Gut Dysbacteriosis Is Associated with an Imbalanced Cytokines Network in Women with Unexplained Miscarriage

Background: A dysregulation of cytokine networks has been suggested to be involved in the pathogenesis of unexplained pregnancy loss. Gut microbiota affects host immune response and induces an imbalance in cytokine levels. However, how gut microbial dysbiosis affects the disturbance of cellular immune function in spontaneous abortion (SA) patients remains inconclusive. Results: Serum proinflammatory cytokine levels were significantly increased in SA patients. Microbial diversity in patients with SA was reduced compared with that in the controls, as were the relative abundances of Prevotella_1 , Prevotellaceae_UCG_003 , and Selenomonas_1 were significantly reduced in the cases. Fecal metabolomics profiles were altered in SA patients, and the correlation analyses indicated that some microbe-associated metabolites, which were significantly enriched in bile secretion and histidine metabolism pathways, were positively associated with changes in levels of IL-17A, IL-17F, TNF-α, and IFN-γ in the SA group. Moreover, the microbial-associated metabolites, imidazolepropionic acid and 1, 4-Methylimidazoleacetic acid, are associated with recurrent pregnancy loss. Conclusions: Our study highlights the network among gut microbiome, fecal metabolites and Th1/Th17-mediated immune response in miscarriage patients and identifies potential novel biomarkers and an intervention target for recurrent miscarriage. IFN-γ: Interferon-γ; SCFAs: ROC: Receiver operating characteristic; LPS: Lipopolysaccharides; HMP: microbiome VIP: projection; HCA: clustering


Background
Spontaneous abortion (SA) is the most common complication during pregnancy and affects approximately 15% of clinically recognized pregnancies [1]. After ruling out chromosomal, endocrine, infectious, anatomic, and autoimmune factors, more than 50% of SA cases remain unexplained [2]. In recent years, immune cell dysfunction has been shown to be a risk factor for the pathogenesis of unexplained pregnancy loss [3]. This dysfunction may involve imbalances in cytokines, growth factors and immunosuppressive factors at the maternal-fetal interface [4]. The role of cytokines in pregnancy loss is an emerging research field, which contributes to the understanding of many SA without obvious etiologies.
Among the leukocytes that populate the maternal-fetal interface, CD4 + helper T cells (Th cells) can be activated and then differentiate into Th1, Th2 and Th17 cells and produce corresponding types of cytokines. It has been hypothesized that during pregnancy there is a subtle immunological shift to the Th2-type cytokine responses that would suppress the potential harmful effects of the cell-mediated (Th1-type) immune system [5]. Thus, pregnancy has been labeled a 'Th2 phenomenon'. Blois et al. showed that Th1 cytokines led to pregnancy loss in mice [6], and Chaouat et al. demonstrated that this loss can be prevented by Th2 cytokines [7]. Inflammatory processes alter the balance of Th1 and Th2 cytokines and lead to a shift toward Th1 predominance. This abnormal shift of Th1/Th2 cytokines during early pregnancy initiates and intensifies the cascade of proinflammatory cytokine production involved in SA. However, this concept cannot be applied to the whole steps of pregnancy, such as implantation and the preparation to parturition. In addition to the classical Th1 and Th2 cells, several novel effector T cell subsets have been recently identified, including Th17 cells. Wang et al. found an accumulation of Th17 cells in the peripheral blood of women with unexplained recurrent pregnancy loss (RPL) as compared to normal pregnant women [8]. These Th17 cells, which secrete interleukin-17A (IL-17A), IL-17F, IL-22, and IL-26, are thought to play a role in autoimmune diseases, allograft rejection, and inflammatory immune responses. It is now evident that Th17, Th1 and Th2 immunity, and regulatory T cell (Tregs) mediated immune regulation are essential for successful embryo implantation and establishment of pregnancy [9,10].
The gut microbiota plays a crucial role in shaping and modulating the immune system and immune responses. Disruption of the commensal microbiota may alter the gut homeostatic balance and lead to gut microbial dysbiosis, which has been increasingly recognized as one of the risk factors in the development of inflammation, as well as autoimmune and immune-mediated diseases [11][12][13]. Gut microbial dysbiosis is linked to aberrant immune responses, in large part by producing small molecules that are often accompanied by abnormal production of inflammatory cytokines, such as IL-1β, IL-6, IL-17, IL-23, and interferon-γ (IFN-γ) [14,15]. The small molecules including their metabolites and components are not only necessary for immune homeostasis, but also influence the susceptibility of the host to many immune-mediated diseases and disorders. Short-chain fatty acids (SCFAs), the bacterially-derived molecules, have been reported to participate in the modulation of cytokines 4 production and Tregs expansion [16,17]. Moreover, microbial metabolites can penetrate the epithelial barrier, allowing them to enter and accumulate in the host circulatory system where they are sensed by immune cells [18]. In pregnancy, many of the immunological and metabolic changes that occurr at the placental interface serve to inhibit rejection of the fetus. Koren et al. have reported that the gut microbial community composition and structure are profoundly altered in the third trimester (T3), and the transfer of T3 microbiota induces inflammation in germ-free recipient mice to a great extent than that in first trimester (T1) microbiota [19]. However, the contribution of gut hostmicrobial interactions in promoting inflammatory cytokines production and other immune changes at early stages of pregnancy remains to be evaluated. Further, the underlying mechanisms of the correlation between gut microbial dysbiosis and the disturbance of cellular immune function in SA patients remain inconclusive.
Given the effect of gut microbiota modulation on immune responses, we hypothesized that the imbalanced gut microbiota and their metabolism can be linked to the immune dysfunction in pregnancy loss. In the current study, we aimed to examine the network between gut microbial community composition, microbial metabolites, and proinflammatory cytokine responses by performing multi-omics analyses of subjects that had unexplained pregnancy loss versus controls that underwent elected abortion.

Characteristics of study participants
Participants who indicated a reproductive tract infection and recent antibiotic treatment on the questionnaire were removed from the microbiome study. In total, 41 SA patients and 19 controls were included for further analysis. The general characteristics of the study participants, including age, prepregnancy body mass index (BMI), education level, smoking status, alcohol-drinking status, reproductive tract infection, antibodies screen and medical history are presented in Table 1. No observable differences in these characteristics were found between the SA and control groups.

Decreased bacterial diversity in fecal microbiota associated with SA patients
Alpha diversity (α-diversity) of the samples, which reflects the local scale of the microbial species, 5 was measured by Chao1 estimators and the Shannon index. Chao 1 is an index of species richness, unrelated to abundance and evenness [20]. The Shannon index is related to not only species richness but also species evenness. Both Chao1 estimators and Shannon index were significantly decreased in the SA group relative to the control group (p < 0.001 and p < 0.01, respectively), indicating a lower richness and evenness of gut bacteria in SA patients (Fig. 1A). We then analyzed the beta diversity of the two groups. Both the unweighted and weighted Principal coordinate analysis (PCoA) plots revealed that the gut microbiota in subjects with SA clustered significantly compared to that of controls ( Fig. 1B and C). These results indicate that the diversity of gut microbiota is significantly lower with different microbiota profile, in SA patients compared with controls.

Alterations in the composition of fecal microflora associated with SA
The filtered data set contained 60 samples ( families, and 57 genera were significantly changed in the SA group compared to the control group. Bacteroidetes was the most predominant phylum, accounting for 53.3% and 51.9% of the OTUs in the SA and control groups, respectively. In addition, Firmicutes was enriched in the SA group compared to the control group, whereas Proteobacteria was enriched in the control group ( Fig. 2A). Given that an upregulated Firmicutes/Bacteroidetes ratio has been suggested as an indicator of several pathological conditions [21], the ratio was of 0.65 in controls and 0.80 in cases (p = 0.039), indicating a pathological change occurred in SA patients (Fig. 2B).
We further compared the differences in fecal microflora between the two groups. At the phylum level, Spirochaetae (p < 0.001), Fibrobacteres (p < 0.001), and Tenericutes (p < 0.001) were significantly more abundant in the control group than in the SA group (Fig. 2C). Fifty-seven genera of bacteria changed in abundance in SA patients (Supplementary Table S1). Specifically, the relative abundance of 55 genera of bacteria, including Prevotella_1, Prevotellaceae_UCG_003, Roseburia, and Selenomonas_1, were significantly reduced in the SA group, and Helicobacter and Lachnospiraceae_UCG_001 were markedly increased (Fig. 2D). Considering that this discriminant analysis did not distinguish the predominant taxon, Linear discriminant analysis coupled with effect size measurements (LEfSe) was used to generate a cladogram to identify the specific bacteria associated with SA (Fig. 3A). It was shown that several opportunistic pathogens including Prevotellaceae_NK3B31_group, Bacteroidales_S24_7_group, and Eubacterium ruminantium_group were all significantly overrepresented (all LDA scores (log10) > 3.0) in the feces of SA patients, whereas Prevotellaceae, Prevotella_1, and Gammaproteobacteria were the most abundant microbiota in the control group (LDA scores (log10) > 4.0) (Fig. 3B). These results indicate that the alterations in the composition of fecal microflora were associated with SA.

Clustering and multivariate analyses reveal distinct metabolites in the SA and control groups
The hierarchical clustering analysis (HCA) of the metabolites that differed between the SA and control groups revealed four large clusters: (i) glycerophosphplipids and prenol lipids, which showed higher abundances in the control group than in the SA group, (ii) steroids and steroid derivatives, (iii) amino acids and derivatives, and (iv) alkaloids, drugs and other metabolites, which showed higher abundances in the SA group than in the control group (Fig. 5A). Variable importance in the projection (VIP) values, which were obtained by OPLS-DA analysis, indicate the importance of metabolites for interpreting the differences. The presence of several metabolites, such as hyocholic acid, methyl dihydrophaseate, cholanoic acid, 3-keto petromyzonol, hydeoxycholic acid, oic acid, oxocholanoic acid, THA, isolithocholoic acid, and chenodeoxycholic acid sulfate were able to differentiate SA patients from control subjects (Fig. 5B). Specifically, the abundances of hyocholic acid, methyl dihydrophaseate, cholanoic acid, oic acid, oxocholanoic acid, and chenodeoxycholic acid sulfate were significantly higher in SA patients than in controls (Fig. 5C). Taken together, our data clearly demonstrated that SA patients with a unique fecal metabolome, suggesting that there are gut microbiota profiles and metabolites that are associated with SA.

Correlation analysis of fecal microbiota, proinflammatory cytokines, and metabolites
Multiplex analysis revealed markedly increased serum levels of IL-2, IL-17A, IL-17F, tumour necrosis factor-α (TNF-α), and IFN-γ in SA patients as compared to control patients (Fig. 6A). Pearson analysis indicates that the Chao1 index was negatively associated with the changes in IL-17A, and IFN-γ, and the Shannon index was negatively associated with the changes in IL-17A (Fig. 6B). Moreover, the Bacteroides abundances were positively associated with the changes in IL-2; the Helicobacter abundances were positively associated with the changes in IFN-γ; the Prevotella_1 and Prevotellaceae_UCG_003 abundances were negatively associated with the changes in IL-17A and IFNγ, and Selenomonas_1 was negatively associated with the changes in IL-17A, TNF-α, and IFN-γ ( Fig.   6C). As shown in the association network (Fig. 6D, and Table 2), the Bacteroides abundances were positively associated with the changes in THA, lucidenic acid J, and ergothioneine; both the Prevotella_1 and Prevotellaceae_UCG_003 abundances were positively associated with the changes in 7-Hydroxy-3-oxocholanoic acid, and cortisone, and negatively correlated with those of 1,4-Methylimidazoleacetic acid, and imidazolepropionic acid; the Selenomonas_1 was positively associated with the changes in 7-Hydroxy-3-oxocholanoic acid, 16,16-dimethyl-6-keto Prostaglandin E1, and cortisone, and negatively associated with the changes in 1,4-Methylimidazoleacetic acid, imidazolepropionic acid, adrenic acid, and chenodeoxycholic acid sulfate.
In addition, we identified some microbe-associated metabolites which were significantly enriched in the bile secretion, histidine metabolism, and arachidonic acid metabolism pathways in SA group, suggesting a correlation with an imbalance in gut microflora or increased serum cytokines. To investigate whether patients with certain characteristics of gut microbiome and their metabolites are more susceptible to repeated pregnancy loss or infertility, we conducted a follow up survey. Our 9 follow-up results demonstrated that 12 participants exhibited with a RPL or unsuccessful pregnancy.
The ROC analysis also showed higher AUCs for imidazolepropionic acid (0.814; Fig. 6I) and 1, 4-Methylimidazoleacetic acid (0.813; Fig. 6J) for RPL. Thus, these results reveal a link between the distinct metabolites (e.g., imidazolepropionic acid and 1, 4-Methylimidazoleacetic acid) and Th17 immunity in SA patients. Furthermore, our data indicate potential roles of these metabolites as 1) biomarkers to identify women potential risk for recurrent miscarriage and 2) potential targets for prophylactics and intervention.

Discussion
Our study found that the diversity and composition of gut microbiota and metabolomic profiles were significantly altered in SA patients. Further analyses revealed that these alterations in gut microbiota were related to the increased Th1-and Th17-related cytokines. The metabolomics profiles revealed potential links between gut microbiota and the changes in cytokines and were associated with recurrent miscarriage or delayed pregnancy in the cohort of SA patients. To the best of our knowledge, this is the first study to investigate the association of gut microbiota, fecal metabolomic profiles and proinflammatory cytokines in SA patients (Fig. 7).
The Th1 response, especially IL-2, TNF-α, and IFN-γ, is harmful to the survival of the conceptus [23,24]. A significantly higher level of the Th1 cytokine IFN-γ was present in women with RPL compared with to these with normal pregnancies [25]. TNF-α has been reported to inhibit trophoblast invasion, and increased levels of TNF-α have been reported in women with SA [26,27]. Moreover, Liu et al.
found that Th17 cells were significantly increased in RPL women compared with women with normal healthy pregnancies [28]. In mice, abnormal elevation of IL-17 at the maternal-fetal interface led to a miscarriage, while administration of an anti-IL-17 antibody prevented unexplained RPL [29]. The current study also found that serum levels of IL-2, IL-17A, IL-17F, TNF-α, and IFN-γ were markedly increased in SA patients compared to the controls. These results suggest that women with SA have a propensity for proinflammation via Th1-and Th17-mediated immunity.
The gut microbiota of a host is a crucial factor for shaping and modulating the immune responses [30][31][32]. Gut microbial dysbiosis is a risk factor in the development of inflammation [33,34]. Reduction of gut microbial diversity has been linked to an increased risk of gastrointestinal diseases and proinflammatory characteristics [35,36], and low gut bacterial richness is a common hallmark of chronic disease [37]. During pregnancy, the gut microbial within-subject (α) diversity is similar to that of the non-pregnant state at T1, but reduced at T3, after comparing to the 16S rRNA gene sequence data from the Human Microbiome Project (HMP) [19]. This shift from T1 to T3 includes an increase in the levels of the proinflammatory cytokines IL-2, IL-6, IFN-γ, and TNF-α [19,38]. Here we report a negative association between systemic Th1-and Th17-associated cytokines and reduced gut microbial diversity during T1, indicating a correlation between gut microbial diversity and the increased proinflammatory cytokines in this cohort of SA patients. One previous study reported that the increased ratio of Firmicutes to Bacteroidetes is related to chronic inflammation [39]. In our study, we observed both the reduced gut microbiome and the increased ratio of Firmicutes to Bacteroidetes in the SA group during T1, indicating that the proinflammatory effects of the microbiome in patients with SA are likely caused by holistic dysbiosis, rather than by a specific pathogen.
In addition to the decreased diversity, we also found that Prevotella_1, Prevotellaceae_UCG_003, and  [45]. Although no significant decrease of Bifidobacterium abundance was found in SA patients in our study, a positive correlation of Prevotella_1 and Bifidobacterium (Supplementary Figure S1) might indicate a regulatory role for Prevotella_1 on Th17 cytokines. Moreover, we found that both Prevotella_1 and Prevotellaceae_UCG_003 correlated highly with enriched imidazolepropionic acid and 1, 4-Methylimidazoleacetic acid, both of which had higher abundance in the SA group. And these two metabolites were positively associated with the level of IL-17A and IL-17F. These data suggested that the gut microbiome has biologically relevant effects on the modulation of Th17 cytokine production in women with early miscarriage.
Modulation of host defense by the microbiota may be exerted mainly through the release of intermediary common mediators (such as metabolites) rather than direct interaction between specific microorganisms and immune cells. An important role for metabolites in microbiome-cytokine interactions is supported by the fact that a large proportion of the metabolites in the blood originate from the gut [46,47], and our findings of the strong impact of microbial metabolic processes on cytokines production. We observed that several microbiota-associated metabolites that were enriched in the bile secretion pathway, were also significantly altered in the feces of SA patients. It is welldocumented that bile acid plays a potent regulation role on intestinal immunity cells [48]. A recently published paper in Nature demonstrated that bile acid metabolites can control Th17 and Treg cell differentiation [49]. In the present study, we found negative associations between hyodeoxycholic acid, isolithocholic acid, and TXB2, which were significantly decreased in the SA group, with serum cytokines, IL-17, TNF-α, and IFN-γ, and positive associations for chenodeoxycholic acid sulfate, L-Carnitine, acetylcholine, and cholic acid metabolites. Our results suggest that the bile acid metabolism of the gut microbiota strongly influences Th17-associated cytokine production. Moreover, we found that the fecal concentrations of 1, 4-Methylimidazoleacetic acid, and imidazolepropionic acid that were enriched in histidine metabolism were also significantly increased in the SA group and aligned well with the increased IL-17, TNF-α, and IFN-γ in this group. Microbially produced imidazole propionate, which has been shown to have systemic effects and is present at higher concentrations in subjects with type 2 diabetes, impairs insulin signaling through mammalian target of rapamycin1 (mTORC1) [50]. A high level of imidazole propionate was reported to be involved in immune activation and low-grade inflammation [51,52]. Intriguingly, ROC analysis indicated that the two imidazole propionic acids we examined are associated with SA and have a positive predictive value for RPL.
Collectively, these results indicate that the gut microbiome has biologically relevant effects on the modulation of Th17 cytokine production through their metabolites in miscarriage. Our results also identified two imidazole propionic acids as not only potential novel biomarkers or risk factors for recurrent miscarriage but targets for prophylactics and intervention.
The strength of this study is that we applied 16S rRNA amplicon profiling, bacterial metabolomics and host serum cytokines networks, which allowed us to gain more information about host-gut microbiota metabolic interactions in response to the imbalanced cytokines network in unexplained miscarriage.
One limitation of this study is that we did not obtain completed questionnaires about the dietary habits of the participants, and therefore could not determine if the diet was a factor in the gut bacterial dysbiosis. In addition, we cannot rule out other etiologies of SA patients such as PCOS and IR, which might be indirectly associated with the proinflammatory cytokines. Another limitation is that we cannot determine the causal relationship between gut microbiota and fecal metabolomics and the associated changes in cytokines. Our follow up results support the causal relationship although the sample size is small.

Conclusions
Th1/Th17-mediated proinflammatory state in SA patients with unknown etiology. This study provided insights into the potential roles of gut microbiome in the pathogenesis of some unexplained miscarriages and the potential underlying mechanisms. Our study highlighted the network among gut microbiota, fecal metabolites and immune response in SA patients and identified potential novel biomarkers and an intervention target for recurrent miscarriage. Further study is warrant to determine the casual relationship between gut microbiota and inflammation induced miscarriage and whether the microbiota associated metabolites are the underlying mechanism of the cytokine network imbalance through larger prospective cohort studies and animal studies.

Participant enrollment
13 Eligible cases were women (i) who were less than 35 years of age with a pre-pregnancy BMI of 18.5-23.9 kg/m 2 (normal reference value standards of BMI for Chinese); (ii) who had no successful pregnancies (their abortions occurred before 20 weeks of gestation); and (iii) whose current male partner had normal semen testing by computer-assisted analysis. Age and gestational week-matched controls were women with normal early pregnancy who elected to have an abortion. All controls had at least one successful pregnancy and no history of SA, preterm labor, or pre-eclampsia. Subjects with a verifiable cause of miscarriage were excluded. Exclusion criteria were: (i) abnormality of the uterus Informed consent was obtained from all participants.

Stool sampling, DNA extraction and 16S rRNA sequencing
Participants collected stool samples at home using sterilized containers the day before or the day of curettage. Participants were asked to store their stool sample in the refrigerator right after collection and before delivering the samples to the hospital where they were aliquoted and stored at −80 °C until DNA extraction. Genomic DNA was extracted from stool samples using the QIAamp DNA Stool Mini Kit (QIAGEN, Dusseldorf, Germany) according to the manufacturer's instructions. The quality of genome DNA was evaluated with 1% agarose gel. The V3-V4 variable regions of the 16S rRNA gene were amplified using bacterium-specific primers 343F (5'-TACGGRAGGCAGCAG-3') and 798R (5'-AGGGTATCTAATCCT-3') [53]. The PCR products were purified, and the concentrations were adjusted for sequencing on an Illumina MiSeq PE300 (OEbiotech Co., Ltd., Shanghai, China).

Bioinformatic analysis of the fecal microbiome
The raw data obtained from high-throughput sequencing were stored in FASTQ format. Low-quality sequences that had an average quality score below 20 were cut off using Trimmomatic software [54].
After trimming, paired-end reads were assembled with FLASH software [55]. The parameters of assembly were as follows: 10 bp of minimal overlapping, 200 bp of maximum overlapping and a 20% maximum mismatch rate. Further de-noising including the removing of ambiguous base (N), sequences below 200 bp, and chimeras was performed with QIIME software (version 1.8.0) [56]. The obtained clean reads were then subjected to primer-sequence removal and clustering to generate OTUs with a 97% similarity cutoff using Vsearch software [57]. The most abundant one was selected as the representative read of each OTU. All representative reads were annotated and blasted against the Silva database (v. 123) or Greengens (16s rDNA) using RDP classifier (at a confidence threshold of 70%). The α-diversity indices evaluating gut microbial community richness and evenness were performed by mother [58]. PCoA based on the Binary-Jaccard distance was performed to compare the global microbiota composition (β-diversity) between the SA and control groups, and statistically significant differences between the two groups were calculated by Adonis analyses. LEfSe was applied to identify the bacterial species that differed between samples.

Fecal metabolic profiling and data analysis
Ten SA patients and 10 matched control subjects were randomly chosen for the metabolomics study.
Accurately weighed 60 mg fecal sample was transferred to a 1.5-mL Eppendorf tube, and then 20 μL each of L-2-chlorophenylalanine solution (0.3 mg/mL) and Lyso PC17:0 (0.01 mg/mL) were added as an internal standard. The QC samples (n = 4) were pooled ones in which aliquots of each sample (n = 20) were mixed together. Following ultrasonication and centrifugation, the supernatants from each tube were collected using crystal syringes, filtered through 0.22 μm microfilters, transferred into a LC/MS glass vial, and analyzed by an ACQUITY UHPLC system (Waters Corporation, Milford, USA) coupled with an AB SCIEX Triple TOF 6600 System (AB SCIEX, Framingham, MA) as described previously [59].
The acquired raw data were analyzed by the progenesis QI software (Waters Corporation, Milford, USA) using the following parameters: precusor tolerance at 5 ppm, fragment tolerance at 10 ppm, and product ion threshold at 5%. Metabolites were identified based on public databases such as the Human Metabolome Database (HMDB), Lipidmaps (v2.3), and METLIN. PCA and OPLS-DA models were carried out to visualize the metabolic alterations among experimental groups. A permutation test (n = 200) was performed to validate the model and avoid over fitting. HCA was applied on Pearson distances using PermutMatrix [60]. Differential metabolites contributing to the separation were identified using VIP value, fold change values and the corresponding p values. In general, metabolites with VIP > 1 were considered as relevant for interpreting the discrimination, a fold change value ≥1.5 or ≤ 0.667 was the cutoff for up or down-regulation in concentration, and p value < 0.05 was believed to be a significant difference.

Cytokines quantification by flow cytometry
A bead-based multiplex panel assay (Biolegend, San Diego, CA, USA) was used for simultaneous quantification of 13 human cytokines including IL-2, -4, -5, -6, -9, -10, -13, -17A, -17F, -21, -22, IFN-γ, and TNF-α in the corresponding 50 vs 30 subjects. Briefly, the peripheral blood was collected and centrifuged for 20 min at 1,000 × g to collect serum. Serum samples were diluted two fold with Assay Buffer before being tested. Next, 25 μL of mixed beads, detection antibodies, and streptavidinphycoerythrin were added stepwise to each well. The plate was placed on a plate shaker for shaking at 500 rpm for 30 min at room temperature. After washing the beads three times with 200 μL of 1×Wash Buffer, the beads were resuspended in 150 μL of 1×Wash Buffer, placed on a plate shaker for 1 min and read on a flow cytometer.

Statistical analysis
Data are presented as the mean ± standard deviation (SD). Significant differences in clinical characteristics were evaluated with Pearson's Chi-square test or Fisher's exact test. A one-way ANOVA or Kruskal-Wallis was used for the significant difference analysis in genus relative abundance.
Student's t test was performed to analyze the differential in metabolites abundance between the SA and control groups. Spearman's rank test or Pearson's correlation test was used to analyze the correlation between changes in genus relative abundance, fecal metabolites, and the levels of serum 16 cytokines. P < 0.05 was considered statistically significant.a

Availability of data and materials
Some data used in the study appear in the supplementary materials. Some data used during the study are available from the corresponding author by request.

Acknowledgements
We thank the obstetrician in the Department of Obstetrics and Gynecology, Xinhua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine for helpful sampling assistance.    IL, interleukin; TNF-α, tumour necrosis factor-α; IFN-γ, interferon-γ. The asterisk represents the statistical significance between metabolites and cytokines. * p < 0.05, ** p < 0.01.      Figure S1.pdf