Severe preeclampsia is associated with a higher relative abundance of Prevotella bivia in the vaginal microbiota

We sought to compare the vaginal microbiota profiles of Taiwanese women with severe preeclampsia (SPE) and normotensive control pregnancies. In a discovery cohort, vaginal swab samples and paired blood specimens were simultaneously obtained at the time of caesarean delivery from 30 women with SPE and 30 controls. The composition of vaginal microbiota was characterised by 16S ribosomal RNA gene sequencing of the V3–V4 region. Results were subsequently validated by real-time qPCR. We sought confirmation of our findings in an expanded cohort consisting of 58 women with SPE and 55 controls. In both the discovery and confirmation cohorts, women with SPE had higher relative abundance of Prevotella bivia in their vaginal microbial community (P = 0.006 and 0.011, respectively). Plasma levels of tumour necrosis factor alpha (TNF-α) were higher when compared with controls (P = 0.031) in the confirmation cohort. Three variables (vaginal Prevotella bivia, plasma TNF-α, and body mass index [BMI]) were included in a prediction panel for SPE. Of these, BMI was the most predictive variable. The area under the curve (AUC) of predicted probability values for the three-variable panel revealed that it can discriminate between SPE and normotensive pregnancies with good accuracy (AUC = 0.797, P < 0.001). We conclude that enrichment of Prevotella bivia in vaginal microbiota, which is tightly regulated by BMI, may be involved in the pathogenesis of SPE.

www.nature.com/scientificreports/ that allows for the assessment of bacterial diversity within clinical samples and taxonomical classification of bacterial sequences 20 . However, the question as to whether vaginal dysbiosis may play a role in the pathogenesis of SPE remains unanswered. We therefore designed the current study to compare the vaginal microbiota profiles of Taiwanese women with SPE and with normotensive pregnancies. On the basis of previous research 21 , we also hypothesised that SPE could be associated with higher plasma levels of proinflammatory molecules.

Results
General characteristics of the study participants. The general characteristics of the women included in the discovery cohort are summarised in Table 1. Cases with SPE were found to differ significantly from control women with respect to four variables: gestational age (lower in cases with SPE), body mass index (BMI; higher in cases with SPE), betamethasone use for foetal lung immaturity (higher in cases with SPE) and neonatal birth weight (lower in cases with SPE). The general characteristics of the expanded validation cohort are reported in Supplementary Table 1.
Diversity and richness of vaginal microbiota in the study participants. The mean number of raw reads per sample was 2.2 × 10 5 for all participants. After exclusion of chimera reads and selection of qualified reads, the mean effective number of reads per sample was 1.24 × 10 5 and 0.90 × 10 5 for cases with SPE and controls, respectively. The rarefaction curve revealed that the sequencing depth in each sample was sufficient to capture most of the microbial diversity after reaching a saturation plateau phase (data not shown). Species richness in the vaginal microbiota did not differ significantly between cases with SPE and controls. However, both Shannon (P = 0.001) and Gini-Simpson (P < 0.001) indices were higher in cases with SPE, indicating that their vaginal microbiota was characterised by a greater diversity than that of controls (Fig. 1a). The unweighted principal component analysis (PCA) plot revealed similar vaginal microbiota between the two study groups (Fig. 1b).

Differences in relative abundances of vaginal bacteria in SPE.
The relative abundance of bacteria was analysed in the two study groups at the main taxonomic ranks (phylum, genus and species) using gene sequencing of the V3-V4 region on a MiSeq platform. The five most common phyla identified in the vaginal microbiota of control women were Firmicutes (69.12%), Actinobacteria (25.44%), Proteobacteria (5.18%), Bacteroidetes (0.18%) and Tenericutes (0.07%), which altogether accounted for 99% of all detected bacteria (Fig. 2a). Notably, the relative abundance of Bacteroidetes was significantly higher in cases with SPE (3.13%) than in control women (0.18%; P = 0.015; Fig. 2b), although no differences were observed with respect to the remaining four phyla. As far as the genus rank is concerned, the 15  www.nature.com/scientificreports/ www.nature.com/scientificreports/ Halomonas, Rhodococcus, Dialister, Anaerococcus, Cryptobacterium and Aerococcus (Fig. 2c). Among them, we found that the relative abundances of Prevotella (P = 0.013), Atopobium (P = 0.020) and Aerococcus (P = 0.041) were significantly higher in cases with SPE than in controls (Fig. 2d). A similar-albeit not statistically significant-trend was observed for Anaerococcus (P = 0.064). When the four genera were subjected to multivariable logistic regression analysis, only Prevotella maintained a significant independent association with SPE (P = 0.008; Table 2). We subsequently analysed the distribution of the 15 most common bacterial species in the two study groups (Fig. 3a). Compared with controls, cases with SPE showed a higher relative abundance of Prevotella bivia (P = 0.006), Atopobium vaginae (P = 0.046), Aerococcus christensenii (P = 0.041) and Anaerococcus tetradius (P = 0.037, Fig. 3b). Multivariable logistic regression analysis revealed that Prevotella bivia was the only bacterial species that showed an independent association with SPE (P < 0.001; Table 3). Finally, we conducted a linear discriminant analysis (LDA) of effect size (LEfSe) to identify the differences in taxonomic distributions between cases with SPE and control women. SPE samples were found to have significantly higher levels of Prevotella bivia and Anaerococcus prevotii (Fig. 4, marked with a and f). Moreover, the results confirmed that the Prevotella bivia species, Prevotellaceae genus, Prevotella family, Bacteroidales order, Bacteriodia class and Bacteroidetes phylum (Fig. 4A, marked with a, b, c, d, and e, respectively) were significantly enriched in the vaginal microbiota of cases with SPE. Collectively, these findings are in agreement with the above observations that link vaginal microbiota composition to the pathogenesis of SPE through the enrichment of Prevotella bivia.

Validation of sequencing results by real-time qPCR.
We sought to validate the results pertaining to the two bacterial species showing the highest relative abundances in cases with SPE versus control women (Prevotella bivia and Atopobium vaginae) using real-time qPCR. In the discovery cohort, the relative abundance of both Prevotella bivia (r = 0.79, P < 0.001; Fig. 5a) and Atopobium vaginae (r = 0.86, P < 0.001) as determined by qPCR showed a significant positive association with the results obtained using gene sequencing. We therefore applied real-time qPCR to investigate the relative abundance of the two bacterial species in the expanded validation cohort. The results confirmed a higher relative abundance of Prevotella bivia in the vaginal microbiota of cases with SPE (P = 0.011), although significant differences were no longer evident for Atopobium vaginae (Fig. 5b). Also, Prevotella bivia showed an independent association with SPE (P = 0.009) in the validation cohort by multivariable logistic regression analysis (Table 3).

Plasma cytokine levels.
Data from the multiplex ELISA-based chemiluminescent assay conducted in the expanded validation cohort revealed that plasma levels of tumour necrosis factor alpha (TNF-α)-but not of other cytokines-were significantly higher in cases with SPE compared with control women (P = 0.031). This difference was consistently observed in both univariate and multivariable logistic regression analyses ( Fig. 6; Supplementary Table 2).

Three-variable panel to discriminate between SPE and normotensive uncomplicated pregnancies. Of the different variables under study, three (vaginal Prevotella bivia, plasma TNF-α and BMI)
were included in a prediction panel for SPE ( Table 4). The area under the receiver operating characteristic curve (AUC) of predicted probability values for the three-variable panel revealed that it can discriminate between SPE and normotensive pregnancies with slightly better accuracy than BMI, Prevotella bivia and TNF-α alone (AUC = 0.797, sensitivity = 0.655; specificity = 0.764; accuracy = 0.708; Supplementary Fig. 1).
To remove the confounding factor of gestational diabetes mellitus, we performed a subgroup analysis excluding 20 women with concomitant gestational diabetes from the group of women with SPE. The higher relative abundance of Prevotella bivia, elevated TNF-α levels and higher BMI in SPE as well as the model of prediction combining Prevotella bivia, TNF-α and BMI remained valid for discriminating between SPE and normotensive pregnancies (Supplementary Table 3).

Discussion
The main results of our study indicate that women with SPE have a higher relative abundance of Prevotella bivia in their vaginal microbiota and higher plasma levels of the proinflammatory cytokine TNF-α-the latter finding being in line with the published literature 22 . Since obese women have a higher risk of preeclampsia, BMI is one of the most important predictive variables. A logistic regression model adjusted for BMI was constructed on log-transformed data to identify differences in taxonomic distributions. After adjusting for BMI, the associations of Atopobium vaginae, Aerococcus christensenii and Anaerococcus tetradius with the disease were removed. Only Table 2. Bacteria genera identified in the discovery cohort (n = 60). a Data are presented as mean log10 (abundance [%] ± standard deviation). Abbreviations: OR = odds ratio; CI = confidence interval. *P < 0.05.

Genera
Control women (n = 30) www.nature.com/scientificreports/ www.nature.com/scientificreports/ Prevotella bivia retained independent significance for the prediction of SPE in the discovery cohort. However, the higher relative abundance of Prevotella bivia in women with SPE was only marginally significant after adjustment for BMI (P = 0.07) in the confirmation cohort, suggesting that the role of Prevotella bivia in SPE may be modified by BMI. We further combined these two parameters with maternal BMI to devise a three-variable panel that discriminates between SPE and normotensive pregnancies with good accuracy. Prevotella bivia is an anaerobic gram-negative bacterial species that has been previously associated with pelvic inflammatory disease and bacterial vaginosis 23,24 . A shift from the physiological lactobacilli-dominated vaginal microbiota to a relative abundance of anaerobic and facultative anaerobic bacteria has been shown to play a role in different adverse obstetric and gynaecologic outcomes, including miscarriage, premature labour, preterm birth, preterm premature rupture of membranes, chorioamnionitis, intrauterine infection, post-caesarean endometritis, upper genital tract infections and pelvic inflammatory disease 3,7,10,12,25,26 . Interestingly, inflammation of the chorionic plate has been associated with the presence of several bacterial species in the preeclamptic placenta-including Bacillus cereus, Listeria, Salmonella, Escherichia, Klebsiella pneumonia, Anoxybacillus, Variovorax, Prevotella, Porphyromonas and Dialister 27 . The question as to whether bacteria are capable of ascending from the vagina to the chorionic plate during the course of preeclamptic pregnancies remains unanswered and requires further scrutiny.
Previous research demonstrated that obesity is a risk factor for preeclampsia 28 and that the vaginal microbial community of obese women is characterised by an increased diversity with a relative preponderance of Prevotella spp. 17 Here, we show that both BMI and a higher relative abundance of Prevotella bivia in the vaginal microbiota were associated with SPE. Notably, the preponderance of Prevotella oralis in the oral microbiome has been previously associated with blood pressure outcomes in older women 29 . It is thus possible that the Prevotella genus may play a role not only in the pathogenesis of hypertension in general but also in hypertensive disorders of pregnancy. In our study, vaginal swab samples used for microbiota analysis were obtained at the time of caesarean section-a procedure which has been related to a lower abundance of Bacteroidetes in the intestinal microbiota of infants 30 . Although a previous study demonstrated that maternal overweight and obesity do not affect infant gut microbiota composition and diversity in caesarean-delivered infants 31 . we believe that further research on the associated between preeclampsia, obesity, maternal vaginal microbiota and infant gut microbiota after caesarean section is warranted. Accordingly, no information on the occurrence of preeclampsia was provided by Singh et al. 31 .
Preeclampsia is accompanied by a systemic inflammatory response 32 , and the ischemic placental injury occurring in this condition has been associated with an increased release of TNF-α in the maternal bloodstream 33 . Using a multiplex assay for eight different cytokines, TNF-α was the only inflammatory biomarker found to be increased in the plasma of our cases with SPE. Notably, we devised a three-variable panel-comprising BMI, vaginal Prevotella bivia and plasma TNF-α-that is capable of discriminating between SPE and normotensive uncomplicated pregnancies with good accuracy. The complex relationships between the three variables deserve further scrutiny in independent investigations.
Several limitations of our study merit comment. First, the sample size was small and our results should be considered as preliminary and hypothesis-generating. All of the study participants were of Taiwanese descent, potentially limiting the generalizability of our findings. Second, we did not specifically investigate whether the maternal vaginal microbiome at the time of caesarean delivery could affect the gut microbiota of newborns. Third, we did not specifically screen for common genital infections-including bacterial vaginosis, vulvovaginal candidiasis, and Chlamydia trachomatis infection. Nonetheless, all participants had national health insurance to cover human immunodeficiency virus and syphilis screening at the first prenatal appointment. Moreover, all newborns tested negative for Chlamydia trachomatis on urine samples collected immediately after delivery. There was no evidence of clinical vaginitis with profuse abnormal discharge suggestive of either bacterial vaginosis or candidiasis when collecting the vaginal swabs. Despite these limitations, we believe that our study may pave the way for clinical testing for Prevotella bivia vaginal colonization during pregnancy. Moreover, this assay may be used to determine whether a relative abundance of Prevotella bivia in vaginal microbiota along with BMI and plasma TNF-α during pregnancy can more accurately predict future development of SPE.
In summary, our results indicate that the higher relative abundance of Prevotella bivia in the vaginal microbial community, which is related to increased BMI, may be involved in the pathogenesis of SPE. Our findings may Table 3. Bacteria species identified in the discovery cohort and validation cohort. a Data are presented as mean log10 (abundance [%] ± standard deviation). b Data are displayed as mean −∆∆Ct ± standard deviation. Abbreviations: OR = odds ratio; CI = confidence interval. **P < 0.001, *P < 0.05.

Materials and methods
Design, participants and study variables. This single-centre, prospective cohort study was conducted at the Perinatal Care Clinic and Delivery Room of the Chang Gung Memorial Hospital (Linkou, Taiwan) between November 2017 and June 2019. Women with SPE were recruited during the third trimester of pregnancy after 28 weeks of gestation. Clinical management of SPE followed the American College of Obstetricians and Gynecologists guidelines 2 and consisted of MgSO 4 administration to prevent seizure attacks, steroid use to improve foetal lung maturity (when the gestational age was < 34 weeks), anti-hypertensive drugs to reduce blood pressure and electronic foetal heartbeat tracing. Prenatal antibiotics were not used. The control group consisted of women with normotensive term pregnancies who were scheduled for caesarean section (because of a history of previous caesarean section or foetal malpresentation). Preterm delivery was defined as delivery before 37 weeks of gestation. Cases with a gestational age of less than 28 weeks, smoking or chronic illnesses (e.g., asthma, systemic lupus erythematosus or rheumatoid arthritis) were excluded. The general characteristics of mothers and www.nature.com/scientificreports/ children were collected from clinical and birth records. All participants were of Taiwanese descent. According to the criteria of the National Health Bureau of Taiwan, BMI was categorized as follows: normal (< 24 kg/m 2 ), overweight (24-27 kg/m 2 ) and obese (> 27 kg/m 2 ). Newborn birth weight was classified into the following categories: small for gestational age (SGA; birth weight below the 10th percentile), appropriate for gestational age (AGA; birth weight from 10 to 90th percentile) and large for gestational age (LGA; birth weight above the 90th percentile) 1 . The discovery cohort consisted of 30 women with SPE (eight with concomitant gestational diabetes) and 30 control women. We sought confirmation of our findings in an expanded cohort consisting of 58 women with SPE (20 with concomitant gestational diabetes) and 55 controls. The study followed the tenets of the Declaration of Helsinki and was granted ethical approval by the Chang Gung Memorial Hospital Institutional Review Board (approval number: 201701371A3). Informed consent was obtained from all participants.   Bioinformatics analysis of amplicon library sequences. Sequencing reads were initially de-multiplexed using the MiSeq Reporter software v2.6 (Illumina) based on sample barcodes. The Usearch tool v11 (https ://drive 5.com/) was used to perform raw reads. In brief, pair reads were joined, and quality filtering for expected errors was implemented (maximum threshold: 1.0). Samples with < 60,000 merged reads were excluded from the analysis. In each dataset, the Cutadapt v1.14 package 34 was used to remove forward and reverse sequencing primers from the merged reads. Merged sequences < 400 bp in length were not included in the analysis. Denoising of effective reads from de-replication reads was conducted using the UNOISE tool 35 with the goal of identifying all correct biological sequences in the zero-radius operational taxonomic unit, i.e., amplicon sequence variants (ASVs). The final taxonomic assignment was performed using the SINTAX classifier 36 . Species richness and diversity of the vaginal microbiota were determined from the number of bacterial species assigned by the ASVs. Specifically, richness was estimated with the Observed ASV and Chao1 indices, whereas diversity was determined by calculating the Shannon index and the Gini-Simpson index. Finally, LEfSe 20 was used to identify the taxonomy of vaginal bacteria that were most likely to distinguish between cases of SPE and controls. Data analysis. The general characteristics and the biochemical data of the study participants are presented using descriptive statistics and compared with non-parametric tests. Univariate and multivariable analyses of variables associated with SPE were conducted using the SPSS statistical software (version 22.0; IBM, Armonk, NY, USA). The ' ASVs relative abundances' were log-transformed to improve normality 37 . Variables independently associated with SPE were used to devise a model to discriminate between SPE and normotensive pregnancies. The sensitivity, specificity, accuracy and AUC of predicted probability values for the panel were calculated using the MATLAB package (2015a) with the integrated 'perfcurve' function. Other analyses were conducted in the R environment (https ://www.r-proje ct.org/). We performed a logistical regression adjusted for BMI on logtransformed data to identify differences in taxonomic distributions. www.nature.com/scientificreports/ Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creat iveco mmons .org/licen ses/by/4.0/.