INTRODUCTION
Approximately 10% of all pregnancies in the United States are preterm, over 14% among Black women (
1). Intrauterine infection is widely speculated to underlie some portion of preterm deliveries; however, the benefit of prophylactic antibiotic therapy for the prevention of preterm birth (PTB) is not universal and appears to depend at least in part on timing of treatment in pregnancy, route (oral versus intravenous) of antibiotic administration, and clinical presentation (e.g., preterm premature rupture of membranes versus intact membranes) (
2–5), which may reflect etiologic heterogeneity among preterm births. One pathway through which pathogenic microorganisms may gain access to the amniotic cavity is by ascending from the vagina and the cervix (
6).
The vaginal microbiome plays an important role in the health of the female reproductive tract, and a
Lactobacillus species-dominated microbiome has been considered important to maintain a healthy state by producing lactic acid and lowering the vaginal pH. In contrast, an increase of bacterial diversity of non-
Lactobacillus species in the vaginal microbiome, as in bacterial vaginosis (BV), is reported to be an independent risk factor for sexually transmitted infections (STI), PTB, and pelvic inflammatory disease (
7,
8). However, there are diverse species of
Lactobacillus present in the vagina, and these species may produce various levels of lactic acid and have different tolerances for anaerobic members of the microbial community (
9). Moreover, some relatively common vaginal microbiome profiles contain few
Lactobacillus spp. (
10,
11). Vaginal microbiomes lacking
Lactobacillus species of any type tend to have higher pH, more taxon diversity, and often prominently represent organisms that make up the BV diagnosis by Nugent score (
7,
11,
12). This profile, which is associated with adverse outcomes, tends to occur more frequently in Black women (
11–13). Furthermore, vaginal douching, the practice of intravaginal cleansing with a liquid solution, is more often practiced by Black women (
14) and associated with BV (
15,
16), although whether BV leads to douching or the reverse is uncertain.
Several studies have attempted to determine whether associations of the vaginal microbiome with PTB differed by race (
13,
17–22). However, to date, the majority of these studies have been hampered by insufficiently large sample sizes in all of the included racial/ethnic subgroups. In general,
L. crispatus is often reported to be associated with lower risk of PTB, while the taxa related to higher risk of PTB are less consistent between studies and possibly driven by the different racial composition of recruited study participants. Differences in the associations of taxa or community patterns with PTB by race have been underpowered, and it remains uncertain whether a lack of association in any given subgroup is due to inadequate power or the absence of effect.
The present literature appears to support that Black women often have more diverse vaginal microbial community patterns (
11,
12) and that women with higher abundance of
L. crispatus tend to have overall lower risk of PTB (
17–19), although this is not always statistically significant (
20). What is not yet clear, given the known racial disparity of PTB (
23,
24), is whether microbial community patterns are associated with PTB independent of racial differences in microbial community structure or whether there are differences in the patterns of association between the vaginal microbiome and PTB by race. Furthermore, vaginal douching, a practice more commonly reported by Black women (
14), is often associated with vaginal dysbiosis (
15,
25,
26) and thus is a critical factor to consider when examining vaginal microbial patterns and preterm birth. To address these questions, we utilized the Pregnancy, Infection, and Nutrition (PIN) Study, a prospectively enrolled pregnancy cohort of women in central North Carolina in the United States, to investigate the relationship between second trimester vaginal microbial community patterns and PTB, and to examine differences in associations by race, after considering important covariates. This well-characterized cohort of women represents the ideal setting to answer this question, given the rich social, behavioral, demographic, and clinical data assembled that describe in detail known determinants of PTB and the large population of Black and White women (
27).
DISCUSSION
In this large, prospective pregnancy cohort, we analyzed the association between the midpregnancy vaginal microbiome, race, and sPTB. We found that vaginal microbiomes were significantly associated with sPTB, race, douching, and other maternal factors. Many of these maternal factors, like poverty, education, marital status, age, douching, and race, have stronger associations with the vaginal microbiome than the vaginal microbiome has with sPTB (
Fig. 1c). Consistent with previous studies (
11,
28), we found that the vaginal microbiomes of Black and White women were significantly different, with higher alpha diversity, higher abundance of
L. iners, and lower abundance of
L. crispatus for Black women. The microbial difference between sPTB and term controls is mainly driven by a higher
L. crispatus abundance in term controls, similar to previous reports (
17,
18). Because of the strong intercorrelations between maternal factors such as race, poverty, education, marital status, and douching (
Fig. S2), we stratified the data set by race and douching with the aim of uncovering potentially stronger sPTB microbial signatures that are independent of race and douching.
With the community state types assigned based on the most abundant taxon, the sPTB risk associated with the
L. crispatus-dominated community state is about 60% of that for the
L. iners-dominated microbiome (
Fig. 2d). The alpha diversity of the
L. crispatus-dominated microbiome is significantly lower than that of the
L. iners-dominated microbiome, indicating that
L. crispatus suppresses the colonization and development of a BV-like microbiome while
L. iners does not. This observation is consistent with previous literature that has found that a vaginal microbiome dominated by
L. iners also more often shifts toward a diverse community compared to
L. crispatus (
29). Similarly, it has been shown that
L. iners enhances the adhesion of
Gardnerella spp. to cervical epithelial cells, and
Gardnerella spp. displaced adherent
L. crispatus but not
L. iners from epithelial cells (
30). Previous research also suggests that
L. crispatus and
Gardnerella tend to be exclusive members of the microbial community while
L. iners and
Gardnerella often show patterns of coexistence (
18). This exclusion hypothesis may explain the higher sPTB risk associated with the
L. iners-dominated microbiome than
L. crispatus in our population. In further support of this idea, the sPTB risk associated with the community state “Lacto_other” dominated by other
Lactobacillus species (mostly
L. gasseri and
L. jensenii/fornicalis/psittaci) is also significantly higher than that of
L. crispatus-dominated community state, indicating that these species also are not as protective as
L. crispatus.
With a similar number of Black and White participants in our study, we analyzed the associations between the microbiome and sPTB risk in each race separately and found that risk of sPTB associated with
L. crispatus and
L. iners are similar for Black and White women (
Fig. 2g). Although at the U.S. population level Black women have substantially higher risk of PTB (
23), in the PIN study specifically, Black race is only marginally associated with PTB (OR, 1.3; 95% CI, 1.0, 1.6) (
27). Our findings that race does not modify the association between
L. crispatus and
L. iners and sPTB (
Fig. 2g) suggest that the disparity in PTB rates at the U.S. population level in part are due to the lower prevalence of
L. crispatus-dominated microbiome among Black women. In terms of the impact of taxa beyond
L. crispatus and
L. iners, sample size considerations prevented us from determining whether race can modify the sPTB risk associated with other species. For example, the vagitype “Others” consisted of many BV-related species and had a higher percentage of sPTBs, but the modest number of women in this group relative to the large number of different species represented by the Others group makes attempts to measure the impact of individual species unreliable. Future studies will be needed to further investigate the other species.
Douching is often associated with BV (
15,
25,
26), although it is difficult to determine whether douching increases the risk of BV or BV leads to douching, or whether these are correlated but not causally linked phenomena. In a prior investigation within the PIN cohort, douching prior to pregnancy and BV were independently but not jointly associated with increased risk of PTB, i.e., douching did not modify the relationship between BV and preterm birth, and BV did not modify the relationship between douching and PTB (
31). However, this prior study did not consider the impact of self-reported race.
In our study, we found that douching in the 12 months before pregnancy played an important role in the structure of the vaginal microbiome during pregnancy among White women in particular. Among women who did not douche, Black and White women had clearly different microbiomes (
Fig. 3a and
b), and both were associated with sPTB (
Fig. 3f). Among women who did douche, however, the microbiomes of Black and White women were similar (
Fig. 3a and
b) and not associated with sPTB (
Fig. 3f). Our findings suggest that had we conducted a study of only women who douche, race would have played no role in differentiating microbiome composition, as all taxa had
P values close to 1 compared to Black and White women that douche (
Fig. 3d,
x axis). Similarly, had we conducted a study of only Black women, our findings suggest that douching would have played no role differentiating microbiome composition, as when comparing douching and nondouching Black women, all taxa in the microbiomes had
P values close to 1 (
Fig. 3e,
x axis). The effect displayed by these figures is striking, but the immediate causal implications for the relationships between race, microbiome, sPTB, and douching are unclear. One hypothesis is that douching disrupts the healthier
L. crispatus-dominated microbiome, which then shifts to a higher-risk microbiome, ultimately leading to sPTB. However, we see evidence for this hypothesis in White women but not Black women, who do not show substantial shifts in microbiome in relation to douching (
Fig. 3b). Another possible explanation for this discrepancy is that a preexisting dysbiotic state causes both douching behavior and sPTB and that this preexisting state is more commonly found in Black women. Future studies with longitudinal vaginal tract sampling, and longitudinal information on douching behavior, will be required to understand why douching behavior is significantly associated with the vaginal microbiome only in White women in our study and why there are no race-related differences in the vaginal microbiomes among women who report douching.
In summary, in this prospective study of midpregnancy microbiome and sPTB in a well-characterized cohort of Black and White women, we found that the vaginal microbiome of Black women was characterized by higher diversity, lower abundance of L. crispatus, and higher abundance of L. iners. These differences were obscured once maternal douching behavior was considered; specifically, among women who douche, there were no material differences in microbiome by race. Additionally, we found that women with microbiome dominated by L. crispatus had lower risk of sPTB than those dominated by L. iners, and these associations were the same for Black and White women. To our knowledge, this is the first study of the vaginal microbiome and sPTB to consider the impact of douching, and we found that douching is an important factor to consider in future studies.
Our study has a number of limitations. As with all association studies, the observed associations do not necessarily demonstrate or reflect mechanistic causality where differences in the microbial community directly cause risk of preterm birth. It is important to note that while we present differences in microbial community patterns by race that are consistent with the prior literature (
11,
12), we observed strong intercorrelations across a number of maternal factors whose effects cannot easily be separated. These intercorrelated factors include race, douching, poverty level, psychosocial stress, education, marital status, and maternal age. In this, as with many other medical research studies, maternally self-classified race only crudely captures complex social determinants of health (
32); thus, disparities in microbial community patterns that we observe in relation to race may actually result from factors such as diet, access to high-quality medical care, behaviors like vaginal douching, social support and life experiences, psychosocial stress, and experiences of discrimination. Therefore, factors that may explain differences in vaginal microbial community patterns by race need further investigation and elucidation. In a recent investigation of Black women enrolled at a single study site, Dunlop and colleagues found low levels of education, but not prenatal insurance, marital status, or antibiotic use, was associated with Shannon diversity of the vaginal microbiome (
33). We also found that education, and even more so, poverty, accounted for substantial amounts of microbiome variation that were higher than the variations explained by race or douching alone (
Fig. 1c). We further found that differences in the microbiome by race were erased among women who douche. In the PIN study, self-reported douching is more commonly reported among Black women, women who are not married at the time of their pregnancy, women with lower levels of education, women with higher prepregnancy BMIs, and those who smoke in the first 6 months of pregnancy (
31). Disentangling these correlated factors will require large and diverse populations, which to date do not exist in the literature. In addition, it is also important to note that studies that recruit women from prenatal clinics may not be representative of all pregnancies, thus presenting the opportunity for selection bias. In a previous examination of the PIN cohort, women recruited into the study from prenatal clinics were compared to women who resided in the geographical area of the study (
34). Among other features, participants in PIN were more likely to be Black, younger, have lower education, be unmarried, and smoke more cigarettes during pregnancy. While we accounted for these characteristics in our multivariable adjusted models, such differences are important to remain conscious of when interpreting results from clinic-based populations.
Another limitation of our study is that we did not sample longitudinally, and therefore we are unable to evaluate the stability of the association between the microbiome and PTB across gestation. Previous studies have suggested a longitudinal signal with some taxa changing with time in ways that are different in different populations (
17). The relationship between time, sPTB, and the microbiome is an important question that should be addressed in future studies. Our study also did not explicitly consider the impact of biomolecules such as defensins. It is likely that interindividual differences in host response, such as through promotion of inflammatory cytokine or chemokine production (
35,
36), or interactions with vaginal defensins (
20), influence the overall association of vaginal microbial community patterns and sPTB. Previous research within PIN has found that vaginal fluid neutrophil defensin concentration was not strongly associated with PTB (
37), but another study found that defensin levels were associated with intermediate BV (
38). Much remains unknown as to the biological mechanisms linking the vaginal microbiome and preterm birth, and future studies should consider the extent to which variation in host defense mechanisms plays an important mediating role.
Although our study is one of the largest studies of the associations between the vaginal microbiome and sPTB, it still lacks power for analyzing less abundant microbes and vagitypes and whether the combination of race or other social factors and douching influences the consistency of microbial signatures. A strength of our study is that it demonstrates how not accounting for social and behavioral factors could lead to discrepant results across studies of the vaginal microbiome. For example, our results suggest that two studies with very different percentages of White and Black women would likely report contradictory results in the association of the microbiome with douching. Pooled studies across cohorts with similar metagenomics data may enable a more precise investigation of rare species as well as the association of maternal factors that may explain or modify effects of the vaginal microbiome on sPTB.
ACKNOWLEDGEMENTS
This study was funded in part by a grant from NIH/NIMHD (R01MD011504) and NIH/NIEHS (P30 ES010126). E.R. was supported by T32ES007018. The Pregnancy, Infection, and Nutrition study was supported by grants HD37584, HD39373, and DK61981. The General Clinic Research Center was supported by the National Institutes of Health General Clinical Research Centers program of the Division of Research Resources grant RR00046.
S.M.E., A.A.F., and G.A.B. contributed to all aspects of the study, including conception, design, data acquisition, analysis, and supervision. M.G.S., J.M.F., P.B., K.L., N.D., J.T., and A.M.S.R. contributed to acquisition of data and interpretation of results. S.S., J.M.F., E.R., A.A.S., I.C.B., M.C.W., A.A.F., and S.M.E. contributed to analysis and interpretation of data. All authors contributed to writing, review, and/or revision of the manuscript and approved the final manuscript.
We declare that we have no competing interests.