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

Abstract

The gut microbiome plays a critical role in human health through its influence on numerous physiological functions such as metabolism and immunity, with disruptions in microbial communities increasingly linked to metabolic disorders. In a large-scale cohort study in Japan, we investigated the association between gut microbiome profiles and metabolic health. Using 16S rRNA gene sequencing, four-enterotype clustering revealed that the Bacteroides 2 (B2) enterotype was associated with lower alpha-diversity and increased risk of metabolic diseases, particularly obesity (OR = 1.51) and hypertension (OR = 1.49). Refined seven-enterotype clustering further stratified the Ruminococcus, Prevotella, and Bacteroides enterotypes into distinct subtypes, uncovering a novel high-risk Prevotella 2 (P2) enterotype associated with nearly two-fold increased risk of obesity and diabetes mellitus. The B2 and P2 enterotypes were characterized by reduced abundance of beneficial short-chain fatty acid (SCFA) producers (Faecalibacterium, Anaerostipes) and enrichment of opportunistic pathogens (Fusobacterium and Veillonella for B2, Megamonas and Megasphaera for P2). Microbial metabolic influence network analysis revealed enterotype-specific interaction patterns, with R1, R2, and P1 enterotypes demonstrating cooperative production of SCFAs and other metabolites, while B enterotypes displayed synergy in the production of a range of sugar compounds. These findings underscore the utility of refined enterotype clustering as a powerful tool to reveal previously unrecognized gut microbial patterns linked to metabolic risk. By identifying B2 and the newly characterized P2 enterotypes as high-risk microbial profiles, this study opens new avenues for microbiome-based stratification and early intervention in metabolic disease management.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13099-025-00730-3.

Background

The human gastrointestinal tract, comprising the gut and its intricate microbiome, represents a complex ecosystem within the body, long recognised for its invaluable role in shaping human health and disease [1]. These gut microorganisms produce various metabolites within the intestinal lumen that influence host physiology through multiple pathways, including energy metabolism, nutrient absorption, and immune system regulation [2]. Recent technological advances in metagenomics, metabolomics, and culturomics have significantly enhanced our understanding of these metabolic dialogues between the microbiota and host, as well as among different microbial communities. This growing knowledge sheds light on the essential role of gut microbiota in maintaining health and its potential involvement in various diseases, particularly metabolic disorders.

Despite the considerable inter-individual variation in gut microbiome composition in humans, researchers have identified distinct patterns of bacterial communities, known as enterotypes. Initially, three main enterotypes were proposed: Bacteroides-dominant, Prevotella-dominant, and Ruminococcus-dominant [3]. Recent studies have suggested a more nuanced classification, particularly within the Bacteroides-dominant group, leading to the identification of a fourth enterotype [4–6]. While enterotypes are not discrete but are proposed as a method of stratification to reduce the complexity of microbial communities, they provide a valuable framework for understanding the relationship between gut microbiota and human health. Notably, the B2 enterotype has been associated with various disease conditions, including obesity and inflammatory bowel disease, highlighting the clinical relevance of these microbial community patterns. [7, 8].

In this study, we conducted an analysis of gut microbiome profiles in a large population cohort in Japan to investigate the association between gut microbial patterns and metabolic health. The large scale of this study using 16S rRNA gene sequencing enabled us to identify novel microbial clustering patterns associated with obesity and metabolic health beyond the previously described B2 enterotype. Moreover, using a microbial metabolic influence network database, we showed that microbial metabolic interactions were distinct among different enterotypes, suggesting the existence of context-specific microbial metabolic networks that may influence host metabolism in enterotype-dependent ways. Collectively, these findings suggest that gut microbiome profiling could serve as a potential biomarker for health status assessment and disease risk prediction.

Methods

Study population

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 sampling, DNA extraction, and sequencing

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).

16S rRNA gene sequencing analysis

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.

Gut microbiome enterotyping

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.

Microbial metabolic influence network (MIN) analysis

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.

Statistical analysis

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.

Results

Four-enterotype clustering method revealed an elevated risk of metabolic diseases in B2 enterotype.

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.

Fig. 1.

Bacteroides 2 enterotype associates with metabolic risks. A Principal coordinate analysis plot based on the Bray–Curtis dissimilarity index. Each point represents an individual sample, colour-coded by four enterotypes; Bacteroides 1 (B1), Bacteroides 2 (B2), Ruminococcus (R), and Prevotella (P). B Violin plots showing the distribution of alpha diversity (i.e., Shannon Diversity Index) across different enterotypes. Boxes represent the interquartile range (IQR; Q1 and Q3), with the line inside marking the median. Wilcoxon test followed by pairwise comparison with Bonferroni correction was conducted. *; p value < 2.2 × 10^–308. C Stacked bar plots illustrating the relative abundance of bacterial genera. The thirteen most abundant genera are colour-coded as indicated in the legend and all other genera are grouped in “Others”. D The Linear discriminant analysis Effect Size (LEfSe) plots identifying distinct gut bacterial genera in each enterotype. E Stacked bar plots illustrating the relative abundance of participants with (Cases) and without (Control) cardiometabolic risks (χ² = 105.867, df = 3, p = 8.5 × 10^–23). F Heatmap of odds ratios from logistic regression analysis showing associations between different gut microbiome enterotypes (B1, B2, R and P) and metabolic risks. Metabolic conditions assessed include obesity (OB), hypertension (HT), dyslipidaemia (DL), and diabetes mellitus (DM). *; p value < 0.05

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.

Table 1.

Characteristics of subjects in two groups

	Control (N = 6680)	Cases (N = 3093)
Age (years)	49.19 (± 6.45)	54.40 (± 6.18)
Sex (female), n (%)	4040 (60.5%)	1534 (49.6%)
Body Mass Index (kg/m2)	22.49 (± 3.41)	24.98 (± 4.24)
Obesity (BMI >  = 30)	207 (3.1%)	362 (11.7%)
Diabetes	0	523
Hypertension	0	1753
Dyslipidaemia	0	1685

^*Detailed information on comorbidities was not available due to their self-reported nature.

Seven-enterotype clustering refined the stratification of gut microbiota

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.

Fig. 2.

Seven-enterotype clustering. A Sankey diagram illustrating the reclassification of gut microbiome enterotypes from four clusters (left) to seven clusters (right). The width of each path represents the proportion of samples transitioning from the four-cluster classification to the seven- cluster classification. B Principal coordinate analysis plot using the Bray–Curtis dissimilarity index reveals the degree of separation among seven enterotypes. Each point represents an individual sample, colour-coded by enterotypes. C Violin plots showing the distribution of alpha diversity across seven different gut microbiome enterotypes. Boxes represent the interquartile range (IQR; Q1 and Q3), with the line inside marking the median. *; p value < 2.2 × 10^–308. D Stacked bar plots illustrating the relative abundance of bacterial genera in each of the seven enterotypes. The thirteen most abundant genera are colour-coded as indicated in the legend and all other genera are grouped in “Others”.

B2 and P2 enterotypes were associated with an increased risk of metabolic diseases

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.

Fig. 3.

Prevotella 2 enterotype associates with metabolic risks. A The LEfSe analysis was performed to identify distinct gut bacterial genera in each enterotype. B, C, F Violin plots showing the centred log ratio- transformed relative abundance of each taxa across the seven enterotypes. Each violin plot displays the distribution of abundance values within each enterotype. D Stacked bar plots illustrating the relative abundance of participants with (Cases) and without (Control) metabolic risks factor across different enterotypes (χ² = 105.053, df = 6, p = 2.21 × 10^–20). E Heatmap of odds ratios from logistic regression analysis showing associations between different gut microbiome enterotypes and metabolic risks, which includes obesity (OB), hypertension (HT), dyslipidaemia (DL), and diabetes mellitus (DM). *; p value < 0.05

Different enterotypes had distinct microbial interaction networks for metabolite production or utilization

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.

Fig. 4.

Microbial metabolic influence network (MIN) analysis shows association between metabolites and enterotypes. For the metabolites that showed differences in total metabolic specific influence (TMSI) index between enterotypes, TSMI index were compared using a heatmap. A Metabolites involved in cooperative interspecies interactions. B Metabolites involved exclusively in competitive interspecies interactions.

Discussion

Due to the complexity of the human gut microbiota, stratifying disease risk based on gut microbiota composition holds potential for advancing precision medicine by linking specific gut microbiota patterns to health outcomes. However, the feasibility of this approach and the technical challenges associated with enterotype analyses remain topics of debate [17]. This study examined the large Japanese population cohort to explore the association between gut microbiota and metabolic disease risk factors and to identify enterotypes that could serve as predictive biomarkers for metabolic diseases. Our findings revealed that two enterotypes, B2 and P2, were significantly associated with risk factors for metabolic diseases, including obesity, diabetes mellitus, and hypertension. Leveraging a robust dataset of about ten thousand participants, we characterized the gut microbiota composition of the population and established meaningful correlations with reliable statistical power.

Since the concept of enterotypes was introduced by Arumugam et al. in 2011, numerous studies have employed various categorization methods, yielding heterogeneous findings on gut microbiome enterotypes in human [18–22]. Our study utilized the DMM method as a modelling approach, classifying human gut microbiome communities into four major enterotypes, consistent with earlier reports [4, 5]. Associations between specific taxonomic drivers and diseases have been investigated, with the B2 enterotype frequently characterized as dysbiotic due to its lower bacterial diversity and links to various diseases [23–27]. For instance, an increase in Bacteroides or B2 enterotype itself, has been associated with non-alcoholic steatohepatitis [28], colorectal cancer [29, 30], low-grade inflammation state in obesity [31], and increased frailty [32]. Periodontal bacteria such as Fusobacterium spp. and Veilonella spp., characteristic taxa of the B2 enterotype, were reported to correlate with human diseases; where Fusobacterium nucleatum contributed to the development of colorectal cancer [33] and Veillonella spp. was enriched in the gut of subjects with inflammatory bowel disease[34]. In addition, Streptococcus spp., another distinguishable taxon in B2, is an independent risk factor for coronary atherosclerosis and systemic inflammation [35]. Our findings demonstrated that the B2 enterotype exhibits reduced microbial diversity, is enriched in opportunistic pathogens, and is associated with an elevated risk of metabolic diseases. However, further research is needed to assess the clinical utility of the B2 enterotype as a predictive biomarker for metabolic disorders.

Interestingly, our seven-enterotype clustering approach uncovered novel enterotypes with distinct dysbiotic features. For example, the P2 enterotype, a subset of the Prevotella-dominant enterotype identified through this refined classification, is characterized by a reduction in SCFA-producing bacteria such as Alistipes and Ruminococcus [15, 36], alongside an increased abundance of opportunistic bacteria like Megamonas and Megasphaera. Megamonas was previously linked to non-alcoholic fatty liver disease and obesity by enhancing lipid absorption through gut bacterial-dependent metabolism of a sugar compound [37, 38]. Despite being an SCFA producer, Megasphaera was previously reported to be associated with dyslipidaemia and increased metabolic risks [39, 40]. Our odds ratio analyses revealed that the association between the Prevotella enterotype and metabolic risk factors in the four-clustering method is primarily attributable to the P2 enterotype, as the P1 enterotype showed no significant association with metabolic disease risks. This highlights the value of our enhanced clustering approach in improving stratification by revealing unique microbial communities with distinct characteristics that were previously masked in the four-enterotype approach.

An interesting observation from our analysis is the identification of Bifidobacterium as a discriminating taxon for the B2 enterotype, which was associated with increased metabolic risk. This finding may appear counterintuitive, as Bifidobacterium is widely recognized for its health-promoting effects and is frequently used as a probiotic [41, 42]. However, a large-scale study involving 1,803 Japanese individuals similarly reported that Bifidobacterium-enriched microbiota was significantly associated with a higher prevalence of inflammatory bowel disease, cardiovascular disease, and diabetes [43]. These seemingly contradictory findings highlight the importance of functional profiling and strain-specific effects when interpreting the role of Bifidobacterium in health and disease. Moreover, the B2 enterotype was also enriched in other potentially pathogenic or opportunistic taxa such as Fusobacterium and Streptococcus, which may contribute to its association with metabolic risks, potentially modulating or overriding any beneficial effects of Bifidobacterium.

Another noteworthy finding is that the subtypes within each enterotype may represent a spectrum rather than discrete groups. For example, B1-1, B1-2, and B2 appear to form a continuum, as evidenced by the gradual changes in alpha diversity, bacterial abundances, and associations with metabolic disorder risk. This aligns with previous findings suggesting that the gut microbiome across populations exists as a continuous gradient of dominant taxa, rather than discrete enterotype groups [44]. Further supporting this gradient-like structure, our metabolic interaction network analysis revealed distinct patterns of SCFA production and sugar compound utilization across enterotypes. Notably, within the Bacteroides-dominant enterotypes, a progression in metabolic potential was observed from B1-1 to B1-2 to B2, mirroring the compositional and clinical shifts noted above.

Despite potential overlaps at enterotype boundaries, dominant microbial community structures remain clinically useful. Enterotypes—whether conceptualized as distinct groups or positions along a spectrum—can serve as useful tools for stratifying individuals by microbiome composition and associated health risks. For instance, Keller et al. recently proposed an"enterotype dysbiosis score", which quantifies an individual's deviation from healthy reference enterotypes along a continuous axis [19]. This underscores the translational utility of enterotype-based frameworks, even in the absence of strictly discrete boundaries, and reinforces their potential application in personalized microbiome-based interventions.

To address the population specificity of our findings, we compared enterotype–disease associations across global populations with those reported in previous studies (Supplementary Table 10). Notably, the Prevotella-enriched P2 enterotype—linked to metabolic risk in our Japanese cohort—shows similar associations in other Asian populations [37, 45], but appears protective and associated with high fibre intake in Western cohorts [3, 46]. This divergence highlights the influence of population-specific factors such as diet, host genetics, and age distribution. In addition, the identification of B2 and P2 as distinct risk-associated enterotypes may stem from methodological differences and the characteristics of our cohort, including a relatively homogeneous lifestyle and dietary background. These findings highlight the importance of contextualizing enterotype interpretations within population and methodological frameworks, and the need for further cross-cohort validation.

To conclude, this study underscores the significance of gut microbiota stratification in understanding metabolic disease risks. The identification of dysbiotic enterotypes, particularly B2 and P2, highlights their potential role as biomarkers and therapeutic targets for metabolic disorders. This stratification is particularly valuable in the context of personalized medicine and nutrition, as individuals with different enterotypes may respond uniquely to dietary interventions and medical treatments [47, 48]. However, this study has several limitations. First, it relies on self-reported medical history, which may introduce recall bias. However, previous studies have shown that self-reported metabolic conditions generally show moderate to high agreement with clinical diagnoses [49, 50]. Any potential misclassification is likely non-differential, which would bias the associations toward the null and may lead to underestimation of the true risks. Second, the cross-sectional study design prevents causal inference between enterotypes and disease outcomes. Third, the absence of biochemical data limits our ability to correlate microbial profiles with specific metabolic parameters and perform confounding factor adjustments in the logistic regression analyses. Additionally, although medications are known to influence the gut microbiome, this study did not account for participants’ medication use, which may confound the observed associations. Future longitudinal studies incorporating biochemical assessments and mechanistic investigations are needed to validate these findings and explore interventions aimed at modulating dysbiotic enterotypes to restore gut microbiota balance and mitigate metabolic disease risk. In particular, testable interventions emerging from our findings could include SCFA supplementation to counteract Fusobacterium-driven inflammation in individuals with the B2 enterotype, and dietary sugar reduction to disrupt Megamonas-associated metabolic synergy in those with the P2 enterotype. Such targeted approaches hold promise for precision nutrition and microbiome-based therapies in metabolic health.

Conclusions

In conclusion, our study identified an association between gut enterotypes, specifically B2 and P2, and metabolic risks including obesity, hypertension and dyslipidaemia in the large Japanese cohort. These findings highlight the potential use of enterotypes as disease biomarker and therapeutic target for metabolic diseases. Future research is needed to validate their clinical utility and explore their application in disease prevention and management.

Abbreviations

BMI

Body mass index

DMM

Dirichlet multinomial mixture

LEfSe

Linear discriminant analysis effect size

MSI

Metabolic specific influence

PCoA

Principal coordinate analysis

SCFA

Short-chain fatty acid

TMSI

Total metabolic specific influence

Author contributions

SW, AKT, TY, and KK conceived and supervised the work. BYR, KB, DNA, SW and KK performed data analysis and wrote the manuscript. All named authors meet the International Committee of Medical Journal Editors (ICMJE) criteria for authorship for this manuscript, take responsibility for the integrity of the work as a whole, and have given final approval for the version to be published. All authors read and approved the final manuscript.

Funding

K.K. was supported by the Ministry of Education (Singapore) under its Academic Research Fund Tier 2 (MOE-T2EP30223-0008) and the MOE Start-up Grant, the Singapore Ministry of Health's National Medical Research Council under its CS-IRG-NIG (MOH-001379), and Vascular Research Initiative, LKCMedicine, Nanyang Technological University.

Availability of data and materials

All data generated or analysed during this study are included in this published article and its supplementary materials. The sequencing data used are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

All participants read and signed an informed consent document with the description of the testing procedures 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) in accordance with the principles of the Declaration of Helsinki.

Consent for publication

This article does not contain any individual person’s data in any form.

Competing interests

The authors declare no competing interests.