Bacterial Vaginosis (BV), a condition in which the normal Lactobacillus-rich vaginal microbiome becomes dominated by polymicrobial anaerobic bacterial species under non-acidic pH, remains the most common cause of abnormal vaginal discharge in reproductive-age women, with an estimated prevalence rate of 29% in North America (Abou Chacra et al., 2021; Elnaggar et al., 2023). The normal vaginal microbiome is dominated by three major Lactobacillus species, L. crispatus, L. jensenii, and L. gasseri, with low abundance of L. acidophilus, which protects the vagina by producing lactic acid, hydrogen peroxide, and bacteriocins that suppress bacterial growth (Pacha-Herrera et al., 2022; Wu et al., 2022). Lactobacillus iners, however, is quite enigmatic as it occurs in both healthy and unhealthy vaginal environments; L. iners, unlike L. crispatus, L. jensenii, and L. gasseri, produces the human non-functional L form of lactic acid (Basavaprabhu et al., 2020).

To produce lactic acid, which is necessary to maintain an acidic pH of < 4.5, Lactobacillus sp. uses glycogen deposited unto the vaginal walls from the epithelial cells (Shen et al., 2022; Navarro et al., 2023). The acidic pH keeps away most microbial species and allows Lactobacillus sp. to proliferate, keeping the species diversity in the normal vaginal microbiota low. Lactic acid is metabolized from glycogen, whose production and deposition are mediated by circulating estrogen in a dose-dependent manner. In the absence of glycogen or lactic-acid–producing species, the pH rises, allowing other microbes to proliferate, colonize the vagina, and increase the microbiota diversity. Hence, physiological factors disrupting or increasing estrogen levels indirectly affect Lactobacillus sp. and BV-associated bacterial species’ prevalence (Shen et al., 2022; Navarro et al., 2023). The etiology and pathogenesis of BV are still not fully understood, making it important to study the involved microbes for timely diagnosis and treatment. Particularly, the interactions between species within the BV microbiome that cause pathologies, treatment failure and recurrence, or healing need further interrogations to enable a species biomarker-based diagnosis of BV.

The colonization of the vaginal microbiota by anaerobic bacterial and fungal species causes BV or vulvovaginal candidiasis, respectively, characterized by increased vaginal discharge with a fishy odor (Swidsinski et al., 2023). Dysbiosis also predisposes the vagina to sequelae of other sexually transmitted infections (STIs) and obstetric disorders. Although BV is asymptomatic in many women, it is associated with the development of common obstetric and gynecologic complications (Javed et al., 2019; Cheng et al., 2020; Gustin et al., 2021). It also increases the risk of contracting HIV and other STIs and pelvic inflammatory disease (PID) (Unemo et al., 2017). Notably, BV recurrence after treatment is common (estimated to affect 50% of women annually) (Hilbert et al., 2016; Coudray and Madhivanan, 2020; Coudray et al., 2020) and may be due to re-infections from sexual partners or failure of current treatment options (Coudray et al., 2020; Vodstrcil et al., 2021). Besides the health impacts of recurrence, BV treatment costs are increasing annually, specifically in the USA: this economic impact is particularly pronounced in BV-associated preterm births and other obstetric complications (Peebles et al., 2019; Watkins et al., 2024).

Species such as Gardnerella vaginalis, Fannyhessea vaginae, Prevotella bivia, Megasphaera sp., and BVAB are commonly implicated in the etiology of BV, with G. vaginalis and F. vaginae being common in most BV infections (Swidsinski et al., 2023) as sessiles. Swidsinksi et al. (2023) observed that vaginal epithelial cells from females with BV were covered with G. vaginalis biofilms that encapsulated other species, forming a polymicrobial biofilm, as well as by non-adherent planktonic bacterial cells. The biofilms associated with BV explain their ability to escape both the immune response and antimicrobial chemotherapy as biofilms prevent the immune cells and antimicrobial agents from reaching the pathogens (Cangui-Panchi et al., 2023; Swidsinski et al., 2023). This leads to high treatment failure and recurrence rates of about 50% (Muñoz-Barreno et al., 2021); indeed, the ability of these biofilm-coated polymicrobial species to spread into the uterus and fallopian tube explains their resistance to antimicrobials and the immune response (Swidsinski et al., 2023).

To diagnose BV, clinicians rely mostly on the classical clinical signs and symptoms outlined in Amsel’s criteria (Amsel et al., 1983) or on the microscopically based Nugent score (Nugent et al., 1991). While these standard diagnostic methods have been effective over the years, they have been confronted with limitations such as being labor-intensive and time-consuming, subjective and unable to accurately identify pathogens (Muzny et al., 2023). As recently discussed by Muzny et al. (2023), nucleic acid tests ease the burden of laboriously going through the clinical testing criteria, saving clinicians time and effort for other duties. The development of culture-independent molecular diagnostics has enabled the detection of non-cultivable bacterial species associated with BV and continues to revolutionize infectious disease diagnosis (Hilbert et al., 2016; van den Munckhof et al., 2019; Bhujel et al., 2021). Molecular tests such as real-time polymerase chain reaction (RT-PCR), combined with the traditional clinical diagnostic criteria, have greatly improved the sensitivity and specificity of detecting BV pathogens (Hilbert et al., 2016; van den Munckhof et al., 2019; Bhujel et al., 2021). They are also applicable in monitoring patient response to antibiotic therapy (Onderdonk et al., 2016; Elnaggar et al., 2023). This approach has proven more useful for identifying patients at risk for recurrent BV (Jones, 2019; Coudray and Madhivanan, 2020; Vodstrcil et al., 2021; Wu et al., 2022; Sobel and Vempati, 2024; Watkins et al., 2024).

Besides next-generation sequencing (NGS)-based metagenomics that target all genomes, current PCR diagnostics for BV mainly focus on a smaller spectrum of bacterial species. Hence, to increase the resolution and diagnostic power of PCR-based BV diagnostics (Balashov et al., 2014), we designed a new 22-species quantitative Real-Time PCR assay (qRT-PCR) that quantitatively detects 22 vaginal bacterial species within 8 hours (from sample reception). This newly designed proprietary qRT-PCR assay (Bacterial Vaginosis (with Lactobacillus Profiling) Panel^®, Medical Diagnostic Laboratories, L.L.C. (MDL), New Jersey, USA) was used to screen 946 vaginal samples routinely obtained from different health centers across the United States. We further used the results, with assistance from machine-learning algorithms (Decision Trees and Random Forests) to design, train, and test an index (herein termed the MDL-BV index) that used a relative-abundance and species-based markers to classify the vaginal microbiome in three categories: BV-negative (healthy microbiome), transitional BV (between healthy and unhealthy vaginal microbiome), and BV-positive (abnormal microbiome).

Unlike other infections with single etiological agents, BV is a polymicrobial infection with no clear consensus on the specific bacterial species responsible for the altered vaginal microbiota state (Mondal et al., 2023; Sobel and Vempati, 2024). The exact contributions of BV-associated microbes to the pathogenesis of BV, which may be relevant for accurate diagnosis and therapeutics, remain unresolved. Nevertheless, the occurrence of normal, transitional, and abnormal vaginal microbiota states is dependent on the composition and interactions of the different Lactobacillus and anaerobic species present (Onderdonk et al., 2016; Lynch et al., 2019; Chee et al., 2020; Drew et al., 2020). To address this gap between the vaginal microbiome dynamics and the current BV diagnostic tests’ limitations, a new qRT-PCR test that detects and measures the gDNA concentrations of 22 species found in all the various conditions of the vaginal microbiome was designed. The relatively shorter turnaround time of this test and the ease of adoption in a routine diagnostic laboratory makes it possibly more valuable than the current classical tests (Muzny et al., 2023; Swidsinski et al., 2023).

With such a broad spectrum of bacterial species, the resolution of the test is enhanced to efficiently distinguish between normal, abnormal, and transitional microbiota. The test has been validated with 95 – 100% sensitivity and specificity with a short turnaround time of 8 hours (from sample reception). In studies where results from PCR or qPCR studies targeting 2 – 13 species have been compared with Amsel’s criteria or Nugent’s score, the results have been highly sensitive and specific (Menard et al., 2008; Malaguti et al., 2015; Lynch et al., 2019; Drew et al., 2020; Vodstrcil et al., 2021). We are therefore confident that subsequent studies with our test, using Nugent score or Amsel’s criteria-diagnosed samples, may equally yield similar or better results with its larger bacteria target spectrum. Although this study is limited by the inability to compare the current results with clinical presentation data, its sensitivity and specificity in detecting the controls used confirm its efficiency.

In fact, the test’s diagnostic capability is further appreciated when the make-up and sample size of our vaginal specimens are considered: 946 vaginal samples from a wide spectrum of ages (18 – 83), races (White, Black, Asian, American Indian, and “Other (Hispanics and Pacific Islanders)”, not to mention the races of most women who failed to state their race), and pregnancy status. The Bray-Curtis dissimilarity matrix and the principal component analysis (PCoA) showed how dissimilar the samples were as few of the samples clustered together ( Figure 6 ). The sample differences were further highlighted by the species richness (α-diversity) of the samples in which the number of species per sample ranged from 2 – 12, with 50% of the samples having 2 – 6 species per sample. Instructively, the Shannon index further clarified this observation by showing that the species diversity per sample was low, with most samples having an index between 0.8 and 1.8 and the most diverse sample having an index of 3.2 (out of 3.5).

Thus, although the samples were dissimilar in terms of composition, their diversity was relatively low, which could be characteristic of a healthy vaginal microbiome if the relative abundance of Lactobacillus sp. is high (Balashov et al., 2014). However, it is worth noting that the Shannon diversity is affected by the number of species sampled and the vaginal microbiome is naturally not as diverse as the gut microbiome (Javed et al., 2019; Muzny et al., 2020; Abou Chacra and Fenollar, 2021). Hence, although the 22 species used are more than any PCR test, it may not represent all the species in the vaginal microbiota. For instance, the alpha diversity of the samples was calculated through a direct count of the number of species observed in each sample without using any estimators or indices. Hence, it does not account for undetected species or attempt to estimate the total species richness in the samples, which methods like the Chao1 index aim to do. Notwithstanding, the representative nature of the samples used in this study is self-evident.

To our knowledge, no qPCR assay has the same broad-spectrum species target, making this assay an important innovation and the first to do so. Whereas other assays have used only two or up to 13 species (Malaguti et al., 2015), not all of the species were quantitatively detected as was done in this assay (Brotman and Ravel, 2008; Menard et al., 2008; Malaguti et al., 2015; Drew et al., 2020). This is despite the calls for increasing the bacterial spectrum for molecular BV diagnosis, focusing on Megasphaera, G. vaginalis, F. vaginae, and other anaerobes (Brotman and Ravel, 2008; Menard et al., 2008; Lynch et al., 2019). Moreover, owing to the biofilm-forming nature of these BV-associated species, which leads to immune evasion, high treatment failure rates and recurrence, a shorter-turnaround diagnostic tool is merited (Cangui-Panchi et al., 2023; Muzny et al., 2023). This study answers that call.

By both detecting and quantifying the 22 species in vaginal samples, we were able to obtain a clearer picture of the vaginal microbiome’s composition and make a better diagnosis of its condition; a feat only achievable with whole-genome metagenomics (Asante and Osei Sekyere, 2019; Osei Sekyere et al., 2020; Osei Sekyere et al., 2023). For instance, although BVAB-1, -2, and -3 were detected together in the same study (Fredricks et al., 2005), BVAB-2 has been commonly tested as a marker of BV positivity without BVAB-1 and -3. Yet, this study showed that BVAB-2 is the least prevalent/abundant among these three species, with BVAB-1, followed by BVAB-3, being the most common. A Brazilian study using 223 BV-positive samples also reported the same findings regarding the relative abundance of BVAB-1, BVAB-2, and BVAB-3, affirming our observations (Malaguti et al., 2015). A positive association between BV and high-risk Human Papilloma Virus (HPV) genotypes has been reported, owing to the occurrence of BVAB 1 and 3, and other BV-associated- bacteria in women co-infected with HIV and HPV (Naidoo et al., 2022). The same study also showed that the presence of BVAB 1 and 3 had an elevated likelihood of increasing the severity of cervical neoplasia in this population (Naidoo et al., 2022). Nevertheless, these species are not as closely related as initially thought as a recent phylogenetic analysis identified BVAB 1, 2, and 3 to be “Clostridiales genomosp.”, Oscillospiraceae bacterium strain CHIC02, and Mageeibacillus indolicus, respectively (Osei Sekyere et al., 2023).

Additionally, the relative abundance of the various species identified by this test mirrors what has been reported (Malaguti et al., 2015; Bayigga et al., 2019; Abou Chacra et al., 2021; Zhang et al., 2022; Powell et al., 2023), with Lactobacillus sp. (except L. acidophilus), G. vaginalis, A. (F.) vaginae, P. bivia, and Megasphaera sp. type 1 being the most dominant and widely distributed ( Figures 2 , 4 ; Supplementary Figures S1 – S22 ). The absence of M. genitalium [which was found in low abundance in other studies using BV-positive samples (Malaguti et al., 2015)], and lower abundance and prevalence of L. acidophilus, M. hominis, U. urealyticum, Megasphaera sp. type 2, Mobiluncus sp., S. sanguinegens, B. breve, and S. anginosus are also consistent with other findings (Atassi et al., 2019; Chee et al., 2020; Argentini et al., 2022; Wu et al., 2022), further crediting the diagnostic efficiency of this test.

Overall, the Lactobacillus species had higher mean abundances in BV-negative samples, lower abundances in BV-positive samples, and in-between for transitional BV, indicating their potential protective role against BV ( Figure 9 ). The higher relative abundance of L. iners in pregnant women and transitional BV samples could further suggest that L. iners play a crucial role in transitioning the vaginal microbiome from BV-negative to BV-positive. The transitioning effect of L. iners has been investigated and reported earlier and these observations further throws more light on the veracity of this conclusion (Zheng et al., 2021), albeit other studies provide conflicting findings and call for additional studies (Carter et al., 2023).

It was observed from the correlation, and co-existence analysis that the five Lactobacillus sp. mostly co-existed together while G. vaginalis, A. (F.) vaginae, P. bivia, and Megasphaera sp. type 1 also co-existed in the same samples ( Figures 4 , 7 ; Supplementary Figures S1 – S23 ). The Chi-square and T-tests also largely agreed with the significant association between the Lactobacillus sp. and the non-Lactobacillus species, with few exceptions ( Figure 7 ). This was used to form two separate groups of species-based biomarkers. Furthermore, the BVABs, S. anginosus, and B. fragilis, which are known to be mainly associated with BV microbiomes and absent in normal vaginal microbiomes (Weyers et al., 2009; Matu et al., 2010; Malaguti et al., 2015; Schmidt et al., 2015; Barrientos-Durán et al., 2020), were also teased from the remaining non-Lactobacillus species and grouped into another biomarker group. The remaining eight species were then also bundled together into another fourth group ( Table 1 ). Using the relative abundance distributions of each species ( Figure 4 ; Supplementary Figures S1 – S22 ), the relative abundance range for each of the four species biomarker groups was set to form the MDL-BV index, which was then further trained and tested on the data using machine-learning algorithms (Decision Trees and Random Forests) to diagnose BV. A relative abundance-based approach was adopted over a nominal concentration value because concentrations vary from sample to sample, and the swabbing method of sample collection among clinicians is not standardized; hence, samples collected by each swab is not quantitative.

The MDL-BV index has a high cut-off range for negative BV status (≥70% Lactobacillus sp. relative abundance) to ensure that samples diagnosed as normal were BV-negative in verity. It also errs on the side of caution by classifying samples that meets the criteria of both BV-negative and transitional or transitional and BV-positive, as transitional BV or BV-positive respectively. The training of the model on the data further revealed certain intricacies and nuances in the distribution of the species and their relative abundances, which were used to refine it by including more details and rules. For instance, in situations where only two of the four biomarkers had a relative abundance score, the ratios were adjusted ( Table 1 ). This strengthened the diagnostic efficiency of the MDL-BV index, resulting in a final classification of the samples as 491 BV-positive, 318 BV-negative, and 137 transitional BV. A manual verification of the results confirmed the veracity of the classifications ( Supplementary Table S3 ).

Instead of using just two biomarkers involving all Lactobacillus sp. and all non-Lactobacillus species (Balashov et al., 2014), this index uses four biomarkers, which further refines the ability of the index to correctly distinguish between transitional BV or BV-positive microbiomes. Evidently, a correct molecular diagnosis of BV is critical to its treatment and monitoring, making this test and index very important in gynecology and obstetrics. The next stage of this research is to train the model on large datasets with Nugent scores/Amsel criteria as a means of further enhancing it and strengthening its proof of concept in BV diagnosis.

The distribution of BV among the different demographic variables viz., age, race, and pregnancy status showed important differences, albeit none of the differences were statistically significant ( Figure 8 ). Specifically, although the number of vaginal samples from Black people was lower than that from White people, there were more BV-positive samples from the former than from the latter. While most of the vaginal samples from White people were BV-negative, those from Black people were either BV-positive or transitional. Notably, the samples from the “Other” races (which includes Hispanics/Latinos and Pacific Islanders), Asians, and Native American Indians, although fewer in number, were also BV-positive, with none being negative. Furthermore, the mean age and age range for White people with transitional BV were higher than those of all the other races, including the mean age of White people with abnormal vaginal microbiomes ( Figure 8 ). Hence, although these differences were not statistically significant, they concur with other studies that show a higher prevalence of BV among Black and Hispanics than among White people and Asians (Peebles et al., 2019). This agreement between our data and other studies further confirms the accuracy of our MDL-BV index.

A further analysis of the racial vaginal microbiome composition showed that Black women had a higher prevalence of many of the anaerobes implicated in the BV pathogenesis, including G. vaginalis, F. vaginae, and Megasphaera sp., followed by Hispanics and Asian women who also had higher abundance of these BV-associated species than Caucasians ( Figure 15 ; Supplementary Figures S32 , S33 ). These findings are not singular as other studies have also identified the same. Particularly, Borgdoff et al. (2017) found a higher abundance of L. crispatus among Caucasians than Black and Asian women and a higher G. vaginalis and L. iners in Ghanaian and Asian (and Mediterranean) women, respectively (Borgdorff et al., 2017). Contrarily however, Roachford et al. (2022) also found a higher abundance of L. crispatus and L. iners in African Americans than in other ethnicities (Roachford et al., 2022); nevertheless, most studies agree that Africans have lower Lactobacillus-rich microbiomes compared with White women. These racial and ethnic variations were examined recently by Xin Wei et al. (2024) and found that the nucleotide identities of these vaginal denizens found in different ethnicities varied from each other, underscoring an evolutionary mechanism that explains these observed variations (Wei et al., 2024).

In effect, the genomes of the same species had differences when isolated from different races. While they also observed similarly higher abundance of BV-associated bacteria in Black pregnant women than other races (higher alpha diversity and Shannon index), they established that the alpha diversity remained fairly constant over gestational time and across ethnicities (Wei et al., 2024). This is similar to our finding in which the differences between pregnant and non-pregnant cohorts’ vaginal microbiome variations were marginal. Notwithstanding, this study brings out certain unique findings that merit further investigation: the absence of B. fragilis, Megasphaera sp. type 2, and B. breve in pregnant women; the higher differential abundance of G. vaginalis, BVAB-1 and -2, U. urealyticum, Megasphaera sp. type 1, and S. anginosus in pregnant women; and the higher abundance of S. sanguinegens in non-pregnant women compared with pregnant women.

Albeit pregnancy had no significant effect on BV, we posit that the observed marginal variations of higher diversity and higher relative abundance in pregnant women than in non-pregnant women can be due to the hormonal changes that occur in the former. Hormonal variations in pregnancy also explain the lower relative abundance of Lactobacillus sp. (except L. iners) and anaerobes among non-pregnant women. Notably, L. acidophilus presented the widest variations between cohorts, suggesting that it is more sensitive to these changes than the other species and could be adopted as a biomarker. Indeed, the higher abundance of L. crispatus among non-pregnant and menopausal women, the absence of L. acidophilus in menopausal women, and the higher abundance of L. iners among pregnant women further show how the variations in hormones affect these species’ relative abundance through changes in the hosts estrogen-medicated metabolisms (Auriemma et al., 2021). The higher abundance of L. crispatus and L. iners was been reported previously (Borgdorff et al., 2017; Wei et al., 2024).

Although BV among non-pregnant women was more prevalent than among pregnant women, BV-negative or transitional BV samples were not obtained from pregnant women ( Figure 8 ). This is a concerning observation as BV in pregnancy is associated with preterm labor, low birth weight, premature rupture of membranes, miscarriage, chorioamnionitis, and birth asphyxia (Bayigga et al., 2019; Basavaprabhu et al., 2020; Vodstrcil et al., 2021). Although the pregnancy stage of the women from whom the samples were obtained is unknown, it is known that the first-trimester vaginal microbiome is similar to a BV-positive microbiome while the 2^nd and 3^rd trimesters are similar to a normal vaginal microbiome (Kaur et al., 2020). Hence, the stage of the pregnancy and vaginal symptoms should be considered in making a final therapeutic decision regarding pregnant women when using molecular-based tests. Furthermore, the difference in the number of pregnant women (n = 53) vis-a-viz the number of non-pregnant (or unknown pregnancy status) women is large and can skew the data. This is a major limitation in our data set and should be considered when analyzing the pregnancy data.

Although BV-positive samples were found in all age groups (18 – 83 years), they were mostly prevalent among women who were between 20 and 60 years, with a higher number of BV-positive and transitional BV samples being found among women between 41 – 50 years. The species variations per age group further support this observation as the anaerobic species diversity and abundance increase from after 20 years, plateau until age 50, and begin to decline again ( Figures 12 , 13 ; Supplementary Figures S28 , S29 ). Indeed, the median age found in this study’s population is similar to what was found in Brazil (Malaguti et al., 2015) and in the USA (Watkins et al., 2024), with women above 21 years being more likely to suffer recurrence of BV (Coudray and Madhivanan, 2020; Coudray et al., 2020). Hence, the age groups found in our data concur with that of other studies, which confirms that BV is common among reproductive-age women and lower before puberty and after menopause (Auriemma et al., 2021). However, age was not significantly associated with BV occurrence in this study.

The new qRT-PCR assay developed by MDL is the first of its kind to quantitatively detect 22 species in the vaginal microbiome. We developed a novel classification system, the new MDL-BV index, which was trained on large sets of data and based on four species-based and relative abundance-based biomarkers. This multifaceted two-tier approach of using species and relative abundance provides a better diagnostic resolution for a polymicrobial infection such as BV. We are working to extend this approach to apply to other microbiome-based diagnostics to make disease diagnosis reflective of the clinical presentations. Our study is, however, limited by the absence of a clinical Nugent score or Amsel’s criteria; however, we are working to include these in the next studies that will involve the MDL-BV index and the 22-species qRT-PCR BV test. The clinical and socio-economic importance of this novel proprietary BV diagnostic test in obstetrics and gynecology is notable.

The original contributions presented in the study are included in the article/ Supplementary Material . Further inquiries can be directed to the corresponding authors.

Ethics approval was not required for this study as the samples used were leftover, discarded samples provided by healthcare providers and were anonymized, posing no risk of identifying individual patients. According to the Common Rule (45 CFR 46) and HIPAA (45 CFR 164.514(b)(2)), research involving anonymized data does not constitute human subjects research and thus does not require IRB approval. Written informed consent was not obtained as the study used leftover, discarded samples, and the study did not involve direct interaction with patients.

AO: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. RH: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing. JT: Investigation, Methodology, Supervision, Writing – review & editing. SF: Supervision, Validation, Writing – review & editing. EM: Funding acquisition, Resources, Supervision, Validation, Visualization, Writing – review & editing. MA: Funding acquisition, Project administration, Resources, Supervision, Validation, Visualization, Writing – review & editing. JO: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Project administration, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing.