Gut microbial carbohydrate metabolism contributes to insulin resistance

Abstract

Insulin resistance is the primary pathophysiology underlying metabolic syndrome and type 2 diabetes^1,2. Previous metagenomic studies have described the characteristics of gut microbiota and their roles in metabolizing major nutrients in insulin resistance^{3,4,5,6,7,8,9}. In particular, carbohydrate metabolism of commensals has been proposed to contribute up to 10% of the host’s overall energy extraction¹⁰, thereby playing a role in the pathogenesis of obesity and prediabetes^3,4,6. Nevertheless, the underlying mechanism remains unclear. Here we investigate this relationship using a comprehensive multi-omics strategy in humans. We combine unbiased faecal metabolomics with metagenomics, host metabolomics and transcriptomics data to profile the involvement of the microbiome in insulin resistance. These data reveal that faecal carbohydrates, particularly host-accessible monosaccharides, are increased in individuals with insulin resistance and are associated with microbial carbohydrate metabolisms and host inflammatory cytokines. We identify gut bacteria associated with insulin resistance and insulin sensitivity that show a distinct pattern of carbohydrate metabolism, and demonstrate that insulin-sensitivity-associated bacteria ameliorate host phenotypes of insulin resistance in a mouse model. Our study, which provides a comprehensive view of the host–microorganism relationships in insulin resistance, reveals the impact of carbohydrate metabolism by microbiota, suggesting a potential therapeutic target for ameliorating insulin resistance.

Main

We analysed 306 individuals (71% male) aged from 20 to 75 years (median age, 61 years), who were recruited during their annual health check-ups (Extended Data Fig. 1a). Individuals diagnosed with diabetes were excluded to avoid any long-lasting effects of hyperglycaemia^5,6. Consequently, our study included relatively healthy individuals compared with most of the previous metagenomic studies of diabetes and obesity^{5,6,7,8,11,12}; the median (interquartile range (IQR)) body mass index (BMI) and glycated haemoglobin (HbA1c) were 24.9 kg m⁻² (22.2–27.1 kg m⁻²) and 5.8% (5.5–6.1%), respectively (Supplementary Table 1). The main clinical phenotype analysed in this study was insulin resistance (IR), which we defined as a homeostatic model assessment of IR (HOMA-IR) score of at least 2.5 (ref. ¹³). We also analysed the associations between faecal metabolites and metabolic syndrome (MetS), an IR-related pathology. The clinical characteristics of IR and MetS largely overlapped except for blood pressure and sex ratio, for which there was no difference between individuals with IR versus normal insulin sensitivity (IS) (Supplementary Table 1). Untargeted metabolomics analysis using two mass spectrometry (MS)-based analytical platforms identified 195 and 100 annotated faecal and plasma hydrophilic metabolites, and 2,654 and 635 annotated faecal and plasma lipid metabolites, respectively (Extended Data Fig. 1a). To identify the overall difference in microbial functions, faecal metabolites and predicted genes were summarized into co-abundance groups (CAGs) and KEGG categories, respectively (Extended Data Fig. 1b). Transcriptomic information of peripheral blood mononuclear cells (PBMCs) was obtained using the cap analysis of gene expression (CAGE) method¹⁴, which can measure gene expression at the transcription-start-site resolution.

To examine how omics data of faecal samples can predict IR, we first compared the area under the curve (AUC) of receiver operating characteristic (ROC) curves on the basis of random-forest classifiers. Predictor variables for the models were selected using the minimum-redundancy maximum-relevance algorithm¹⁵ from the faecal 16S, metabolome, metagenome and their merged datasets (Supplementary Table 2). We found that the selected features of faecal metabolomic data generally outperformed those of 16S and metagenomics in predicting IR (Fig. 1a), suggesting that faecal metabolomics could be used to study IR pathogenesis.

Faecal carbohydrates are increased in IR

We next searched for the associations between clinical phenotypes and faecal metabolite CAGs (Fig. 1b and Supplementary Tables 3–8). Major confounding factors, namely sex and age, were adjusted throughout the correlation and regression analyses with clinical markers. Among the hydrophilic metabolites, most of the CAGs showing significant associations with IR were those of carbohydrate metabolites, mainly monosaccharides (hydrophilic CAGs 5, 12 and 15; Fig. 1b, top). Short-chain fatty acids (SCFAs), which are known as carbohydrate fermentation products, were also increased in IR (hydrophilic CAG 8). Hydrophilic CAG 18 remained unannotated as it included metabolites from different pathways (Supplementary Table 5). KEGG pathway enrichment analysis of the metabolites in these IR-related hydrophilic CAGs revealed that these metabolites were indeed involved in carbohydrate metabolism (Extended Data Fig. 2a). Specifically, we found that the major monosaccharides such as fructose, galactose, mannose and xylose significantly correlated with IR (Fig. 1c). Among the SCFAs, propionate was particularly increased in IR (Extended Data Fig. 2b), consistent with its role in gluconeogenesis¹⁶. Faecal monosaccharides were similarly increased in MetS, obesity and prediabetes (Fig. 1d and Extended Data Fig. 2c,d). By contrast, disaccharides showed weak or no association (Extended Data Fig. 2b–d). These findings show that the end products of carbohydrate degradation—such as monosaccharides, which are readily absorbed and used by the host—are particularly increased in the faeces of individuals with IR and MetS. Supporting these findings, our analysis of previously published faecal metabolomics data from the TwinsUK cohort¹⁷ showed that faecal monosaccharides, notably glucose and arabinose, were positively associated with obesity and HOMA-IR, both of which relate to IR (Extended Data Fig. 3a–c and Supplementary Table 9). Similarly, the peak intensity of faecal fructose, glucose and galactose was associated with BMI in a small number of individuals without inflammatory bowel disease (IBD) from HMP2 data¹⁸ (Extended Data Fig. 3d). Together, these findings indicate that faecal carbohydrates are increased in IR and related pathologies and that this alteration is consistently observed across populations.

In addition to hydrophilic metabolites, faecal lipid CAGs were also associated with IR (Fig. 1b). Lysophospholipids, bile acids and acylcarnitine were associated with IR and MetS as reported previously¹⁹. Among them, a lipid CAG largely consisting of digalactosyl/glucosyldiacylglycerol (DGDG) (lipid CAG 11) came to our attention as DGDG is reportedly derived from bacteria^20,21. These lipids contain glucose and/or galactose in their structures, although their biological functions in mammals are largely unclear. Most of the DGDGs in this cluster showed positive correlations with some of the precursor diacylglycerols and monosaccharides (that is, glucose and galactose) (Extended Data Fig. 4a). As diacylglycerols are deeply involved in IR pathogenesis²², the biological functions of this metabolite class are of particular interest. Notably, DGDGs with different acyl chains in lipid CAG 41 showed no association with IR (Supplementary Table 7), implying that the differences in acyl chains of lipids may have a physiological importance as reported previously²³.

Microorganism–metabolite relationships in IR

We next investigated the alteration in gut microbiota and the functions of gut microbiota that are associated with IR. Gut microbiota diversity varied among individuals (Extended Data Fig. 5a–e). We then profiled the genus-level microbial composition of the study participants using 16S rRNA sequencing data²⁴ and identified four bacterial groups (Extended Data Fig. 5f). Group 1 was dominated by the Lachnospiraceae family such as Blautia and Dorea, whereas group 2 was characterized by Bacteroidales (such as Bacteroides, Parabacteroides and Alistipes) and Faecalibacterium. Group 3 contained Actinobacteria genera. Group 4 did not form a distinct network. We could further classify the study participants into four clusters, A to D, on the basis of their taxonomic profiles (Fig. 2a). Individuals in cluster C distinctly harboured group 2 with Bacteroidales, whereas those in cluster D showed a higher abundance of group 1 and 3 bacteria (Extended Data Fig. 5g). Notably, the proportion of IR (Fig. 2a; P = 0.0071) was significantly lower in cluster C. Other metabolic parameters associated with IR and MetS such as HOMA-IR, BMI, triglycerides, HDL-cholesterol (HDL-C) and adiponectin were also different between cluster C (with the lowest proportion of IR) and the other three clusters (Fig. 2b and Supplementary Table 10). The proportion of IR among individuals with abundant group 1 and 3 bacteria was consistently higher than those with abundant group 2 bacteria, as identified on the basis of shotgun metagenomics data (Extended Data Fig. 5h). HOMA-IR showed negative associations with the genus Alistipes in the Rikenellaceae family and several species from Bacteroides, Bifidobacterium and Ruminococcus (Extended Data Fig. 5i and Supplementary Tables 11 and 12), partly recapitulating previous reports regarding individuals with obesity^25,26,27. Notably, different genera and species correlated with other clinical markers, suggesting that the individual association between microbial taxa and clinical manifestation is not as robust as in the co-abundance analysis.