Large-scale characterisation of the pregnancy vaginal microbiome and sialidase activity in a low-risk Chinese population

Vaginal microbiota-host interactions are linked to preterm birth (PTB), which continues to be the primary cause of global childhood mortality. Due to population size, the majority of PTB occurs in Asia, yet there have been few studies of the pregnancy vaginal microbiota in Asian populations. Here, we characterized the vaginal microbiome of 2689 pregnant Chinese women using metataxonomics and in a subset (n = 819), the relationship between vaginal microbiota composition, sialidase activity and leukocyte presence and pregnancy outcomes. Vaginal microbiota were most frequently dominated by Lactobacillus crispatus or L. iners, with the latter associated with vaginal leukocyte presence. Women with high sialidase activity were enriched for bacterial vaginosis-associated genera including Gardnerella, Atopobium and Prevotella. Vaginal microbiota composition, high sialidase activity and/or leukocyte presence was not associated with PTB risk suggesting underlying differences in the vaginal microbiota and/or host immune responses of Chinese women, possibly accounting for low PTB rates in this population.


INTRODUCTION
There is substantial evidence implicating the pregnancy vaginal microbiota in shaping maternal and neonatal health outcomes [1][2][3] . The dominance of the vaginal niche by commensal Lactobacillus species is often considered "optimal" due to their ability to prevent pathogen colonization through competitive exclusion, in part achieved through the production of antimicrobial compounds and production of lactic acid 4,5 . Recent studies have highlighted L. crispatus dominance as being protective against preterm birth (PTB) 2,6-9 and neonatal sepsis following preterm prelabour rupture of membranes 1 . By contrast, colonization by L.iners 7,10-12 or Lactobacillus species depleted, high diversity compositions are associated with an increased risk of PTB 2,7,9,10,[13][14][15] . Despite the majority of PTBs (60%) occurring in Asia 16 , molecular-based characterization of the vaginal microbiota in pregnancy and its relationship with PTB has largely been restricted to Northern American and European populations. Moreover, ethnicity is now recognized as a potential confounder of the relationship between the vaginal microbiome and PTB, particularly between Caucasian and women of African-descent in North American or European populations 2,9,10,17 .
The Amsel criteria 18 and Nugent scoring system 19 are commonly used to diagnose Bacterial Vaginosis (BV), a condition characterized by a loss of vaginal lactobacilli and overgrowth of anaerobes, which is associated with a two-fold increased risk of PTB [20][21][22][23] . The Amsel criteria requires microscopy and is subject to potential interobserver bias 24 , whereas Nugent scoring is limited by the requirement for laboratory access. Enzymatic-based assays for rapid BV diagnosis may offer an objective, point-of-care alternative to BV diagnosis in clinical settings, including during pregnancy 25,26 . These assays often work by measuring microbial sialidase (neuraminidase), produced by BV-associated bacteria such as Gardnerella vaginalis, which removes sialic acid from sialoglycoconjugates including those on the surface of vaginal epithelial cells, providing a nutrient source and exposing glycan-binding sites for bacterial adhesion. Additionally, sialidase is thought to mediate biofilm formation and the establishment of suboptimal vaginal microbiota compositions [27][28][29] . High vaginal sialidase levels have previously been associated with an increased risk of PTB 30 and with failure of cervical cerclage 31 , a procedure used to reinforce the cervical opening in women at risk of preterm delivery due to cervical shortening. Sialidase-producing taxa associated with BV have also been implicated in chorioamnionitis, a risk factor for PTB that is characterized by inflammation of the fetal membranes 32 . Although BV is not classified as an inflammatory syndrome, the disease has been associated with the presence of vaginal leukocytes, which have been purported to offer predictive value in identifying upper reproductive tract infections [33][34][35] and PTB [36][37][38] . Quantification of leukocyte counts using vaginal wet mount microscopy could therefore represent an easily accessible and cost-effective method to determine cervicovaginal inflammation 39 .
In this study, we characterized the bacterial component of the vaginal microbiome in 2689 Chinese women sampled at midpregnancy and in a subset (n = 819), explored the relationship between vaginal microbiota composition, sialidase activity, and leukocyte presence with risk of PTB. recruited between November 2015 to December 2018. The median (IQR) time of sampling was 16 +4 weeks +days gestation (range 15 +6 -17 +6 ) after exclusion of women with samples collected in the third trimester. Of these, 1397 delivered at the same hospital and maternal and neonatal outcomes data were obtained. The median (IQR) gestation at delivery was 39 +3 (38 +5 -40 +1 ) weeks +days . The PTB (<37 weeks) rate in the cohort was 5.44% (76/1397) (Supplementary Data 1). The remaining 1356 women delivered elsewhere and due to data protection, pregnancy outcome data were not available. There was no significant difference in the gestation of sampling between women with and without available outcome data (Mann-Whitney Test, p > 0.05; Supplementary Fig. 1).

The pregnancy vaginal microbiota of Chinese women
The 2796 sequenced vaginal samples generated a total of 58,582,840 reads with a mean read count of 20,952 per sample. Of these samples, 2689 passed library size and microbiome classification criteria ( Supplementary Fig. 1). Only those samples collected in the first and second trimester were included in subsequent analyses (2646/2689, 98.4%) After removal of kit and reagent contaminants, a total of 82 taxa were detected and included in subsequent analyses. Vaginal microbiota profiles were classified into 19 groups based upon the dominant (>30% relative abundance) taxa observed within each sample 2 . At the genus level, the majority of samples were dominated by Lactobacillus (2275/2646, 85.98%), which were predominately L. crispatus (1058/ 2646, 39.98%) or L. iners (952/2646, 35.98%) dominated (Fig. 1a). The median relative abundance of L. crispatus and L. iners in these samples was 96.37% and 96.19% respectively and the alpha diversity as measured by the Shannon Index was similar between the two groups (median, 0. 22 Fig. 3) were similar (Fisher's Exact Test, p = 0.24). Additional analyses also showed that vaginal microbiota compositional characteristics and their relationship with clinical outcomes were not impacted by the type of extraction kits used (see Supplementary Fig. 4).
Vaginal microbiota, sialidase activity, and leukocyte presence are not associated with risk of PTB and chorioamnionitis in pregnant Chinese women In women with available outcome data (n = 1379), the proportion of Lactobacillus dominated or depleted vaginal microbiota (p = 0.57; Fisher's Exact test) and VMGs (p = 0.05; Fisher's Exact test) were similar between women delivering preterm or term, including when preterm births were divided into early (<34 weeks gestation) and late (≥34 weeks gestation)( Supplementary Fig. 6). Likewise, proportions of high or low sialidase activity (p = 0.45; Fisher's Exact test) and leukocyte high/low (p = 0.74; Fisher's Exact test) were comparable between women subsequently experiencing preterm or term deliveries (Fig. 4a). Apart from women who had Lactobacillus dominated microbiomes, high sialidase activity, and leukocyte low which had insufficient sample size (n = 2), no significant difference was observed in birth gestation based on Lactobacillus abundance, VMG prevalence, sialidase activity, and leukocyte wet mount results (Kruskal-Wallis Test, p > 0.05) ( Fig. 4b and Supplementary Table 4). Logistic regression results similarly showed vaginal microbiome Lactobacillus composition (p = 0.83), sialidase activity (p = 0.45) and leukocyte high or low (p = 0.70) did not significantly contribute to birth outcome (Supplementary  Table 5). Finally, no relationships were observed between vaginal microbiota composition, sialidase activity and/or leukocyte presence, and chorioamnionitis ( Supplementary Fig. 7).

DISCUSSION
In this large cross-sectional study of 2646 pregnant Chinese women, the vaginal microbiota in mid-pregnancy was characterized by Lactobacillus spp. dominance and low diversity. This observed profiles. In previous work, we have observed that the most prevalent vaginal microbiota community compositions in European Caucasian women are those dominated by L. crispatus (50%) followed by L. iners (25%), whereas the inverse has been reported for Asian and Black women (L. iners, 48% and 50% v L. crispatus, 26% and 20%, respectively) 6 . Higher frequency of L. iners dominated vaginal microbiota has also been reported in Indian 43 , Karen and Burman 44 , Hispanic and African American pregnant women 9,10,17 . Our results therefore provide further evidence that ethnic and/or geographical differences are a major source of variation in the underlying structure and composition of the vaginal microbiome during pregnancy.
Correlation analyses in our patient cohort showed that L. crispatus was negatively correlated with almost all other vaginal taxa, highlighting its well-described exclusionary behavior in the vaginal niche that appears to be broadly consistent across different ethnic groups 6,13,17,45 . However, we also observed a strong negative correlation between L. iners and most other vaginal taxa. This was surprising given that L.iners has been shown to co-occur with various BV-taxa in other ethnic groups including Caucasian and African American women 9,46 . The highly negative correlation observed between L. iners and L. crispatus might be attributable to overlapping ecological functions in the vagina 5,47 . Compared to L. crispatus, L. iners has a substantially smaller genome, thought to be indicative of a symbiotic and/or parasitic role in the vaginal niche 5,47,48 . Similar to G. vaginalis, L. iners can produce pore-forming toxins such as inerolysin and vaginolysin, which can lead to lysis of host cells and release of carbon sources, particularly during times of nutrient scarcity 5,47,48 . This behavior may explain the observed positive relationship between L.iners and leukocyte presence in our study cohort. Consistent with these findings, a recent study of 83 healthy Chinese pregnant women reported a positive association between leukocyte esterase concentrations and increased L. iners levels determined using quantitative real-time PCR 49 . These women were also reported to have increased presence of white blood cells and other non-Lactobacillus morphotypes observed under microscopy. In vitro experiments using human vaginal epithelial cells have also demonstrated that L. iners stimulates increased pro-inflammatory cytokine production compared to L. crispatus 50,51 . Despite these findings, we did not observe any relationship between L. iners dominance or any vaginal microbiota profile and subsequent risk of PTB in this cohort. This is in contrast to previous studies of predominantly Caucasian women by ourselves and others that have reported a relationship between L.iners and increased risk of PTB or cervical shortening, which is clinically used as a marker of PTB risk 6,7,12,52 . Our data are consistent with earlier reports in Hispanic and African American women indicating that L. iners dominance is not a risk factor for PTB in these ethnic groups.
In our sub-cohort analysis, high vaginal sialidase activity was strongly associated with Lactobacillus depleted microbiota enriched with BV-associated taxa including Gardnerella, Prevotella, Atopobium, and Megasphaera species. This finding is consistent with previous studies in both pregnant and nonpregnant women where the production of sialidase by BV-associated taxa, such as Gardnerella spp. and Prevotella spp., is thought to be important for pathogenesis 27,53,54 . In this context, desialylation of glycolipids and/or glycoproteins (e.g. immunoglobulins, cytokines, cellular receptors, mucins, and antimicrobial molecules) can decrease the ability of host defence responses to recognize and bind to microbes whilst increasing bacterial adherence, invasion, and tissue breakdown 27,30,54,55 . Interestingly, high sialidase activity was observed in some Lactobacillus-dominated vaginal microbiota. These often contained low relative aundance of BV-associated taxa indicating that perhaps even low levels of sialidase-producing bacteria may be sufficient to produce high sialidase activity. It is also possible that sialidase was produced by other microorganisms including viruses and fungi that were not assessed in our study 56,57 . In contrast to these findings, a proportion of samples (23%) harboring Lactobacillus spp. depleted vaginal microbiota enriched with BV-associated taxa did not have high sialidase activity. This was supported by our findings indicating sialidase activity has high specificity (99%) but low sensitivity (14%) in predicting high-diversity, Lactobacillus spp. depleted vaginal microbiota. Low sialidase activity in these samples could be due to several factors: 1) the production of sialidase by BV-taxa, including Gardnerella spp., is strain-dependent; 27,30,31 2) binding of sialidase to the BV test kit substrate may be influenced by vaginal pH; 26,58 3) the amount of sialidase produced by BV-taxa maybe insufficient to reach the detection limit of the BV test kit; and/or 4) the bacterial load of BV-taxa, which was not quantified in our study, may be inadequate to produce sufficient sialidase for detection by the BV test kit. Overall, our findings suggest that high sialidase activity is significantly associated with BV-taxa but cannot be used as an accurate proxy for high-diversity, Lactobacillus spp. depleted vaginal microbiota in pregnant women.
Elevated sialidase activity has previously been associated with spontaneous PTB and late miscarriage in a North American study of 1806 pregnant women when measured at 12 weeks gestation 30 , and with early PTB and low birth weight in a mid-trimester study of 579 Danish women 59 . Additionally, in a recent study of 85 Chinese women, sialidase activity was associated with subsequent cervical cerclage failure, a risk factor for PTB 31 . However, in our study, no significant association was found between elevated sialidase activity and PTB. Our findings are similar to a previous nested case-control study of 126 pregnant women, where elevated sialidase activity measured at mid-pregnancy was not associated with PTB risk 53 . The observed differences between studies could be due to different methods for measuring sialidase activity where a single cut-off was used rather than a quantitative measurement. An association between vaginal leukocytes and preterm labor or risk factors of PTB including reproductive tract infections, histologic chorioamnionitis and inflammatory cytokines have also previously been reported 33,34,39,[60][61][62] . Here, we classified leukocyte presence based on vaginal wet mount microscopy, which has been shown to be useful for identification of cervicovaginal inflammation in pregnant women 39,63 . Logistic regression analyses indicated that neither leukocyte presence, sialidase activity nor vaginal microbiota contributed to preterm risk in our study cohort. The rate of PTB in our study population was 5.4% which is slightly lower than the background rate in China (7.2%) 64 , but comparable to published preterm birth rates Fig. 2 Comparison between women in sialidase sub-cohort. a Microbiome composition, diversity (Shannon Index) and richness (species observed) for all women with high (n = 36) and low (n = 783) sialidase activity. b Proportion of Lactobacillus abundance groups (Lactobacillus dominant and Lactobacillus deplete) and vaginal microbiome groups (VMG) for women with high or low sialidase activity. Statistical significance based on Fisher's Exact Test. c LDA showing effect size of differentially abundant taxa associated with microbiota of women with high (red) or low (green) sialidase activity. d Comparison of relative abundance between main BV-associated bacteria identified as significantly different (Mann-Whitney Test, p < 0.05) between women with high and low sialidase activity (see Supplementary Table 3). The bounds of the box represent the first and third quartiles, center line represents the median, and whiskers show min-to-max values. e Confusion matrix with Lactobacillus abundance (dominant as positive/deplete as negative) as the reference test and sialidase result (High/Low) as the predicted test. The sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated.
for the local population in Guangzhou (4.15% and 6.0%) [65][66][67] . These rates are considerably less than the global average of 10% 68 . Our findings therefore highlight differences in the vaginal microbiota and/or host immune response as a potential contributing factor to reduced incidence of PTB in Chinese women.
There are several limitations to our study that should be considered when interpreting the results. Birth outcome data was unavailable for 48% of the women in our cohort. These women did not return to the recruitment site of Nanfang Hospital and due to data protection considerations, were unable to be contacted for follow-up. Although vaginal microbiota composition in these women was comparable to those with available outcome data, non-observed outcomes may represent a potential source of bias. Further, the pregnancy vaginal microbiota can be impacted by a variety of factors including pregnancy history, smoking status, genetics and hygiene habits 21,69,70 . This information was not available for our patient cohort and cross-sectional sampling of the vaginal microbiota did not permit us to assess the potential role of vaginal microbiota dynamics in shaping clinical outcomes 17,71,72 . Our study was also limited to the assessment of the bacterial component of the vaginal microbiota via 16S rRNA gene sequencing, which does not provide sufficient resolution to Boxplots showing relative abundance of L. crispatus, L. gasseri, L. iners, L. jenseni, L. delbruecki, and Lactobacillus spp. in women with leukocyte low (grades I and II, green) and high (grades III and IV, red). The bounds of the box represent the first and third quartiles, center line represents the median, and whiskers show min-to-max values. c LDA of differentially abundant taxa and cladogram displaying bacterial clades and nodes identified as differentialy abundance from LDA analysis in vaginal microbiota of women with leukocyte low (grades I-II, green) and high (grades III-IV, red). Statistical significance of categorical variables based on Fisher's Exact Test and continuous variables based on Kruskal-Wallis Test. Fig. 4 Association between the vaginal microbiome, sialidase activity and leukocyte presence with birth outcome. a Stacked bar plots showing proportion of Lactobacillus abundance (deplete, red; or dominant, blue), vaginal microbiome groups (VMG), sialidase activity (high, pink; or low, red) and leukocyte low (grades I-II, green) or high (grades III-IV, red) in preterm and term groups. Statistical significance based on Fisher's Exact Test. b Boxplots showing birth gestation in weeks grouped by Lactobacillus abundance, sialidase activity, and leukocyte presence or absence with dot points colored by VMG. The bounds of the box represent the first and third quartiles, center line represents the median, and whiskers show min-to-max values. No significant difference between boxplots with n ≥ 3 based on Kruskal-Wallis Test, p > 0.05. account for strain-level variations associated with potential function of microbial communities 73,74 . Future studies could apply function-based profiling approaches to enable a deeper insight into potential relationships between predicted function of microbiome communities within different ethnic groups and preterm birth risk phenotype.
In conclusion, our study provides new insight into the vaginal microbial composition and structure of Chinese pregnant women. Although sialidase activity was predictive of high diversity vaginal microbiota compositions, neither were associated with increased risk of PTB. Lactobacillus spp. depleted vaginal microbiomes with high sialidase activity and leukocyte presence are not associated with higher risk of PTB in Chinese women. Our results provide further evidence that ethnicity is an important determinant of microbiota-host interactions during pregnancy and highlight the need for further investigations into the mechanisms underpinning these relationships.

Study design
This prospective study was reviewed and approved ([2013] EC (100)) by the Ethical Committee of Nanfang Hospital, Southern Medical University, Guangzhou, China. Written, informed consent was obtained from all participants. Pregnant women at their first prenatal visit were recruited from the outpatient clinic of Nanfang Hospital of Southern Medical University in Guangzhou, China, from January 2015 to December 2018. All women were recruited consecutively, with no disruption to recruitment during this period. Women who received antibiotics, prebiotics or probiotics within 30 days prior to vaginal swab collection and/or who had sexual activity within 48 h of sample collection were excluded. Metadata collected included maternal age, mode of labor, gestation at delivery in days (calculated using the last menstrual period and/or ultrasound data) and chorioamnionitis (Supplementary Data 1).

Sample collection and processing
Vaginal samples were collected using a sterile swab that was inserted into the vaginal posterior fornix, before being gently rotated 360°for approximately 20 rotations prior to removal. All swabs were placed immediately on ice and were then stored at −20°C within 1 h of collection before being transferred to −80°C within 24 h for long-term storage. For DNA extraction, swabs were immersed and vigorously mixed in 500 μL of sterile water before the solution was transferred to a clean 2 mL centrifuge tube. The sample was then vortexed for 5 sec before being centrifuged at 13,800 g for 10 min. The supernatant was removed, and the pellet retained for DNA extraction. For PCR negative controls (n = 82), 2 μL of diethylpyrocarbonate (DEPC) water was used as non-template control. For DNA extraction kit negative controls (n = 21), 500 μL of sterile water was used instead of test samples for the DNA extraction process.
DNA extraction and 16S rRNA (V4) amplicon sequencing DNA was extracted from vaginal swabs manually using the BioTeke bacterial DNA extraction kit (BioTeke Corporation, Cat #DP7001) per manufacturer's instructions (n = 1176) or via an automated protocol using the Bioeasy bacterial DNA extraction kit (Shenzhen Bioeasy Biotechnology Co. Ltd., Cat #YRMBN7001) on a Thermo Scientific TM KingFisher TM Flex Purification System (n = 1620). The V4 region of bacterial 16S rRNA was amplified using barcoded V4F 5' GTGCCAGCMGCCGCGGTAA 3' forward and V4R 5' CTACCNGGGTATCTAAT 3' reverse primers. The PCR condition included an initial denaturation step at 94°C for 5 min, 30

Sialidase enzyme activity detection
A total of 848 women had an additional vaginal swab collected for detection of sialidase enzyme activity. Sialidase enzyme activity was measured as per manufacturer's instructions using a single-enzyme BV kit based on the colorimetric method (Zhuhai DL Biotech. Co. Ltd). Briefly, each swab sample was immersed into a BV test bottle solution containing 5-bromo-4-chloro-3-indolyl-α-D-N-acetylneuraminic acid (BCIN) as a substrate 75 . The BCIN substrate hydrolyzes and reacts with the added 1 -2 drops of BV chromogenic solution containing hydroxide and potassium acetate, resulting in a color reaction that was measured using the BV-10 analyser (Zhuhai DL Biotech. Co. Ltd). The resulting sialidase activity is defined based on enzyme unit (U), where one U is the amount of sialidase required to catalyze 1 nmol of BCIN per minute in a ml. The amount of sialidase activity corresponded to color changes in the BV test bottle with ≥7.8U/ml for a positive result (blue or green) and 0 -7.8U/ml for a negative result (yellow). The samples were defined as having high or low sialidase activity based on ≥7.8U/ml or 0 -7.8U/ml, respectively 26 .

Leukocyte wet mount
Preparation of leukocyte wet mounts was performed by agitating vaginal swabs in 1 mL of saline solution and a single drop of the fluid was then placed on a glass slide 76,77 . A cover slip was applied on the droplet and the slides were then inspected under 400x maginification. The average white blood cell count (non-clumped) in 10 nonadjacent fields of magnificant were determined. Leukocyte counts were categorized into grades I (0-5 counts/Hp), II (5-10 counts/Hp), III (10-15 counts/Hp) or IV (>30 counts/ Hp). Grades I and II were further classified as leukocyte low (<15 counts/ Hp) and grades III-IV as high (>15 counts/Hp) 63 .

Chorioamnionitis diagnosis
Chorioamnionitis was diagnosed using a combination of clinical symptoms and histological assessment. Clinical symptoms included the presence of high maternal fever (≥39°C), maternal tachycardia (>100 beats/min) and malodorous amniotic fluid. Histological chorioamnionitis was diagnosed based on the presence of inflammatory cells in the chorionic plate and/or the chorioamniotic membranes consistent with acute inflammation and the presence of neutrophils in the wall of the umbilical vessels and Wharton's jelly 78,79 . Staging and grading of chorioamnionitis was performed following the guidelines of the Amsterdam Placental Workshop Group Consensus Statement 80 .

Bioinformatics and statistical analyses
The raw sequencing reads were processed using the DADA2 package (v1.6.0) in R (v3.4.3) 81 . The raw paired-end reads were assigned to samples based on their unique barcodes and truncated by cutting off the barcodes and primer sequences. The reads were then processed using the DADA2 package (v1.6.0) in R (v3.4.3) according to the following steps: quality controls, dereplication, error rate calculations, sequences denoising, paired-end reads merging, amplicon sequence variant (ASV) table constructing and removal of chimeras. The sequences in the ASV table were annotated using the RDP Classifier 82 with the GreenGenes database 83 . The sequences assigned to Lactobacillus were further classified into species level using UCLUST 84 with the customized Lactobacillus database (V4 region extracted from representative sequences of Lactobacillus species from RDP database).
Each vaginal bacterial profile in the cohort was initially classified on the basis of the highest relative abundance of the dominant species taxon within the sample 2,89 . Dominant species with <30% abundance were