Search tips
Search criteria 


Logo of jcmPermissionsJournals.ASM.orgJournalJCM ArticleJournal InfoAuthorsReviewers
J Clin Microbiol. 2012 July; 50(7): 2321–2329.
PMCID: PMC3405607

Development and Validation of a Semiquantitative, Multitarget PCR Assay for Diagnosis of Bacterial Vaginosis


Quantitative PCR assays were developed for 4 organisms reported previously to be useful positive indicators for the diagnosis of bacterial vaginosis (BV)—Atopobium vaginae, Bacterial Vaginosis-Associated Bacterium 2 (BVAB-2), Gardnerella vaginalis, and Megasphaera-1—and a single organism (Lactobacillus crispatus) that has been implicated as a negative indicator for BV. Vaginal samples (n = 169), classified as positive (n = 108) or negative (n = 61) for BV based on a combination of the Nugent Gram stain score and Amsel clinical criteria, were analyzed for the presence and quantity of each of the marker organisms, and the results were used to construct a semiquantitative, multiplex PCR assay for BV based on detection of 3 positive indicator organisms (A. vaginae, BVAB-2, and Megasphaera-1) and classification of samples using a combinatorial scoring system. The prototype BV PCR assay was then used to analyze the 169-member developmental sample set and, in a prospective, blinded manner, an additional 227 BV-classified vaginal samples (110 BV-positive samples and 117 BV-negative samples). The BV PCR assay demonstrated a sensitivity of 96.7% (202/209), a specificity of 92.2% (153/166), a positive predictive value of 94.0%, and a negative predictive value of 95.6%, with 21 samples (5.3%) classified as indeterminate for BV. This assay provides a reproducible and objective means of evaluating critical components of the vaginal microflora in women with signs and symptoms of vaginitis and is comparable in diagnostic accuracy to the conventional gold standard for diagnosis of BV.


Bacterial vaginosis (BV) is a common medical syndrome that is typically associated with a shift in the vaginal flora from a homogeneous, lactobacillus-dominated state to a heterogeneous state containing a complex population of anaerobic and microaerophilic organisms (6, 13). Symptoms of BV include thin, whitish-gray discharge with an unpleasant odor, and this condition is associated with upper genital tract infections, pelvic inflammatory disease, adverse pregnancy outcomes, and an increased risk of sexually transmitted infections (6, 13, 15, 24). The precise pathophysiology and epidemiology of BV, as well as the optimal medical management of the condition, are far from clear, with much of this lack of understanding occurring as a direct result of the difficulty in establishing a diagnostic standard for the syndrome (3).

Conventional microbiological approaches have only limited utility in evaluating patients for BV. Since the hallmark of the condition is a complex perturbation of the normal vaginal microflora, culture-based identification of single “marker” organisms lacks both sensitivity and specificity. Many putative BV-associated organisms, such as Gardnerella vaginalis, Mobiluncus spp., Mycoplasma hominis, and Bacteroides spp., can comprise variable fractions of the vaginal microflora in women without clinically defined BV, compromising the specificity of culture-based testing (21). In addition, many of the key organisms associated with BV are obligate anaerobes and are either difficult to recover or unrecoverable using conventional culture methods, which makes a true evaluation of vaginal microflora by culture impossible (9, 18).

The syndrome of BV was first characterized using clinical criteria and simple laboratory tests applied to vaginal samples (1). Together, this constellation of evaluations became known as the “Amsel criteria.” A diagnosis of BV requires that at least 3 of 4 Amsel criteria be positive (abnormal gray discharge, pH of >4.5, a positive amine test, and presence of epithelial “clue” cells). Although generally regarded as a relatively specific method for identifying patients with BV, Amsel scoring requires considerable clinical acumen and has been demonstrated to be relatively insensitive (17). A more accurate approach to BV diagnosis was proposed in the early 1990s (17) and involved the use of semiquantitative evaluation of vaginal microflora (0 to 3, normal; 4 to 6, intermediate; and 7 to 10, abnormal) based on observation of different bacterial morphotypes in Gram-stained preparations of vaginal samples. This so-called “Nugent score” (NS) or a simplified variant of it, the “Hay-Ison score” (14), has since become accepted as the de facto gold standard for BV diagnosis (12, 23). Many of the key morphotypes are, however, difficult to differentiate from noncontributory organisms of similar appearance, and interpretation of the slide is inevitably somewhat subjective. In addition, quantitative Gram stain examination is laborious and impractical for routine clinical use, and intermediate scores of uncertain clinical significance are reported in 10 to 25% of samples tested (4, 22).

The use of a variety of DNA-based analysis tools, such as broad-range and quantitative PCR (qPCR), has identified novel bacteria associated with BV while also providing more objective, quantitative measures of bacterial presence (79, 13, 18, 22, 25). It has also enabled a greater awareness of the complexity of microflora alterations underlying BV and provided more probative tools for developing improved diagnostic tests. A number of studies have been published describing the use of quantitative or semiquantitative PCR methodologies for diagnosing BV (2, 5, 8, 10, 11, 16, 19, 20, 22). The precise identities of the marker organisms used in these studies differ, as do the cutoff values described as optimal for differentiating abnormal samples from normal samples, and there is as yet no unified approach to using PCR technology for BV diagnosis. There is, however, a recognition that the use of multivariate analysis of the quantity (as determined by nucleic acid amplification) of a set of BV-associated marker organisms present in vaginal samples represents the best approach to obtaining a truly objective and accurate option for diagnosing women with this condition (3, 5, 11).

The current study describes the development and subsequent validation of a BV PCR construct that builds on the foundational framework established in previous studies. By analyzing the presence and concentrations of a number of well-recognized BV-associated marker organisms, a construct was established that permits molecular diagnosis of BV to be performed in the clinical laboratory setting using a combination of two relatively simple and robust PCR assays. A comparative analysis of the results of testing vaginal samples using this construct with designation of samples based on a combination of Nugent and Amsel criteria is presented.


Patient population.

A total of 402 women presenting for clinical evaluation at either the Sexually Transmitted Diseases Clinic, Jefferson County Department of Public Health (JCDH), Birmingham, AL (n = 299), or the Personal Health Clinic (PHC), University of Alabama—Birmingham, Birmingham, AL (n = 103), between April and October 2011 were enrolled in the study. All enrollees were >18 years of age and had not received antibiotics or used vaginal medications for at least 14 days prior to enrollment. The median age of the participants was 25 years (range, 19 to 67 years); 87.1% (350/402) of enrollees were African-American, 12.7% (50/402) were White non-Hispanic, and 0.2% (1/402) were Asian-American. Evaluations could not be completed for 6 enrollees; thus, results for a total of 396 patients were available for data analysis.

Sample collection.

After informed consent was obtained, a series of vaginal samples were obtained to enable comprehensive evaluation of patients for markers of vaginosis (bacterial, Candida spp., and Trichomonas vaginalis) and to enable testing for Chlamydia trachomatis and Neisseria gonorrhoeae. This sample series consisted of a vaginal swab that was utilized for Gram stain preparation and subsequently placed in an Affirm VPIII transport system (Becton Dickinson, Sparks, MD), 2 ESwab (Copan Diagnostics Inc., Murrieta, CA) collections for culture and confirmatory Gram stain evaluation, and 2 Aptima vaginal swab collections (GenProbe Inc., San Diego, CA) for nucleic acid amplification testing.

Conventional diagnostic assessment.

Vaginal secretions were collected and evaluated in the respective clinics according to the Amsel criteria (1). These criteria form a basis for differentiating BV from other etiologies of vaginosis, and the presence of 3 of the following 4 criteria is considered presumptive evidence of BV: a positive “whiff test” (“fishy” odor from KOH-treated wet-mount material); the presence of clue cells upon microscopic examination; a vaginal pH value greater than 4.5; and thin, whitish, homogeneous discharge. Vaginal samples were also evaluated by quantitative Gram staining at the University of Alabama—Birmingham, as previously described (17). In brief, swabs were rolled across glass microscope slides, air dried, and then fixed in methanol prior to Gram staining. The stained preparations were then examined for specific bacterial morphologies, and Nugent scores (range, 0 to 10) were generated. An NS of 0 to 3 is interpreted as normal or negative for BV, a score of 7 to 10 as abnormal or positive for BV, and a score of 4 to 6 as intermediate for BV. To enable categorization of all samples analyzed in the study as positive or negative for BV, samples with intermediate NSs that met the Amsel criteria for BV were considered BV positive and samples with intermediate NSs that failed to meet the Amsel criteria for BV as BV negative.

Nucleic acid isolation.

Nucleic acid was extracted from vaginal swab suspensions (Aptima collection system) using the MagNA Pure LC Total Nucleic Acid Isolation Kit on MagNA Pure LC instruments (Roche Applied Science, Indianapolis, IN), following the manufacturer's instructions. Prior to extraction, DNA Sample Processing Reagent (DNA-SPR) (EraGen Biosciences, Madison, WI; 5 μl) was added to each specimen. SPR contains a proprietary, extractable DNA target that is used as an internal control in PCR assays to monitor recovery of nucleic acid and elimination of PCR inhibitors through sample preparation. Nucleic acid was eluted in a final volume of 50 μl and either analyzed immediately or stored at less than −20°C and analyzed within 7 days.

qPCR assays.

Primer designs were developed for each of the 5 initial target organisms (Atopobium vaginae, Bacterial Vaginosis-Associated Bacterium 2 [BVAB-2], G. vaginalis, Lactobacillus crispatus, and Megasphaera-1) based on in silico analysis of published 16S rRNA gene sequences (Table 1) and were screened for multiplex compatibility and lack of cross-reactivity. To determine the exclusivity of amplification, a panel consisting of high concentrations of the following organisms was tested: A. vaginae, BVAB-2, G. vaginalis, L. crispatus, Lactobacillus acidophilus, Lactobacillus jensenii, Megasphaera-1, Megasphaera elsdenii, C. trachomatis, Escherichia coli, Mobiluncus curtisii, Mycoplasma genitalium, M. hominis, N. gonorrhoeae, Prevotella melaninogenica, Pseudomonas aeruginosa, Staphylococcus aureus, Streptococcus agalactiae, Treponema pallidum, Ureaplasma urealyticum, Candida albicans, Candida glabrata, Candida krusei, Candida lusitaniae, Candida tropicalis, Trichomonas vaginalis, herpes simplex virus type 1, herpes simplex virus type 2, HIV-1, hepatitis B virus, and hepatitis C virus. Positive results were obtained only for the specific targets of the individual assays (A. vaginae, BVAB-2, G. vaginalis, L. crispatus, and Megasphaera-1). Assays were designed to use the primer-based MultiCode-RTx system (EraGen Biosciences, Madison, WI) for real-time product detection that, by exploiting the physicochemical properties of two unique synthetic nucleotide bases, enables specific amplification of PCR products to be monitored as concentration-dependent decreases in fluorescence (20), with confirmation of amplification of the desired product accomplished postamplification via determination of peak melting temperatures (Tm) of the amplified product. One member of each primer pair thus contained a 2′-deoxy-5-methyl-isocytidine (iC) base coupled to a fluorescein moiety immediately proximal to the 5′ terminus of the molecule. To enable quantitative measurement of the respective analytes, synthetic oligonucleotide Ultramers (Integrated DNA Technologies Inc., Coralville, IA) were used to construct calibrating material. Each Ultramer contained the target sequence of one of the intended analytes, and quantitative analytical data provided by the manufacturer were used as the basis for value assignment (in DNA copies) of these materials. All qPCR assay runs included a set of 3 ultrameric calibrators, and interpolation of crossing threshold (CT) values generated during PCR amplification of vaginal samples into calibration curves enabled the derivation of DNA concentrations per milliliter of sample for each analyte. A total of 5 μl of extracted nucleic acid (corresponding to 20 μl of original Aptima vaginal sample) was utilized for qPCR amplification reactions, and the dynamic range of all qPCR assays was established at 1 × 103 to 1 × 108 copies/ml. Amplification reactions were performed on RotorGene Q instruments (QiaGen Inc., Chatsworth, CA) under the following conditions: initial denaturation for 2 min at 95°C, 50 cycles of amplification (95°C for 5 s, 58°C for 10 s, and 72°C for 20 s (fluorescence was collected during this step), with postamplification melting analysis (60°C to 95°C ramp, 1.0°C per second).

Table 1
Primer sequences used to detect indicated 16S rRNA regions of target organisms

Multiplexed BV PCR assay design and interpretation.

The final BV PCR design consisted of a pair of multiplexed PCRs, BV-1 and BV-2. The BV-1 master mixture contained (in a 20-μl volume) AvFP-BV1/AvRP-BV-1 primers (0.2 μM final concentration), EraGen Internal Control primer set 1 (0.1 μM final concentration; sequences proprietary to EraGen Biosciences; 1 internal control [IC] primer labeled with 6-hexachlorofluorescein [HEX]), Isolution (1×, consisting of PCR amplification buffer, MgCl2, and deoxynucleoside triphosphates [dNTPs] [including 4-[4-(dimethylamino)phenylazo]benzoic acid (DABCYL)-labeled deoxyisoGTP]; a component of the EraGen Biosciences DNA Reaction Kit), and Titanium Taq polymerase (0.5 μl; Clontech Laboratories Inc., Mountain View, CA). The BV-2 master mixture contained (in a 20-μl reaction volume) BvabFP-BV2/BvabRP-BV2 primers (0.2 μM final concentration), MegaFP-BV2/MegaRP-BV2 primers (0.2 μM final concentration), 1× Isolution, and 0.5 μl Titanium Taq polymerase. Following addition of 5 μl of extracted nucleic acid to each reaction mixture (obtained as described in Materials and Methods), amplification was performed using a RotorGene Q instrument. The amplification parameters were as described for the qPCR assays, except that only 40 cycles of amplification were performed prior to the postamplification melting analysis, because of the lack of a requirement for low-end sensitivity based on analysis of qPCR results. Positive samples were identified based on the generation of CT values during amplification and appropriate product peak Tm signatures in postamplification analysis. Median peak Tm values for the respective amplicons were as follows: A. vaginae, 80.4°C; IC, 79.0°C; BVAB-2, 79.7°C; Megasphaera-1, 79.1°C. The peak Tm values for amplicons were required to be within 1°C of values generated by the positive control for the appropriate analyte on each individual assay run for the result to be considered positive. Each BV PCR run contained 2 ultrameric calibrators (Cal-1 and Cal-2) for the A. vaginae reaction (set at 7.0 log10 copies/ml for Cal-1 and 5.5 log10 copies/ml for Cal-2), 2 ultrameric calibrators for the BVAB-2 reaction (set at 6.0 log10 copies/ml for Cal-1 and 4.5 log10 copies/ml for Cal-2), and a single ultrameric calibrator for Megasphaera-1 (set at 6.0 log10 copies/ml). The number and concentration of these cutoff calibrators were established based on analysis of frequency distributions of analyte concentrations obtained during qPCR testing of the developmental sample set (see Results). Each MagNAPure extraction tray contained an Aptima swab collection fluid sample as a negative control and a positive control manufactured from known-positive samples and required to generate a CT value within a defined range for each of the 3 analytes. Appropriate amplification of the IC amplicon served to ensure the elimination of PCR inhibitors and recovery of nucleic acid through sample preparation. Upon completion of testing, the results were exported into a Microsoft Excel worksheet for scoring according to the scheme shown in Table 2. Composite scores (sums of 3 individual analyte scores) were then compiled, and the final interpretation was generated as follows: BV negative, scores of 0 to 1; BV indeterminate, score of 2; BV positive, scores of 3 to 6.

Table 2
Scoring system used to categorize samples in the BV PCR assay


Sample characteristics.

In the initial assay development phase of the study, 169 samples were utilized, of which 108 (63.9%) were positive and 61 (36.1%) were negative for BV. Of the BV-positive samples, 96 (88.1%) had NSs of 7 to 10, with 12 samples having NSs of 4 to 6 and positive Amsel results. Of the BV-negative samples, 47 (77.0%) had NSs of 0 to 3, with the remaining 14 samples being Amsel negative and NS intermediate. Of the samples tested in this phase of the study, 168/169 (99.4%) were collected from patients attending the JCDH clinic. An additional 227 samples were used to validate the performance characteristics of the BV PCR assay. One hundred thirty-one (57.7%) of these samples were obtained from patients attending JCDH and 96 (42.3%) from patients attending PHC. Of the validation phase samples, 110 (48.5%) were positive and 107 (51.5%) were negative for BV. Unambiguous NSs were obtained for 200/227 (88.1%) validation samples, and of the 27 samples with intermediate NSs, 8 were resolved as positive and 19 as negative using Amsel criteria. The overall prevalence of BV in the study population, therefore, was 55.1% (218/396), with an NS-intermediate rate of 13.6% (54/396). Of the 54 NS-intermediate samples, 21 (38.9%) resolved as positive for BV using the Amsel criteria.

qPCR results.

Results obtained upon qPCR testing of the 169 assay development phase samples are shown in Table 3. All 4 of the previously reported positive marker organisms were, as expected, frequently present in samples designated BV positive (Table 3). Using an assay cutoff value of 1 × 103 copies/ml, 98.1% (106/108) of BV-positive samples were positive for G. vaginalis, 98.2% (106/108) for A. vaginae, 89.8% (97/108) for BVAB-2, and 87.0% (94/108) for Megasphaera-1. The frequencies with which these organisms were found in BV-negative samples differed significantly, however, with 60.7% (37/61) of such samples being positive for G. vaginalis, 52.5% (32/61) for A. vaginae, 18.0% (11/61) for BVAB-2, and 16.4% (10/61) for Megasphaera-1. Analysis of the distribution of quantitative values also demonstrated significant differences in organism concentrations in BV-positive and -negative sample populations (Table 3). The median concentrations observed in BV-positive and -negative populations differed by 3.5 log10 copies/ml for G. vaginalis, 3.8 log10 copies/ml for A. vaginae, >3.1 log10 copies/ml for BVAB-2, and >3.9 log10 copies/ml for Megasphaera-1. The sole negative marker organism evaluated, L. crispatus, demonstrated a relatively low positivity rate in the study population, with only 37/169 (21.9%) samples having >1 × 103 copies/ml of the organism. The positivity rate for L. crispatus in BV-positive samples was significantly lower than that observed in BV-negative samples (10.2% versus 42.6%; P < 0.01); however, a majority (57.4%) of BV-negative samples in the population lacked detectable levels of the organism.

Table 3
Results of quantitative PCR measurements of marker organisms in vaginal samples identified as positive or negative for BV

Statistical analysis of qPCR data.

The results obtained by qPCR analysis indicated that no single marker organism could reliably differentiate BV-positive from BV-negative samples. G. vaginalis and A. vaginae each demonstrated high sensitivity, but limited concentration discrimination between the 4th quartile of the negative population and the 1st quartile of the positive sample population compromised specificity (Table 3). BVAB-2 and Megasphaera-1, in contrast, demonstrated lower overall sensitivity but greater concentration discrimination between positive and negative populations. In order to objectively determine which combination of 2, 3, or 4 markers provided the optimal approach for further assay development, logistic regression analysis (MedCalc Software Suite) was performed on all marker combinations using the 169 development phase sample results (Table 4). The BV status of the patients, based on NS plus Amsel criteria, was used as the dependent variable (negative = 0, positive = 1), and the qPCR results of the combinations of marker organisms were used as the independent variable. The results of this analysis demonstrated that combinations of 3 marker organisms, either A. vaginae/BVAB-2/Megasphaera-1 or G. vaginalis/BVAB-2/Megasphaera-1, appeared to offer the optimal combination of sensitivity and specificity, with both parameters exceeding 90% for these combinations (Table 4). Inclusion of results obtained by qPCR for the putative negative predictive marker organism, L. crispatus, in the logistic regression models did not impact the sensitivity or specificity of either of the optimal 3-marker combinations (results not shown). This was entirely expected, given the low rate of detection of the organism in the developmental sample set. The combination of PCR tests for A. vaginae, BVAB-2, and Megasphaera-1was selected for development of the diagnostic assay construct. This selection was made, at least in part, because of a concern over possible misinterpretation of the significance of results reported for G. vaginalis. The organism has an extensive history of being utilized as a standalone marker for BV diagnosis, and its mere identification in the final test report might be misconstrued by users of the test as indicating the presence of BV. Since such a conclusion is not supported by our data, and given that this risk can be entirely eliminated by including A. vaginae as a high-sensitivity, low-specificity marker, the combination of A. vaginae, BVAB-2, and Megasphaera-1 was chosen for the BV PCR construct.

Table 4
Results of logistic regression analysis of qPCR results for marker organism combinations

BV PCR assay construct.

To assist in the creation of a simple, interpretive, test for BV based on the data generated by qPCR analysis, frequency distributions of the values generated by qPCR for the 3 selected marker organisms (A. vaginae, BVAB-2, and Megasphaera-1) were compared to designation of samples by NS (Fig. 1). Breakpoint values were selected that best differentiated sample populations by correlating qPCR values with high, intermediate, and low NSs. For A. vaginae and BVAB-2, the breakpoint value delineating the high population was selected to discriminate NS-positive from NS-negative samples with high specificity (Fig. 1a and andb),b), with the assumption that samples containing concentrations of either analyte in excess of that threshold would likely be scored BV positive based on NS alone. Since a significant subset of NS-positive and -intermediate samples contained concentrations of A. vaginae or BVAB-2 lower than that delineated by the high breakpoint concentration; however, a second medium breakpoint was also selected. The interbreakpoint population for A. vaginae and BVAB-2 contained samples belonging to all 3 NS categories, and thus, it was envisaged that multiple analytes generating values in this medium range would be required to define a sample as BV positive. For Megasphaera-1 (Fig. 1c), it was not possible to create a medium category because of the sharp breakpoint associated with the transition from low to high qPCR values; thus, only a single breakpoint was utilized to differentiate NS-positive from -intermediate or -negative samples. Each category was then assigned a numerical value for each marker, from 0 to 2 (Table 2), and the composite value of the 3 organism scores was compared to the previously determined BV designations of the samples (Table 5). The simplified scoring system demonstrated good correlation with the BV designation (by NS and Amsel) for a majority of the samples. Scores of 3 to 6 were highly correlated with positive samples, with 96.1% (98/102) of samples yielding these scores being BV positive. Similarly, a score of 0 was highly correlated with negative samples, with 93.9% (46/49) of samples yielding this score being BV negative. The remaining 18 samples (10.7%; 11 BV negative, 7 BV positive) yielded composite scores of 1 or 2. Of the 10 samples with a composite score of 1, 2 had NSs of ≤3, 6 had NSs of 4 to 6, and 2 had NSs of ≥7. Of the 8 samples with a composite score of 2, 2 had NSs of ≤3, 3 had NSs of 4 to 6, and 2 had NSs of ≥7. This subpopulation (1 or 2 composite scores), therefore, appeared to represent the PCR equivalent of the NS-intermediate category, with correlation of BV PCR results with NS and Amsel findings somewhat variable and difficult to reconcile. For the purposes of the validation study, it was decided to designate only samples with composite scores of 2 as indeterminate, since this score required the presence of any of the 3 marker organisms at high concentration, or at least 2 of them (A.vaginae and BVAB-1) at moderate concentration, findings clearly not representative of a population with normal vaginal flora, even though only a subset of these samples could be classified as BV using conventional test methods. Samples generating a composite score of 1 were designated negative pending the generation of further data, with the expectation that the inclusion of more BV-negative samples in the final validation sample set would enable a definitive conclusion about the appropriateness of this categorization. After excluding samples generating a score of 2 from the final analysis, the predicted sensitivity and specificity of the BV PCR construct for the remaining 161 samples were 93.3% (98/105) and 92.9% (52/56), respectively, with an indeterminate rate of 4.7% (8/169).

Fig 1
Distribution of qPCR values of development phase samples by Nugent score category for markers used in final BV PCR constructs. (a) A. vaginae. (b) BVAB-2. (c) Megasphaera-1. The positions of cutoff calibrators Cal-1 (lines a) and Cal-2 (lines b) are shown. ...
Table 5
Distribution of composite PCR scores by BV status

BV PCR assay validation.

Following the establishment of the BV PCR assay construct, the final assay design was formulated as described in Materials and Methods, utilizing a multiplex approach to analyze samples for the presence of the 3 target markers. Incorporation of nucleic acid calibrators in each run, at the concentrations determined by qPCR analysis to be most probative in differentiating populations of samples, enabled categorization of samples without the need for fully quantitative PCR analysis. The development phase samples were reanalyzed using this prototype BV PCR assay, and the results are shown in Table 5, with the interpretive assignment of results shown in Table 6. These results were highly congruent with those obtained by analyzing the individual qPCR assay results, with only 5 of 169 samples (2.9%) generating categorically different results. These changes resulted in 1 additional BV-positive sample being classified as positive by BV PCR, 1 additional BV-negative sample being classified as negative by BV PCR, 2 BV-negative samples moving from an indeterminate to a negative PCR score, and 1 BV-positive sample moving from a positive to an indeterminate PCR score. Overall, therefore, 162/169 (95.9%) samples generated interpretable composite PCR scores in the BV PCR assay construct, with the sensitivity of the assay being 94.2% (98/104) and the specificity 94.8% (55/58).

Table 6
Correlation of interpretive results generated from BV PCR assay analysis of study samples with consensus gold standard results

To confirm and extend these preliminary findings, an additional 227 samples were evaluated in the BV PCR construct. As described in “Sample characteristics” above, a substantial minority of these samples were collected at the lower-prevalence PHC location; thus, the overall prevalence of BV in this sample set was 48.5% as opposed to the 64.5% in the initial development phase sample set. The score assignments generated by BV PCR testing of these samples are shown in Table 5, and the interpretive results in Table 6. The results obtained for these samples were largely comparable to those obtained for the developmental sample set and supported the scoring system created from the development phase results. Of the 227 samples in the validation sample set, 14 (6.2%) yielded a composite score of 2 and were thus deemed indeterminate for BV, of which 9 were BV-negative samples and 5 were BV-positive samples. In total, 21 samples tested generated a BV PCR score of 2; 9 (42.9%) of these were BV-positive samples, and 12 (57.1%) were BV-negative samples (Table 5), supporting the use of indeterminate as a categorical designation for specimens generating this composite PCR score. Of the 213 samples that generated an evaluable PCR score in the validation sample set, 104 of 105 (99.0%) BV-positive samples generated a positive PCR score and 98 of 108 (90.7%) BV-negative samples generated a negative score. Of the 17 samples in the validation sample set that generated a composite PCR score of 1, only 1 was a BV-positive sample (Table 5), confirming the appropriateness of categorizing samples with this score as BV negative. Analysis of the entire 396 member data set, therefore, demonstrated that the BV PCR assay had a sensitivity of 96.7% (202/209), a specificity of 92.2% (153/166), a positive predictive value of 94.0%, and a negative predictive value of 95.6%, with an indeterminate rate of 5.3% (21/396).


The present report describes the development and validation of a multianalyte, semiquantitative PCR assay for the diagnosis of BV. Although previous studies have demonstrated the potential value of using multiple molecular markers either qualitatively or quantitatively for diagnosing this condition (5, 8, 10, 16, 19, 22, 25), the assay described here is the first to provide a simple, readily interpretable corollary to a unified gold standard definition for BV diagnosis, combining the NS and Amsel criteria. The definition of BV positivity utilized in this study was necessary because use of NS alone results in a significant minority of samples (10 to 20%) being designated intermediate. Such samples clearly are atypical but are insufficiently aberrant to be deemed indicative of BV without supplemental patient-specific information (17, 21). Previously published studies have generally excluded NS-indeterminate samples in their determination of the performance characteristics of molecular BV assays (5, 16) or have considered them uniformly equivalent to NS-negative samples (8, 10), making the overall value of such assays in clinical practice difficult to ascertain. Given the frequency of NS-intermediate samples in populations with clinical symptoms consistent with BV, we decided to utilize the Amsel clinical criteria to resolve such samples, with NS-intermediate, Amsel-positive samples deemed to be BV positive. The use of a unique combination of 3 marker organisms, A. vaginae, BVAB-2, and Megasphaera-1, each of which has previously been identified as independently associated with BV (5, 7, 10, 16, 18, 22, 25), in an assay format that segregated samples into marker-specific populations (high, medium, and low) based on their relationship to breakpoint DNA concentrations, enabled 94.7% (375/396) of samples to be categorized with respect to the presence or absence of BV with an overall accuracy of 94.7% (355/375). Analysis of the 20 samples generating discordant results between BV PCR and BV status by NS and Amsel criteria revealed the challenge of attempting to correlate molecular data for this condition with conventional techniques. Of the 13 BV-negative samples that were positive by BV PCR, only 3 had NSs of ≤3, and of the 7 BV-positive samples that were negative by BV PCR, only 1 had an NS of ≥7, and this patient was negative by Amsel criteria. Thus, only 4/20 (20%) BV PCR discordant results could be unambiguously categorized by NS; the remainder were NS-intermediate samples, a cohort likely to have significant, patient-specific variation in the extent of correlation of symptoms with specific changes in microflora.

Analysis of data generated from the individual PCR results demonstrated findings similar to those reported by previous investigators (5, 10, 11, 16, 22, 25). Both A. vaginae and G. vaginalis were highly sensitive markers for BV (5, 7, 10, 16, 25). In our study, they were found in 98.2% (106/108) of BV-positive samples in the development phase sample set (Table 3), but their utility as individual markers was limited by the frequency with which they were identified in samples from patients not classified as BV positive: A. vaginae in 53.3% and G. vaginalis in 60.0% (Table 3). Consistent with the findings of Fredricks et al. (10), BVAB-2 and Megasphaera-1 were somewhat more specific indicators of BV positivity than either G. vaginalis or A. vaginae, but at least one of these organisms was present in 24.6% (15/61) of BV-negative study subjects. We did not find that simple qualitative analysis of any individual positive predictive marker or combination of such markers was sufficiently accurate to be of value as a diagnostic assay for BV, since no combination generated sensitivity and specificity parameters of >90% in the development phase sample cohort. This contrasts with the findings of Dumonceaux and colleagues (5), who somewhat surprisingly reported a reasonable specificity for BV using a qualitative combination of A. vaginae and G. vaginalis in a small cohort of patients. This combination of organisms yielded the lowest specificity of any pair of analytes in our study (77%) (Table 4), and our findings were consistent with those of other investigators who have reported these markers to be sensitive but not qualitatively specific for BV (10, 25). Fredricks et al. (10), reported that qualitative detection of either BVAB-2 or Megasphaera-1 was highly sensitive and specific compared to either NS or Amsel criteria. Our data indicate that this combination also lacks a high level of specificity, even if only samples with NS values of ≥7 are considered positive and intermediate NS samples are excluded from analysis. Of the 47 samples in the development phase cohort with NSs of ≤3, 9 (19.1%) had positive results for either BVAB-2 (2 samples), Megasphaera-1 (4 samples), or both organisms (3 samples), using a threshold for positivity of 1 × 103 copies/ml. The performance of this combination using only the Amsel criteria for BV designation was even less impressive, with 26/72 (36.1%) Amsel-negative samples in the development phase cohort having either Megasphaera-1 or BVAB-2 DNA detected. These results were somewhat surprising, given that the median quantitative values for each of the markers in BV-positive patients reported by the same authors in a related publication (11) were only 0.5 log10 DNA copies per sample higher than those reported here, suggesting comparability of value assignment between our assays and those utilized by Fredricks and colleagues. Menard et al. (16) reported that a combination of A. vaginae and G. vaginalis analyzed quantitatively could be used to accurately differentiate BV-positive from BV-negative samples, using threshold concentrations of 108 copies/ml and 109 copies/ml, respectively. Logistic regression analysis of the development phase sample set (Table 4) demonstrated that quantitative analysis of these 2 markers in combination could not reliably differentiate BV-positive from BV-negative individuals and that inclusion of markers with a higher degree of specificity, for example, BVAB-2 and Megasphaera-1, is necessary to produce an assay with adequate positive predictive value. Quantitative PCR analysis of samples for A. vaginae, BVAB-2, and Megasphaera-1 appeared to provide the optimal combination of sensitivity and specificity, achieving 93.5% concordance (158/169) with the combined NS and Amsel gold standard (Table 4).

Relative quantitation of large Gram-positive rods, a morphotype consistent with a number of peroxide-producing Lactobacillus spp. (e.g., L. crispatus, Lactobacillus iners, and L. jensenii), organisms believed to be important for maintaining a healthy vaginal environment (6, 17, 21), constitutes a significant component (40%) of the total NS. It seemed reasonable, therefore, to include a qPCR assay for at least one of these organisms in the development phase of the BV PCR test, allowing the potential utility of such a marker in improving the positive predictive value of the final test design to be assessed. L. crispatus was selected for evaluation, since the organism has been identified as commonly present in low-NS samples across diverse patient populations (10, 18, 22). In the development phase sample set, L. crispatus DNA was detected in 42.6% of BV-negative samples and only 10.2% of BV-positive samples (Table 3), and median quantitative values of L. crispatus in samples containing this organism were significantly higher in the BV-negative population (8.9 × 107 copies/ml versus 4.1 × 104 copies/ml; P < 0.01), consistent with previous reports that the presence of the organism is inversely associated with BV (10, 22). Examination of the results for other marker organisms in L. crispatus-positive samples, however, revealed that in only a single instance was L. crispatus detected in a BV-negative sample that was scored as positive based on the combination of A. vaginae, BVAB-2, and Megasphaera-1 qPCR results. In addition, high L. crispatus DNA levels were strongly correlated with the absence of significant concentrations of positive predictive marker organisms. No samples in the development phase cohort containing an L. crispatus concentration of >5.1 × 105 copies/ml (n = 23) generated a composite BV PCR score in the positive range based on the criteria described in Table 2. The inclusion of the organism in the final assay design, therefore, would not have improved the accuracy of positive results, and these data suggest that molecular determination of BV can be achieved with acceptable reliability using solely positive predictive markers.

The final design of the BV PCR construct enabled a semiquantitative assessment of the DNA concentrations of key positive predictive markers, maintaining the probative advantage of stratifying samples by concentration afforded by qPCR, but doing so in a simplified and highly reproducible assay format. The categorical boundaries were driven by the frequency distribution data shown in Fig. 1, and the data in Table 6 demonstrate the validity of the categorical classification of samples based on a blinded comparison with the combined NS and Amsel gold standard used in the study. One clear advantage of the assay format used here is that, since it generates results on a discrete numerical scale, its performance characteristics in any given population can be estimated by comparing the frequency distribution of composite scores in that population with the accuracy of each score derived from the data presented here (Table 5). Since a score of 0 had a negative predictive value of 97.6% (125/128) while a score of 5 or 6 had a positive predictive value of 99.2% (123/124), the proportion of samples generating these 3 values will strongly influence the overall predictive value of the BV PCR construct. We retrospectively examined the frequency distribution of composite scores generated by the first 6,000 samples submitted for the assay after its introduction into routine testing (data not shown), and this demonstrated that in this unselected clinical population with a prevalence of BV (as determined by BV PCR) of 26.3% and an indeterminate rate of 6.2%, 81.2% of samples tested generated scores of 0, 5, or 6, resulting in estimated positive and negative predictive values of 95.1% and 96.8%, respectively.

In conclusion, the detection of a combination of 3 marker organisms (A. vaginae, BVAB-2, and Megasphaera-1) and differentiation of vaginal populations based on assignment of a numerical score directly related to critical concentrations of these organisms enabled highly concordant results between the BV PCR construct described here and conventional techniques for diagnosing BV to be obtained. The BV PCR construct utilizes a relatively simple and robust design that enables accurate differentiation of BV-positive and -negative samples to be performed in a standardized and objective manner. The availability of such a diagnostic tool could provide the impetus for future investigations seeking to develop improved management options for this common and problematic syndrome.


We thank Tom Scholl and Zhaohui Wang for assisting with design and initial testing of PCR primer pairs, Rebecca Amos for cataloging and organizing specimens, and Matt Brown for assistance in preparing the manuscript.

ViroMed Laboratories Inc. is a wholly owned subsidiary of Laboratory Corporation of America Holdings.


Published ahead of print 25 April 2012


1. Amsel R, et al. 1983. Nonspecific vaginitis. Am. J. Med. 74:14–22 [PubMed]
2. Biagi E, et al. 2009. Quantitative variations in the vaginal bacterial population associated wih asymptomatic infections: a real-time polymerase chain reaction study. Eur. J. Clin. Microbiol. Infect. Dis. 28:281–285 [PubMed]
3. Brotman RM, Ravel J. 2008. Ready or not: the molecular diagnosis of bacterial vaginosis. Clin. Infect. Dis. 47:44–46 [PubMed]
4. Cauci S, et al. 2002. Prevalence of bacterial vaginosis and vaginal flora changes in peri- and post-menopausal women. J. Clin. Microbiol. 40:2147–2152 [PMC free article] [PubMed]
5. Dumonceaux TJ, et al. 2009. Multiplex detection of bacteria associated with normal microbiota and with bacterial vaginosis in vaginal swabs by use of oligonucleotide-coupled fluorescent microspheres. J. Clin. Microbiol. 47:4967–4977 [PMC free article] [PubMed]
6. Eckert LO. 2006. Acute vulvovaginitis. N. Engl. J. Med. 355:1244–1252 [PubMed]
7. Ferris MJ, et al. 2004. Association of Atopobium vaginae, a recently described metronidazole resistant anaerobe, with bacterial vaginosis. BMC Infect. Dis. 4:5. [PMC free article] [PubMed]
8. Ferris MJ, Norori J, Zozaya-Hinchliffe M, Martin DH. 2007. Cultivation-independent analysis of changes in bacterial vaginosis flora following metronidazole treatment. J. Clin. Microbiol. 45:1016–1018 [PMC free article] [PubMed]
9. Fredricks DN, Fiedler TL, Marrazzo JM. 2005. Molecular identification of bacteria associated with bacterial vaginosis. N. Engl. J. Med. 353:1899–1911 [PubMed]
10. Fredricks DN, Fiedler TL, Thomas KK, Oakley BB, Marrazzo JM. 2007. Targeted PCR for detection of bacteria associated with bacterial vaginosis. J. Clin. Microbiol. 45:3270–3276 [PMC free article] [PubMed]
11. Fredricks DN, Fiedler TL, Thomas KK, Mitchell CM, Marrazzo JM. 2009. Changes in bacterial vaginal concentrations with intravaginal metronidazole therapy for bacterial vaginosis as assessed by quantitative PCR. J. Clin. Microbiol. 47:721–726 [PMC free article] [PubMed]
12. Hay PE. 2006. National guideline for the management of bacterial vaginosis. Royal Society of Medicine, London, United Kingdom:
13. Hillier SL. 2005. The complexity of microbial diversity in bacterial vaginosis. N. Engl. J. Med. 353:1886–1887 [PubMed]
14. 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 [PMC free article] [PubMed]
15. Leitich H, et al. 2003. Bacterial vaginosis as a risk factor for preterm delivery: a meta-analysis. Am. J. Obstet. Gynecol. 189:139–147 [PubMed]
16. Menard J-P, Fenollar F, Henry M, Bretelle F, Raoult D. 2008. Molecular quantification of Gardnerella vaginalis and Atopobium vaginae loads to predict bacterial vaginosis. Clin. Infect. Dis. 47:33–43 [PubMed]
17. 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 [PMC free article] [PubMed]
18. Ravel J, et al. 2011. Vaginal microbiome of reproductive-age women. Proc. Natl. Acad. Sci. U. S. A. 108:4680–4687 [PubMed]
19. Sha BE, et al. 2005. Utility of Amsel criteria, Nugent score, and quantitative PCR for Gardnerella vaginalis, Mycoplasma hominis, and Lactobacillus spp. for diagnosis of bacterial vaginosis in human immunodeficiency virus-infected women. J. Clin. Microbiol. 43:4607–4612 [PMC free article] [PubMed]
20. Sherrill CB, et al. 2004. Nucleic-acid analysis using an expanded genetic alphabet to quench fluorescence. J. Am. Chem. Soc. 126:4550–4556 [PubMed]
21. Spiegel CA. 1991. Bacterial vaginosis. Clin. Microbiol. Rev. 4:485–502 [PMC free article] [PubMed]
22. Tamrakar R, et al. 2007. Association between Lactobacillus species and bacterial-vaginosis associated bacteria, and bacterial vaginosis scores in pregnant Japanese women. BMC Infect. Dis. 7:128. [PMC free article] [PubMed]
23. U.S. Preventative Services Task Force 2008. Screening for bacterial vaginosis in pregnancy to prevent preterm delivery: US Preventative Services Task Force recommendation. Ann. Intern. Med. 148:214–219 [PubMed]
24. Wiesenfeld HC, Hillier SL, Krohn MA, Landers DV, Sweet RL. 2003. Bacterial vaginosis is a strong predictor of Neisseria gonorhoeae and Chlamydia trachomatis infection. Clin. Infect. Dis. 36:663–668 [PubMed]
25. 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 [PMC free article] [PubMed]

Articles from Journal of Clinical Microbiology are provided here courtesy of American Society for Microbiology (ASM)