Compositional and functional differences in human gut microbiome with respect to equol production and its association with blood lipid level: a cross-sectional study

The present study included a total of 99 with an average age of 36 years old, including 46 males and 53 females, 59 out of which were classified to be EP according to equol excretion status in urine. A 24-h urine sample was collected after oral administration of a capsule of SI for 3 days and equol excretion was determined by high performance liquid chromatography (HPLC).

To identify the association of gut microbiome with equol phenotype, we performed shotgun metagenomics sequencing of fecal samples from study participants. The DNA sequencing data have been deposited in the BIG Data Center (accession numbers: CRA001481). Metagenomic reads from 99 fecal samples were processed with MetPhlAn2 (Metagenomic Phylogenetic Analysis) to determine composition of gut microbiome and to calculate the relative abundances of species. The microbial composition at phylum level was shown in Additional file 1: Figure S1. Phyla and species with relative abundance equal to or larger than 0.0001 of the average level in each group (EP and NP) were included in further analyses. We compared the difference in microbial composition between EP and NP adjusted for age, gender, BMI, equol phenotype and smoking habit (Table 2). Only the factor of equol was significantly associated with inter-individual microbial distance and this factor explained 2.05% of the variation in microbial composition (P = 0.02 for equol). We then discovered significant community level microbiota differences, using the Adonis test and Bray–Curtis distance matrix calculated from species-level composition between EP and NP (P = 0.0062). However, no significant difference in bacterial richness and evenness were observed between EP and NP (P = 0.64 for Shannon–Wiener index, P = 0.72 for Chao1, and P = 0.43 for Simpson index in Fig. 1a). PCoA based on the Bray–Curtis distance matrix were conducted to reveal dissimilarity in metagenome-based relative abundances between EP and NP at the species level (Fig. 1d). EP and NP are clustered into different structures of gut microbiome as indicated in Fig. 1d.

At individual taxa level, we used the criteria mentioned above and kept 139 from a total of 351 species for further analyses. With Wilcoxon rank-sum test, 32 species showed significant differences in two groups (Fig. 1c; P < 0.1). Equol producing-related species Adlercreutzia equolifaciens and Bifidobacterium bifidum, showed higher relative abundance in EP than in NP (77.5% vs. 22.5% and 72.0% vs. 28.0%, respectively), despite the significance being only marginal (P = 0.06 and P = 0.08). We also used LEfSe analysis for biomarker discovery within the microbiome at species level between the two groups; 14 biomarkers were defined within the NP group and 10 biomarkers within the EP group as shown in Fig. 1f, g.

To investigate the differences in presence/absence, as well as abundance of metabolic pathways in the gut microbial community between EP and NP, we carried out analysis on the metagenomics sequences and obtained their genomic functional potential using HUMAnN2 (the HMP Unified Metabolic Analysis Network) [19]. A total of 134 metabolic pathways with relative abundance larger than 0.0001 of the average level were included in the analysis. We identified 75 metabolic pathways, which showed significant differences (P < 0.1) between EP and NP by Wilcoxon rank-sum test, 57 of which remained statistically significant after false discovery rate (FDR) adjustment (Q < 0.1). Most of these pathways are biosynthesis pathways (55 out of 75), among which 7 pathways were significantly associated (P < 0.05) with equol concentration in urine by using Spearman correlation analysis (Additional file 1: Figure 2). Chao1 index and evenness in metabolic pathways showed no significant differences between EP and NP, and richness in metabolic pathways was higher in EP than in NP (P = 0.86 for Chao1, P = 0.12 for Shannon–Wiener index, and P = 0.015 for Simpson index in Fig. 1b). Subsequently, we conducted constrained PCoA analysis on the metabolic pathways (Fig. 1e) and statistical significance test (anova.cca) on the result of constrained PCoA, which showed significant dissimilarity between EP and NP for metabolic pathways as well (Permanova P = 0.001). We also established an equol production-related pathway based on the three critical gene sequences in A. equolifaciens 19450T, Eggerthella sp. YY7918, and Lactococcus garvieae to blast the reference dataset of HUMAnN2. We found only the genes from A. equolifaciens 19450T can be found with right annotation, and only these genes can be found in the result of HUMAnN2. So, we separated the relative abundances of these genes from the results of the gene part, then we used the relative abundances of these three genes to calculate the geometrical mean as the equol metabolic-related pathway’s abundance. As shown in Fig. 2 and Additional file 2: Table S1 and Table S2, this equol production-related pathway was mainly identified and showed higher relative abundance in EP in this study.

To assess whether the differences in abundance of metabolic pathways in the gut microbial community may explained by other characteristics of the participants, we further calculated the Bray–Curtis distance of the metabolic pathway using HUMAnN2 with gender, age, BMI, equol phenotype and smoking habit in the model. The result was similar to the composition of gut microbiome, only the equol phenotype had significant correlation to beta diversity (P = 0.0001) and explains 6.58% of the variation, while the rest of factors do not play significant roles (Table 1).

As shown in Table 2, a total of 36 (36.4%) of the participants were classified as having dyslipidemia. EP showed a substantially lower prevalence of dyslipidemia (27.1% vs. 50.0%, P = 0.02) than NP. Results from logistic regression models indicated that the association with dyslipidemia in serum lipid levels remained significant after adjustment for age and BMI (adjusted odds ratio = 0.37 [0.15–0.94], P = 0.036). Serum triglycerides (TG) levels was 12% lower in EP compared to NP (1.18 ± 0.48 mmol/L vs. 1.36 ± 0.57 mmol/L, P = 0.08) (Fig. 3a), although the difference between the two groups were marginally significant due to a relatively small sample size.

We also tested the correlations between bacterial species and serum lipid levels from our dataset and found several species with FDR < 0.05. Negative association of Haemophilus parainfluenzae and positive association of Klebsiella pneumoniae and Lachnospiraceae bacterium-8_1_57FAA with total cholesterol (TC), LDL-C, and apolipoprotein B (ApoB) were observed (Fig. 3b). As is indicated in Fig. 1c, EP shows a higher relative abundance of H. parainfluenzae and lower levels of K. pneumoniae and L. bacterium-8_1_57FAA compared to NP. The findings suggest that these species might contribute to the changes in blood lipid levels via equol production.

Intake of SI was calculated according to a 3-day, 24-h dietary record. We first compared intake of SI (including total SI, daidzein, genistein, and glycitein) and soy products (including tofu, dried bean curd, soy bean milk, soy milk, dried beans, fermented bean curd, fermented soy beans, and other soy products) between EP and NP. No significant differences between the two groups were observed (Additional file 2: Tables S3, S4). Subsequently, we examined the correlation between intake of daidzein (the precursor to equol) and equol concentration in urine/equol excretion in 24 h in EP by Pearson correlation analysis. The results showed that neither equol concentration nor equol excretion in 24 h was associated with the amount of intake of the precursor, daidzein (r = 0.12, P = 0.3 and r = 0.07, P = 0.5, respectively). These findings suggest that equol production is mostly determined by the gut microbiome of the participants, while the intake of soy isoflavones plays a minimal role if any.