Vaginal Microbiota and In Vitro Fertilization Outcomes: Development of a Simple Diagnostic Tool to Predict Patients at Risk of a Poor Reproductive Outcome

Abstract

Background

Female reproductive tract microbiota may affect human reproduction. The current study considered whether a more detailed characterization of the vaginal microbiota could improve prediction of risk of poor reproductive outcome in patients undergoing in vitro fertilization (IVF).

Methods

Vaginal samples from 120 patients undergoing IVF were sequenced using the V4 region of the 16S ribosomal RNA gene with clustering of Gardnerella vaginalis genomic clades. Abnormal vaginal microbiota was defined by microscopy and quantitative polymerase chain reaction (qPCR) for G. vaginalis and/or Atopobium vaginae above a threshold.

Results

Three major community state types with abundance of Lactobacillus crispatus, Lactobacillus iners, and a diverse community type were identified, including 2 subtypes, characterized by a high abundance of L. crispatus and L. iners, respectively, but in combination with common diversity type operational taxonomic units. No significant association between community state type and the reproductive outcome could be demonstrated; however, abnormal vaginal microbiota by qPCR and a grouping based on high Shannon diversity index predicted the reproductive outcome equally well.

Conclusions

The predictive value of 16S ribosomal RNA gene sequencing was not superior to the simpler and less expensive qPCR diagnostic approach in predicting the risk of a poor reproductive outcome in patients undergoing IVF.

Clinical Trials Registration

NCT02042352

Vaginal microbiome, Vaginal microbiota, in vitro fertilization, bacterial vaginosis, infertility, pregnancy

The vaginal microbiota of women in the reproductive age can be stratified into community state types (CSTs) [1]. Four CSTs are dominated by 4 distinct Lactobacillus species and an additional 2—or more—are so-called diversity CSTs, which are not dominated by any single species [1, 2]. A vaginal microbiota dominated by Lactobacillus spp. is generally considered healthy, whereas the diversity groups are more difficult to categorize. The bacterial species in the diversity group CST IV have been reported to be associated with increased vaginal pH and bacterial vaginosis (BV) [1], a dysbiosis that is diagnosed clinically and/or microscopically and that, when symptomatic, causes a fishy odor and/or vaginal discharge [3, 4]; BV has also been linked to increased risk of acquiring sexually transmitted infections, such as Chlamydia trachomatis, human immunodeficiency virus, and human papillomavirus [1, 5, 6].

In human reproduction, BV has been associated with a Low pregnancy rate, spontaneous abortion, and spontaneous preterm delivery [7–12], albeit results are not unequivocal and causality has generally been difficult to demonstrate. Hence, screening and treatment for asymptomatic BV is usually not recommended [13]. The most common vaginal species considered healthy are Lactobacillus spp., especially Lactobacillus crispatus, whereas species associated with vaginal dysbiosis typically include Gardnerella vaginalis and Atopobium vaginae. These have been shown to also be part of the endometrial microbiota, albeit less abundant relative to the vaginal microbiota [14–16]. Although more research is needed within this field, it seems biologically plausible that the vaginal bacteria ascend to the endometrium, where they may negatively affect the implantation process [17].

Until now, only a few studies have used molecular methods to investigate the vaginal microbiota of infertile patients [9, 17–19]. In a previous study, we observed that abnormal vaginal microbiota (AVM) was associated with a poor reproductive outcome in patients undergoing in vitro fertilization (IVF) [9]. However, that study used quantitative polymerase chain reaction (qPCR) assays targeting only 6 vaginal microbial species. In the current article, we report a more detailed analysis using 16S ribosomal RNA (rRNA) gene sequencing on those same specimens to investigate the microbiota in more detail.

METHODS

Patients and Samples

Methods for sampling and initial microbiological interpretation have been described previously [9]. Briefly, a total of 130 infertile women (90% white) attending IVF treatment at 2 centers in Denmark were prospectively included at the respective clinics at their first consultation before initiation of the IVF treatment. A total of 66% (79 of 120) underwent their first IVF cycle, and 34% (41 of 120) had previous failed cycles. Irrespective of cycle day, 2 vaginal samples were taken from the posterior fornix during speculum examination. One sample was used for Gram staining and Nugent scoring [4] as well as grading for inflammation according to vaginal leukocytosis. Vaginal leukocytosis was defined as having more leukocytes than epithelial cells in ≥3 fields in a Gram-stained smear. The second sample was taken for molecular analyses using the Copan ESwab system (Copan Italia, Brescia, Italy). This sample was immediately frozen at −80°C until further use.

The present study was approved by the regional ethical committee of Central Jutland (file No. 1-10-72-325-13) and the Danish Data Protection Agency (file No. 1-16-02-26-14), and the project was registered at clinicaltrials.gov (NCT02042352). All patients signed written informed consent.

Reproductive Outcomes

The reproductive outcomes were calculated per embryo transfer and only for patients receiving an embryo transfer within 2 months from vaginal sampling. We arbitrarily defined this time period owing to potential disturbance from subsequent cyclic variation and potential antibiotic treatment for urinary tract infections (by general practitioners), which was not accounted for in this study. The definition of a biochemical pregnancy was a positive human chorionic gonadotropin result on day 14. Clinical pregnancy was defined as a sonographically demonstrated fetal heartbeat at 7–9 weeks of gestation. National guidelines for single embryo transfer were followed; thus, a maximum of 2 embryos were transferred. Live birth outcomes were registered from self-reported questionnaires.

Molecular Methods

DNA extraction was performed using the FastDNA SPIN kit for Soil (MP Biomedicals), using 200 µL of the ESwab transport medium and elution of the DNA in 100 µL of DNase-free water; qPCR was performed in a 50-µL total reaction volume with 5 µL of template DNA. The qPCR analyses for G. vaginalis, A. vaginae, L. crispatus, Lactobacillus jensenii, Lactobacillus gasseri, and Lactobacillus iners were performed as described elsewhere [20]. The presence of AVM was determined using receiver operating characteristic (ROC) curve analysis of G. vaginalis or A. vaginae load by qPCR in relation to Nugent scores, as described elsewhere [9]. The 16S rRNA gene sequencing was performed on an Illumina Miseq platform using the same DNA extraction used for the qPCR analyses [9]. Primers targeted the V4 region (http://www.earthmicrobiome.org/emp-standard-protocols/16s/), and amplicons were sequenced at the Argonne National Laboratory http://www.anl.gov/. Further molecular methods and taxonomic classification can be seen in the Supplementary Material and sequencing data are available from GenBank SRA (accession No. PRJEB30385).

Statistical Methods

We used the Fisher exact test for binary data and the Mann-Whitney test for continuous data. In case of multiple groups, the Kruskal-Wallis test was used to test for differences in medians with differences between groups evaluated using the Dwass-Steel-Critchlow-Fligner procedure. We used ROC curve analysis to determine the Shannon diversity index, predicting 95% of the pregnancies. Multivariate logistic regression analysis was performed for the outcome clinical pregnancy, adjusting for age, body mass index, and the number of embryos transferred. A twin pregnancy (n = 1) was counted as a single live birth, but the birth weight was excluded from the statistical analysis. All P values were 2 sided, using a significance level of .05. Statistical analyses were performed using Stata (version 12.1; StataCorp) and StatsDirect (version 3.1.17; (StatsDirect) software .

RESULTS

From the original data set [9] comprising 130 samples, 10 samples were missing owing to lack of sufficient specimen material from the original DNA purification for 16S rRNA gene sequencing (n = 8) or an insufficient number of mapped reads, following the Illumina sequencing (n = 2). The number of amplicon sequences classified per sample ranged from 2595 to 33 675, with a mean of 9020. After rarefaction, a total of 311 400 reads from 120 samples were mapped to 483 unique operational taxonomic units (OTUs). No significant effect in PCoA could be demonstrated before and after rarefaction, see Supplementary Figures 1 and 2. Basic patient characteristics are displayed in Table 1.

Table 1.

Baseline Patient Characteristics

Characteristic	Patients, No. (%)a (n = 120)
Age, median (range), y	30 (20–42)
Body mass index, median(range)b	25.6 (17.5–46)
Unemployed	6 (5)
Current smokers before fertility treatment	8 (7)
Intercourse within previous 24 h	7 (6)
Vaginal bleeding within previous 24 h	15 (12)
Cause of infertility
Endometriosis	6 (5)
Single/lesbian	10 (10)
Tubal	6 (5)
Male factor	43 (36)
Ovarian	17 (14)
Unknown	38 (31)
Candida infectionc	4 (3)
More leukocytes than epithelial cellsc	9 (7)
pH, median (range)d	4 (4–7)

^aData represent No. (%) of patients unless otherwise specified.

^bBody mass index is calculated as the weight in kilograms divided by height in meters squared.

^cAccording to presence on Gram-stained smears.

^dData missing in 4 patients.

CST Clustering, Nugent Scoring, and qPCR

Clustering of the 120 samples making up the full data set did not provide all the CSTs described initially by Ravel et al [1]. Three major clusters corresponding to CST I, III, and IV were well defined (Figure 1; heat map). However, CSTs I and III contained clear subclusters (shown on horizontal bar in Figure 1). Twenty-five women had the classic CST I (L. crispatus), and 24 women formed a subcluster denominated CST Ia with L. crispatus and a varying amount of diverse OTUs. The median Shannon diversity index was 0.06 and 0.44 in CST I and Ia, respectively (P < .001). Similarly, CST III (L. iners dominated; 26 women) and IIIa (L. iners and diverse OTUs; 25 women) had median Shannon diversity indexes of 0.13 and 0.72, respectively (P < .001). CST IV was found in 10 women, with a median Shannon diversity index of 2.0, significantly higher than that of CST IIIa (P < .001).

Figure 1.

Heat map of the 43 most common vaginal operational taxonomic units (OTUs) among all 120 patients included in the study. Top and left dendrograms show a hierarchical clustering based on Bray-Curtis dissimilarity of samples and OTUs, respectively. Top bar represents community state type; second bar, Nugent score (0–3 indicates normal findings [green]; 4–6, intermediate [yellow], and 7–10 bacterial vaginosis [red]); and third bar, quantitative polymerase chain reaction (qPCR) diagnosis of abnormal vaginal microbiota (AVM) (red). Bottom graph represents the Shannon diversity index.

CST II (L. gasseri) and CST V (L. jensenii) were not found in any of the patients, although 2 women with high L. gasseri abundance were clustered within CST Ia, and a relatively high abundance of L. jensenii was found in some patients clustering in CST IIIa. Of 20 women with CST IV, 19 (95%) had BV or intermediate microbiota, as determined by the Nugent score [21], and all had AVM as defined by qPCR [9] (see Figure 1 and Table 2). Compared with all other CSTs, CST IV was significantly associated with higher Nugent scores (P < .001). However, CST IIIa was also associated with higher Nugent scores than CSTs I, Ia, and III. The Shannon diversity index across Nugent scores showed an ascending linear trend (r² = 0.6; P < .001); see Supplementary material Supplementary Figure 3.

Table 2.

Correlation between CST Cluster, AVM by qPCR and Nugent Score

	qPCR Finding, No. (%)			Nugent Score, No. (%)
CST Cluster	Normal(n = 88)	AVM(n = 32)	P Value	Normal (0–3) (n = 79)	Intermediate (4–6) (n = 15)	BV (7–10) (n = 26)	P Value (P Value for Trend)
CST I (L. crispatus)	25 (28)	0 (0)	.001	25	0	0	<.001 (<.001)
CST Ia (L. crispatuswith diverse OTU)	19 (22)	5 (16)	NS	19	2	3	NS
CST III (L. iners)	25 (28)	1 (3)	.002	24	1	1	.004 (.002)
CST IIIa (L.iners with diverse OTU)	19 (22)	6 (19)	NS	10	12	3	<.001 (.300)
CST IV (diverse OTUs)	0	20 (63)	<.001	1	0	19	<.001 (<.001)

Abbreviations: AVM, abnormal vaginal microbiota; BV, bacterial vaginosis CST, community state type; L. crispatus, Lactobacillus crispatus; L. iners, Lactobacillus iners; NS, not significant; OTU, operational taxonomic unit; qPCR, quantitative polymerase chain reaction.

The median load of L. iners in CST III was 15-fold higher than that of L. crispatus in CST I (10 233 287 genome equivalents [geq] vs 682 220 geq; P < .001); similarly, L. iners was present with a higher median load in CSTs III and IIIa compared with L. crispatus in CSTs I and Ia (4 940 757 geq vs 578 054; P < .001).

Role of G. vaginalis 16S rRNA Gene Variants

A total of 44 samples had ≥26 G. vaginalis reads (1% of the rarefied count); this sample composition comprised 27 of 32 samples (84%) with AVM by qPCR, 23 of 26 (88%) with BV by Nugent score, 9 of 24 (38%) with CST Ia, 15 of 25 (60%) with CST IIIa, and all 20 samples with CST IV. In the 42 samples where 1 of the variants had ≥50% of the total G. vaginalis reads (see Supplementary material and Supplementary Table 1), a total of 26 (62%) were dominated by variant G1, 15 (36%) by variant G2, and 1 (2%) by variant G3. Variant G1 was significantly associated with BV by Nugent score (P ≤ .001), whereas variant G2 was significantly associated with intermediate Nugent scores (P ≤ .001). This was in agreement with associations in CSTs, because 8 of 9 CST Ia samples (89%) were dominated by variant G1, 12 of 14 CST IIIa samples (86%) were dominated by variant G2 and 16 of 19 CST IV samples (84%) were dominated by variant G1 (P < .001; (Supplementary Table 1). Only 20 of these women underwent embryo transfer (11 in variant G1 with 2 pregnancies, 9 in variant G2 with 3 pregnancies; not significant).

Vaginal Leukocytosis and 16S rRNA Gene Sequencing Results

The prevalence of vaginal leukocytosis was 8% (9 of 120). No significant association was observed between CSTs and vaginal leukocytosis (P = .23). Significantly more patients with vaginal leukocytosis were classified as having an intermediate Nugent score; 4 of 9 (44%) versus 12 of 111 (11%), (P = .02), and among the 4 with intermediate Nugent score, 1 sample was dominated by Streptococcus spp. in 95% of reads (classified as CST Ia), 1 was not dominated by any single species or genus but comprised a highly diverse microbiota, although with >10% of reads from Streptococcus spp. (classified as CST IIIa), and finally 2 samples were dominated by L. iners (>60% of reads) with also minor BV-like microbiota, such as G. vaginalis and Prevotella spp. (both classified as CST IIIa).

Association Between Vaginal Microbiota and Risk of a Poor Reproductive Outcome

To develop a diagnostic tool to classify patients undergoing IVF who were at risk of a poor reproductive outcome with a tentative infectious cause, combined sequencing and pregnancy data were analyzed. A total of 75 patients underwent embryo transfer, and their were analyzed according to pregnancy rate per embryo transfer (Figure 2).

Figure 2.

Heat map of results in the 75 patients receiving embryo transfer. Top and left dendrograms show a hierarchical clustering based on Bray-Curtis dissimilarity of samples and operational taxonomic units, respectively. Top bar represents Nugent score (0–3 indicates normal findings [green]; 4–6, intermediate [yellow], and 7–10 bacterial vaginosis [red]); second bar, abnormal vaginal microbiota (AVM) diagnosed by means of quantitative polymerase chain reaction (qPCR) (red); third bar, clinical pregnancy diagnosed at gestational week 7–9 with transvaginal ultrasonography. Bottom graph represents the Shannon diversity index.

In Table 3, patient characteristics and the biochemical and clinical pregnancy rates are shown stratified according to CST and according to a Shannon diversity index cutoff of 0.93 which was determined to predict 95% of the pregnancies by ROC curve analysis. The group of 14 women with a diversity index above the cutoff comprised all 10 women with CST IV, 1 with CST Ia, and 3 with CST IIIa. All women had AVM by qPCR and a significantly higher vaginal pH than the low diversity group. Moreover, vaginal pH in women with CST IV was significantly higher than that in women with CST Ia (P = .008). No differences were observed between the other CSTs.

Table 3.

Patient Characteristics and Reproductive Outcome Data for Patients Receiving Embryo Transfer Within 2 Months After Vaginal Sampling (n = 75).

Characteristic or Outcome	All CSTs	CST I (L. crispatus)	CST Ia (L. crispatus With Diverse OTU)	CST III (L. iners)	CST IIIa (L. iners With Diverse OTU)	CST IV (Diverse OTUs)	Shannon Diversity Index >0.93	Shannon Diversity Index ≤0.93
No. of patients	75	18	16	17	14	10	14	61
Age, median (range), y	31 (24–42)	30 (27–41)	29 (26–42)	33 (25–39)	30.5 (24–39)	30.5 (26–41)	30.5	31
BMI, median (range), y	25.5 (17.5–35.8)	25.15 (19.7–35.8)	26.25 (18.8–33.1)	24.44 (19.6–32.74)	25.36 (19.4–33.0)	27.9 (17.5–31)	27.9	24.8
Smoking (before fertility treatment), No. (%)	1 (1)	1 (6)	0	0	0	0	0	1 (2)
Cause of infertility, No. (%)
Tubal factor	4 (5)	0	0	0	4 (29)	0	1 (7)	3 (5)
Idiopathic	23 (31)	6 (33)	4 (25)	7 (42)	2 (14)	4 (40)	5 (36)	18 (30)
Endometriosis	5 (7)	2 (11)	1 (6)	1 (6)	1 (7)	0	0	5 (8)
Ovarian factor	12 (16)	1 (6)	4 (25)	2 (12)	3 (21)	2 (20)	2 (14)	10 (16)
Male factor	26 (35)	8 (44)	7 (44)	6 (35)	3 (21)	2 (20)	3 (21)	23 (38)
Single/lesbian	5 (7)	1 (6)	0	1 (6)	1 (7)	2 (20)	3 (21)	2 (3)
No. of previous failed cycles, median (range)	0 (0–7)	2 (0–7)	0 (0–4)	0 (0–4)	0 (0–2)	0 (0–2)	0 (0–2)	0 (0–7)
More leukocytes than epithelial cells, No. (%)	6 (8)	1 (6)	2 (13)	0	3 (21)	0	1 (7)	5 (8)
pH, median (range)	4 (4–7)	4 (4–5.8)	4 (4–4.7)	4 (4–7)	4 (4–7)	4.7 (4–7)	4.7	4.0b
No. of oocytes retrieved, median (range)	11 (1–31)	11 (4–17)	8.5 (4–25)	8 (1–31)	12.5 (3–18)	12 (2–28)	10(1–31)	12 (2–28)
Double embryo transfer, No. (%)	24 (32)	8 (44)	4 (25)	6 (35)	3 (21)	3 (30)	5 (36)	19 (31)
Biochemical pregnancy, No. (%)	36 (48)	8 (44)	9 (56)	10 (59)	6 (43)	3 (30)	4 (29)	32 (53)
Clinical pregnancy, No. (%)	27 (36)	6 (33)	6 (38)	9 (53)	5 (36)	1 (10)	1 (7)b	26 (43)
Live birth rate, No. (%)	27 (36)	6 (33)	6 (38)	9 (53)	5 (36)	1 (10)	1 (7)b	26 (43)
Gestational age at birth, median (range), wka	39 (32–41)	38.5 (37–41)	40 (37–41)	38 (32–41)	40 (39–41)	35	35	39 (32–41)
Birth weight, median (range), ga	3465 (1830–5310)	3360 (3210–3820)	3615 (2620–4490)	3340 (1830–3900)	3890 (3440–5310)	2320	2320	3470 (1830–5310)
Shannon diversity index, median (range)	0.20 (0.01–2.64)	0.08(0.02–0.13)	0.39 (0.08–1.14)	0.14 (0.01–0.74)	0.65 (0.10–2.64)	1.99 (1.84–2.37)	…	…

Abbreviations: BMI, body mass index (calculated as weight in kilograms divided by height in meters squared); CST, community state type; L. crispatus, Lactobacillus crispatus; L. iners, Lactobacillus iners; OTU, operational taxonomic unit.

^aOne twin pregnancy was not included in the calculations.

^bP < .001.

Overall, a biochemical pregnancy rate of 48% (36 of 75) was observed, and after a biochemical pregnancy loss of 25% (9 of 36) the overall clinical pregnancy rate was 36% (27 of 75). There were no differences in either the biochemical or the clinical pregnancy rate between the 5 CSTs. The live birth rate was equal to the clinical pregnancy rate. No statistically significant differences were seen between the 5 CSTs regarding gestational age and birth weight; however, numbers were small. It was remarkable that the single clinical pregnancy in CST IV ended as a preterm delivery in week 35 with a birth weight of 2320 g. Women with CST IV were less likely to experience a clinical pregnancy (10%; 1/of 10) than women with all other CSTs (40%; 26 of 65), although this difference was not statistically significant (P = .08)

No difference in biochemical pregnancy rates was observed between the women with a high or low Shannon diversity indexes, respectively. However, clinical pregnancy and live birth were significantly less likely in women with a Shannon diversity index >0.93 a (P = .01).

Comparison Between Microbiota Classification, Diversity, AVM by qPCR, and Nugent Score in Predicting Poor Reproductive Outcome in Patients Undergoing IVF

No significant associations were observed in reproductive outcomes for Nugent score ≥7 (BV) or ≥4 (intermediate and BV) or for CST IV alone or in combination with CST IIIa or with CST Ia and IIIa. However, AVM diagnosed by means of qPCR was significantly associated with a poor reproductive outcome, with an odds ratio (OR) of 0.16 (95% confidence interval [CI], .02–.69; P = .01). Similarly, a high Shannon diversity index >0.93 was associated with poor reproductive outcome, with an OR of 0.10 (95% CI, .006–.7; P = .01) (Table 4). In the adjusted analysis controlling for age, body mass index, and embryos transferred, the effect estimates remained significant; the adjusted OR (aOR) was 0.07 (95% CI, .01–.59; P = .02) and 0.10 (.01–.83; P = .03) for AVM by qPCR and Shannon diversity index, respectively.

Table 4.

Prediction of Clinical Pregnancy Among 75 Women After Embryo Transfer According to Characterization of the Vaginal Microbiota

	Clinical Pregnancy, No./Total No. of Women		OR (95% CI)	P Value	Adjusted OR (95% CI)a	P Value
Characteristic	With Characteristic	Without Characteristic
Nugent score ≥7	1/11	26/64	0.18 (.008–1.2)	.09	0.76 (.6–1.0)	.08
Nugent score ≥4	6/20	21/55	0.69 (.2–2.1)	.60	0.81 (.58–1.1)	.21
AVM by qPCR	2/18	25/57	0.16 (.02–.69)	.01b	0.07 (.01–.59)	.02 b
CST IV	1/10	26/65	0.17 (.007–1.1)	.08	0.48 (.15–1.5)	.22
CST IV + IIIa	6/24	21/51	0.48 (.2–1.4)	.18	1.01 (.89–1.1)	.92
CST IV + Ia + IIIa	12/40	15/35	0.6 (.2–1.5)	.26	0.39 (.13–1.2)	.09
Shannon diversity index >0.93	1/14	26/61	0.10 (.006–.7)	.01b	0.10 (.01–.83)	.03b

Abbreviations: AVM, abnormal vaginal microbiota; CI, confidence interval; CST, community state type; OR, odds ratio; qPCR, quantitative polymerase chain reaction.

^aORs adjusted for age, body mass index, and number of embryos transferred.

^bSignificant at P < .05.

DISCUSSION

Principal Findings

In the present study, the vaginal microbiota was investigated using 16S rRNA gene sequencing of the V4 region in patients undergoing IVF. The samples were clustered into CSTs and grouped into 2 groups depending on high or low Shannon diversity index. Three major clusters ,corresponding to CST I, III, and IV, were well defined on a high level, albeit CSTs I and III contained clear subclusters, designated Ia and IIIa, characterized by a higher level of diversity. The sequencing method was compared with Nugent score for BV diagnosis and the qPCR assay for AVM diagnosis. In line with previous findings [1], a higher proportion of Lactobacillus spp. was observed in the lower Nugent scores compared with the higher Nugent scores. Similarly, the higher Nugent scores were associated with higher diversity and comprised high loads of OTUs corresponding to G. vaginalis, A. vaginae, Sneathia sanguinegens, and Prevotella spp.

The samples included in the present nested cohort study were previously characterized using a qPCR-based approach in which the presence of G. vaginalis or A. vaginae above thresholds levels was used to define AVM in infertile patients [9]. The majority (63%) of AVM as defined by qPCR was clustered in CST IV, but a significant proportion of patients in CST Ia and IIIa also had AVM (16% and 19%, respectively. This suggests that these subclusters may be less beneficial for the reproductive outcome compared with low-diversity CST I and III. It also points to the fact that not all L. iners–dominated communities are associated with high diversity.

Role of G. vaginalis 16S rRNA Gene Variants

G. vaginalis is a diverse species, and based on whole-genome comparisons it has been suggested that it should be subdivided into 4 separate species [22]. We used a similar approach to that suggested by Callahan et al [23], wherein sequences classified as G. vaginalis are classified into 3 different variants according to sequence variation in 2 positions in the V4 region of the 16S rRNA gene. Variants G2 and G3 represent the genomic clades 1 and 2 and variant G1 represents clades 3 and 4 according to the whole-genome sequencing classification [22]. We found the G1 variant to be strongly associated with BV and with CST IV, whereas G2 was strongly associated with intermediate Nugent scores and CST IIIa. This is in contrast to findings of a previous study in which the G2 variant was the main explanation for preterm birth in a study of predominantly white women [23], and the parent study [24] found a strong association between CST IV and preterm birth. Because the present study also investigated predominantly white women, we would have expected to see the same association between G2 and CST IV. Using G. vaginalis clade–specific PCR [25], clades 1 and 4 seemed to increase more dramatically in loads with increasing Nugent scores, however, although all clades showed an increasing trend [26], none of the clades seemed to be associated with intermediate Nugent scores (Supplementary Table 1).

16S rRNA Gene Sequencing in Comparison With Nugent Score BV

The majority (73%) of samples with BV by Nugent scoring clustered in CST IV, but these samples were almost absent (2%) from the low-diversity CSTs, I and III. Similarly, 80% of samples with intermediate Nugent scores were classified in CST IIIa with mixed L. iners and diverse microbiota, a classification in accordance with the relatively high abundance of Lactobacillus morphotypes, allowing for maximum scores for Gardnerella and Mobiluncus morphotypes in the presence of >30 lactobacilli per high-power microscopy field in this group. On the other hand, L. iners is a particularly difficult species to interpret, because it is sometimes mistaken as Gardnerella morphotypes by microscopy owing to its gram-variable nature and its smaller size.

Some studies have found that L. iners is associated with poor reproductive outcomes, such as preterm birth [27, 28], but it should be noted that L. iners is often found in much higher bacterial loads than the other Lactobacillus spp., which may obscure the importance of other species in the vaginal microbiota. These species may be abundant on an absolute scale as determined, for example, by qPCR, whereas, in contrast, they may represent only a low relative abundance using 16S rRNA gene sequencing owing to the dominance of L. iners. In this study, L. iners was 15-fold more abundant in CST III than was L. crispatus in CST I, which emphasizes the need for a combination of specific qPCR alongside 16S rRNA gene sequencing to investigate the role of genital tract microbiota in adverse reproductive outcomes.

16S rRNA Gene Sequencing and Vaginal Leukocytosis

Vaginal leukocytosis was associated with Nugent intermediate scores, but none of the CSTs were significantly associated with vaginal leukocytosis. Three of the 9 samples contained a relatively high proportion of streptococci, 2 of those in combination with Escherichia coli and other fecal bacteria, suggesting that these women had aerobic vaginitis [29].

Prediction of Reproductive Outcome

The diagnosis of AVM by qPCR has been suggested as a useful tool to identify patients undergoing IVF who are at risk of a poor reproductive outcome [9], and in the present nested cohort it was still highly predictive of clinical pregnancy, with an aOR of 0.07 (95% CI, .01–.59). Although 16S rRNA gene sequencing provided a much more detailed insight into the vaginal microbiota, we could not identify a CST that was significantly associated with clinical pregnancy. However, the present study included women with a variety of infertility diagnoses, and it is possible that a more targeted selection of patients (eg, including only patients with strict idiopathic infertility) might reveal closer associations between CSTs and reproductive outcome.

Owing to the difference in Shannon diversity index between the CSTs, we used ROC curve analysis to select a cutoff that predicted 95% of the pregnancies. Using the taxonomic classification from the present study, we found that a cutoff of 0.93 predicted the reproductive outcome with an aOR of 0.1 (95% CI, .01%–.83), similar to that predicted by the qPCR method for AVM. Thus, microbiota classification in terms of CSTs or according to diversity did not provide a better prediction of an adverse reproductive outcome than did the simple, rapid, and inexpensive qPCR-based method for detection of AVM.

Strengths and Limitations

The main strength of the study is the thorough characterization of the vaginal microbiota using Nugent scoring, 16S rRNA gene sequencing alongside qPCR with validated assays for 6 bacteria: L. crispatus, L. iners, L. jensenii, L. gasseri, A. vaginae, and G. vaginalis in extracted DNA from the same specimens. We also refined the G. vaginalis classification by single-nucleotide polymorphisms in the amplicon, allowing a rough classification according to the genomic clades, and we classified the samples according to a diversity cutoff based on ROC curve analysis. A limitation is the relatively small number of women undergoing embryo transfer, which decreases the power of the reproductive outcome findings. The arbitrary cutoff for analyzing only embryo transfers performed no later than 2 months after vaginal swab sample collection allowed only 1 embryo transfer, and we could not evaluate the effect of cumulative pregnancy rates, which from a clinical point of view might have been important. Finally, although patients are usually have either high or very low loads of AVM bacteria, it is a limitation that we included women at varying time points during their menstrual cycle, because the vaginal microbiota is subject to cyclic fluctuations. How it influences the AVM diagnosis by qPCR is not known.

Conclusions

In the present study, we provide a comprehensive analysis of different diagnostic tools to define AVM in infertile women undergoing IVF treatment. We conclude that the qPCR assay targeting G. vaginalis and A. vaginae is valid and robust, providing the clinician with an easy access yes-or-no diagnosis of AVM. Nevertheless, intervention-based trials are needed to investigate a causal relation between AVM and poor reproductive outcomes before AVM-related infertility can be considered.

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

Author contributions. T. H., P. H.. and J. S. J. designed the study. T. H., P. H., H. O. E., B. A., R. J. L., K. R., and J. S. J. conducted the study at the involved clinics. T. H., T. B. J., P. S. A., K. L. N., B. L., and J. S. J. made the microbiological interpretations. T. H., P. H.. and J. S. J made the clinical interpretations. All authors read, commented on, and agreed on the final manuscript.

Acknowledgments. We extend our sincere gratitude to all staff and participating patients at the Fertility Clinic Skive, Skive Regional Hospital, and Trianglen Fertility Clinic, Copenhagen, Denmark. Special thanks to research nurses Alice Toft Mikkelsen and Laura Karlsson. We also thank the laboratory at Statens Serum Institut, Copenhagen, in particular laboratory technician Susanne Larsson, for excellent technical assistance.

Disclaimer. The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Financial support. This work was supported by the AP Møller Maersk Foundation for the advancement of Medical Science and the Hospital of Central Jutland Research Fund, Denmark.

Potential conflict of interests. Outside this work, T. H., P. H., and J. S. J. received a research grant from Ose. T. H. has received honoraria for lectures from Bifodan and Merck. P. H. has received honoraria for lectures and unrestricted research grants from Ferring, Merck, and MSD. J. S. J. has received speaker’s fee from Hologic, BD, and Cepheid and serves on the scientific advisory boards of Roche Molecular Systems, Abbott Molecular, and Cepheid. 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.