Characterization of Microbial Communities Found in the Human Vagina by Analysis of Terminal Restriction Fragment Length Polymorphisms of 16S rRNA Genes

Construction of vaginal model communities.

Sixteen clinical isolates (Table 1) were obtained from A. Onderdonk (Harvard Medical School), cultivated on various media, and used to prepare cell suspensions that were subsequently combined to produce model vaginal communities. Obligate anaerobes were cultivated on prereduced brucella blood agar with 5% sheep blood, 0.001% (wt/wt) hemin, and vitamin K₁ (Becton Dickinson, Franklin Lakes, NJ). Prior to inoculation, the medium was prereduced by overnight incubation in an anaerobic chamber with an atmosphere of 80% N₂, 10% CO₂, and 10% H₂. All manipulations were carried out in the same anaerobic chamber. Afterwards, the plates were sealed in anaerobic jars and incubated at 37°C. Aerobic or facultatively anaerobic bacterial strains were cultured either on brain heart infusion agar or in brain heart infusion broth and incubated over night at 37°C.

Model communities were prepared in 3 ml of liquid dental transport medium (pH 7.2) (Anaerobe Systems, Morgan City, CA) by adding the appropriate amounts of cell suspensions needed to give the final numbers indicated in Table 1.The numbers of bacteria in cell suspensions of each strain were enumerated by counting DAPI (4′,6′-diamidino-2-phenylindole)-stained cells (41). Model community A consisted of an equal number of cells (10⁷ cells ml⁻¹) of each of the 16 strains, whereas in model community B the abundance of each strain approximated that found in the vaginas of healthy women.

Clinical study.

Study protocol and informed consent documents were reviewed and approved by The Procter & Gamble Corporate Institutional Review Board. Informed consent was obtained from all subjects prior to participation in the study. The study population consisted of five premenopausal, nonpregnant, white women between the ages of 28 and 43 years who were in good general health. They were recruited for the study via advertisement and screened for study eligibility. To be included in the study, the women had to have regular menstrual cycles, be willing to provide informed consent, and refrain from douching during the entire course of the study, bathing within 2 h of the clinic visits, and having sexual intercourse for 48 h prior to study visits. They also agreed to exclusively use Olay moisturizing bar soap for cleansing needs during the study and Always pads (Procter and Gamble, Cincinnati, OH) during menstruation. Women were excluded from the study if they had used antibiotics (orally or by topical application in vulvar/vaginal area) during the prior 6 weeks or if they were immunocompromised (self-reported); were pregnant (self-reported); had body piercings in the vulvar, thigh, or buttock areas; or had a sexually transmitted disease, AIDS, or hepatitis (self-reported). All women completed all five of the study visits. All submitted vaginal swabs were acceptable for further vaginal microbial community analyses.

After enrollment (visit 1), subjects completed a total of four additional visits. Enrolled subjects were given a supply of Always pads (Procter and Gamble, Cincinnati, OH), with instructions to use the pads exclusively while menstruating. This was done in order to standardize the menstrual habits and practices of subjects while they were enrolled in the study. Two of the visits coincided with days 20 to 24 of the menstrual cycle. The other two visits coincided with days 2 to 5 of the menstrual cycle. Each study subject was interviewed before each of the four study visits to ensure that all of the inclusion and none of the exclusion criteria were met. A vaginal swab was obtained from each study participant. Vaginal swab samples were obtained from the mid-vagina by using a speculum lubricated with sterile saline and a swab (Pur-Wraps Harwood Medical Company). Swabs were immediately placed into 3 ml liquid dental transport medium (Anaerobe Systems, Morgan City, CA) and shipped on the same day to A. Onderdonk at Harvard Medical School, Boston, MA. Cells collected on the swab were suspended by vortexing. A small amount of each cell suspension was removed for studies done in his laboratory (data not shown), and the remainder was archived at −80°C. Batches of samples were shipped on dry ice to the University of Groningen and delivered within 24 h, where they were stored at −80°C. Experiments showed that samples shipped to the University of Groningen via Harvard University were no different than those obtained from a replicate sample that was shipped directly to the University of Groningen (data not shown). Once all the samples had been gathered, genomic DNAs were isolated as described below.

Genomic DNA extraction.

Genomic DNA was isolated from 0.5-ml aliquots of the cell suspensions, using a two-step cell lysis procedure. First, bacterial cell walls were disrupted enzymatically by the addition of mutanolysin (50 μg) and lysozyme (500 μg), followed by incubation for 1 h at 37°C. Second, the cells were mechanically disrupted by six freeze-thaw cycles. Each cycle consisted of 2 min of incubation at 100°C, which was immediately followed by 2 min in liquid nitrogen. Between each freeze-thaw cycle, the cell suspensions were incubated for 1 min in an ultrasonic bath. Proteins in the disrupted cell suspension were digested with proteinase K (QIAGEN, Hilden, Germany) during 1 h of incubation at 55°C. Further isolation and purification of the total DNA extract was performed using the Wizard DNA purification kit (Promega, Madison, WI).

T-RFLP analyses.

To prepare samples for terminal restriction fragment length polymorphism (T-RFLP) analyses, regions of the 16S rRNA genes in each sample were amplified using combinations of four fluorescently labeled primers: 8f (5′-AGA GTT TGA TCC TGG CTC AG-3′) (27), 341f (5′-CCT ACG GGA GGC AGC AG-3′) (27), 926r (5′-CCG TCA ATT CCT TTR AGT TT-3′) (1), and 1406r (5′-ACG-GGC GGT GTG TRC-3′) (27). Reaction mixtures for PCR contained 50 ng of genomic DNA, 5 μl of 10× buffer (500 mM KCl; 100 mM Tris-HCl, pH 9.0; and 15 mM MgCl₂) (Amersham Biosciences, Piscataway, NJ), bovine serum albumin (20 μg), each deoxynucleoside triphosphate at a concentration of 200 μM (Amersham Biosciences), each primer at a concentration of 0.4 μM, and 1 U of Taq polymerase (Amersham Biosciences) in a final volume of 50 μl. If PCR products were used for subsequent T-RFLP analysis, the forward primers were labeled with 5-carboxy-fluorescein at the 5′ termini, and the reverse primers were labeled with tetrachlorofluorescein at the 5′ termini (Eurogentec, Seraing, Belgium). The same primers without fluorescent labels were used for PCRs to generate target DNA for subsequent cycle sequencing reactions as described below. DNA amplification was performed with a Geneamp9700 thermocycler (Perkin-Elmer, Norwalk, CT), using the following program: a 5-min initial denaturation at 94°C, followed by 30 cycles consisting of denaturation (1 min at 94°C), primer annealing (1 min at 49.5°C for the primer combination 341f-926r, and 1 min at 55°C for the primer combinations 8f-926 and 8f-1406r), and primer extension (2 min at 72°C). A final extension was performed at 72°C for 10 min. Salts, nucleotides, and primers were removed from PCR products by using Qiaquick PCR purification kits (QIAGEN). Amplification of DNA was verified by electrophoresis of each PCR product in 1.5% agarose in 1× Tris-acetate-EDTA buffer, followed by staining with ethidium bromide and visualization under UV illumination.

Reaction mixtures for the enzymatic digestion of amplified rRNA genes contained 100 ng of PCR product, 1× restriction buffer, 20 μg of bovine serum albumin, and 10 units of restriction enzyme. The mix was adjusted to a final volume of 20 μl with water, and the DNA was digested at 37°C for 3 h. The restriction enzymes used to evaluate model microbial communities were AluI, HhaI, HaeIII, RsaI, MspI, and HinfI (all from Amersham Pharmacia Biotech, Uppsala, Sweden) and MvnI (Roche Applied Science, Indianapolis, IN) with buffers recommended by the manufacturers.

For analysis of terminal restriction fragment length polymorphisms, 1 μl of digested PCR product was mixed with 0.5 μl internal size standards (Tamra 2500; ABI) and deionized formamide. After 3 min of denaturation at 95°C, the lengths of the various T-RFs were analyzed using an ABI 310 Prism automated sequencer (ABI). The sizes of T-RFs are designated by the color used in the electropherogram traces, i.e., green (G) for 5′ T-RFs and blue (B) for 3′ T-RFs, followed by the size of the fragment in base pairs.

DNA sequencing and analysis.

The sequences of the 16S rRNA genes (positions 8 to 926) of reference strains were determined. Each sequencing reaction mixture contained 4 μl of 5× sequencing buffer, 2 μl of Ready Reaction Mix (Applied Biosystems Inc., Foster City, CA), 20 ng of template DNA, and a final concentration of 0.2 μM of primer. Sterile water was added to a final volume of 20 μl. Each cycle sequencing reaction consisted of 25 cycles, and each cycle included a melting step at 96°C for 10 s, followed by primer annealing at 50°C for 5 s and extension at 60°C for 4 min. Prior to sequence analysis, the products were purified using the isopropanol precipitation method as described by the manufacturer. Sequence data were collected using an ABI Prism 310 Genetic Analyzer and analyzed using the AutoAssembler version 2.0 software package (Applied Biosystems Inc.). The 16S rRNA gene sequences obtained were matched with all sequences presently available from the databases of the Ribosomal Database Project (32) and GenBank (www.ncbi.nlm.gov) to identify their closest relatives.

Construction and analysis of 16S rRNA gene clone libraries.

To construct 16S rRNA gene clone libraries, 3 μl of cleaned PCR product was cloned in a TOPO TA vector (Invitrogen, San Diego, CA), using the method recommended by the manufacturer. Competent cells of Escherichia coli (One Shot E. coli; Invitrogen, Inc., San Diego, CA) were transformed with ligated plasmids, and 50 μl of each transformation mixture was spread onto Luria-Bertani (LB) agar plates that contained X-Gal (5-bromo-4-chloro-3-indolyl-β-d-galactopyranoside), IPTG (isopropyl-β-d-thiogalactopyranoside), and 50 μg/ml kanamycin. After incubation overnight at 37°C, 100 white colonies were picked and inoculated into 5-ml aliquots of LB broth containing 50 μg/ml kanamycin. After overnight incubation, the cells were harvested from each culture and plasmid DNAs were extracted. The 16S rRNA gene inserts were individually amplified by PCR using the conditions described above and subjected to terminal restriction fragment analysis as described above. Clones yielding T-RFs that corresponded to those in the T-RFLP profile were sequenced, and the data were analyzed as described above. Ten clones representing each T-RF were sequenced, and all were sequenced if there were fewer than 10 clones for a given T-RF.

Optimization of T-RFLP.

An optimal strategy for characterizing microbial communities found in the human vagina by T-RFLP would employ the fewest combinations of PCR primers and restriction enzymes needed to resolve all the populations likely to be encountered in samples from healthy women. The optimal combination of primers and restriction enzyme(s) would be that which produces 5′ or 3′ T-RFs that differ in size, thus providing a means to resolve all members of the community. Pairs of four commonly used bacterial primers, namely, 8f, 341f, 926r, and 1406r, were used. Seven restriction enzymes that are frequently used in T-RFLP analyses (13, 31) were evaluated; these were AluI, HhaI, HaeIII, MvnI, RsaI, MspI, and HinfI. Combinations of primers and restriction enzymes were tested on axenic cell suspensions of 16 bacterial strains commonly found in the vaginal tracts of healthy women. Primers 8f and 926r were superior to 341f and 1406r when used in combination with HaeIII, HhaI, and MspI, which were equally effective in their ability to resolve populations found in the vagina. The predicted sizes of 5′ and 3′ HaeIII fragments from vaginal bacteria are shown in Table 2.

We opted to use the 8f-926r primer pair for PCR, followed by digestion with HaeIII, for the analysis of model communities and clinical samples. All of the strains tested could be resolved to at least the genus level. For Peptostreptococcus, three species could be distinguished based on difference in the sizes of both 5′ and 3′ fragments. In contrast, Staphylococcus and Corynebacterium could not be resolved at the species level. The predicted 5′ T-RF from Lactobacillus crispatus (245 bp) differed from that of the other two Lactobacillus spp. (327 bp), allowing resolution of this species. The predicted sizes of the 3′ T-RFs of Lactobacillus sp. strains 1 and 2 differed by 1 bp, and thus in theory these two strains could be resolved, since current technology provides resolution of DNA fragments that are ±1 bp in size.

Model vaginal microbial communities.

Model communities (Table 1) were used to determine whether each member of the community could be detected by T-RFLP when the populations were present in equal numbers (community A) or when the abundance of each population was comparable to that found in the vaginas of healthy women and lactobacilli outnumbered other species by at least 2 orders of magnitude (community B). Both model communities were constructed by addition of known amounts of cells that had been quantified bydirect microscopic counts of DAPI-stained cells, rather than by addition of known amounts of DNA. Analyses done on communities constructed on the basis of equal amounts of DNA could be misleading, since genome sizes and 16S rRNA gene copy number vary among bacterial species (15, 16). Furthermore, various factors (including cell wall composition and architecture) can markedly affect the efficiency of cell lysis. For these reasons, it was preferable to construct the model communities by addition of known amounts of intact cells.

According to the T-RFLP profile of model community A (Fig. 1A), strains of Corynebacterium (T-RF pair, G65/B23), Lactobacillus crispatus (G245/B459), Lactobacillus sp. strains 1 and 2 (G327/B458 or B459), Peptostreptococcus harei (G272/B582), Staphylococcus (G310/B613), Streptococcus agalactiae (G311/B87), and Enterococcus faecalis (G288/B592) could be resolved based on differences in the sizes of their 5′ and/or 3′ T-RFs. This was consistent with the outcome predicted based on the 16S rRNA gene sequences (Table 2). For unknown reasons, the 3′ T-RFs from Peptostreptococcus vaginalis and Peptostreptococcus magnus were absent from the profile. Digestion of PCR products from the 16S rRNA genes of Peptostreptococcus vaginalis and Peptostreptococcus magnus resulted in the 210-bp-long 5′ T-RF; however, the 3′ T-RFs were missing. The variation in the quantity (peak area) of the various T-RFs as well as the absence of T-RFs derived from both Prevotella species could be explained by differences in the efficiency of cell lysis or 16S rRNA gene copy number between various bacterial strains (15, 16), preferential amplification of specific 16S rRNA genes, or other unknown causes (26, 50).

Only the T-RFs of the Lactobacillus species were evident in the T-RFLP profile of model community B (Fig. 1B). The total number of Lactobacillus cells (8.2 × 10⁹) in model community B exceeded that of other species by at least 2 orders of magnitude. In experiments in which DNA from 10⁸ cells of Lactobacillus strain 1 was mixed with DNA from a known number of cells from Staphylococcus epidermidis strain 1, the limit of detection of the latter organism was roughly 1% of the total number of organisms (data not shown). This was in agreement with experiments done by Dunbar (10), who found that only populations that constituted between 0.1 and 1% of a bacterial community could be detected in T-RFLP profiles. Thus, it was expected that T-RFs from strains other than lactobacilli would be absent from the T-RFLP profile of model community B.

T-RFLP of vaginal communities.

Samples taken from the left and right sides of the vagina of each woman were used to evaluate the methodology and to compare the outcomes of T-RFLP analyses with those obtained from sequences from 16S rRNA clone libraries prepared from the same samples. The T-RFLP profiles of swab samples taken from the left and right sides of the vagina (Fig. 2) were virtually the same, suggesting that the methodology was reproducible. The simplicity of the profiles was consistent with the notion that comparatively few populations of bacteria dominate vaginal communities. Sequencing and phylogenetic analyses of 16S rRNA clones prepared from the same samples showed that these samples had high numbers of populations closely related to Lactobacillus crispatus (T-RF pair G245/B459) and three other species (T-RF pairs G287/B590, G287/B459, and G335/B590) and that all of the fragments in the T-RFLP profile could be accounted for in the library (data not shown). The three populations were only distantly related to Lactobacillus acidophilus and Lactobacillus johnsonii and were phylogenetically distinct from L. crispatus, Lactobacillus jensenii, and Lactobacillus gasseri, three species of Lactobacillus known to commonly occur in the human vagina (2, 24, 40). The lack of sequence similarity to known species of Lactobacillus suggests these may be phylotypes that were not previously known to reside in the human vagina.

Comparisons of vaginal communities among women.

The T-RFLP profiles of microbial communities in different women and within an individual over a 2-month period of time were compared. Differences and similarities among women were readily apparent (Fig. 3). The vaginal communities of woman 2 and woman 3 had identical simple profiles that contained a single 5′ fragment and a single 3′ fragment, with sizes of 244 and 452 bp, respectively. This suggests that these communities had a limited number of numerically dominant populations, and perhaps only a single population. The profiles of woman 4 and woman 1 differed from each other and from those of women 2 and 3. Like for women 2 and 3, a single 5′ fragment and a single 3′ fragment, with sizes of 220 and 179 bp, respectively, dominated the T-RFLP profile of woman 1. The community of woman 4 had at least two phylogenetically distinct populations, which yielded two 5′ T-RFs that were 286 and 331 bp and a single 3′ T-RF that was 588 bp. Comparisons of the 5′ and 3′ T-RFs found in the communities of women 2, 3, and 4 to those predicted from in silico analyses of known 16S rRNA gene sequences indicated that the numerically dominant populations in these women were species of Lactobacillus. This is in contrast to the results of comparable analyses done on the numerically dominant population(s) present in the vaginal community of woman 1. None of the 16S rRNA gene sequences from species of Lactobacillus would yield the 220-bp 5′ fragment or a 179-bp 3′ fragment. Analyses done using algorithms available on the Microbial Community Analysis website (http://mica.ibest.uidaho.edu/) suggest that these fragments might be produced by Atopobium fossor, which would be expected to yield 221-bp 5′ and 180-bp 3′ fragments, while data obtained by Zhou et al. (55) suggest that the population could be Atopobium vaginae, which would produce 220-bp 5′ and 179-bp 3′ fragments. Of course, the possibility that it is some as-yet-uncharacterized phylotype cannot be excluded. Interestingly, the community of woman 5 appeared to be a composite of those found in women 1 and 4 and contained all the 5′ and 3′ fragments found in these communities. Given this, it appeared that woman 5 also had numerically dominant phylotypes that were not species of Lactobacillus. Analyses of samples collected from each woman four times over a 2-month period indicate that the numerically dominant populations in these communities remained remarkably constant (Fig. 4) even though the samples were collected from each woman during menses as well as at other times during their menstrual cycles, from the left and right sides of the vagina, and with scrapers as well as cotton swabs (data not shown).

Taken in sum, these data suggest the following: (i) there are distinct differences in the identities of the numerically dominant populations, thus suggesting that there is more than one kind of “normal” microbial community found in healthy women; (ii) the structures of the communities (at least as reflected by the most abundant populations) were comparatively invariant over time; and (iii) some communities (women 1 and 5) have phylotypes other than lactobacilli among the dominant populations. Conclusions i and ii are consistent with those of previous studies (3, 19, 30, 53), whereas this represents the first instance where populations other than lactobacilli have been shown to be numerically dominant in the vaginal communities of healthy women. Additional studies are needed to verify this observation, since only a small number of women were included in this study.