Impact of oral metronidazole treatment on the vaginal microbiota and correlates of treatment failure

Background Metronidazole is the first-line treatment for bacterial vaginosis, but cure rates are suboptimal and recurrence rates high. Objectives To evaluate the impact of a standard course of oral metronidazole treatment (500 mg twice per day for 7 days) on the vaginal microbiota of Rwandan bacterial vaginosis patients using microscopy and 16S rRNA gene sequencing, and to evaluate correlates of treatment failure. Study Design HIV-negative, nonpregnant women aged 18–45 years with bacterial vaginosis and/or Trichomonas vaginalis (N=68) were interviewed and sampled before and after metronidazole treatment. They were also screened, and treated if applicable, for other urogenital infections. The vaginal microbiota was assessed by Gram stain Nugent scoring, Illumina 16S rRNA HiSeq sequencing (relative abundances), and BactQuant 16S gene quantitative polymerase chain reaction (estimated concentrations). Only women with a pretreatment Nugent score of 7–10 and a valid posttreatment Nugent score (N=55) were included in metronidazole treatment failure analyses, with treatment failure defined as a posttreatment Nugent score of 4–10. Results The bacterial vaginosis cure rate by Nugent scoring was 54.5%. The mean total vaginal bacterial concentration declined from 6.59 to 5.85 log10/μL (P<.001), which was mostly due to a reduction in mean bacterial vaginosis-associated anaerobes concentration (all bacterial vaginosis-associated anaerobe taxa combined) from 6.23 to 4.55 log10/μL (P<.001). However, only 16.4% of women had a bacterial vaginosis anaerobes concentration reduction of more than 50%, and only 3 women had complete eradication. The mean concentration of lactobacilli (all species combined) increased from 4.98 to 5.56 log10/μL (P=.017), with L. iners being the most common species pre- and posttreatment. The mean concentration of pathobionts (defined as Proteobacteria, streptococci, staphylococci, enterococci, and a few others) did not change significantly: from 1.92 log10/μL pretreatment to 2.01 log10/μL posttreatment (P=.939). Pretreatment pathobionts concentration, and having a pretreatment vaginal microbiota type containing more than 50% Gardnerella vaginalis (compared with less than 50%), were associated with increased likelihood of treatment failure, but the latter did not reach statistical significance (P=.044 and P=.084, respectively). Conclusions Metronidazole alone may not cure women with high G. vaginalis relative abundance, potentially due to biofilm presence, and women with high pathobionts concentration. These women may benefit from additional biofilm-disrupting and/or pathobiont-targeting treatments.

M ost women have an optimal vaginal microbiota (VMB) dominated by lactobacilli, but vaginal dysbiosis is highly prevalent. 1 The most common type of vaginal dysbiosis is bacterial vaginosis (BV), which is characterized by a reduction of lactobacilli and an increase of other anaerobes, usually leading to increased species diversity. 2,3 Some women carry microorganisms that do not necessarily dominate the VMB but have a greater pathogenic potential than BV anaerobes, such as bacterial pathobionts (including most Proteobacteria, streptococci, staphylococci, and enterococci), Candida albicans (which is the main cause of vulvovaginal candidiasis), or Trichomonas vaginalis. 2,3 Vaginal dysbiosis can cause symptoms but is often asymptomatic. 2,3 Both symptomatic and asymptomatic vaginal dysbiosis have been associated with pelvic inflammatory disease, HIV acquisition, and adverse pregnancy outcomes. [3][4][5][6] Symptomatic dysbiosis is most commonly diagnosed empirically or syndromically without laboratory testing. 7 However, in research settings, BV is diagnosed by Nugent scoring of vaginal Gram stains or by the Amsel criteria. 8,9 BV is treated with antibiotics, of which oral and vaginal formulations of metronidazole or clindamycin are most commonly used. 10 Although shortterm BV cure rates of multiday oral and vaginal metronidazole regimens are 65%e90%, 11,12 recurrence rates are high. 13,14 Metronidazole is a nitroimidazoleclass drug. Under anaerobic conditions, anaerobic bacteria and some parasites (including T. vaginalis) metabolize metronidazole into nitroso radicals, which break microbial DNA and cause cell lysis. 15,16 Culture studies have shown that lactobacilli are not sensitive to metronidazole. 17 Metronidazole has thus far been associated with low levels of antimicrobial resistance. 18 The high recurrence rate is therefore often hypothesized to be due to vaginal mucosal biofilm formation by G. vaginalis and other BV-associated anaerobes but this has not been confirmed. 19 We investigated the impact of the most commonly used oral metronidazole treatment regimen, 500 mg twice per day for 7 days, on the VMB of Rwandan women with BV and/or T. vaginalis. We compared their VMB compositions as assessed by microscopy and sequencing pre-and posttreatment, and determined sociodemographic and biological correlates of treatment failure.

Materials and Methods
Data were collected at the Rinda Ubuzima research clinic in Kigali, Rwanda, in 2015 (flow diagram in Appendix A, Figure A.1). HIV-negative, nonpregnant women, aged 18e45 years, and in good overall physical and mental health, were screened at a pretreatment visit. Recruitment targeted women at high risk of BV/T. vaginalis, defined as having had more than 1 sex partner, or having been treated for BV and/or a sexually transmitted infection, in the last 12 months. Women who were BV-positive (by Nugent 7e10 and/or modified Amsel criteria as defined below) and/or T. vaginalis-positive (by wet mount and/ or by culture), regardless of symptomatology, were treated with 7 days of 500 mg generic oral metronidazole (Tricozole; Laboratory & Allied, Ltd, Nairobi, Kenya) twice daily. Women also were tested, and treated or referred, for pregnancy, HIV, syphilis, Chlamydia trachomatis, Neisseria gonorrhoeae, vulvovaginal candidiasis, and urinary tract infections, using local guidelines. 20 Women returned for a posttreatment visit within 3 days after metronidazole treatment completion and were retested for BV, T. vaginalis, and vulvovaginal candidiasis. Dacron vaginal swabs for molecular VMB testing were collected during speculum examinations at both visits, and stored dry at e80 C.

Diagnostic procedures
BV was diagnosed by Gram stain Nugent scoring (a score of 0e3 was considered optimal, 4e6 intermediate microbiota, and 7e10 BV), 8  To evaluate the impact of 7-day oral metronidazole treatment on the vaginal microbiota of Rwandan bacterial vaginosis patients using microscopy and 16S rRNA gene sequencing, and correlates of treatment failure.

Key findings
Metronidazole treatment decreased bacterial vaginosis-associated anaerobes, but not by as much as expected, and increased lactobacilli. Pathobionts (bacteria typically associated with hospital infections and neonatal sepsis) were not affected. Treatment failure was associated with high levels of pretreatment Gardnerella vaginalis or pathobionts.

What does this add to what is known?
The sequencing data explain why metronidazole treatment is associated with suboptimal cure and high recurrence rates. Women with high pretreatment Gardnerella may need biofilm-disrupting treatment, and women with high pretreatment pathobionts may need additional antibiotics.

Statistical analyses
Statistical analyses were performed using Stata, version 13 (StataCorp, College Station, TX) and R. VMB characteristics pre-and posttreatment were compared using the StuarteMaxwell test for matched categorical data, and Wilcoxon signed-rank test for matched continuous data, for all women (N¼68), and women stratified by treatment success/failure (women who had Nugent 7e10 pretreatment and a valid Nugent score at posttreatment; N¼55), pretreatment C. trachomatis/N. gonorrhoeae status (results became available after metronidazole treatment completion; N¼67 due to 1 missing result), having received another antibiotic in addition to metronidazole at the pretreatment visit or not (N¼68), and having reported unusual vaginal discharge pretreatment or not (N¼68). Successful treatment was defined as Nugent 7e10 pretreatment and Nugent 0e3 posttreatment, and treatment failure as Nugent 7e10

Ethical statement
All participants provided written informed consent. The study was conducted in accordance with the Helsinki Declaration, and approved by the National Ethics Committee of Rwanda and the University of Liverpool Research Ethics Subcommittee for Physical Interventions.

Results
We screened 176 women, and 68 women completed metronidazole treatment (ineligibility reasons in Appendix A, Figure Table 2). Pre-and posttreatment 16S microbiota data ( Figure 1) shows a shift toward increased relative abundance of lactobacilli (mainly L. iners) and decreased relative abundances of several BVanaerobes ( Figure 1). The mean bacterial group relative abundance data confirmed this (Table 2, Figure 2, B) and additionally showed that the mean relative abundance of pathobionts increased posttreatment (Wilcoxon signed rank P¼.050). Metronidazole treatment was associated with a significant decrease in the mean concentration of total bacteria from 6.59 log 10 /mL pretreatment to 5.85 log 10 /mL posttreatment (P<.001; Table 2). The mean BV-anaerobes concentration decreased from 6.23 log 10 /mL to 4.55 log 10 /mL (P<.001), the mean Lactobacillus concentration increased from 4.98 log 10 /mL to 5.56 log 10 /mL (P¼.017), and the mean concentrations of pathobionts (1.92 log 10 /mL pretreatment and 2.01 log 10 /mL posttreatment; P¼.939) and "other bacteria" (1.85 log 10 /mL pretreatment and 1.46 log 10 /mL posttreatment; P¼.176) did not change significantly. Among lactobacilli, the concentrations of L. iners, L. crispatus, and "other lactobacilli" (mostly L. jensenii and L. gasseri) all increased, with L. iners having the greatest concentrations before and after treatment, but "other lactobacilli" achieving the greatest concentration increase ( Table 2). The median vaginal pH decreased from 5.3 to 4.4 (P<.001). Among BV-anaerobes, the concentrations of 8 common BV-anaerobes that were individually tested (Gardnerella, Atopobium, Prevotella, Sneathia, Megasphaera, Veillonella, and Fusobacterium species, and BV-associated bacterium type 1) decreased ( Table 2). The   a StuarteMaxwell test for matched categorical data and Wilcoxon signed-rank test for matched continuous data; b Successful treatment was defined as having a Nugent score of 7e10 before treatment and 0e3 after treatment (n¼30), whereas treatment failure was defined as having a Nugent score of 7e10 before treatment and 4e10 after treatment (n¼25). Thirteen women were excluded from these analyses because they did not have Nugent 7e10 at the pretreatment visit (n¼12) or did not have a valid Nugent result at the posttreatment visit (n¼1); c Valid Nugent data available for 67 participants at the pretreatment visit and 66 participants at the posttreatment visit; d The definition of treatment success/failure was based on Nugent scores and these P values are therefore meaningless; e Relative abundance, Simpson diversity indices, and VMB type data available for 67 participants at each visit; f Concentration data may contain at most 5 missing values (see Appendix A Methods); g Includes all amplicon sequence variants attributed to L. crispatus, also those with multiple species assignments; h Includes amplicon sequence variants attributed to L. jensenii, L. delbrueckii, L. fermentum, L. gasseri, L. johnsonii, and Lactobacillus genus, as well as 11 other minority amplicon sequence variants. Verwijs et al. Impact of oral metronidazole treatment on the vaginal microbiota and correlates of treatment failure. Am J Obstet Gynecol 2020.

Original Research GYNECOLOGY
ajog.org concentrations of the 3 most common pathobionts in our dataset (Streptococcus, Staphylococcus, Escherichia/ Shigella species) did not change significantly (Table 2).
Although the mean trends were clear, the interindividual variability was high (Figure 3, AeG). Of the 61 participants for whom pre-and posttreatment concentration data were available, most had decreases in total bacterial concentration (n¼45) and BV-anaerobes (n¼52), but not everyone (Figure 3, E and G). BVanaerobes were completely eradicated in only 3 women (4.9%), and the concentration was reduced by more than 50% in an additional 7 women (11.5%). Pathobiont concentrations showed the most interindividual variability (Figure 3, H).
The mean inverse Simpson diversity index was 0.67 pretreatment and 0.31 posttreatment (P<.001; Table 2). Metronidazole treatment changed the proportions of women with certain VMB types based on hierarchical clustering results: the proportions of women with lactobacilli-dominated VMB types (Li and Lo) and the mixed LA VMB type increased, whereas the proportions of women with the 3 BV-associated VMB types (BV_GV, BV_noGV, and to a lesser extent GV) decreased (Table 2, Figure 2, C). The number of women with a pathobionts-containing VMB type increased from 1 (1.5%) pretreatment to 6 (9.0%) posttreatment. Alluvial diagrams show that the majority of women transitioned from the 3 BV-associated VMB types into VMB types containing  Figure A.2, A).
Participants with treatment failure as defined by Nugent scoring had a lower mean lactobacilli relative abundance, smaller decreases in mean total bacteria and BV-anaerobes concentrations, and less often a lactobacilli-dominated VMB type, posttreatment (Appendix A, Figure A.3, AeD). They also had a greater mean pathobionts concentration pretreatment (Appendix A,  Heatmaps of 16S rRNA gene sequencing data pre-and posttreatment AeB, Heatmaps at the pretreatment (A) and the posttreatment visit (B) depicting the 20 amplicon sequence variants with the greatest mean relative abundance on the y-axis and samples (N¼67 at each visit) on the x-axis. The dendrogram above the heatmap depicts vaginal microbiota clusters based on Euclidean distance. The bar below the dendrogram depicts Nugent score categories (see legend; black means no score available). ajog.org GYNECOLOGY Original Research significant decreases in the mean concentrations of all 8 common BVanaerobes in our dataset that were tested individually, but unsuccessfully treated participants did not have decreases in G. vaginalis and BVAB1 (Table 2). Simpson diversity did not differ by treatment success/failure (Appendix A: Figure A.3, E). In logistic regression models, we did not identify any statistically significant sociodemographic or biological correlates of treatment failure, except for pathobionts concentration pretreatment (P¼.044; Table 3). Pretreatment C. trachomatis/N. gonorrhoeae status, having received another antibiotic in addition to metronidazole, and reporting unusual vaginal discharge at the pretreatment visit, did not modify the effects of metronidazole treatment on the VMB (Appendix A: Tables A.2, A.3, A.4; Figures A.4, A.5, A.6).

Principal findings
In this study among high-risk women in Rwanda diagnosed with BV and/or T. vaginalis, the cure rate of 7-day oral metronidazole treatment by Nugent scoring was only 54.5%. The sequencing data showed a decrease in BV-anaerobes (but a reduction of more than 50% in only 16.4% of women), an increase in lactobacilli, and no change in Nugent scores, mean bacterial group relative abundances, and vaginal microbiota types pre-and posttreatment AeC, Changes in vaginal microbiota characteristics before and after metronidazole treatment. A, Nugent scores, B, mean bacterial group relative abundances, and C, vaginal microbiota types. D, Three-dimensional non-metric multidimensional scaling plot based on rarefied relative abundances of samples before and after metronidazole treatment. The figure shows that samples cluster together by visit (and hence, treatment status). Original Research GYNECOLOGY ajog.org pathobionts. Treatment failure was associated with greater levels of pretreatment Gardnerella vaginalis or pathobionts levels but not with sociodemographic factors.

Results of the study in context of what is known
Metronidazole treatment resulted in a mean BV-anaerobes concentration reduction (as well as T. vaginalis eradication), which is in agreement with a priori knowledge about the mechanism of action of metronidazole. 15,16 However, the extent of BV-anaerobes reduction was more modest than expected, with only 16.4% of women having a reduction of more than 50%. The mean lactobacilli concentration increased, and mean concentrations of pathobionts and "other bacteria" were low pre-and posttreatment, resulting in an overall 5.5-fold reduction of total bacterial concentration (from 6.59 to 5.85 log 10 /mL). The observed increase in lactobacilli is consistent with culture studies showing that lactobacilli are not sensitive to metronidazole 17 but is inconsistent with claims made by some clinical researchers that the high BV recurrence rate may be due to detrimental effects of metronidazole on lactobacilli. 14 The reduction in BV-anaerobes concentration clearly allows lactobacilli, which are not affected by metronidazole, to expand. 26e28 In our study population of high-risk Rwandan women, L. iners was by far the most common Lactobacillus species pre-and posttreatment (4.81 and 5.28 log 10 /mL, respectively), "other lactobacilli" (which includes L. jensenii) increased the most during treatment (from 1.46 to 3.03 log 10 / mL), and L. crispatus was uncommon and increased only slightly (from 0.15 to 0.51 log 10 /mL). A metronidazole study in American women also showed that L. iners and L. jensenii concentrations increased more than the L. crispatus concentration. 29 Women with a pretreatment VMB type containing a relative abundance of >50% G. vaginalis, compared with 50%, were more likely to continue to have a dysbiotic VMB type posttreatment, but pretreatment G. vaginalis concentration (as a continuous variable) was not associated Individual bacterial group concentrations pre-and posttreatment A-D, Changes in total bacterial concentrations and bacterial group concentrations before (n¼66) and after metronidazole treatment (n¼63). A, Total bacterial concentration, B, total lactobacilli, C, total BV-anaerobes, and D, total pathobionts (boxplot not shown because of high proportion of zero values). See Table 2 for mean concentrations and 95% confidence intervals, and statistical significance. E-H, Change in concentrations between pre-and posttreatment, expressed as a percentage for every individual participant with valid quantitative polymerase chain reaction results at both visits (n¼61), for total bacterial concentration (E), total Lactobacillus (F), total BV-anaerobes (G), and total pathobionts (H). In some women, concentration went from zero to non-zero; these increases were set at 100% or the greatest increase observed among the other participants, whichever was greatest.   13,29 and with the G. vaginalis-containing biofilm hypothesis of treatment failure. 10 Metronidazole may eliminate dispersed G. vaginalis at low to modest concentrations but may no longer be able to do so when a biofilm (containing a high concentration of G. vaginalis) has been established. However, other hypotheses have also been posited. A recent metatranscriptomics study showed that the VMB of BV patients with treatment failure contained G. vaginalis with upregulated clustered regularly interspaced short palindromic repeat-associated (CRISPR)-genes, which may protect the bacteria against metronidazole. 30 In our study, the concentrations of all other key BV-associated bacteria, including A. vaginae, were effectively reduced by metronidazole and were not associated with treatment failure. This is in accordance to one study, 29 but in contrast to others that have suggested that pretreatment presence or concentration of A. vaginae was associated with treatment failure. 31e33 The concentrations of the pathobionts bacterial group were low pre-and posttreatment (1.92 and 2.01 log 10 /mL, respectively). However, pathobionts have greater pathogenicity potential than BV-anaerobes 3 and may therefore be clinically relevant even at low concentrations. Unfortunately, it is currently unknown which concentrations of which pathobionts in the vagina should be treated to prevent complications, such as transmission to neonates. In our study, metronidazole treatment did not change the pathobionts concentration but did increase the relative abundance due to the reduction in total bacterial concentration. Furthermore, a greater pretreatment pathobionts concentration was associated with increased likelihood of treatment failure. None of the sociodemographic factors that have been associated with VMB composition in other studies, including menses in the 7 days prior to the posttreatment visit, 34 were associated with treatment failure in our study.

Research and clinical implications
Our study has several implications. Women with persistent or recurrent BV might benefit from vaginal biofilmdisrupting treatment, adjuvant therapy with lactobacilli-based live biotherapeutics, or treatment with drugs that specifically target all G. vaginalis clades. The former 2 are actively researched, 10,35,36 but the latter are not yet available. Whether these strategies are efficacious in real life would have to be evaluated in clinical trials. Furthermore, diagnostic tests to determine the presence of a biofilm or concentrations of G. vaginalis are not yet available to clinicians. Women at risk of complications caused by vaginal pathobionts (not just Group B streptococci), such as pregnant women, might benefit from targeted screening and treatment. We encourage the incorporation of quantitative molecular characterization of both key individual bacteria with pathogenic potential, as well as bacterial communities and biofilms, in future intervention studies.

Strengths and limitations of the study
Limitations of our study include potentially limited generalizability of the results to lower risk and non-African populations, 37 and the lack of vaginal biofilm detection and characterization pre-and posttreatment. Recent studies have shown good correlations between the method that we used to quantify relative abundance data and speciesspecific quantitative polymerase chain reaction results of non-minority species, 38,39 but additional validation studies are desirable. Although Nugentbased studies have shown that oral and  11,12 molecular studies comparing the in-depth microbiological effects of different metronidazole formulations are desirable. A major strength of our study is that we used multiple laboratory and analytic methods to characterize the VMB, including methods that incorporated a priori knowledge about the pathogenic potential of specific microorganisms and the types of communities in which they typically live.

Conclusion
Oral metronidazole treatment alone may not be sufficient for women with recurrent BV or for women at risk of complications caused by pathobionts (such as pregnant women). Additional treatments are urgently needed, including biofilm-disrupting treatments and drugs that specifically target all G. vaginalis clades or pathobionts. n