Cervicovaginal microbiota and metabolome predict preterm birth risk in an ethnically diverse cohort

Abstract

The syndrome of spontaneous preterm birth (sPTB) presents a challenge to mechanistic understanding, effective risk stratification, and clinical management. Individual associations between sPTB, self-reported ethnic ancestry, vaginal microbiota, metabolome, and innate immune response are known but not fully understood, and knowledge has yet to impact clinical practice. Here, we used multi-data type integration and composite statistical models to gain insight into sPTB risk by exploring the cervicovaginal environment of an ethnically heterogenous pregnant population (n = 346 women; n = 60 sPTB < 37 weeks' gestation, including n = 27 sPTB < 34 weeks). Analysis of cervicovaginal samples (10-15+6 weeks) identified potentially novel interactions between risk of sPTB and microbiota, metabolite, and maternal host defense molecules. Statistical modeling identified a composite of metabolites (leucine, tyrosine, aspartate, lactate, betaine, acetate, and Ca2+) associated with risk of sPTB < 37 weeks (AUC 0.752). A combination of glucose, aspartate, Ca2+, Lactobacillus crispatus, and L. acidophilus relative abundance identified risk of early sPTB < 34 weeks (AUC 0.758), improved by stratification by ethnicity (AUC 0.835). Increased relative abundance of L. acidophilus appeared protective against sPTB < 34 weeks. By using cervicovaginal fluid samples, we demonstrate the potential of multi-data type integration for developing composite models toward understanding the contribution of the vaginal environment to risk of sPTB.

Keywords: Bioinformatics; Microbiology; Obstetrics/gynecology; Reproductive Biology; Translation.

Figure 1. Whole cohort cervicovaginal fluid (CVF) components relationships.

CVF metabolites, pH, and host defense peptides explored in relation with bacterial composition based on Principal Coordinates Analysis (PCoA) groups in early pregnancy (10–15⁺⁶ weeks). Wilcoxon comparison shown if P < 0.05. (A–I) lactate, acetate, Ca²⁺, betaine glucose, succinate, pH, elafin, and cathelicidin (high-risk women only). Number of samples (n) per PCoA group comparisons as follows: (A–F) PCoA A = 89, PCoA B = 31, PCoA C = 115, PCoA D = 64, and PCoA E = 6. (G) PCoA A = 55, PCoA B = 19, PCoA C = 62, PCoA D = 39, and PCoA E = 4. (H) PCoA A = 85, PCoA B = 29, PCoA C = 111, PCoA D = 61, and PCoA E = 6. (I) PCoA A = 41, PCoA B = 16, PCoA C = 76, PCoA D = 37, and PCoA E = 4. Y axis represents the normalized NMR peaks (arbitrary units, au) (A–F), pg/μL (H), and ng/mL (I). Horizontal line and boxes represent median and IQR.

Figure 2. Bacterial composition of the cervicovaginal fluid based on Principal Coordinates Analysis (PCoA) groups and stratified by pregnancy outcome.

(A) Distribution of PCoA groups in the whole community based on delivery outcome for term birth and spontaneous preterm birth before 37 weeks’ gestation (sPTB37) in early samples (10–15⁺⁶ weeks). (B–D) Dynamics of PCoA groups during pregnancy as identified in the early and late (16–23⁺⁶ weeks) sampling times in relation to delivery outcome for the whole community (B), for White women (C), and Black women (D). Circles represents early samples, and crosses represent late samples. (A and B) Early samples term, n = 255, and sPTB37, n = 50; late samples term, n = 263, and sPTB37, n = 50. (C) White women early samples term, n = 184, and sPTB37, n = 23; late samples term, n = 190, and sPTB37, n = 28. (D) Black women early samples term, n = 51, and sPTB37, n = 16; late samples term, n = 52, and sPTB37, n = 13.

Figure 3. Spearman’s correlation analyses of early gestation cervicovaginal fluid.

(A and B) OTUs and metabolites (n = 305) and OTUs, metabolites, elafin, cathelicidin, and HNE (n = 161). OTUs selected as follows: showing more than 1% average abundance; identified via LEfSe analyses as associated with spontaneous preterm birth (sPTB < 37 weeks). Only correlations with adjusted P < 0.05 are shown; scale represents correlations values: blue (negative) and red (positive).

Figure 4. CVF L. acidophilus proportion above 20% is associated with term birth.

Relationship between the percentage of L. crispatus and L. acidophilus in cervicovaginal fluid (CVF) of women stratified by preterm delivery < 34 weeks (sPTB34, red) or delivery > 34 weeks (gray). Data from n = 618 samples (10–15⁺⁶ and 16-23⁺⁶ weeks) from n = 341 women.

Figure 5. Receiver operating characteristic (ROC) curves areas and AUC for composite models for spontaneous preterm birth predictions.

(A–F) sPTB37 (A–C) and sPTB34 prediction (D–F) (total number of samples, n = 618; n = 425 samples from White women, n = 132 from Black women, and n = 61 from women reporting other ethnicities; n = 306 from 10–15⁺⁶ weeks or n = 312 from 16–23⁺⁴ weeks). (A–C) a model using a composite of 7 cervicovaginal fluid (CVF) metabolites performs equally for prediction of sPTB37 when stratified by ethnicity and gestation of CVF sampling (10–15⁺⁶ weeks or 16–23⁺⁴ weeks). (D and E) For sPTB34, a model of 3 CVF metabolites and CVF L. crispatus and L. acidophilus proportions shows differences in performance when stratified by ethnicity. (F) sPTB34 model performs similarly when testing samples taken between 10–15⁺⁶ or 16–23⁺⁶ weeks.