Lactobacillus crispatus thrives in pregnancy hormonal milieu in a Nigerian patient cohort

Steroid hormones are one of the presumed modulators of Lactobacillus abundance in the vaginal epithelium. We set out to characterize the vaginal microbiome (VMB) and also provide an in-depth understanding of the relative contribution of estradiol (E2) and progesterone (P1) in shaping the vaginal microbiome of Nigerian women (n = 38) who experienced both uncomplicated term delivery and preterm delivery using samples longitudinally collected during pregnancy (17–21, 27–31, 36–41 weeks gestation) and 6 weeks postpartum. Vaginal swabs and blood samples were aseptically collected. Vaginal swabs were used for microbiome assessment using 16S ribosomal RNA (rRNA) gene sequencing. Blood samples were used for hormonal measurement using a competitive-based enzyme-linked immunosorbent assay (ELISA). Across several maternal covariates, maternal age, pregnancy status and delivery mode were not significantly associated with the vaginal microbiota whereas maternal E2 level (pE2 = 0.006, Omnibus), and P1 level (pP1 = 0.001, Omnibus) were significantly associated with the vaginal microbiome. E2 and P1 concentrations increased throughout pregnancy commensurately with increasing proportions of L. crispatus (pE2 = 0.036, pP1 = 0.034, Linear Mixed Model). An increasing trend of α-diversity was also observed as pregnancy progressed (pobserved ASV = 0.006, LMM). A compositional microbiome shift from Lactobacillus profile to non-Lactobacillus profile was observed in most postnatal women (pCST IV < 0.001, LMM). Analysis of our data shows a species-specific link between pregnancy steroid hormone concentration and L. crispatus abundance.

The vaginal microbiome in pregnancy has been found to be involved in both maternal health outcome 1-3 and disease pathogenesis [4][5][6][7] . Vaginal microbial communities state types (CSTs) are typically defined by the dominance of one of four Lactobacillus species; L. crispatus (CST I), L. gasseri (CST II), L. iners (CST III) and L. jensenni (CST V) while CST IV is defined by heterogenous mixture of strict and facultative anaerobes 8 . In pregnancy, healthy obstetric outcomes are associated with an enrichment of Lactobacillus and low bacterial diversity 1,3,9 . In this context, Lactobacillus are associated with full-term birth and modulate reproductive health 10,11 through the production of lactic acid 12,13 and antimicrobial compounds 14 . The presence of different Lactobacillus species in the vaginal microenvironment is widely believed to enhance stability of the vaginal microbiome in pregnancy as failure of these species to retain dominance over time (remain stable) may result to overgrowth of anaerobic bacteria including bacterial vaginosis associated bacteria (BVAB1), Prevotella species, Gardnerella vaginalis, Atopobium vaginae, Sneathia, Megasphaera 15 . These anaerobes have been shown to be associated with poor pregnancy outcomes such as post-abortal infections, early miscarriage and preterm birth [16][17][18] . It is important to consider the microbiome in health disparities as recent studies also reveal that women of African ancestry are less likely to exhibit vaginal lactobacilli, frequently have less stable VMB, are more likely to exhibit increased vaginal microbial diversity and richness [19][20][21][22] .
The composition of the vaginal microbiome appears hormonally regulated 23 as evidence shows it adapts to hormonal changes in menstruation 24 and menopause 25 . In general, high steroid hormone levels tend to decrease the dominance of anaerobic bacteria and favor the dominance of Lactobacillus on the vaginal epithelium 26,27 . Pregnancy is accompanied by a rise in the level of estrogens and progesterone hormones 28 , thus as pregnancy advances, elevated levels of circulating estrogen are hypothesized to increase Lactobacilli species abundance. The withdrawal of estrogen during the postpartum period is also assumed to precipitate a dramatic shift from a prior Lactobaccillus dominated microbiota to a diverse and non-Lactobacillus dominated microbial profile in postnatal women 2,29 . Several studies of North American 1,6,9,29,30 and European populations 2 have characterized the vaginal microbiota throughout pregnancy. These studies show the vaginal microbiomes of pregnant women to be highly stable and typically dominated by Lactobacillus. Unfortunately, these studies can provide us with little insight about the dynamics of the pregnancy vaginal microbiome with the shifting hormonal milieu during the postpartum period. In addition, ethnicity is also an important determinant of the vaginal microbiome composition during pregnancy as Black ethnicity is a risk factor for bacterial vaginosis 31 . In Africa, the vaginal microbiome during pregnancy has been described in an HIV pregnant cohort 32,33 but only using a cross sectional study design. To our knowledge, this is the first study describing the vaginal microbiota during pregnancy and postpartum longitudinally. Using this data, we set out to characterize the VMB during pregnancy and puerperium in Nigerian women and provide an in-depth understanding of the relative contribution of estradiol (E2) and progesterone (P1) in shaping the vaginal microbiome.

Methods
Ethical statement and participant enrollment. Ethical approval for this study was obtained in accordance with the Nigerian National Code for Health Research Ethics and approved by the Joint University of Ibadan/University College Hospital Health Research Ethics Committee (UI/UCH HREC) with approved IRB number UI/EC/18/0411. All experiments were performed in accordance with the approved guidelines. Written informed consent was obtained from all participants prior to sampling. Women (n = 38) aged between 24 and 41 years who had no medical problems or adverse outcome during any previous pregnancy were enrolled to the longitudinal study at the time of booking for antenatal care. Participants were recruited from 17 weeks' gestation, getting details of their previous pregnancies history and other maternal and fetal covariates and monitoring them until puerperium (6 weeks postpartum). Biological samples were collected thrice during pregnancy: at the recruitment visit occurring between 17 and 21 weeks (first timepoint), at 27-31 weeks (second timepoint), above 36-41 weeks' gestation (third timepoint), and once at 6 weeks postpartum (fourth time point). Women were included in the study if they: self-reported as Nigerian; they were between 17 and 21 weeks' gestation (confirmed by clinical records and ultrasound result); were of reproductive age (18-49 years); have no intercurrent infection requiring treatment with antibiotics, weighed greater than 50 kg; had no medical complication during any previous and current pregnancy including diabetes mellitus, autoimmune disease and hypertension; were not on supplemental progesterone, were able to provide written informed consent; and were willing to participate in all aspects of the study. Women were excluded if they: developed any inter-current infection requiring antibiotic therapy; had sexual activity within 72 h of sampling; used probiotic supplement, medication or probiotic feed in the preceding 2 weeks before sampling, reported vaginal bleeding in the preceding week, used antibiotics in the preceding 2 weeks, had chronic active viral infections, including HIV-1/2, HTLV-1/2, hepatitis B/C or have known autoimmune disease, such as, rheumatoid arthritis or systemic lupus erythematosus; solid organ or transplant recipient as previously described 34 . To reduce incidence of vaginal infection induced by douching agents or preterm birth induced by douching agent we excluded women who use vaginal douching. These foundational data were collected by the research team and participants were followed up at every antenatal visit taking details of important maternal and clinical variables until delivery (both term and preterm).
Sample collection. Blood and vaginal swabs were simultaneously collected during sampling. Under visual inspection during a speculum examination, swabs (LQ AMIES Copan CA, USA) were collected from the posterior fornix 35 . The vaginal swabs for genomic bacterial DNA isolation were transferred to the laboratory in Amies transport medium and stored at − 80 °C until isolation. Three milliliter of blood samples were collected aseptically from the peripheral vein of the arm in an EDTA vacutainer blood collection tube and transported on ice to the laboratory for further analysis. DNA extraction. Genomic DNA was extracted from stored vaginal swabs using the Zymobiomics DNA extraction Kit (Zymo research Irvine, CA USA) adhering to the manufacturer's protocol as previously described 37 . Briefly, frozen swabs were thawed on ice and suspended into a 2 ml bashing bead lysis tube containing lysis solution. Microbial cells were lysed by mechanical disruption with a high-speed cell disruptor (FastPrep Classic Instrument, MP Biomedicals, LLC Irvine, CA, USA) set at 6.0 m/s for 1 min. The Bashing Bead lysis tube were centrifuge at 10,000 × g for 1 min. The resulting lysate (supernatant) was further processed with the Zymobiomics DNA extraction Kit (Zymo research Irvine, CA USA) and the DNA was eluted in 100 μl of TE buffer.
Vaginal 16S rRNA gene amplification and sequencing. Amplification of the V3-V5 hypervariable regions of the 16S rRNA gene was performed using a two step-PCR and then incorporating Illumina flow cell adaptors containing indices as previously described 38  GTC TCG TGG GCT CGG PCR products were diluted to 20 μL with PCR grade water and cleaned up using 1.0× AMPureAP beads (Beckman Coulter, Brea, CA), vacuum-dried, reconstituted in 12 μL of PCR grade water, quantified using a Quant-It dsDNA HS assay kit (Thermo Fisher Scientific Inc., Waltham, MA), normalized and pooled. The sequencing pool was concentrated, cleaned up using 1.8× AMPureAP beads (Beckman Coulter, Brea, CA). Pooled 16S amplicon samples are quantified using the KAPA SYBR FAST qPCR kit (KAPA Biosystems, Woburn, MA), diluted to 2 nM, denatured with an equal volume of 0.2 N NaOH, diluted to 8 pM with Illumina HT1 buffer, spiked with 10% PhiX, heat denatured at 96 °C for 2 min immediately prior to loading and sequenced using the MiSeq 600 cycle v3 kit (Illumina, San Diego, CA) and MCS v2.6.1.
Bioinformatic methods for microbiota analysis. The sequence reads were processed using the QIIME 2 bioinformatic pipeline (2019 v7) 40,41 . Raw paired end demultiplexed sequence were quality filtered and adapter trimmed. Further processing followed the DADA2 workflow where sequencing reads were denoised to form amplicon sequence variants (ASVs) 42 . A reference data base, SILVA database v132 was used for taxonomic assignment 43 . Read counts for ASVs assigned to the same taxonomy were summed for each sample. All samples were analyzed in the same run to reduce batch effects that may arise from run-to-run variability. A phylogenetic tree was constructed using FastTree v2.1 44 . ASVs with otherwise unidentified genus and species were identified by using BLAST to query their corresponding representative genomes. Representative BLAST matches were chosen if they achieved a query cover and percentage identity of 100% respectively. To normalise our data, a sampling depth of 1208 sequence read per sample was used. Artifact, non-bacterial and singleton ASVs were also removed 45 . Community state types (CST) were assigned to each sample based on ASV abundance using hierarchical clustering with Hellinger distance (Euclidean distance on square-root proportion data) and Ward linkage as described previously 46 . Each CST is defined by the dominance of L. crispatus (CST I), L. gasseri (CST II), L. iners (CST III), an heterogenous mixture of strict and facultative anaerobes (CST IV) comprising of CST IV-A and CST IV-B and L. jensenni (CST V) as previously described 8,47 . Tables comprising of distribution of samples after quality control and total sequence data for each sample with more than 1208 sequences were generated and used for statistical analyses. A total of 1,067,231 sequences were generated from 87 samples for an average of 12,039 reads per sample ( Figure S1). Statistical analysis. Custom scripts written in the statistical language R were used for statistical analysis 48 .
Significant differences between participants clinical characteristics were determined using two-tailed Student's www.nature.com/scientificreports/ t test, Wilcoxon rank sum test or Fisher's-exact test where appropriate. Analysis of microbiome data was performed for α-diversity, β-diversity and taxa abundances. For α-diversity and β-diversity analyses we rarefied the sequence data to a common depth of 1208 reads. Briefly, α-diversity reflects species richness and evenness within bacterial populations 49 . Two α-diversity metrics, the observed ASV number and the Shannon index, were calculated on the rarefied dataset. The observed ASV number reflects species richness, whereas the Shannon index reflects species' richness and evenness. To test the longitudinal association between the covariates and α-diversity, a linear mixed effect model (LMM) was used while adjusting for potential confounders where necessary. β diversity reflects the similarities or shared diversity between bacterial communities in terms of ecological distance between samples 49 . Four β diversity measures (unweighted, generalized (α = 0.5), weighted UniFrac distances and Bray-Curtis distance) were calculated 50 . The unweighted UniFrac reflects differences in community membership (i.e., the presence or absence of an ASV), whereas the weighted UniFrac captures both differences in community membership and also reflects differences in abundance. The Bray-Curtis quantifies the compositional dissimilarity between the two groups based on counts of each sample across each group 51 . The Augmented Dickey-Fuller Test, was used to identify stable taxa across all pregnancy samples 52 , previously described in 53 . The quantitative measure of community stability was estimated with the UniFrac distance measure (both weighted and unweighted). To test the association between the steroid hormones (estradiol and progesterone), other covariates including age groups (age above 35 and age below 35), delivery mode (ceaserian section and vaginal delivery, pregnancy status (multigravida and primigravid) and the vaginal microbiome, PERMANOVA, a distance-based analysis of variance method was used (999 within-subject permutations, "adonis" function in the R "vegan" package 1.17-4 54 . An Omnibus test which takes the minimum of the PERMANOVA p-values of individual β-diversity measures as test statistic, was used to combine association evidence from different β-diversity measures and an overall association p-value was reported ("PermanovaG" function in the R "GUni-Frac" package v1.1) 55 . Ordination plots were generated using classic multi-dimensional scaling (MDS) in R with the command described ("cmdscale" function in the R "stats" package). Significance at differences in taxa abundance between pregnancy and postpartum samples was assessed using normalized abundance data at each taxonomic rank (phylum, class, order, family, genus and ASV level) using a permutation test (999 within-subject permutations) 55 . The count data was normalized by dividing by the Geometric Mean of Paired Rations (GMPR) size factor, which has been shown to be a robust estimation of the sampling effort 56 . The permutation test uses the F-statistic of a linear model (square-root normalized taxa abundance as the response variable) as the test statistic. Taxa with prevalence less than 10% or with a maximum proportion (relative abundance) less than 0.2% were excluded from testing to reduce the numbers of tests. Next, false discovery rate (FDR) control (Benjamini-Hochberg procedure) 57 was used to correct for multiple testing at each taxonomic level and FDR-adjusted p-values or q-values < 0.10 were considered significant. Similar permutation tests were used to identify stable taxa, whose within-subject variability was smaller than the between-subject variability by treating the subject ID as a covariate in the permutation test.
Ethics approval and consent to participate. The Table S1. Data presentation of participants across all sampling timepoint during pregnancy and postpartum is illustrated in the flow chart in Figure S2.  www.nature.com/scientificreports/ between-subjects (q < 0.01, Permutation test) (Fig. 2). Between first and second timepoint during pregnancy, the vaginal CST I (L. crispatus dominated) remained stable but transitioned to CST IV-B in one participant. A similar trend was equally observed with CST II (L. gasseri dominated) and CST III (L. iners dominated) (Table S2).
Tracking the non-lactobacillus dominated vagitypes (CST IV), our result show that CST IV vagitypes observed at the first sampling timepoint remained persistent at subsequent visits. In particular, vaginal samples of participants who delivered preterm were dominated by Atopobium vaginae from the first sampling time point and remained persistent until preterm delivery. Further, to demonstrate how stable individual microbial taxa were across the pregnancy sampling time points, we performed the Augmented Dickey Fuller (ADF) test, which evaluated the null hypothesis that a given ASV is non-stationary (that is, the ASV tends to return to an equilibrium value). L. crispatus, A. vaginae, and L. iners were stationary according to the ADF test (P < 0.05). In particular, L. iners had the highest index of stability from our ADF model (Table 2).

Trend in diversity across longitudinal vaginal samples in term and preterm group.
In our Nigerian cohort, we observed a significant trend of increasing α-diversity with increasing gestational age. At the first time point (middle trimester), a trend of lower α-diversity of the vaginal microbiome was observed in the majority of women who delivered at term whereas a higher α-diversity was observed at second and third timepoints (p = 0.006 and 0.07; LMM for Observed ASV and Shannon diversity respectively) (Fig. 3A,B).  www.nature.com/scientificreports/ The microbiome structure of the preterm group across pregnancy follows a predictive pattern of dissimilarity in community composition which mirrors the dichotomy observed in the community structure. Figure 3 illustrates a longitudinal β-diversity measure across pregnancy samples (term and preterm samples) which shows different community structures across term and preterm groups (p = 0.001, PERMANOVA) for both weighted and unweighted UniFrac measure. Principal coordinate analysis also shows clear separation between the microbial communities of term and preterm delivery groups across samples collected longitudinally (Fig. 3C,D).    (Fig. 4A). With increasing gestational age, pregnant women were also more likely to steadily maintain a Lactobacilli profile (Fig. 4B). A taxa analysis of these profiles confirmed a significant shift in the postpartum microbiome. While pregnancy was associated with a microbiome largely dominated by Firmicutes (Lactobacillus), puerperium was characterized by a shift in bacteria phylum structure to become dominated by anaerobes especially Actinobacteria (Gardnerella and Atopobium) (Fig. 4C).
Increased vaginal microbiota richness and diversity at postpartum. According to α-diversity, the observed amplicon sequence variants (ASVs) and Shannon index were significantly higher in the postpartum vaginal samples compared to pregnancy samples (p = 0.001 and 0.05 respectively, LMM) (Fig. 5A,B). With respect to β diversity, principal coordinate analysis also shows clear separation between the microbial communities of pregnancy and postpartum samples (Fig. 5C). This separation is consistent across all the β diversity measures (Fig. 5D-F).
Shift in vaginal microbiomes at postpartum. All CST I switched to CST IV, while other CSTs showed more resilience to change (Table S3). We also show herein that changes between pregnancy timepoints were much smaller than changes observed at postpartum. This difference reached statistical significance for all the measures described (Fig. 6A). Notably, this shift in microbiome at postpartum was equally observed with women who delivered preterm. The vaginal microflora of four PTB participants dominated by Atopobium during pregnancy persisted with increasing gestational age. However, we observed a compositional shift from Atopobium vaginae (CST IV) to Gardnerella vaginalis (CST IV) in PTB subjects who returned for postpartum sampling (Fig. 6B,C).

Association between vaginal microbiota and maternal covariates.
With our sources of variability model, we explored the relationship between the vaginal microbiota and steroid hormones, and identified factors that could impact the vaginal microbiome while adjusting for potential confounders. Age (women below 35 years old or women above 35 years old), pregnancy status (multigravid or primigravid) and delivery mode (caesarian section or vaginal delivery) were not significantly associated with the vaginal microbiota whereas maternal estradiol level (p estradiol = 0.006, Omnibus), and maternal progesterone level (p progesterone = 0.001, Omnibus) were significantly associated with the vaginal microbiome ( Table 3).

Dynamics of the vaginal microbiome with steroid hormones.
Hormone concentrations significantly increased throughout pregnancy and reduced postpartum (p < 0.001, LMM) (Fig. 7A,B). Commensurately, with a shift in hormonal milieu (both estradiol and progesterone concentration) pregnancy samples dominated by high proportions of Lactobacillus depleted to become dominated by BVAB (p Lactobacillus < 0.001 and p CST IV < 0.001; spearman correlation test) for both estradiol and progesterone respectively (Fig. 7C-F, Figure S3A-D). As estradiol and progesterone concentration increase during pregnancy, we observed an increasing trend in non-Lactobacillus dominated microbes ( Figure S4A-D). This trend, however, did not apply to all participants with Lactobacillus dominated vagitypes. Conducting a longitudinal (within subject) association of the vaginal microbiome and hormones with our permutation model, we observed that the proportion of L. crispatus increased commensurately with increasing hormone concentration (p estradiol = 0.036, p progesterone = 0.034; permutation test) (Fig. 7G,H) while other vagitypes including L. iners, and A. vaginae did not (Fig. 7I-L).

Relationship between postpartum vaginal microbiome and delivery mode.
To determine the association between postpartum vaginal microbiome and delivery mode, we explored the postpartum microbiome of women that delivered via Ceasarian section and vaginally. We demonstrate that delivery mode had no significant impact on the vaginal microbiome. (p richness = 0.981, LMM) (p shannon = 0.624, LMM) (Fig. 8A,B).

Discussion
Here we present an exploratory, high-throughput and longitudinal assessment of the vaginal microbiome during pregnancy and postpartum in a Nigerian cohort. The dominant taxon in the vaginal microbiome of pregnant Nigerian women that delivered at term is Lactobacillus. The findings support the hypothesis that having a Lactobacillus spp.-dominated vagitype is strongly associated with healthy pregnancy outcomes 1,2,6,22 . As shown here and in previous studies 1,3,29 in pregnancy, a larger proportion of women with healthy pregnancy outcome have Lactobacillus spp. in high relative abundance in the vaginal microbiota, whereas the majority of women lacking Lactobacillus spp. in the vaginal microbiota delivered preterm as also reported in previous studies 7,22,58,59 . In particular, Atopobium (CST IV) remained persistent throughout pregnancy in women who delivered preterm,  www.nature.com/scientificreports/ buttressing a cause-effect relationship between Atopobium and preterm birth 7,60-62 . Our data further reveal that the vaginal microbiome of Nigerian women was typically dominated by L. iners. This lends important clues to the question of whether L. iners is indicative of a healthy or dysbiotic vaginal microbiota 63,64 . The data presented here is consistent with studies which support the notion that L. iners is part of a healthy microbiota, as most women in our cohort with L. iners-dominated vagitype delivered at term. Consistent with previous studies 2, 65 , the structure and composition of the vaginal microbiome at postpartum are significantly altered to comprise mostly of CST IV vagitypes. This variation could be explained by the presence of lochia (alkaline vaginal discharge during postpartum period) 66 . A major mechanism by which Lactobacillus restricts the growth of possible pathobionts is by creating an acidic vaginal microenvironment 12,13 . This led to the hypothesis 2 that lochia alkaline discharge tends to impede Lactobacillus spp. growth and increase the propensity for several CST IV microbes to thrive.   2,29,65 , our data reveals a shift in the core structure and composition of the pregnancy vaginal microbiome during the postpartum period. This shift has led to the hypothesis 26,27 that hormones drive the vaginal microbiome in pregnancy. While a compelling notion, delivery comes with many confounding factors. Our work verified this trend using time-longitudinal statistical associations during pregnancy, thereby assessing the role of hormones without the confounders of delivery. Our results support the notion that estrogen and progesterone enhance the presence of L. crispatus, but not other Lactobacilli species. This supports the notion that sex steroids steer the maturation of the vaginal epithelium giving rise to accumulation of glycogen 67 , which is then catabolized to products including maltose, maltriose, and maltatetraose by alpha amylase present in the host vaginal mucosa and that continuously supports Lactobaccillus colonization 68 . This finding further demonstrates that L. crispatus is a significant biomarker for healthy pregnancy progression 69 .
Our longitudinal analysis of the pregnancy vaginal microbiome revealed a higher initial microbiome diversity from some pregnant women in our cohorts compared to other reported populations 6,70 . Since this study is a pilot longitudinal assessment, we posit that the high initial microbiome diversity is individualized and requires further investigation with a larger cohort as we are limited by a small sample size. Overall, our analysis of the pregnancy and postpartum vaginal microbiome confirms that healthy pregnancy outcome is characterized by an enrichment of Lactobacillus spp. while PTB and puerperium vaginal microbiome is mostly characterized by CST IV microbes.
Large multicenter studies are needed to evaluate the scope of our findings. Undoubtedly there exists a relationship between age, parity and hormones as younger and primiparous women tend to have higher hormone concentration during pregnancy than older and multiparous women 71 . Further, some vagitype are more predominant in older women than younger women. For example, Lactobacillus iners was found dominate in women above 35 years compared to younger women 1 however there was no significant association between both age, parity and the vaginal microbiome in this study. Indeed, a limitation of this study, is that direct matching and comparison of vagitypes across women of similar age, similar parity and similar hormone concentration were not obtained and thus it is difficult to establish a direct link between hormonal regulation of the vaginal microbiota across women of different age group and parity. This represents an important research question for future study. Nonetheless, our data merits consideration for studies investigating the pregnancy vaginal microbiome during pregnancy and postpartum and the relative contribution of and estradiol and progesterone concentration".
In conclusion, we report a longitudinal study of the vaginal microbiome during pregnancy, puerperium and provide insights into the relative contribution of estradiol (E2) and progesterone (P1) in shaping of the vaginal microbiome during pregnancy and post-delivery in a Nigerian cohort. We provide evidence that the vaginal microbiome is associated with obstetric outcome. Our work shows that increasing hormone concentrations in pregnancy favors the dominance and high relative abundance of L. crispatus in the vaginal microbiota. Furthermore, the shift from a Lactobacillus dominated microbiome to a non-Lactobacillus dominated microbiome during postpartum is associated with a drastic fall in hormone levels. This has implications for future studies designed to explore relationships between the steroid hormones and vaginal microbiome in pregnancy, postpartum, and obstetric outcome.

Data availability
The metadata supporting this article has been provided and the sequence dataset supporting the results of this article has been publicly deposited and are available at the NCBI Sequence Read Archive, with BioProject accession no. PRJNA757760.