalye LA et al. (1998) Bacterial vaginosis and disturbances of vaginal flora: association with increased acquisition of HIV. AIDS 12: 1699-1706. doi:10.1097/00002030-199813000-00019. PubMed: 9764791.
- 4. Cohen CR, Lingappa JR, Baeten JM, Ngayo MO, Spiegel CA et al. (2012) Bacterial vaginosis associated with increased risk of female-to-male HIV-1 transmission: a prospective cohort analysis among African couples. PLOS Med 9: e1001251: Article No.: e1001251. PubMed: 22745608.
- 5. Bradshaw CS, Morton AN, Hocking J, Garland SM, Morris MB et al. (2006) High recurrence rates of bacterial vaginosis over the course of 12 months after oral metronidazole therapy and factors associated with recurrence. J Infect Dis 193: 1478-1486. doi:10.1086/503780. PubMed: 16652274.
- 6. Amsel R, Totten PA, Spiegel CA, Chen KC, Eschenbach D et al. (1983) Nonspecific vaginitis: diagnostic criteria and microbial and epidemiologic associations. Am J Med 74: 14-22. doi:10.1016/0002-9343(83)91112-9. PubMed: 6600371.
- 7. Ison CA, Hay PE (2002) Validation of a simplified grading of Gram stained vaginal smears for use in genitourinary medicine clinics. Sex Transm Infect 78: 413-415. doi:10.1136/sti.78.6.413. PubMed: 12473800.
- 8. Nugent RP, Krohn MA, Hillier SL (1991) Reliability of diagnosing bacterial vaginosis is improved by a standardized method of Gram stain interpretation. J Clin Microbiol 29: 297-301. PubMed: 1706728.
- 9. Bingen E, Lambert-Zechovsky N, Mariani-Kurkdjian P, Doit C, Aujard Y et al. (1990) Bacterial counts in cerebrospinal fluid of children with meningitis. Eur J Clin Microbiol Infect Dis 9: 278-281. doi:10.1007/BF01968060. PubMed: 2112465.
- 10. Fredricks DN, Fiedler TL, Marrazzo JM (2005) Molecular identification of bacteria associated with bacterial vaginosis. N Engl J Med 353: 1899-1911. doi:10.1056/NEJMoa043802. PubMed: 16267321.
- 11. Hummelen R, Fernandes AD, Macklaim JM, Dickson RJ, Changalucha J et al. (2010) Deep sequencing of the vaginal microbiota of women with HIV. PLOS ONE 5: e12078: Article No.: e12078. PubMed: 20711427.
- 12. Ravel J, Gajer P, Abdo Z, Schneider GM, Koenig SS et al. (2010) Microbes and Health Sackler Colloquium: Vaginal microbiome of reproductive-age women. Proc Natl Acad Sci U S A 108(Suppl1): 4680-4687. PubMed: 20534435.
- 13. Srinivasan S, Hoffman NG, Morgan MT, Matsen FA, Fiedler TL et al. (2012) Bacterial communities in women with bacterial vaginosis: high resolution phylogenetic analyses reveal relationships of microbiota to clinical criteria. PLOS ONE 7: e37818: Article No.: e37818. PubMed: 22719852.
- 14. Nyirjesy P, McIntosh MJ, Steinmetz JI, Schumacher RJ, Joffrion JL (2007) The effects of intravaginal clindamycin and metronidazole therapy on vaginal Mobiluncus morphotypes in patients with bacterial vaginosis. Sex Transm Dis 34: 197-202. doi:10.1097/01.olq.0000235152.98601.d7. PubMed: 17179772.
- 15. Schwebke JR, Desmond RA (2007) A randomized trial of the duration of therapy with metronidazole plus or minus azithromycin for treatment of symptomatic bacterial vaginosis. Clin Infect Dis 44: 213-219. doi:10.1086/509577. PubMed: 17173219.
- 16. Thoma ME, Gray RH, Kiwanuka N, Wang M-C, Sewankambo N et al. (2011) The natural history of bacterial vaginosis diagnosed by Gram stain among women in Rakai, Uganda. Sex Transm Dis 38: 1040-1045. doi:10.1097/OLQ.0b013e3182275499. PubMed: 21992981.
- 17. Sobel J, Peipert JF, McGregor JA, Livengood C, Martin M et al. (2001) Efficacy of clindamycin vaginal ovule (3-day treatment) vs. clindamycin vaginal cream (7-day treatment) in bacterial vaginosis. Infect Dis Obstet Gynecol 9: 9-15. doi:10.1155/S1064744901000035. PubMed: 11368263.
- 18. Matu; Matu. MN, Orinda GO, Njagi EN, Cohen CR, Bukusi EAMN, Orinda GO, Njagi ENM, Cohen CR, Bukusi EA (2010) In vitro inhibitory activity of human vaginal lactobacilli against pathogenic bacteria associated with bacterial vaginosis in Kenyan women. Anaerobe 16: 210-215. doi:10.1016/j.anaerobe.2009.11.002. PubMed: 19925874. doi:10.1016/j.anaerobe.2009.11.002. PubMed: 19925874.
- 19. Verhelst R, Verstraelen H, Claeys G, Verschraegen G, Delanghe J et al. (2004) Cloning of 16S rRNA genes amplified from normal and disturbed vaginal microflora suggests a strong association between Atopobium vaginae, Gardnerella vaginalis and bacterial vaginosis. BMC Microbiol 4: 16. doi:10.1186/1471-2180-4-16. PubMed: 15102329.
- 20. Fredricks DN, Fiedler TL, Thomas KK, Mitchell CM, Marrazzo JM (2009) Changes in vaginal bacterial concentrations with intravaginal metronidazole therapy for bacterial vaginosis as assessed by quantitative PCR. J Clin Microbiol 47: 721-726. doi:10.1128/JCM.01384-08. PubMed: 19144794.
- 21. Srinivasan S, Liu C, Mitchell CM, Fiedler TL, Thomas KK et al. (2010) Temporal variability of human vaginal bacteria and relationship with bacterial vaginosis. PLOS ONE 5: e10197. doi:10.1371/journal.pone.0010197. PubMed: 20419168.
- 22. Matsen FA, Kodner RB, Armbrust EV (2010) pplacer: linear time maximum-likelihood and Bayesian phylogenetic placement of sequences onto a fixed reference tree. BMC Bioinformatics 11: 538. doi:10.1186/1471-2105-11-538. PubMed: 21034504.
- 23. Schneider CA, Rasband WS, Eliceiri KW (2012) NIH Image to ImageJ: 25 years of image analysis. Nat Methods 9: 671-675. doi:10.1038/nmeth.2089. PubMed: 22930834.
- 24. Holmes I, Harris K, Quince C (2012) Dirichlet Multinomial Mixtures: Generative models for microbial metagenomics. PLOS ONE 7: e30126. doi:10.1371/journal.pone.0030126. PubMed: 22319561.
- 25. Morgan M (2012) Dirichlet Multinomial: Dirichlet-Multinomial Mixture Model Machine Learning for Microbiome Data. 1.1.0 ed. pp. R package.
- 26. Zeileis A, Kleiber C, Jackman S (2008) Regression models for count data in R. J Stat Softw 27.
- 27. Roberts MC, Hillier SL, Schoenknecht FD, Holmes KK (1985) Comparison of Gram stain DNA probe and culture for the identification of species of Mobiluncus in female genital specimens. J Infect Dis 152: 74-77. doi:10.1093/infdis/152.1.74. PubMed: 2409187.
- 28. Spiegel CA, Amsel R, Holmes KK (1983) Diagnosis of bacterial vaginosis by direct Gram stain of vaginal fluid. J Clin Microbiol 18: 170-177. PubMed: 6193137.
- 29. Verhelst R, Verstraelen H, Claeys G, Verschraegen G, Van Simaey L et al. (2005) Comparison between Gram stain and culture for the characterization of vaginal microflora: Definition of a distinct grade that resembles grade I microflora and revised categorization of grade I microflora. BMC Microbiol 5: 61-. PubMed: 16225680.
- 30. Holst E, Hofmann H, Mårdh PA (1984) Anaerobic curved rods in genital samples of women. Performance of different selective media, comparison of detection by microscopy and culture studies, and recovery from different sampling sites. Scand J Urol Nephrol Suppl 86: 117-124. PubMed: 6399403.
- 31. Zozaya-Hinchliffe M, Lillis R, Martin DH, Ferris MJ (2010) Quantitative PCR assessments of bacterial species in women with and without bacterial vaginosis. J Clin Microbiol 48: 1812-1819. doi:10.1128/JCM.00851-09. PubMed: 20305015.
- 32. Acinas SG, Marcelino LA, Klepac-Ceraj V, Polz MF (2004) Divergence and redundancy of 16S rRNA sequences in genomes with multiple rrn operons. J Bacteriol 186: 2629-2635. doi:10.1128/JB.186.9.2629-2635.2004. PubMed: 15090503.
- 33. Klappenbach JA, Saxman PR, Cole JR, Schmidt TM (2001) rrndb: the Ribosomal RNA Operon Copy Number Database. Nucleic Acids Res 29: 181-184. doi:10.1093/nar/29.1.181. PubMed: 11125085.
- 34. Spiegel CA, Amsel R, Eschenbach D, Schoenknecht F, Holmes KK (1980) Anaerobic bacteria in nonspecific vaginitis. N Engl J Med 303: 601-607. doi:10.1056/NEJM198009113031102. PubMed: 6967562.
- 35. Paster BJ, Dewhirst FE, Olsen I, Fraser GJ (1994) Phylogeny of Bacteroides, Prevotella, and Porphyromonas spp. and related bacteria. J Bacteriol 176: 725-732. PubMed: 8300528.
- 36. Shah HN, Collins DM (1990) Prevotella new genus to include Bacteroides melaninogenicus and related species formerly classified in the genus Bacteroides. Int J Syst Bacteriol 40: 205-208. doi:10.1099/00207713-40-2-205. PubMed: 2223612.
- 37. Shah HN, Collins MD (1988) Proposal for reclassification of Bacteroides asaccharolyticus, Bacteroides gingivalis and Bacteroides endodontalis in a new genus Porphyromonas. Int J Syst Bacteriol 38: 128-131. doi:10.1099/00207713-38-1-128.
- 38. Shah HN, Collins MD (1989) Proposal to restrict the genus Bacteroides castellani and chalmers to Bacteroides fragilis and closely related species. Int J Syst Bacteriol 39: 85-87. doi:10.1099/00207713-39-1-85.
- 39. Hillier S, Marrazzo JM, Holmes KK (2008) Bacterial vaginosis. In: KK HolmesP-A Sparling. Sexually transmitted diseases. New York: McGraw-Hill. pp. 737-768.