Impact of Periodic Presumptive Treatment for Bacterial Vaginosis on the Vaginal Microbiome among Women Participating in the Preventing Vaginal Infections Trial

Abstract

Background.

Evidence suggests that specific vaginal bacteria associated with bacterial vaginosis (BV) may increase the risk of adverse health outcomes in women. Among women participating in a randomized, double-blinded trial, we assessed the effect of periodic presumptive treatment (PPT) on detection of select vaginal bacteria.

Methods.

High-risk women from the United States and Kenya with a recent vaginal infection received intravaginal metronidazole 750 mg plus miconazole 200 mg or placebo for 5 consecutive nights each month for 12 months. Vaginal fluid specimens were collected via polyester/polyethylene terephthalate swabs every other month and tested for bacteria, using quantitative polymerase chain reaction (PCR) assays targeting the 16S ribosomal RNA gene. The effect of PPT on bacterium detection was assessed among all participants and stratified by country.

Results.

Of 234 women enrolled, 221 had specimens available for analysis. The proportion of follow-up visits with detectable quantities was lower in the PPT arm versus the placebo arm for the following bacteria: BVAB1, BVAB2, Atopobium vaginae, Leptotrichia/Sneathia, and Megasphaera. The magnitude of reductions was greater among Kenyan participants as compared to US participants.

Conclusions.

Use of monthly PPT for 1 year reduced colonization with several bacteria strongly associated with BV. The role of PPT to improve vaginal health should be considered, and efforts to improve the impact of PPT regimens are warranted.

Bacterial vaginosis (BV) is a highly prevalent vaginal infection, affecting hundreds of millions of reproductive-aged women globally every year [1–3]. It is a polymicrobial condition that frequently recurs following successful treatment [4–6] and is associated with several adverse outcomes, including human immunodeficiency virus type 1 (HIV-1) infection, other sexually transmitted infections, and preterm birth [7–10]. A number of bacterial species are associated with BV. However, the exact etiology remains unknown [3]. Over the past decade, molecular techniques that detect bacterial genes, such as the gene encoding 16S ribosomal RNA (rRNA), have been used to characterize the vaginal microbiota. These molecular approaches highlight the immense bacterial diversity in the human vagina, including several new BV-associated species [11–13]. In addition, recent evidence suggests that specific BV-associated bacteria are independently associated with adverse health outcomes in women, such as HIV-1 acquisition, preterm birth, and cervicitis [14–17].

Given the high BV prevalence globally, frequency of recurrence, and associations with multiple adverse health outcomes in women, innovative treatment and prevention approaches are needed to reduce the burden of BV. Suppressive therapies, including periodic presumptive treatment (PPT), have been evaluated as strategies to reduce BV prevalence and incidence [5, 18–20]. However, there is limited understanding of the impact of suppressive therapy on specific bacterial species in the vaginal microbiome. We recently completed a randomized trial of monthly PPT, which reduced BV by 35%, compared with placebo [19]. The objective of the present analysis was to further assess the effect of the intervention on the detection and quantity of key vaginal bacterial species.

METHODS

Study Population and Procedures

This is a secondary analysis of data from women participating in the Preventing Vaginal Infections (PVI) trial, a double-blinded, randomized, controlled trial that assessed the effect of monthly periodic presumptive treatment (PPT) using topical metronidazole 750 mg with miconazole 200 mg intravaginal suppositories versus matching placebo nightly for five consecutive nights each month for 12 months to reduce rates of BV and vulvovaginal candidiasis (clinical trials registration NCT01230814). Detailed methods and results for the trial have been published elsewhere [19]. Briefly, 234 high-risk women were enrolled between May 2011 and August 2012 from 3 sites in Kenya and 1 in the United States. Eligible women were HIV-1 uninfected, 18–45 years of age, not pregnant or breastfeeding, and sexually active and had a vaginal infection at screening (BV, confirmed on the basis of the Nugent score [21]; vulvovaginal candidiasis, on the basis of saline/potassium hydroxide wet preparation plus positive culture for yeast on Sabouraud’s agar; and/or Trichomonas vaginalis infection, on the basis of a saline wet preparation). Women with symptomatic BV or with vulvovaginal candidiasis or T. vaginalis infection regardless of symptoms were provided treatment and returned for enrollment 7–14 days later. Eligible participants were randomly assigned in equal proportions to receive either the intervention or the matching placebo. All participants provided written informed consent for participation. Participants were asked to provide separate written consent for the storage and future testing of biological specimens, including those used for the present analysis. The trial was approved by the human subjects research committees at the University of Washington (Seattle), the University of Alabama at Birmingham, and the Kenyatta National Hospital (Nairobi).

At enrollment, structured face-to-face interviews were conducted to collect data on demographic, behavioral, and clinical characteristics. At monthly follow-up visits, data were collected on study product use, genital tract symptoms, sexual behaviors, intravaginal practices, and contraceptive use, and a urine pregnancy test was performed. Nonpregnant participants received a 1-month supply of study product and free male condoms. At enrollment and during follow-up visits at months 2, 4, 6, 8, 10, and 12, participants underwent a physical examination, including pelvic speculum examination, with collection of genital swabs for diagnosis of genital tract infections. Vaginal fluid was collected for polymerase chain reaction (PCR) testing of vaginal bacteria, using a push-off polyester/polyethylene terephthalate swab (FitzCo), and stored at −80°C. If a participant missed an examination visit, a physical examination was performed at her next follow-up visit.

Laboratory Procedures

Vaginal Gram-stained slides were evaluated for BV by trained technicians, using the Nugent score [21]. Vaginal swabs for bacterial quantitative PCR (qPCR) assays were transported on dry ice to the Fred Hutchinson Cancer Research Center (Seattle) for analysis. A MoBio BiOstic Bacteremia DNA Isolation kit (Carlsbad, CA) was used to extract and purify DNA. This protocol uses bead beating and chaotropic lysis to break apart bacterial cells and recover DNA that is free of PCR inhibitors. Sham extraction controls that did not contact a mucosal surface were included to monitor contamination from extraction reagents and collection swabs. All samples were subjected to 2 quality control assays. They were evaluated for PCR inhibitors, using exogenously added DNA (aequorin plasmid) and aequorin gene qPCR [22]. To verify contact of the vaginal swab with a human mucosal surface, we used a broad-range 16S rRNA qPCR assay to measure total bacterial concentrations on all samples [12]. Bacterium-specific qPCR assays were performed using purified DNA and targeted the following bacteria: Lactobacillus crispatus, Lactobacillus jensenii, Lactobacillus iners, BV-associated bacterium 1 (BVAB1), BVAB2, Mageeibacillus indolicus, Atopobium vaginae, Leptotrichia/Sneathia species, Megasphaera species, and Gardnerella vaginalis, as previously described [23, 24]. No-template water controls were included with all PCR runs, to monitor contaminants from PCR reagents.

Statistical Analysis

Our primary objectives were to assess the effect of the PVI trial intervention on detection and concentrations of vaginal bacteria associated with vaginal health or BV, using a modified intent-to-treat approach. Participants were excluded from the analysis if they did not consent to future testing of stored specimens or if they did not return for any follow-up visits. Enrollment characteristics, stratified by study arm, were summarized using descriptive statistics, and differences were assessed using χ² analysis or the Fisher exact test (where appropriate) for categorical factors and the Wilcoxon rank sum test for continuous factors. For each bacterium, the proportion of follow-up visits with detectable quantities (ie, values above the lower limit of detection [LLD]) was calculated by arm. Relative risks (RRs) were generated using generalized estimating equations with a log link and exchangeable correlation structure to separately assess the effect of the intervention on species detection. Evidence suggests that the vaginal microbiota may differ by geographic region [25]. In addition, there were differences by country in key baseline factors known to be associated with BV in the primary analysis. To investigate the potential impact of these differences on our primary results, we performed a secondary analysis stratifying by country. Per protocol, study product was withheld if a participant tested positive for pregnancy; therefore, we performed a sensitivity analysis that excluded visits where pregnancy was detected.

Since we used qPCR assays, which provide bacterial concentrations (rather than broad-range PCR with sequencing, which provides relative abundance), we planned to assess the impact of the intervention on bacterial concentration in addition to detection. The qPCR assays used in this study are highly sensitive and can detect bacterial quantities as low as 94 (log₁₀ 1.97) copies/swab. However, for many of the bacteria assessed, a large proportion of specimens had values below the LLD, precluding analyses using linear regression models, owing to an excessive number of samples with values below the LLD for some species. In addition, the range of quantities detected was quite large (Supplementary Table 1). Data are limited concerning the relationship between BV and the quantity of specific bacterial species. Therefore, we performed an exploratory analysis to evaluate changes in species quantities that may have a clinically meaningful relationship to BV. We used baseline data from all participants to generate receiver operating characteristic (ROC) curves of the log₁₀ quantity of species versus the presence of BV by Nugent score at enrollment in order identify a quantity threshold that was predictive of BV. For each species, we noted the log₁₀ quantity corresponding to the highest proportion correctly classified as being BV positive or BV negative on the basis of the Nugent score and calculated the sensitivity and specificity for BV at that quantity value. We use the term “ROC cutoff” to refer to this value. A similar approach has been used in other areas of public health to assess relationships between organism burden and clinical disease [26]. The same statistical methods used for the primary analysis were implemented to repeat our analyses, using the ROC cutoff as the outcome. All statistical tests were assessed using a 2-sided significance level of 0.05. Analyses were conducted using Stata, version 14·0 (StataCorp, College Station, TX).

RESULTS

Of the 234 participants enrolled, 232 (99%) returned for at least 1 follow-up visit, of whom 221 (95%; 111 [50%] in the intervention arm and 110 [50%] in the placebo arm) provided consent for future testing of stored specimens. Demographic, behavioral, and clinical characteristics at enrollment are presented by study arm and by country in Table 1. Characteristics of participants in the intervention arm were generally similar to those in the placebo arm. Participant follow-up did not differ by study arm, with 87% of participants in the intervention arm attending ≥80% scheduled study visits, compared with 89% of participants in the placebo arm (P = .55). In contrast, a number of baseline characteristics differed between US and Kenyan participants (Table 1).

Table 1.

Baseline Characteristics of Participants in Microbiome Analysis, by Study Arm and by Country

Characteristic	Placebo (n = 110)	Intervention (n = 111)	P a	US (n = 53)	Kenya (n = 168)	P a
Age, y	29 (23–34)	29 (2–34)	.47	30 (23–36)	29 (24–34)	.99
Education duration, y	11 (8–12)	10 (8–13)	.61	13 (12–15)	9 (8–12)	<.001
Black raceb	106 (96)	111 (100)	.06	49 (92)	168 (100)	.003
Partnership status			.43			<.001
Married or living with a partner	29 (26)	34 (31)		15 (28)	48 (29)
Separated, divorced, or widowed	48 (44)	39 (35)		8 (15)	79 (47)
Never married	33 (30)	38 (34)		30 (57)	21 (24)
No. of live births	2 (1–3)	2 (1–3)	.97	1 (0–2)	2 (1–3)	<.001
Current family planning method			.46			<.001
None	16 (15)	23 (21)		12 (23)	27 (16)
Condoms only	30 (27)	31 (28)		11 (21)	50 (30)
Oral contraceptives	12 (11)	11 (10)		5 (9)	18 (11)
Injectable contraceptives	25 (23)	24 (22)		3 (6)	46 (27)
Intrauterine device	10 (9)	4 (4)		4 (8)	10 (6)
Implant	9 (8)	5 (5)		1 (2)	13 (8)
Tubal ligation	5 (5)	10 (9)		14 (26)	1 (1)
Otherc	3 (3)	3 (3)		3 (6)	3 (2)
Currently smoke cigarettes	10 (9)	20 (18)	.05	18 (34)	12 (7)	<.001
Vaginal washing in the past month	55 (50)	56 (50)	.95	20 (38)	91 (54)	.04
Ever had sex in exchange for goods/money/ servicesd	60 (55)	59 (53)	.84	1 (2)	118 (70)	<.001
Sexual behaviors in the past week
Vaginal sex	2 (1–4)	2 (1–3)	.93	1 (1–3)	2 (1–4)	<.001
Unprotected sex			.40			.13
No sex	20 (18)	21 (19)		10 (19)	31 (18)
100% condom use	53 (48)	42 (38)		19 (36)	76 (45)
Intermittent condom use	11 (10)	12 (11)		3 (6)	20 (12)
No condom use	26 (24)	36 (32)		21 (40)	41 (24)
No. of partners	1 (1–2)	1 (1–2)	.25	1 (1–1)	1 (1–3)	<.001
New partner	23 (21)	22 (20)	.84	1 (2)	44 (26)	<.001
History of anal sex	13 (12)	12 (11)	.84	21 (40)	4 (2)	<.001
Sex with a woman in the last 12 months	3 (3)	1 (1)	.31	4 (8)	0 (0)	<.001
Clinical
Gonorrhea	0 (0)	3 (3)	.35	0 (0)	3 (3)	1.00
Chlamydia	8 (7)	8 (7)	.99	2 (4)	14 (8)	.37
HSV-2e	68 (62)	71 (64)	.74	29 (55)	110 (65)	.16
Trichomonas vaginalis f	6 (5)	10 (9)	.33	3 (6)	13 (8)	.77
Vulvovaginal candidiasis	24 (22)	28 (25)	.55	15 (28)	37 (22)	.35
Bacterial vaginosis, by Nugent score (7–10)	40 (36)	41 (37)	.76	25 (47)	56 (33)	.02
Intermediate (4–6)	20 (18)	24 (22)		4 (8)	40 (24)
Normal (0–3)	50 (45)	46 (41)		24 (45)	74 (43)
Cervicitisg	14 (13)	18 (16)	.48	29 (56)	3 (2)	<.001

Data are no. (%) of women or median value (interquartile range).

^aBy the Mantel-Haenszel test, stratified by site, or by the χ² test or Fisher exact test (for categorical factors) or Wilcoxon rank sum test (for continuous factors).

^bAll participants enrolled in Kenya reported black race. Of those enrolled at the US site, 49 of 53 women reported identifying as black or African American race.

^cOther includes Essure, Bayer (n = 1), fertility awareness method (n = 2), herbal pill (n = 1), partner vasectomy (n = 1), and withdrawal (n = 1).

^dTwo of the sites in Kenya recruited participants from longitudinal cohort studies following women who reported a history of engaging in transactional sex.

^eKenyan participants with an optical density >2.1 were considered positive for herpes simplex virus type 2 (HSV-2).

^fOne participant from Mombasa site missing baseline TV result. TV by NAAT.

^gDefined as ≥30 polymorphonuclear cells per high-powered field. One participant from the US was missing a result.

At enrollment, the proportion of participants with bacterial quantities below the LLD cutoff did not differ by arm (Table 2, Figure 1). Overall, L. iners and G. vaginalis were the most prevalent species at enrollment (88.2% and 92.8% of visits, respectively), while L. jensenii and BVAB1 were the least prevalent (20.8% and 21.7%, respectively). Bacterial detection at enrollment differed by country, with a higher proportion of US participants having detectable levels of L. jensenii, L. iners, BVAB1, and M. indolicus as compared to Kenyan participants (Table 2, Figure 1).

Table 2.

Proportion of Participants with Bacteria Detected at Enrollment, by Study Arm and Country

Bacteria	Bacterial Level Above LLD			Bacterial Level at or Greater Than ROC Cutoff
	Intervention (n = 111)	Placebo (n = 110)	P a	Log 10 ROC Cutoff	Intervention (n = 111)	Placebo (n = 110)	P a
L. crispatus	26 (23.4)	28 (25.5)	.73	2.15	10 (9.0)	15 (13.6)	.73
L. jensenii	20 (18.0)	26 (23.6)	.30	2.30	10 (9.0)	18 (16.4)	.30
L. iners	99 (89.2)	96 (87.3)	.66	2.24	90 (81.1)	84 (76.4)	.44
BVAB1	23 (20.7)	25 (22.7)	.72	6.80	18 (16.2)	18 (16.4)	.98
BVAB2	58 (52.2)	55 (50.0)	.74	7.64	42 (37.8)	40 (36.4)	.62
M. indolicus	35 (31.5)	34 (30.9)	.92	6.59	25 (22.5)	24 (21.8)	.97
A. vaginae	84 (75.7)	83 (75.5)	.97	8.68	67 (60.4)	59 (53.6)	.94
Leptotrichia/Sneathia species	83 (74.8)	75 (68.2)	.28	8.73	63 (56.8)	58 (52.7)	.95
Megasphaera species	56 (50.5)	50 (45.5)	.46	8.01	49 (44.1)	42 (38.2)	.36
G. vaginalis	106 (95.5)	99 (90.0)	.12	8.84	92 (82.9)	83 (75.5)	.62
	US (n = 53)	Kenya (n = 168)	P a		US (n = 53)	Kenya (n = 168)	P b
L. crispatus	18 (34.0)	36 (21.4)	.06	2.15	18 (34.0)	36 (21.4)	.06
L. jensenii	20 (37.7)	26 (15.5)	.001	2.30	20 (37.7)	26 (15.5)	.001
L. iners	51 (96.2)	144 (85.7)	.04	2.24	44 (83.0)	110 (65.5)	.02
BVAB1	25 (47.2)	23 (13.7)	<.001	6.80	17 (32.1)	17 (10.1)	<.001
BVAB2	30 (56.6)	83 (49.4)	.36	7.64	24 (45.3)	66 (39.3)	.44
M. indolicus	24 (45.3)	45 (26.8)	.01	6.59	19 (35.9)	31 (18.5)	.008
A. vaginae	45 (84.9)	122 (72.6)	.07	8.68	25 (47.2)	76 (45.2)	.81
Leptotrichia/Sneathia species	35 (66.0)	123 (73.2)	.31	8.73	24 (45.3)	72 (42.9)	.76
Megasphaera species	28 (52.8)	78 (46.4)	.42	8.01	23 (43.4)	62 (36.9)	.40
G. vaginalis	48 (90.6)	157 (93.5)	.48	8.84	24 (45.3)	66 (39.3)	.44

Data no. (%) of women. ROC curves, proportion correctly classified at the specified cutoff, sensitivity, and specificity are presented in Supplementary Figure 1 and Supplementary Table 1.

M-H was used for the P value between study arms and chi-square was used for P value by country. This applies to both LLD and ROC cut-offs

Abbreviations: A. vaginae, Atopobium vaginae; BVAB, bacterial vaginosis–associated bacterium; G. vaginalis, Gardnerella vaginalis; L. crispatus, Lactobacillus crispatus; L. iners, Lactobacillus iners; L. jensenii, Lactobacillus jensenii; LLD, lower limit of detection; M. indolicus, Mageeibacillus indolicus; ROC, receiver operating characteristic.

^aBy the Mantel-Haenszel test, stratified by site.

^bBy the χ² test.

Figure 1.

Bacterial detection at enrollment, by study arm and country. A, Detection at levels above the lower limit of detection (LLD), by study arm. B, Detection at levels at or above the receiver operating characteristic (ROC) curve cutoff, by study arm. C, Detection at levels above the LLD, by country. D, Detection at levels at or above the ROC cutoff, by study arm. Abbreviations: A. vaginae, Atopobium vaginae; BVAB, bacterial vaginosis–associated bacterium; G. vaginalis, Gardnerella vaginalis; L. crispatus, Lactobacillus crispatus; L. iners, Lactobacillus iners; L. jensenii, Lactobacillus jensenii; M. indolicus, Mageeibacillus indolicus.

In unadjusted bacterium-specific models, the proportion of follow-up visits with BV-associated species at levels below the LLD was lower in the PPT arm, compared with the placebo arm, for the following bacteria: BVAB1, BVAB2, A. vaginae, Leptotrichia/Sneathia species, and Megasphaera species (Table 3). The proportion of visits with L. crispatus, L. jensenii, or L. iners was higher for each species, but these increases were smaller in magnitude and not statistically significant. Results were similar after adjustment for study site (Table 3) and were also similar in a sensitivity analysis excluding visits where participants were pregnant and did not receive study product (data not shown). In analyses stratified by country, among Kenyan participants we observed reductions in the proportion of visits with values below the LLD in the intervention arm, compared with the placebo arm, for BVAB1, BVAB2, A. vaginae, Leptotrichia/Sneathia species, Megasphaera species, and M. indolicus. However, among US participants the effect sizes were smaller in magnitude, with no significant differences by arm (Figure 2 and Supplementary Table 2). In light of these differences by country, we explored self-reported adherence by country and observed that 99% of Kenyan participants reported using all 5 doses of study product at each visit where product was dispensed, compared with 88% of US participants (P < .001).

Table 3.

Prevalence of Bacteria at Levels Above the Lower Limit of Detection (LLD) and at or Above the Receiving Operating Characteristic (ROC) Curve Cutoff, by Study Arm During Follow-up

Bacteria, by Outcome	Placebo a (n = 630)	Intervention a (n = 616)	RR b (95% CI)	P	aRR c (95% CI)	P
Bacterial quantity above LLD
L. crispatus	173 (27.5)	199 (32.3)	1.20 (.91–1.56)	.19	1.19 (.91–1.54)	.20
L. jensenii	161 (25.6)	197 (32.0)	1.30 (0.96, 1.76)	.09	1.29 (0.96, 1.73)	.09
L. inersd	558 (88.6)	571 (92.7)	1.05 (0.98, 1.12)	.15	…
BVAB1	149 (23.7)	89 (14.5)	0.63 (0.41, 0.96)	.03	0.65 (0.47, 0.90)	.009
BVAB2	267 (42.4)	194 (31.5)	0.74 (0.56, 0.97)	.03	0.78 (0.60, 1.00)	.05
M. indolicus	198 (31.4)	142 (23.1)	0.74 (0.54, 1.02)	.07	0.77 (0.57, 1.04)	.09
A. vaginae	458 (72.7)	366 (59.4)	0.81 (0.71, 0.93)	.004	0.84 (0.73, 0.96)	.009
Leptotrichia/Sneathia species	384 (61.0)	304 (49.4)	0.81 (0.67, 0.96)	.02	0.82 (0.69, 0.97)	.02
Megasphaera species	276 (43.8)	167 (27.1)	0.62 (0.46, 0.82)	.001	0.67 (0.51, 0.87)	.003
G. vaginalis	572 (90.8)	541 (87.8)	0.96 (0.91, 1.03)	.25	0.95 (0.90, 1.01)	.08
Bacterial quantity at or above ROC cutoff
L. crispatus	173 (27.5)	199 (32.3)	1.20 (0.91, 1.56)	.19	1.19 (0.91, 1.54)	.20
L. jensenii	157 (24.9)	194 (31.5)	1.32 (0.97, 1.79)	.08	1.31 (0.97, 1.77)	.08
L. iners	438 (69.5)	489 (79.4)	1.14 (1.04, 1.26)	.008	1.13 (1.03, 1.24)	.01
BVAB1	92 (14.6)	47 (7.6)	0.50 (0.28, 0.91)	.02	0.54 (0.33, 0.90)	.02
BVAB2	203 (32.2)	133 (21.6)	0.66 (0.48, 0.92)	.02	0.70 (0.52, 0.96)	.03
M. indolicus	127 (20.2)	92 (14.9)	0.73 (0.49, 1.08)	.12	0.75 (0.51, 1.09)	.13
A. vaginae	283 (44.9)	174 (28.3)	0.62 (0.48, 0.80)	<.001	0.63 (0.50, 0.81)	<.001
Leptotrichia/Sneathia species	174 (27.6)	123 (20.0)	0.70 (0.51, 0.97)	.03	0.72 (0.53, 1.00)	.05
Megasphaera species	205 (32.5)	115 (18.7)	0.57 (0.40, 0.81)	.001	0.60 (0.43, 0.84)	.003
G. vaginalis	209 (33.2)	162 (26.3)	0.78 (0.61, 1.01)	.06	0.80 (0.63, 1.03)	.08

Abbreviations: aRR, adjusted relative risk; A. vaginae, Atopobium vaginae; BVAB, bacterial vaginosis–associated bacterium; CI, confidence interval; G. vaginalis, Gardnerella vaginalis; L. crispatus, Lactobacillus crispatus; L. iners, Lactobacillus iners; L. jensenii, Lactobacillus jensenii; M. indolicus, Mageeibacillus indolicus; RR, relative risk.

^aThe numerator represents the number of follow-up visits where the bacteria of interest was detected at levels above the LLD, and the denominator is the number of follow-up visits with microbiome testing.

^bData are from generalized estimating equations with a log link and exchangeable correlation structure and clustered by participant.

^cData are from generalized estimating equations with a log link and exchangeable correlation structure and clustered by participant, with adjustment for study site.

^dConvergence not achieved for the L. iners model with adjustment for site.

Figure 2.

Forest plot of the effect of the Preventing Vaginal Infections trial intervention on detection of bacterial species above different thresholds among all participants and by country. Abbreviations: A. vaginae, Atopobium vaginae; BVAB, bacterial vaginosis–associated bacterium; CI, confidence interval; G. vaginalis, Gardnerella vaginalis; L. crispatus, Lactobacillus crispatus; L. iners, Lactobacillus iners; L. jensenii, Lactobacillus jensenii; LLD, lower limit of detection; M. indolicus, Mageeibacillus indolicus; ROC, receiver operating characteristic.

In our exploratory analysis that assessed the impact of the intervention on the proportion of species with bacterial quantities at or above the ROC cutoff, we noted a wide range of cutoffs (Table 2). Consistent with the primary analysis, no differences were noted by study arm at enrollment. However, a higher proportion of US participants also had quantities at or above the ROC cutoff for L. jensenii, L. iners, BVAB1, and M. indolicus, compared with Kenyan participants. Similar to the primary analysis, we observed significant decreases in the proportion of visits with bacterial quantities at or above the ROC cutoff among women in the intervention arm, compared with those in the placebo arm. The effect sizes were generally larger for the ROC cutoff analysis as compared to the analysis of quantities above the LLD. After we stratified by country, again there was an increase in the proportion of samples in the intervention arm with an L. iners value at or above the ROC cutoff and decreases in almost all BV-associated species among Kenyan participants in the intervention arm, compared with no differences among US participants (Figure 2 and Supplementary Table 2).

DISCUSSION

Use of monthly periodic presumptive treatment for 1 year significantly reduced colonization with several bacteria strongly associated with BV. In addition, the magnitude of the decrease in the proportion of women in the intervention arm with quantities at or above the ROC cutoff for BV-associated species was greater than the magnitude of the decrease observed using the LLD. Given that the ROC cutoffs for BV-associated species represented high quantities, ranging from log₁₀ 6.59 to log₁₀ 8.84, these decreases are indicative of reductions in bacterial quantity (ie, although the bacterium was still present, fewer women had higher quantities of bacteria). The intervention reduced detection and quantities of BVAB2 and Megasphaera species, 2 bacteria associated with both BV persistence and recurrence [24, 27]. Other studies of suppressive therapy have noted that the treatment effect wanes following cessation of the intervention [5, 6]. Further research is required to understand the durability of changes in bacterial concentrations following cessation of the intervention and their impact on subsequent BV risk. Consistent with observations following standard treatment for BV, decreases in detection and quantity (at or above the ROC cutoff) of A. vaginae were also observed. It is hypothesized that A. vaginae may have limited susceptibility to metronidazole, depending on the strain [28, 29]. In the absence of susceptibility testing, our data suggest that A. vaginae may be susceptible to metronidazole or that A. vaginae may be reliant on the presence of other species in the community and their metabolites for survival. As a result, alterations in the larger community composition may impact A. vaginae detection and quantity. Our analysis shows that metronidazole use is associated with decreased concentrations of A. vaginae.

Based on observations from prior studies suggesting differences in the vaginal microbiota in different geographic regions and the fact that a number of demographic and behavioral factors associated with BV differed by country, we repeated our analyses, stratified by country. At enrollment, we observed notable differences in species colonization, with lower frequencies of Lactobacillus species and some BV-associated species among Kenyan women as compared to US women (the majority of whom were African American). With regard to the intervention effect, more-pronounced decreases in BV-associated bacteria were observed among Kenyan women, while the effect sizes among US women were attenuated and no longer statistically significant. It is possible that differences by geographic region in BV-associated risk behaviors and bacterial colonization contributed to the differential effectiveness of the intervention [30]. Alternatively, uptake of the intervention is likely to influence the vaginal microbiome. When study product use was evaluated, it was noted that a higher proportion of Kenyan women reported using all 5 does at visits where study product was dispensed, compared with findings observed for US women. It is possible that differences in adherence and in BV-associated behaviors by country modified the impact of the intervention.

The prevalence of BV has been observed to be higher among African and African American women, compared with women from other regions and women of other races, respectively [1, 31]. Broad-range PCR analysis of vaginal samples has also demonstrated differences in bacterial communities by racial category, with African American women having a higher prevalence of communities dominated by anaerobic species, compared with women of European ancestry [32, 33]. Few studies have directly compared the vaginal microbiota of women living in different geographic regions, using molecular methods [25, 34]. The prevalence of L. crispatus, L. jensenii, and L. iners among Kenyan women was generally similar to the prevalence reported in a cross-sectional study of South African, Rwandan, and Kenyan women [34]. In contrast, at enrollment, nearly 95% of Kenyan women in our study had measureable quantities of G. vaginalis, compared with 45%–73% in the multicountry African study, despite having a similar prevalence of BV in both studies. Of note, we observed a higher prevalence of BVAB1 among US women, which is consistent with reports from others, who noted that BVAB1 was detected more frequently among African American women as compared to women of European ancestry [33]. Interestingly, BVAB1 was detected less frequently among Kenyan women, suggesting that geographic region (and factors associated with region) also influence species prevalence. However, with these comparisons, it is important to note that differences in laboratory methods between published studies may impact the ability to directly compare findings.

Our analysis included several strengths, including data collected as part of a multisite, multicountry clinical trial, which allowed for comparisons across 2 distinct geographic regions. The intervention was evaluated among women with a recent vaginal infection, as they would be most likely to benefit from a vaginal health intervention. Nonetheless, our findings should be interpreted in the context of several limitations. Study participants were predominantly African or African American. As a result, we had limited ability to explore the impact of race, beyond comparing African versus African American, as an effect modifier. Several additional features of our study population are important to consider and could impact the generalizability of our findings, including diagnosis of a vaginal infection at screening and frequent reporting of transactional sex history. This analysis assessed the impact of the intervention on select species that are strongly associated with the presence or absence of BV. However, the intervention likely affected the overall vaginal microbiota, which is not well assessed using qPCR assays. Many samples had values below the LLD, limiting our ability to evaluate changes in bacterial quantity by study arm. To address this limitation, the use of predictive methods to determine thresholds based on ROC curves provided a novel approach to addressing the issue of changes in bacterial quantity associated with a clinical outcome. The PVI trial was powered to assess the effect of the PPT intervention on BV and vulvovaginal candidiasis. As a result, statistical power was insufficient to detect smaller shifts in the prevalence of individual bacterial species, especially in analyses stratified by country.

In summary, monthly periodic presumptive treatment for 1 year has previously been shown to reduce overall detection of BV. The present analysis extends the findings of the original trial by showing that this PPT regimen also reduces detection and concentrations of species that are highly predictive of BV persistence and recurrence, including BVAB2, A. vaginae, Leptotrichia/Sneathia species, and Megasphaera species. Several approaches were used to assess the impact of the intervention on vaginal bacteria with similar results, highlighting the robustness of our findings. In light of recent findings demonstrating associations between detection of specific vaginal bacteria and increased risk of HIV-1 infection and preterm birth [14, 26], the role of PPT to improve vaginal health and reduce these adverse health outcomes in women should be explored.

Supplementary Data

Supplementary materials are available at The Journal of Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.

Notes

Acknowledgments. We thank the women who participated in this study; the PVI study team and study sites, for their tireless work on data and sample collection; and FHI 360, for their work on data management and study operations.

J. E. B., R. S. M., and D. F. conceptualized the study and analysis plan. C. A. generated all of the PCR data. J. E. B. drafted the initial report. D. F., S. S., and R. S. M. contributed to the content and revisions. J. E. B., R. S. M., O. A., J. K., and J. S. contributed to data collection. All authors contributed to the article’s contents and approved the final manuscript.

Disclaimer. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Financial support. This work was supported by the National Institute of Allergy and Infectious Diseases (contract HHSN266200400073C [through the Sexually Transmitted Infections Clinical Trials Group] and grant R01-AI099106) and the University of Washington Center for AIDS Research (grant P30-AI27757).

Potential conflicts of interest. J. E. B. has received donated assay reagents from Hologic/Gen-Probe and honoraria from Symbiomix for consulting. R. S. M. has received honoraria for invited lectures and consulting, as well as donated study product for the PVI trial, from Embil Pharmaceutical and currently receives research funding from Hologic/Gen-Probe. J. S. has received consultancy payments from Akesis, Hologic, Symbiomix, and Starpharma and has grants/pending grants from Akesis, BD Diagnostic, Hologic, Cepheid, Quidel, Symbiomix, Starpharma, and Viamet. All other authors report no potential conflicts. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.