Flagellin in blood and Bifidobacterium pseudocatenulatum in gut are associated with live birth upon IVF-frozen embryo transfer.

Conceptualization, C.W.W.; methodology, K.T.C.C., J.H.C.Y., C.W.W., and R.H.W.L.; validation, K.T.C.C., J.H.C.Y., C.W.W., and R.H.W.L.; formal analysis, K.T.C.C., J.H.C.Y., C.W.W., and R.H.W.L.; investigation, L.C.Y., W.Y.-L.W., J.C., S.W.M.C., and K.W.C. resources, K.W.C, R.H.W.L., and C.W.W.; data curation, R.H.W.L. and C.W.W.; writing—original draft, R.H.W.L. and C.W.W.; writing—review and editing, R.H.W.L. and C.W.W.; visualization, C.W.W.; supervision, R.H.W.L. and C.W.W.; funding acquisition, C.W.W., K.T.C.C., and J.H.C.Y. contributed equally to this study.

The demographic and clinical characteristics of the participants with and without attainment of live birth in the index FET cycle are shown in Table 1 . There was no significant difference in all the listed parameters ( Table 1 ). Various cytokines that commonly indicate systemic inflammation and bacterial components in serum samples were measured, and no significant difference in the serum levels of C-reactive protein (CRP), interferon-γ (IFN-γ), and monocyte attractant protein-1 (MCP1) between women with and without live birth after FET was observed, except serum amyloid A (SAA) level ( Table 2 , Figure S1 ). The group with live birth showed a higher serum SAA level ( Table 2 , Figure S1 ). By contrast, serum flagellin concentration was significantly lower in the group with live birth than those without, whereas no difference in serum LPS concentration was observed ( Table 2 , Figure S1 A). To examine whether the increase in serum flagellin level was owing to alterations in the specific antibodies, we measured anti-flagellin immunoglobin G (IgG) and immunoglobin A (IgA) antibodies in the serum. Regardless of the source of flagellins used as the antigens for antibody measurement, there was no difference in either anti-flagellin IgG or IgA between the two groups ( Table 2 , Figure S1 A). On the other hand, the analysis using a univariate binary logistic regression model demonstrated that although both serum flagellin and SAA levels were able to predict pregnancy, only serum flagellin in the prediction of live birth reached statistical significance ( Table 3 ). The odds ratios (OR) indicated that a 10-fold increase in serum flagellin level was associated with an 82.9% decrease in the odds of a woman attaining live birth upon FET, while a 10-fold increase in serum amyloid A level was 3.9-fold more likely to have a successful pregnancy ( Table 3 , Figure S1 B). Receiver-operating characteristic curve analysis for prediction of live birth by serum flagellin concentration revealed an area under the curve of 0.729 (95% CI 0.605–0.851, p  < 0.001) ( Figure S2 ). Furthermore, the best cut-off of serum flagellin concentration at 0.6 ng/mL based on the Youden’s index gave a sensitivity of 71.9% and a specificity of 68.6% in predicting successful live birth. Table 1 Comparison of demographic characteristics and clinical parameters between women with and without live birth after the index frozen embryo transfer cycle Parameter Women with live birth ( N  = 32) Women without live birth ( N  = 35) p value a Age of women (years) 36 (34–38) 36 (35–38) 0.594 Body mass index (kg/m 2 ) 22.9 (20.2–24.4) 23.2 (21.6–26.1) 0.196 Smoking status 0.115  Non-smoker 32 (100.0%) 31 (88.6%)  Ex-smoker 0 (0.0%) 4 (11.4%) Type of infertility 0.527  No infertility 1 (3.1%) 0 (0.0%)  Primary 21 (65.6%) 21 (60.0%)  Secondary 10 (31.3%) 14 (40.0%) Duration of infertility (years) 3 (3–5) 4 (3–5) 0.557 Cause of infertility 0.807  Tubal factor 6 (18.8%) 5 (14.3%)  Endometriosis 2 (6.3%) 1 (2.9%)  Male factor 9 (28.1%) 8 (22.9%)  Unexplained 5 (15.6%) 9 (25.7%)  Mixed causes 10 (31.3%) 12 (34.3%) Parity 0.755  0 27 (84.4%) 28 (80.0%)  1 5 (15.6%) 7 (20.0%) Frozen embryo transfer protocol 0.551  Natural cycle 24 (75.0%) 28 (80.0%)  Letrozole cycle 0 (0.0%) 1 (2.9%)  Hormone replacement cycle 8 (25.0%) 6 (17.1%) Pre-implantation genetic testing 0.246  No 27 (84.4%) 33 (94.3%)  Yes 5 (15.6%) 2 (5.7%) Stage of embryo transferred 0.159  Cleavage stage 5 (15.6%) 11 (31.4%)  Blastocyst 27 (84.4%) 24 (68.6%) Data presented as number (percentage) or median (25th – 75th percentile). a Mann-Whitney U test for continuous variables, Fisher’s exact test for categorical variables. Table 2 Comparison of various bacterial constituent and cytokines levels in serum between women with and without live birth after the index frozen embryo transfer cycle Parameter Women with live birth ( N  = 32) Women without live birth ( N  = 35) p value a Flagellin (ng/mL) 0.40 (0.16–0.605) 0.71 (0.45–1.02) <0.001 b Lipopolysaccharides (LPS, EU/ml) 10.46 (7.79–13.95) 9.12 (7.13–11.45) 0.120 Interleukin-1β (IL-1β, pg/ml) ND c ND c N/A Interferon-γ (IFNγ, pg/ml) 24.57 (0–78.71) 28.06 (0–77.80) 0.682 Monocyte chemoattractant protein-1 (MCP1, pg/ml) 121.61 (90.48–154.23) 131.99 (97.89–188.82) 0.283 C-reaction protein (CRP, μg/ml) 0.72 (0.23–2.03) 1.21 (0.31–2.24) 0.441 Serum amyloid A (SAA, μg/ml) 1.77 (0.87–2.70) 0.96 (0.45–1.94) 0.039 b Anti-flagellin IgA ( S . typhimurium) (R.U.) 133.68 (86.26–192.66) 135.36 (101.85–201.71) 0.743 Anti-flagellin IgG ( S . typhimurium) (R.U.) 144.52 (104.30–234.60) 124.82 (95.74–207.94) 0.598 Anti-flagellin IgA ( B . subtilis ) (R.U.) 89.88 (77.68–121.25) 90.2 (67.07–133.92) 0.715 Anti-flagellin IgG ( B . subtilis ) (R.U.) 219.00 (136.45–380.18) 211.19 (162.11–361.77) 0.762 Data presented as median (25th – 75th percentile), R.U., Relative unit. See also Figure S1 . a Mann-Whitney U test. b Statistically significant ( p  < 0.05). c Detectable only in 6 women with live birth and 6 women without live birth. Table 3 Odd ratios of serum flagellin and serum amyloid A levels to predict different outcomes of FET Predictor a Outcomes Odd ratio Confidence interval (95%) p value Flagellin (ng/mL) Pregnancy b 0.290 0.076–0.892 0.045 c Live birth 0.171 0.040–0.553 0.007 c Serum amyloid A (SAA) (μg/mL) Pregnancy 3.900 1.171–14.939 0.034 c Live birth 3.174 0.976–11.910 0.067 See also Figures S1 and S2 . a Log 10 -transformed. b Detection on the initial pregnancy test. c Statistically significant. Comparison of demographic characteristics and clinical parameters between women with and without live birth after the index frozen embryo transfer cycle Data presented as number (percentage) or median (25th – 75th percentile). Mann-Whitney U test for continuous variables, Fisher’s exact test for categorical variables. Comparison of various bacterial constituent and cytokines levels in serum between women with and without live birth after the index frozen embryo transfer cycle Data presented as median (25th – 75th percentile), R.U., Relative unit. See also Figure S1 . Mann-Whitney U test. Statistically significant ( p  < 0.05). Detectable only in 6 women with live birth and 6 women without live birth. Odd ratios of serum flagellin and serum amyloid A levels to predict different outcomes of FET See also Figures S1 and S2 . Log 10 -transformed. Detection on the initial pregnancy test. Statistically significant. The differences in gut microbiota between the participants with and without live birth after FET were then examined. The two groups displayed the same degree of diversity in gut bacteria, as indicated by the Shannon index ( Figure 1 A). When we compared the composition of the gut microbiota as intra-sample diversity (β-diversity), there was a significantly distinctive pattern between subjects with and without live birth (PERMANOVA, p  < 0.05, Figure 1 A). The linear discriminant analysis effect size (LEfSe) revealed that Roseburia inulinivorans and Bifidobacterium pseudocatenulatum were the species with the most significant differences in abundance between the two groups, where R. inulinivorans was enriched in the group without live birth and B. pseudocatenulatum in the group with live birth ( Figure 1 B). By applying the same analysis as for flagellin, we found that the best cut-off of B. pseudocatenulatum abundance at 0.371% gave a sensitivity of 50% and a specificity of 87.1% in predicting successful live birth, and the best cut-off of R. inulinivorans abundance at 0.597% gave a sensitivity of 45.2% and 93.3% in predicting the absence of live birth attainment. Figure 1 The importance of serum flagellin levels and gut bacteria in determination of live birth outcome of FET (A) β and α diversities of the gut microbiota from subjects grouped as with (yes) or without (no) live birth evaluated by non-metric multidimensional scaling (NMDS) and Shannon index, respectively. (B) Mean relative abundances of gut bacteria with a significant difference in the linear discriminant analysis effect size between the two groups. (C) Ranking of the importance of serum flagellins and the abundance of different gut bacteria in determining the live birth outcome of FET using random forest classification. The prevalence of these bacteria in the cohort is shown in the right panel. (D) Receiver operating characteristic curves and the values of the corresponding area under curve (AUC) yielded by different combinations of classifiers. Data are represented as mean ± s.e.m. ( N  = 30 for with live birth and N  = 31 for with live birth group.). p < 0.05 statisically significant. The importance of serum flagellin levels and gut bacteria in determination of live birth outcome of FET (A) β and α diversities of the gut microbiota from subjects grouped as with (yes) or without (no) live birth evaluated by non-metric multidimensional scaling (NMDS) and Shannon index, respectively. (B) Mean relative abundances of gut bacteria with a significant difference in the linear discriminant analysis effect size between the two groups. (C) Ranking of the importance of serum flagellins and the abundance of different gut bacteria in determining the live birth outcome of FET using random forest classification. The prevalence of these bacteria in the cohort is shown in the right panel. (D) Receiver operating characteristic curves and the values of the corresponding area under curve (AUC) yielded by different combinations of classifiers. Data are represented as mean ± s.e.m. ( N  = 30 for with live birth and N  = 31 for with live birth group.). p < 0.05 statisically significant. We then evaluated the importance of these species and serum flagellin level in predicting successful live birth with random forest classification and repeated cross-validation due to the small sample size ( Figure 1 C). The abundance of R. inulinivorans and B. pseudocatenulatum , along with serum flagellin levels, consistently ranked as the most important factors to predict the outcome of live birth ( Figure 1 C). These two bacterial species were present in more than 70% of subjects in this cohort ( Figure 1 C). The predictive performance of the combination of R. inulinivorans abundance and serum flagellin level was better than each alone, and the addition of B. pseudocatenulatum abundance to the combination enhanced the predictive performance further ( Figure 1 D). Taken together, serum flagellin level and the abundances of R. inulinivorans and B. pseudocatenulatum are the three most important factors for predicting live birth outcomes of FET. R. inulinivorans is a flagellated species, so we evaluated whether serum flagellin level was directly associated with the abundance of flagellated microbiota in feces. The functional profile of the gut microbiota was generated from Gene Ontology (GO) functional enrichment analysis on the genes identified from the microbiome analysis. No difference in the abundances of the flagellum-related pathways was observed between the participants with and without live birth after FET ( Figure 2 A). Moreover, the abundance of flagellum-related GO terms derived from R. inulinivorans was not correlated with the serum flagellin level of the hosts ( Table S1 ). Conversely, the abundance of B. pseudocatenulatum but not R. inulinivorans was significantly and negatively correlated with serum flagellum level ( Figure 2 B). Upon dividing the cohort into quartiles according to serum flagellin level, a higher abundance of B. pseudocatenulatum was observed at the 1 st and 2 nd quartiles ( Figures 2 B and 2C). The abundance of R. inulinivorans was notably elevated in the 4 th quartile ( Figures 2 B and 2C). Taken together, the abundance of B. pseudocatenulatum is associated with the suppression of serum flagellin levels and live birth attainment after FET. Figure 2 Correlations between the abundances of various gut bacteria and serum flagellin level (A) Relative abundances of flagellum-related gene ontology (GO terms) of gut microbiota in women with (yes) and without (no) attaining live birth upon FET. (B) Scatterplot depicting spearman correlations between the abundances of R. inulinivorans and B. pseudocatenulatum , and serum flagellin level adjusted with age and BMI. (C) The percentage of subjects with (yes) and without (no) live birth and the abundance of these two bacteria among the cohorts divided by quartiles based on serum flagellin level. The cutoffs and the number of subjects with and without live birth in each quartile in the upper panel. Data are represented as mean ± s.e.m. ∗Statistically significant. Correlations between the abundances of various gut bacteria and serum flagellin level (A) Relative abundances of flagellum-related gene ontology (GO terms) of gut microbiota in women with (yes) and without (no) attaining live birth upon FET. (B) Scatterplot depicting spearman correlations between the abundances of R. inulinivorans and B. pseudocatenulatum , and serum flagellin level adjusted with age and BMI. (C) The percentage of subjects with (yes) and without (no) live birth and the abundance of these two bacteria among the cohorts divided by quartiles based on serum flagellin level. The cutoffs and the number of subjects with and without live birth in each quartile in the upper panel. Data are represented as mean ± s.e.m. ∗Statistically significant. The GO terms from the functional profiles under the category of biological processes of the gut microbiota were subjected to correlation analysis with serum flagellin level ( Table S2 ), and the GO terms contributed by bacteria in Figure 1 B were further examined ( Table S3 ). We separated these GO terms according to whether the contributing bacteria were enriched in the with or without live birth group; the common terms were excluded from further analysis ( Figure 3 A, Table S3 A). Among the 57 terms unique to the with live birth group, 18 biological processes were negatively correlated with serum flagellin level with an unadjusted p value <0.05 ( Figure 3 A, Table S3 B). By contrast, 94 terms were exclusive to the without live birth group, and only 2 of them were correlated with serum flagellin level with an unadjusted p value <0.05 ( Figure 3 A, Table S3 C). In those 18 processes, 6 processes represented biosynthetic processes, in which 3 processes showed significant correlations with serum flagellin level after adjustment for multiple comparisons, namely, the cysteine biosynthetic process via cystathionine (GO:0019343), GDP-mannose (GO:0009298), and glucan (GO:0009250) biosynthesis ( Table S3 B). The cysteine biosynthetic process via cystathionine (GO:0019343) showed the strongest correlation with serum flagellin level with a correlation coefficient of −0.475 ( Figure 3 B, Table S3 B, Figure S3 ), and B. pseudocatenulatum was the key bacteria contributing to its readout ( Table S3 B). Figure 3 Correlations between the GO terms readouts and serum flagellin level (A) Venn diagram showing the number of gene ontology (GO) terms derived from bacteria whose abundance in feces was higher in with (w/) or without (w/o) live birth groups. (B) Spearman correlations of GO terms representing the enzymes related to cystathionine metabolism and serum flagellin level adjusted with age and BMI. (C) Schematic diagram illustrating transsulfuration pathways in gut bacteria. (D) Chemical reactions related to the biosynthesis and degradation of cystathionine. ∗Statistically significant. Correlations between the GO terms readouts and serum flagellin level (A) Venn diagram showing the number of gene ontology (GO) terms derived from bacteria whose abundance in feces was higher in with (w/) or without (w/o) live birth groups. (B) Spearman correlations of GO terms representing the enzymes related to cystathionine metabolism and serum flagellin level adjusted with age and BMI. (C) Schematic diagram illustrating transsulfuration pathways in gut bacteria. (D) Chemical reactions related to the biosynthesis and degradation of cystathionine. ∗Statistically significant. Besides, B. pseudocatenulatum was also the major contributor to α-glucan (GO:0030979) and the creatine (GO:0006601) biosynthesis ( Table S3 B). Among the remaining 12 biological processes, only creatinine catabolic process (GO:0006602) and carboxylic acid metabolic process (GO:0019752) were contributed mainly by B. pseudocatenulatum . However, only ∼11% of the readout of the carboxylic acid metabolic process (GO:0019752) was derived from B. pseudocatenulatum, and this process was concurrently contributed by 76 identified bacteria in total ( Table S3 B). Next, we further examined the GO terms representing enzymes related to the three biosynthetic processes showing significant correlations with serum flagellin level, only the activities of cystathionine β-synthase (GO:0004122) under GO:0019343 and phosphomannomutase (GO:0004615) under GO:0009298 were significantly and negatively correlated with serum flagellin level ( Figure 3 B, Table S3 D, Figure S3 ). Unlike cystathionine β-synthase activity, which showed a comparable correlation coefficient as GO:0019343, the correlation of phosphomannomutase activity with serum flagellin level was weaker than that of GO:0009298 ( Table S3 D), suggesting the uniqueness of cystathionine β-synthase. Cysteine biosynthesis via cystathionine is the key process of the transsulfuration pathway ( Figure 3 C). Gut bacteria can utilize cysteine to produce hydrogen sulfide, (H 2 S) which in adequate concentration, helps to maintain gut integrity. 17 , 18 The GO terms representing other enzymes related to the transsulfuration pathway, including cystathionine β-lyase (GO:0004121) and cystathionine γ-synthase (GO:0003962), were also examined ( Figure 3 C). The cystathionine β-lyase activity (GO: 0004121) showed a moderate but insignificant correlation with serum flagellin levels ( Figure 3 B). The comparison of the products yielded by the reactions catalyzed by cystathionine β-synthase and cystathionine β-lyase showed that H 2 S was another common product beside cystathionine ( Figure 3 D). The taxonomical and functional profiles that indicated the abundance of B. pseudocatenulatum and cysteine biosynthesis via cystathionine in the gut microbiota were the most significant parameters that were negatively correlated with serum flagellin level, and, therefore, the relationship between these two was further explored. The readout of GO terms, cysteine biosynthetic process via cystathionine (GO:0019343), and cystathionine β-synthase activity (GO:0004122), were attributed to 17 bacterial species detected in this cohort, and 6 of them belonged to the genus Bifidobacterium ( Figure 4 A). Among these 17 species, only B. pseudocatenulatum , B. adolescentis , and  Escherichia coli were present in more than half of the subjects in this cohort ( Figure 4 A). The correlation coefficient between the abundance of GO:0004122 derived from B. pseudocatenulatum and serum flagellin level was −0.319 ( Table S4 ) and was strengthened to −0.462 upon the addition of those readings contributed by B. adolescentis ( Figure 4 B). The combination of readouts from all the detectable species in Bifidobacterium similarly enhanced the correlation ( Figure 4 B). Furthermore, moderate but significant negative correlations between the abundance of and readout of GO:0004122 derived from B. pseudocatenulatum and serum MCP-1 level were observed ( Figures 4 C and S4 ). Figure 4 GO terms of cystathionine β-synthase in the genus of Bifidobacterium (A) Species contributing to the readouts of gene ontology (GO) terms for cysteine biosynthetic process via cystathionine (GO:0019343) and cystathionine β-synthase activity (GO:0004122) and their prevalences in the cohort. (B) Spearman correlation between serum flagellin level and the readouts of GO:0019343 and GO:0004122 attributed by B. pseudocatenulatum alone or in combination with B. adolescentis , all Bifidobacterium or E. coli adjusted with age and BMI. (C) Spearman correlation between the serum levels of various cytokines and flagellin and the abundance and readout of GO:0004122 attributed by B. pseudocatenulatum adjusted with age and BMI. (D) The prevalences of all and the most prevalent species contributing to the readouts of GO:0019343 and GO:0004122 in this cohort. CRP, C-reactive protein; IFNγ, interferon-γ; MCP1, monocyte chemoattractant protein-1; SAA, serum amyloid A. ∗Statistically significant. GO terms of cystathionine β-synthase in the genus of Bifidobacterium (A) Species contributing to the readouts of gene ontology (GO) terms for cysteine biosynthetic process via cystathionine (GO:0019343) and cystathionine β-synthase activity (GO:0004122) and their prevalences in the cohort. (B) Spearman correlation between serum flagellin level and the readouts of GO:0019343 and GO:0004122 attributed by B. pseudocatenulatum alone or in combination with B. adolescentis , all Bifidobacterium or E. coli adjusted with age and BMI. (C) Spearman correlation between the serum levels of various cytokines and flagellin and the abundance and readout of GO:0004122 attributed by B. pseudocatenulatum adjusted with age and BMI. (D) The prevalences of all and the most prevalent species contributing to the readouts of GO:0019343 and GO:0004122 in this cohort. CRP, C-reactive protein; IFNγ, interferon-γ; MCP1, monocyte chemoattractant protein-1; SAA, serum amyloid A. ∗Statistically significant. As a comparison, the correlations of serum MCP-1 and flagellin levels with GO terms representing cysteine biosynthesis via serine (GO:0006535 and GO:0009001) and utilization of cysteine via glutathione biosynthesis (GO:0006750 and GO:0004357) derived from B. pseudocatenulatum were also evaluated, but no significant correlation was observed ( Figure S5 ). Moreover, although Escherichia. coli was a highly prevalent species in this cohort ( Figure 4 A), the contribution of E. coli to the readout of GO:0004122 was detected in less than 30% of the total participants ( Figures 4 B and 4D). Taken together, the cystathionine β-synthase activity of Bifidobacterium is associated with a decreased serum flagellin level, thus decreasing the risk of unsuccessful live birth upon FET.

Requests for further information and resources should be directed to and will be fulfilled by the lead contact, Connie W. Woo ( [email protected] ). This study did not generate new unique reagents. • The gut microbiota data have been deposited at NCBI Sequence Read Archive (SRA) BioProject database and are publicly available as of the date of publication. Accession number is listed in the key resources table . • All data reported in this paper will be shared by the lead contact upon request. • This paper does not report original code. The gut microbiota data have been deposited at NCBI Sequence Read Archive (SRA) BioProject database and are publicly available as of the date of publication. Accession number is listed in the key resources table . All data reported in this paper will be shared by the lead contact upon request. This paper does not report original code.

This observational study indicates that serum concentration of flagellin and fecal abundance of R. inulinivorans and B. pseudocatenulatum are important factors for predicting the live birth outcome of a FET cycle. The activity of cystathionine β-synthase in B. pseudocatenulatum is significantly and negatively correlated to flagellin and MCP-1 levels in the circulation. Although the causative relationship between these three factors and live birth attainment after FET requires further investigation, our findings suggest that gut microbiota may be a suitable treatment target for improving the live birth rate upon FET. An elevation of bacterial constituents in the circulation often reflects a decreased integrity of the gut, where trillions of microbes inhabit. The gut barrier is upheld by epithelial tight junctions and mucosal immunity. 10 Owing to the different natures of flagellin, a bacterial protein, and LPS, a polysaccharide, the mechanisms of their infiltration into the circulation are different. Women without live birth after FET in this study showed higher serum levels of flagellin but not LPS, which implicated a subtle change in mucosal immunity rather than systemic damage at the gut epithelium. Given the fact that the participants in this cohort were generally healthy individuals, damage to the gut epithelium was not anticipated. Flagellin as an antigen can trigger adaptive immunity, and the penetration of flagellated bacteria is usually blocked by the induced anti-flagellin antibodies. 19 However, our previous and other studies have found that certain types of flagellin derived from gut microbiota can escape host surveillance and enter the circulation. 15 , 20 On the other hand, a significantly higher SAA level was observed in women with live births after FET, and a negative correlation between serum SAA and flagellin levels was observed in this study (r = −0.235, p  = 0.066). Although it did not reach statistical significance, it is noteworthy to investigate whether SAA can bind to flagellin in circulation like it does to LPS, which has been previously reported to promote LPS clearance. 21 Roseburia species are recognized as beneficial species due to their capacity to produce butyrate. 22 , 23 Their roles as flagellated species in the gut microbiota are also suggested to be favorable characteristics in disease treatment. 24 The activation of TLR5 in intestinal dendritic cells by R. intestinalis enhances the differentiation of regulatory T (Treg) cells, resulting in immunomodulation against Crohn’s disease. 24 By contrast, in this study, we observed a higher abundance of R. inulinivorans in women who failed to attain live birth after FET ( Figure 1 B). The individuals in the 4 th quartile of serum flagellin level had the highest abundance of R. inulinivorans. Nonetheless, our data do not conclude whether the increased abundance of R. inulinivorans was a cause or consequence of elevated serum flagellin levels. Compared with R. intestinalis , R. inulinivorans is less frequently studied. A study reported several distinctive features of R. inulinivorans that differ from other Roseburia species and complement other species to maintain the gut ecosystem. 25 The detection of flagellin in circulation implicates the penetration of flagellins, which can be caused by weakened mucosal immunity. Whether elevation of R. inulinivorans is an indicator of weakened intestinal immunity or a result of the compensatory mechanism of an altered gut ecosystem in subjects without live birth attainment from FET requires further investigation. Bifidobacterium species have been widely used as probiotics. Although many diseases are associated with the reduced abundance of these species, the mechanisms by which they exert beneficial effects are largely unknown. 26 Our data suggest that the abundance of B. pseudocatenulatum in the gut is negatively correlated with serum flagellin level and higher in women attaining live birth after FET. Such a beneficial relationship appears to be strongly associated with cystathionine β-synthase activity. The chemical reactions catalyzed by cystathionine β-synthase produce not only cystathionine but also hydrogen sulfide. There are studies suggesting the role of hydrogen sulfide in preserving gut integrity. 17 , 18 Other than the cystathionine β-synthase activity, the α-glucan biosynthetic process that showed a significant negative correlation with serum flagellin level ( Table S3 B) was mainly contributed by Bifidobacterium species ( Tables S3 B and S5 ). Glucans are also suggested to be beneficial for maintaining gut symbiosis. 27 , 28 We speculate that the beneficial effect of B. pseudocatenulatum toward live birth attainment upon FET is possibly mediated by the production of hydrogen sulfide and glucans, resulting in improved gut integrity and limited system inflammation caused by the penetration of bacterial materials. The potential causative relationship is noteworthy for future studies. There are limitations of this observational study, including the inability to prove the causative nature of flagellin in preventing live birth attainment and the relatively small number of subjects. It is possible that the serum level of flagellin is an indicator of a change in immunity. Hence, further investigations with larger sample sizes and experimental approaches to delineate the role of flagellin in FET will validate whether circulating flagellin can serve as a treatment target to improve the live birth rate upon FET or is merely an indicator of systemic inflammation.

Infertility, a problem affecting around 17.5% of the adult population, is a major medical challenge globally. 1 The costs of treating infertility remain higher than the GDP per capita in many countries. 2 In vitro fertilization (IVF) is the most commonly employed assisted reproductive technology, and the global IVF market size has been growing since the birth of the first baby using this technique in 1978. Despite the advanced technology in recent decades that has improved the success rates of IVF, it only stays at around 45% among women aged less than 35 years and drops to less than 10% for those aged 40 years and older. 3 , 4 In particular, implantation failure of apparently good-quality embryos is a major obstacle, for which there is no good solution at present. Ongoing research is being conducted in an attempt to elucidate the factors that predict or modulate embryo implantation. Uncovering such factors would greatly help to alleviate the physiological and psychological stress of the patients. The microbiome has been increasingly studied in obstetrical care. In regard to its relevance to IVF, studies have been heavily focused on the vaginal and endometrial microbiota. 5 , 6 However, the use of vaginal probiotics to improve pregnancy outcomes shows limited evidence. 7 , 8 Low-grade inflammation is known to negatively impact the live birth rate in women undergoing IVF treatment. 9 Although it remains unclear how low-grade inflammation evolves, gut microbiota is suggested to be one of the contributors. 10 Gut bacteria are the major source of bacterial components in the circulation of individuals free of infection. 11 , 12 Unhealthy diet and disease conditions can disrupt the normal ecology of the gut microflora, leading to a state referred to as gut dysbiosis, which is often associated with the impairment of gut integrity resulting in the penetration of bacteria products. 10 , 13 The leakage of bacterial constituents, such as lipopolysaccharides, cell membrane components of gram(−) bacteria, and flagellins, structural proteins of flagella on flagellated bacteria, from the gut into the circulation can activate immune cells, resulting in production of proinflammatory molecules and consequently low-grade systemic inflammation. 10 , 14 Moreover, flagellins consist of conservative and variable regions, and these variable regions are species-depending and vary in immunogenicity. 15 Special diets, such as Mediterranean diet, that lower systemic inflammation have been shown to improve fecundability. 16 We hypothesize that gut dysbiosis may affect the live birth rate in IVF treatment by modifying inflammatory status through the infiltration of bacterial constituents into the uterine environment. This study aims to assess the relationship between gut microbiota and the live birth outcome of IVF treatment. We recruited women undergoing frozen embryo transfer (FET), a procedure in which a frozen embryo from a previous fresh IVF cycle is thawed and transferred back into the uterus, followed by analyzing bacterial constituents in their serum and gut microbiota samples that were collected prior to FET. Our results reveal that Bifidobacterium in the gut microbiota and lower serum flagellin levels are associated with live birth attainment after FET.

The authors declare no competing interests.

REAGENT or RESOURCE SOURCE IDENTIFIER Antibodies anti-human IgG secondary horseradish peroxidase-conjugated antibodies Jackson ImmunoResearch Cat#109-035-088; RRID: AB_2337584 anti-human IgA secondary horseradish peroxidase-conjugated antibodies Jackson ImmunoResearch Cat#109-035-011; RRID: AB_2337580 Biological samples Human serum The University of Hong Kong-Queen Mary Hospital N/A Human feces The University of Hong Kong-Queen Mary Hospital N/A Chemicals, peptides, and recombinant proteins ZymoBIOMICS Microbial Community Standard Zymo Research Cat#D6300 QUANTI-Blue™ InvivoGen Cat#rep-qbs2 Purified flagellin from Salmonella typhimurium Enzo Life Sciences Cat #ALX-522-058-3010 Critical commercial assays OMNIgene·GUT Collection Kits DNA Genotek Cat#OM-200 QIAamp PowerFecal Pro DNA Kit Qiagen Cat#51804 LAL assay HyCult Biotech Cat#HIT302 IL-1β R&D System Cat#DY201 IFNγ R&D System Cat#DY285B MCP1 R&D System Cat#DY279 CRP R&D System Cat#DY1707 SAA R&D System Cat#DY3019 ultrapure flagellin from Bacillus subtilis InvivoGen Cat#tlrl-pbsfla ultrapure flagellin from Salmonella typhimurium InvivoGen Cat#tlrl-epstfla Deposited data Raw sequences of gut microbiota analysis NCBI Sequence Read Archive (SRA) https://www.ncbi.nlm.nih.gov/bioproject/?term=991649 PRJNA991649 Experimental models: Cell lines HEK-Dual™-hTLR5 reporter cell line InvivoGen Cat. #hkd-htlr5ni Software and algorithms Kneaddata Harvard School of Public Health https://github.com/biobakery/biobakery/wiki/kneaddata N/A MetaPhlAn Duy Tin Truong, Nicola Segata and Curtis Huttenhower https://github.com/biobakery/biobakery/wiki/metaphlan3 Version 3.0 HUMAnN Harvard School of Public Health https://github.com/biobakery/humann Version 3.0 R studio R Project for Statistical Computing and Graphing https://www.r-project.org Version 3.6.2 STAMP Dalhousie University https://beikolab.cs.dal.ca/software/STAMP Version 2.1.3 SPSS IBM Version 26 MedCal MedCalc Software Ltd Version 22 This study was approved by the Institutional Review Board, the University of Hong Kong – Hospital Authority Hong Kong West Cluster (reference number: UW 19-340) in April 2019. Written informed consent was obtained from all participating women. This observational study recruited 67 women who underwent FET at the Centre of Assisted Reproduction and Embryology, The University of Hong Kong- Queen Mary Hospital (HKU-QMH CARE), which was an assisted reproduction centre in a university-affiliated tertiary hospital in Hong Kong. The participants were recruited between June 2019 and September 2021. Exclusion criteria included a history of antibiotics intake or infection within 7 days, and the presence of known structural pathologies that could reduce the conception rate after FET, such as hydrosalpinx or endometrial polyps. Serum and fecal samples were collected prior to FET. Six participants who did not provide fecal samples prior to FET or provided problematic samples were excluded from the gut microbiota study. The primary outcome was live birth rate, while secondary outcomes included pregnancy rate and miscarriage rate. Samples were allocated to “with live birth” and “without live birth” groups for analysis. Serum flagellin level was measured using the HEK-Dual™-hTLR5 reporter cell line (InvivoGen, San Diego, USA) according to the manufacturer’s instructions as described previously. 29 20 μl of 10-times-diluted serum was incubated with 180 μl of cell suspension (65,000 cells/well) for 24 hours. The secreted alkaline phosphatase activity was measured using QUANTI-Blue™ (InvivoGen, San Diego, USA), and the absorbance at 660 nm was measured with a spectrophotometer. Purified flagellin from Salmonella typhimurium (Enzo Life Sciences, USA) was used as a standard. Serum lipopolysaccharide level was measured using limulus amoebocyte lysate (LAL) assay (HyCult Biotech, Uden, Netherlands) according to the manufacturer’s protocol. The proinflammatory cytokines in the serum including IL-1β, IFNγ, MCP1, CRP, and SAA were measured using commercial ELISA kits according to the manufacturer’s instructions (R&D System, Minneapolis, USA). The measurement of anti-flagellin antibodies in circulation was performed as previously described, with modifications. 30 ELISA microplates were coated with 1 μg/ml ultrapure flagellin from Bacillus subtilis (InvivoGen) or Salmonella typhimurium (InvivoGen) with 0.1M carbonate buffer overnight at 4°C, respectively. The wells were then blocked using 3% BSA in tris-buffered saline containing 0.05% tween-20 for 2 hours at room temperature. After blocking, coated wells were incubated with diluted serum samples for an hour, followed by incubation with anti-human IgG or IgA secondary horseradish peroxidase-conjugated antibodies (Jackson ImmunoResearch) for an hour. TMB substrate solution was added for antibody detection. The absorbance was measured at 450 nm. Feces were collected using OMNIgene·GUT Collection Kits (DNA Genotek, Canada). Fecal DNA was extracted using the QIAamp PowerFecal Pro DNA Kit (Qiagen, Venlo, Netherlands) according to the manufacturer’s protocol, and subjected to whole genome shotgun metagenomic sequencing by the Genomics Core of the Centre for PanorOmic Sciences (CPOS), LKS Faculty of Medicine, the University of Hong Kong. The efficiency of the isolation protocol was verified using a mock microbial community, the ZymoBIOMICS Microbial Community Standard, purchased from Zymo Research (CA, USA). DNA fragments were sequenced using Illumina NovaSeq 6000 (151bp; pair-end). The shotgun metagenomic sequence files were first preprocessed using KneadData (Harvard University, USA) with the default settings, including the removal of human (hg37dec_v0.1) contaminant sequences identified by Bowtie2 using the very-sensitive mode, reads with low quality reads (Q < 20), and fragmented short reads (<50 bp). The taxonomic profile was determined using MetaPhlAn 3.0 (Harvard University, USA). This version was built using 99,237 reference genomes representing 16,797 species retrieved from Genbank as of January 2019. The functional profiling was performed using HUMAnN 3.0 (Harvard University, USA) with a non-redundant database, Uniref90 (Version 201901b). 31 Statistical analyses were performed using R software Version 3.6.2, STAMP (Version 2.1.3), SPSS Statistics (Version 26), and MedCalc (Version 22). 32 A 95% power was achieved for this study on the two independent group designs, to detect a 20% difference in measured parameters at 5% significance. Normality was checked by the Shapiro-Wilk test. Continuous variables were expressed as medians (25 th  – 75 th percentile) and compared between groups using the Mann-Whitney U test. Categorical variables were expressed as counts (%) and compared between groups by the Fisher’s Exact test. Univariate binary logistic regression was used to study the prediction of live birth by flagellin. Repeated k-fold cross validation and a random forest model using classification were applied to determine the importance of bacterial species and serum flagellin levels to predict live birth. The receiver-operating characteristics curve was used to determine the predictive performance on live birth rate. The difference in gut microbiota pattern was evaluated by permutational analysis of variance (PERMANOVA) and non-metric multidimensional scaling (NMDS) using the vegan package in R programme. Spearman partial correlation analysis was performed using the ppcor package in R programme. P values were adjusted by the Benjamini-Hochberg method or false discovery rate (FDR) for multiple comparisons. All significance tests were two-tailed, and the statistical significance level was defined by P values of <0.05.

This work was supported by internal seed funding from the 10.13039/501100003803 University of Hong Kong (ref. no. 104005818 ) and the Hong Kong Research Grant Council ( 17102920 , AoE/M-707/18). Jensen Yiu was supported by the Hong Kong RGC Postdoctoral Fellowship Scheme ( PDFS 2021-7S06 ).