Bacteroides- and Prevotella-enriched gut microbial clusters associate with metabolic risks

The study initially enrolled 16,472 stool samples from individuals aged 40 to 64 years who participated in a microbiome testing service. To ensure the inclusion of only unique participants, multiple submissions from the same individuals were excluded beyond the first submission. Furthermore, individuals with non-lifestyle-related diseases or those undergoing medical treatment were also excluded. Based on these criteria, 6699 individuals were removed from the final analysis, and 9773 individuals were enrolled in the study. The presence or absence of comorbidities is determined based on questionnaire responses, and no biochemical analysis, such as blood tests, has been conducted. All participants provided written informed consent for enrolment. The study was conducted in accordance with the principles of the Declaration of Helsinki, approved by the Institutional Review Board (no LD-001-06 and LD-002-06), and registered in the UMIN Clinical Trials Registry System (UMIN000028887 and UMIN000028888).

Faecal samples were collected using brush-type collection kits containing guanidine thiocyanate solution (TechnoSuruga Laboratory, Japan), transported at room temperature, and stored at 4 °C. DNA extraction was performed using an automated DNA extraction system (GEBE PREP STAR PI-480, Kurabo Industries Ltd, Japan) according to the manufacturer’s protocol. The V1-V2 region of the 16S rRNA gene was amplified using a forward primer (16S_27Fmod: TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG AGR GTT TGA TYM TGG CTC AG) and a reverse primer (16S_338R: GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG ACA GTG CTG CCT CCC GTA GGA GT) along with KAPA HiFi HotStart ReadyMix (Roche, Switzerland). We selected the primers for V1–V2 hypervariable regions because we previously conducted a study comparing V1–V2 and V3–V4 primers [9], which showed that 16S rRNA sequencing using the V1–V2 primers provided robust results. Sequencing libraries were prepared according to the 16S library preparation protocol provided by Illumina. Dual index adapters for sequencing on the Illumina MiSeq platform were attached using Nextera XT Index kit (Illumina, San Diego). Each sequencing library was diluted to 5 ng/μL. Equal volumes of each library was mixed to obtain a final concentration of 4 nM. The DNA concentrations of the mixed libraries were measured by quantitative polymerase chain reaction (qPCR) using KAPA SYBR FAST qPCR Master Mix (Roche), primer 1 (AAT GAT ACG GCG ACC ACC), and primer 2 (CAA GCA GAA GAC GGC ATA CGA). These libraries were sequenced in a 250-bp paired-end run using MiSeq Reagent Kit v2 (500 cycles).

The data processing and assignment based on the QIIME2 pipeline [10] were performed using the following steps: (1) joining paired end reads, filtering, and denoising with DADA2 and (2) assigning taxonomic information to each ASV using a naive Bayes classifier in the QIIME2 classifier. The classifier was trained with SILVA 138 [11], with slight modifications on the taxonomic nomenclature for improved clarity of reports (i.e., removed labels with little information such as “D_6__unclutured bacteria” and corrected duplicate entries such as “D_0__Bacteria;D_1__Bacteria Firmicutes” to “D_0__Bacteria;D_1__Firmicutes”). We examined beta diversity using principal coordinate analysis based on the Bray–Curtis dissimilarity index. We also determined alpha diversity by computing a Shannon index.

We calculated two enterotype classifications using Dirichlet multinomial mixture clustering. First, we divided samples into 4 types to examine whether previous findings could be reproduced in our cohort. To determine optimal clustering, we explored cluster numbers based on Laplace approximation and analyzed each bacterial genus’ contribution by computing the mean absolute difference between relative abundances derived from the selected number of components versus a single component model. For characterization of each enterotype, we identified major bacterial genera (defined as those present at > 3% in any type) and visualized their composition. We then assessed separation between enterotypes using principal coordinate analysis based on Bray–Curtis distances.

Considering that differences in disease risk among subtypes in the 7 enterotypes could be attributed to differences in metabolites that are produced or consumed, we conducted microbial metabolic interaction analysis to investigate this. A previous paper reported a literature-curated interspecies network of the human gut microbiota, called NJS16 [12]. Using this database, we estimated microbial metabolic interactions for each sample, bacterial combination, and metabolite. To identify metabolites that differ between subtypes, we aggregated the data to obtain total values for each enterotype and metabolite, and calculated effect sizes (Hedge’s g) based on subtype comparisons as sample sizes were large and conventional statistical tests would all reach the lower limit of p values. Subsequently, we selected metabolites with effect sizes of 0.5 or greater in any of the B, R, or P groups (Supplementary Table 9). For directionality of microbial metabolic interactions, we separately visualized metabolites with cooperative effects and ones with only competitive effects.

For the 4 and 7 types of enterotypes, we conducted statistical analyses on several parameters. For alpha diversity based on the Shannon index, we performed Wilcoxon test and multiple comparisons based on Holm’s methods. Chi-square test was performed for the comparison of Cases vs Controls in each enterotype. To determine biomarkers for each enterotype, LefSe analysis (p < 0.05 and LDA score > 3.0) was conducted. For association with medical history, we performed odds ratio analysis based on logistic regression, adjusted for age and sex.

A total of 9773 subjects from our population cohort were included in this study. To investigate whether a certain enterotype is associated with an increased risk of metabolic diseases, we introduced Dirichlet Multinomial Mixture model to stratify the cohort into enterotypes. The Principal Coordinate Analysis (PCoA) plot of Bray–Curtis distances representative of beta diversity (Fig. 1A) showed that the subjects were clustered into four groups according to the predominantly enriched genera, namely Bacteroides 1 (B1), Bacteroides 2 (B2), Ruminococcaceae (R) and Prevotella (P). Based on the Shannon index, we observed that the B2 enterotype had a lower alpha-diversity compared to the other three enterotypes (Fig. 1B, Supplementary Table 1). Reduced alpha-diversity has been shown to be associated with dysbiosis and elevated risks for diseases [13, 14]. B1 and B2 enterotypes were predominantly enriched with Bacteroides genus, and B2 was distinguishable from B1 by a higher abundance of Prevotella genus (Fig. 1C). Consistent with previous studies, the B2 enterotype contains lower abundance of SCFA producers such as Faecalibacterium and Ruminococcus [8]; while some of the potential opportunistic bacteria taxa like Megamonas, Streptococcus and Fusobacterium were in greater relative abundance compared to the other enterotypes. The R and P enterotypes were predominantly enriched with Ruminococcus and Prevotella genus respectively. Figure 1D showed the linear discriminant analysis effect size (LEfSe), which depicts the genera that most likely explain the differences between enterotypes. Similarly, the distinction of B2 enterotype from the other enterotypes were attributable to the enrichment of Fusobacterium, Megamonas and Streptococcus (Fig. 1D, Supplementary Table 2). Interestingly, Bifidobacterium, which is a conventional probiotic and is generally considered beneficial, was the highest discerning taxa in B2 enterotype.

Bild vergrößern

Next, the subjects were categorised into the control (n = 6680) and case (n = 3093) groups based on the self-reported history of metabolic diseases, including diabetes, hypertension and dyslipidaemia. The baseline characteristics of the subjects were summarised in Table 1. The cases group had significantly higher age compared to the control (54.40 ± 6.18 vs 49.19 ± 6.45 years), lower proportion of females (49.6 vs 60.5%), higher BMI (24.98 ± 4.24 vs 22.49 ± 3.41 kg/m²), and a higher prevalence of obesity (BMI over 30, 11.7% vs 3.1%). Subjects with B2 enterotype have a higher proportion of metabolic risk factors depicted by the portion of cases (Fig. 1E), indicating that the B2 enterotype is associated with metabolic diseases. The logistic regression of the four enterotypes against individual risk factors for metabolic diseases demonstrated that individuals with B2 enterotype have a greater likelihood of developing metabolic abnormalities, with obesity [OR = 1.51] and hypertension [OR = 1.49] among the risk factors with the highest odds ratios (Fig. 1F, Supplementary Table 3). We also observed that P enterotype was slightly associated with a higher risk of diabetes, hyperlipidaemia and hypertension, although it did not reach statistical significance. Together, these findings validate the previously reported association between B2 enterotype and metabolic dysfunction in Western population within a Japanese cohort, and suggest that it may serve as a potential microbial biomarker for increased metabolic disease risk.

Using the Laplace approximation as a measure of model fit in the cohort, we found that seven enterotypes was a better stratification model in characterising gut microbiota for this cohort compared to four enterotypes in the cohort (Supplementary Table 4). Figure 2A showed the relationship between four-enterotype clusters and seven-enterotype clusters. In comparison to four-enterotype, the seven-enterotype further stratified R, P and B1 enterotypes into two subtypes respectively; each enterotype was clearly distinct from each other as evidenced by the PCoA plot (Fig. 2B). Similar to the four-enterotype clustering, B2 enterotype had the lowest alpha-diversity compared to other enterotypes based on Shannon index (Fig. 2C, Supplementary Table 5). Interestingly, the second subtype of each enterotype (B1-2, R2, P2) had a lower alpha-diversity compared to their original counterparts (B1-1, R1, P1), indicating that the seven-enterotype clustering refined the stratification to uncover unique clusters with distinct characteristics that were previously masked in the four-enterotype. The predominant bacterial genera remained aligned with each major enterotype: Bacteroides (B1-1, B1-2, B2), Ruminococcus (R1, R2) and Prevotella (P1, P2) (Fig. 2D). It is worth mentioning that other than B2 enterotype, there was another possible dysbiotic enterotype, P2, characterised by the increased abundance of opportunistic pathogens such as Megamonas, Streptococcus and Fusobacterium, and a reduction of “beneficial strains” like Faecalibacterium and Ruminococcus relative to P1. In addition, B1-2 enterotype appeared to be the intermediate enterotype between B1-1 and B2 in terms of bacterial composition, depicted by lower Faecalibacterium and higher Fusobacterium composition.

Bild vergrößern

SCFAs, such as acetate, propionate and butyrate, play a key role in human health and diseases, including maintenance of gut integrity and provision of anti-inflammatory properties. The major source of SCFAs is through the degradation of dietary fibres by SCFA-producing bacteria, including Anaerostipes and Faecalibacterium [15, 16]. In the LEfSe plot of the seven-clustering method, Anaerostipes and Faecalibacterium were differentially enriched in B1-1 and R1 enterotypes respectively (Fig. 3A, Supplementary Table 6). The relative abundance of Anaerostipes and Faecalibacterium were significantly higher in B1-1 and R1 enterotypes and were lower in B2 enterotype (Fig. 3B, Supplementary Table 7). On the other hand, opportunistic pathogens, such as Veillonella and Fusobacterium, distinguished the B2 enterotype from the rest, and the relative abundance of these taxa was remarkably increased in B2 (Fig. 3C, Supplementary Table 7). Correspondingly, the number of subjects with metabolic risk factors constituted a larger proportion of the dysbiotic B2 and P2 enterotypes (Fig. 3D). Assessing the risk of individual metabolic diseases, the B2 enterotype was associated with a significantly increased risk of obesity [OR = 1.52] and diabetes mellitus [OR = 1.53]; while the P2 enterotype was associated with nearly two-fold increased risk of obesity [OR = 2.2] and diabetes mellitus [OR = 1.75] (Fig. 3E, Supplementary Table 8). On the other hand, R1 was associated with significantly lower risks of hypertension [OR = 0.72] and dyslipidaemia [OR = 0.70]; R2 associated with reduced risk of obesity [OR = 0.43] and dyslipidaemia [OR = 0.77]. To investigate the distinguishing bacterial taxa in the P2 enterotype, we found that the relative abundance of Megamonas and Megasphaera were significantly elevated in P2, while SCFA-producing bacteria such as Alistipes and Ruminococcus were reduced (Fig. 3F, Supplementary Table 7). While Megamonas and Megasphaera are well-known opportunistic pathogens, the enrichment of these genera in P2 enterotype suggest associations with metabolic diseases. These results highlight that the dysbiotic features and enrichment of opportunistic pathogens in the B2 and P2 enterotypes underlie their strong associations with increased metabolic disease risk.

Bild vergrößern

To understand the metabolic interactions between gut microbes within each enterotype, specifically on the production or utilization of a specific metabolite, a microbial metabolic influence network analysis was performed [12]. In short, the metabolic specific influence (MSI) index was calculated for each possible bacterial pair in the community, where MSI > 0 represents cooperative interactions and MSI < 0 represents competition. The total MSI (TMSI) signifies the overall direction and magnitude of interactions across the entire microbiome for each metabolite. We found that there is a distinct variation in the microbial interaction networks between different enterotypes for each metabolite. For instance, gut communities in R1, R2 and P1 enterotypes adopt a cooperative approach in producing SCFAs like acetate, butyrate, isobutyrate and isovalerate, while members of the B2 enterotype compete for these SCFAs (Fig. 4A, Supplementary Table 9). On the other hand, all subsets of B enterotypes, especially B2, displayed synergistic interactions in the production of various monosaccharides, disaccharides and oligosaccharides, while bacterial communities in the P and R enterotypes exhibited competitive interactions for these sugar compounds (Fig. 4B, Supplementary Table 9). These findings highlight how microbial communities within different enterotypes adopt distinct interaction strategies, either cooperative or competitive, to produce different metabolites that may influence metabolic risk.

Bild vergrößern