Insulin resistance is the main pathophysiological basis of metabolic syndrome and type 2 diabetes. Previous metagenomic studies have characterized the gut microbiota and its role in metabolizing major nutrients in insulin resistance. In particular, carbohydrate metabolism by commensal bacteria is thought to provide up to 10% of the total energy to the host, thus playing a role in the pathogenesis of obesity and prediabetes.
However, the underlying mechanism remains unclear. Here, we use a comprehensive multi-omics strategy to study this relationship in humans. We combine unbiased fecal metabolomics and metagenomics, host metabolomics, and transcriptomics data to characterize the involvement of the gut microbiota in insulin resistance. These data suggest that carbohydrates in feces, particularly host-available monosaccharides, increase in insulin-resistant individuals and are associated with carbohydrate metabolism in microbes and inflammatory cytokines in the host.
We identified gut bacteria associated with insulin resistance and insulin sensitivity and found that they showed different patterns of carbohydrate metabolism and demonstrated that bacteria associated with insulin sensitivity can ameliorate the host phenotype of insulin resistance in a mouse model. Our study provides a comprehensive perspective on the host-microbial relationship in insulin resistance, reveals the effects of microbial carbohydrate metabolism, and provides a potential target for improving insulin resistance.
We analysed 306 individuals (71% male) aged between 20 and 75 years (median 61 years) who were recruited during annual health check-ups (Extended Data Figure 1a). Diabetic patients were excluded to avoid the long-term effects of hyperglycemia. Therefore, compared to most previous metagenomics studies on diabetes and obesity, our study included relatively healthy individuals;The median (interquartile range) for body mass index (BMI) and glycosylated hemoglobin (HbA1C) were 24., respectively9 kg m−2（22.2–27.1 kg m 2) and 58%（5.5–6.1%) (Supplementary Table 1). The primary clinical phenotype analysed in this study was insulin resistance (IR), which we defined as a Homeostatic Model Assessment of Insulin Resistance (HOMA-IR) score of at least 25 (Ref. 13).
We also analysed associations between fecal metabolites and metabolic syndrome (METS), a pathology associated with IR. The clinical features of IR and METS largely overlapped in terms other than blood pressure and sex ratio, and there were no differences between individuals with IR and normal insulin sensitivity (IS) (Supplementary Table 1). Untargeted metabolomics analysis using two mass spectrometry (MS)-based analysis platforms identified 195 and 100 annotated fecal and plasma hydrophilic metabolites, and 2654 and 635 annotated fecal and plasma lipid metabolites, respectively (Extended Data Figure 1A).
To determine the overall differences in microbial function, fecal metabolites, and ** genes, the co-abundance groups (CAGs) and KEGG categories were summarized, respectively. Transcriptome information of peripheral blood mononuclear cells (PBMCs) was obtained using the CAGE method, which allows the measurement of gene expression at the resolution of the transcription start site.
To examine how the omics data of stool samples are, we first compared the receiver operating characteristic (ROC) and area under the curve (AUC) based on a random forest classifier. The ** variables of the model were selected from fecal 16s, metabolome, metagenome, and their pooled datasets using the least redundancy maximum correlation algorithm (Supplementary Table 2). We found that selected features of fecal metabolomics data were generally superior to those of 16S and metagenomics in terms of **IR (Figure 1A), suggesting that fecal metabolomics could be used to study the pathogenesis of IR.
Next, we investigated the association between the clinical phenotype and the fecal metabolite CAGS (Figure 1B and Supplementary Tables 3-8). In correlation and regression analyses, the main confounders, i.e., gender and age, were adjusted. Among the hydrophilic metabolites, most of the CAGS that are significantly associated with IR are carbohydrate metabolites, mainly monosaccharides (hydrophilic CAGS and 15;).Figure 1b, top). Short-chain fatty acids (SCFAS), which are known to be products of carbohydrate fermentation, are also increased in IR (hydrophilic CAG 8). Hydrophilic CAG remains unannotated because it includes metabolites from different pathways (Supplementary Table 5). Using KEGG pathway enrichment analysis, it was found that metabolites in these IR-associated hydrophilic CAGS were indeed involved in carbohydrate metabolism (Extended Data Figure 2A). Specifically, we found that the major monosaccharides, such as sugar, galactose, mannose, and xylose, were significantly associated with IR (Figure 1C). In SCFAS, propionic acid is specifically increased in IR (Extended Data Figure 2B), which is consistent with its role in gluconeogenesis. There was a similar increase in monosaccharides in stool in METS, obesity, and prediabetes (Figure 1D and Extended Data Figure 2C,D). Conversely, disaccharides showed weak or no association (extended data figures 2b-d). These findings suggest that the end products of carbohydrate degradation, such as monosaccharides, can be directly absorbed and utilized by the host, especially increased in the feces of individuals with IR and Mets. Supporting these findings, our analysis of previously published fecal metabolomics data from the Twins UK cohort revealed that monosaccharides in feces, especially glucose and arabinose, were positively associated with obesity and HoMA-IR, both of which were associated with IR (Extended Data Figures 3A-C and Supplementary Table 9). Similarly, in a small number of individuals without inflammatory bowel disease (IBD) from HMP2 data, the peak intensities of fructose, glucose, and galactose in feces correlated with BMI (Extended Data Figure 3D). These findings suggest that carbohydrates in feces increase in IR and associated pathologies, and this alteration is consistently observed across different populations.
In addition to hydrophilic metabolites, fecal lipid CAGS are also associated with IR (Figure 1B). Lysophospholipids, bile acids, and acylcarnitine are associated with IR and METS, which is consistent with previous reports. Among them, a lipid CAG composed primarily of bisgalactose glucodiacylglycerol (DGDG) (lipid CAG 11) caught the author's attention because DGDG** was reported to be associated with bacteria. These lipids contain glucose and/or galactose in their structure, although their biological function in mammals is largely unknown. Most of the DGDGs in this category are positively correlated with some precursors, diacylglycerol, and monosaccharides (i.e., glucose and galactose) (Extended Data Figure 4A). Since diacylglycerol plays an important role in the pathogenesis of IR, the biological function of such metabolites is particularly important. Notably, DGDGs with different acyl chains in lipid CAG 41 are not associated with IR (Table 7), implying that differences in lipid acyl chains may have previously reported physiological importance.
Figure 1Fecal carbohydrate metabolites are markedly changed in IR. A, left: Insulin resistance (IR) based on genus-level 16S data (n=282), Kegg Orthologue (KO) level metagenomic data (N=266), fecal metabolome, and metagenomic (Kegg Orthologue)+ fecal metabolome data (N=266) using AUC values from the random forest classifier. On the x-axis, the number of selected feature markers gradually increases. On the right, the boxplot shows the AUC values obtained for the selected feature. B, fecal hydrophilic metabolites (Hydrocag, top) and lipid metabolites (LipiDCaG, bottom) and clinical phenotypes and markers (n=282). The two-column heatmap on the left represents the correlation of the primary clinical phenotypes (IR and METS) using ranking-based linear regression analysis, while the main heatmap shows the partial Spearman correlation (PSC) adjusted for age and sex with representative metabolic markers. Only padj < 005 cags coloring. Psc between C, Homa-IR and fecal monosaccharide levels. Coefficients (PSC) and PADJ values (n=282) are described. d, fecal monosaccharide levels in mets (n=306).
We went on to look at the alterations of the gut microbiota associated with IR and their function. The diversity of the gut microbiota varies between individuals (Extended Data Fig. 5A-E). We analyzed the genus-level microbial composition of the study participants using 16S rRNA sequencing data and identified four bacterial groups (Extended Data Figure 5F). Group 1 is dominated by bacteria of the Lachnospiraceae family, such as Blautia and Dorea, while group 2 is characterized by bacteroidales (such as bacteroides, parabacteroides, and alistipes) and faecalibacterium. Group 3 contains the genera of Actinomycetes. Group 4 does not form a distinct network. We can also divide the study participants into four clusters, A to D, based on their taxonomic characteristics (Figure 2A). Individuals in cluster C are clearly enriched in group 2 of bacteroidales, while individuals in cluster D show higher abundance of bacteria in groups 1 and 3 (extended data Figure 5g). It is worth noting that the scale of IR (Figure 2A;p = 0.0071) was significantly lower in cluster c. Other metabolic parameters associated with IR and METS, such as HoMA-IR, BMI, triglycerides, high-density lipoprotein cholesterol (HDL-C), and adiponectin, also differed between cluster C (the cluster with the lowest IR ratio) and the other three clusters (Figure 2B and Supplementary Table 10). The proportion of IR was consistently higher in individuals enriched with group 1 and 3 bacteria than in individuals enriched with group 2 bacteria, which was identified based on metagenomic sequencing data (extended data Figure 5h). Homa-IR was inversely correlated with the genus Alistipes of the family Rikenellaceae and several species from Bacteroides, Bifidobacterium and Ruminococcus (Extended Data Figure 5i and Supplementary Tables 11 and 12), partially reproducing previous reports of obese individuals. Notably, different genera and species correlate with other clinical markers, suggesting that individual associations between microbial taxa and clinical manifestations are not as robust as those of coabundance analysis.
Next, we constructed a microbial-metabolite network based on significant positive or negative correlations (Supplementary Table 13). Although SCFas and lipids in feces, such as DGDG, are associated with both IR- and IS-related bacterial groups, IR-related fecal carbohydrates are mainly associated with genera in groups 1 and 4, the most prominent of which is Dorea in the Lachnospiraceae family (Fig. 2C, D). In contrast, most of these carbohydrates are inversely correlated with is-related genera in group 2 bacteria, such as bacteroides, alistipes, and fl**onifractor (Figure 2d and extended data Figure 5j), with little correlation with bacteria in group 1. As a result, there was a clear difference in carbohydrate levels in the feces between clusters of participants (Extended Data Figure 5k). Previous studies have shown that some species of the Lachnospiraceae family are involved in the fermentation of polysaccharides, while alistipes are increased in animal-based diets rather than diets rich in polysaccharides. These findings highlight the strong link between carbohydrate degradation products and IR- and IS-related bacteria, suggesting that these bacteria may be involved in abnormalities in fecal carbohydrate profiles in IR.
IR-associated fecal carbohydrates are also associated with KEGG pathways associated with carbohydrate metabolism and transport, such as phosphotransfer system (PTS), starch and sucrose metabolism, galactose metabolism, while negatively correlated with pathways associated with carbohydrate breakdown, such as glycolysis and pyruvate metabolism (Figure 2E and Supplementary Tables 14 and 15). These pathways also have a clear correlation with the participant clusters defined in Figure 2a and the carbohydrate-related genera defined in Figure 2d. Amino acid metabolism also varied, especially between clusters C and D, while lipid metabolism did not show a clear association with the microbiota (Extended Data Fig. 6a,b and Supplementary Table 16). Although carbohydrate pathways such as PTS and starch and sucrose metabolism are strongly positively correlated with HbA1c and -GTP, the correlation with other IR markers is relatively sparse (Extended Data Figure 6C and Supplementary Table 17), suggesting that metabolites are more sensitive to clinical manifestations, as shown in Figure 1A. PTS is an essential component for bacteria to transfer sugars as energy** to themselves. Detailed analysis of KEGG congeners revealed that fecal carbohydrates and participant clusters were primarily associated with PTS associated with disaccharides and amino sugars (Extended Data Figure 6D,E and Supplementary Table 18), suggesting that microbial preference for glycohydrate utilization through PTS may affect metabolite levels. Glycosidases that hydrolyze oligosaccharides and disaccharides are also associated with fecal monosaccharides (Extended Data Figure 6f). Extracellular glucosidases, such as -fructosidase (K01193, KEGG Homolog Database), starch sucrase (K05341, KEGG Homolog Database), and oligosaccharide-1,6-glucosidase (K01182, KEGG Homolog Database)* can hydrolyze sucrose and dextrin to glucose and fructose (Extended Data Figure 6g, H), showing the highest positive correlation, especially with fecal glucose. In contrast, glucosidases associated with starch utilization, such as -amylase (K01176 and K07405, KEGG homolog database), are inversely correlated with fecal carbohydrates. Importantly, the abundance of these glycosidase genes differed significantly between the participant cluster c and the other three clusters defined in Figure 2A, suggesting that taxonomic signatures largely explained the variation in glycosidase (Figure 2f, extended data in Figure 6h and Supplementary Table 18). Consistently, genes that break down disaccharides are predominantly retained in the genomes of Blautia and Dorea that are more abundant in cluster D, while these genes are almost absent in the genomes of bacteroidales, which are more abundant in cluster C (Extended Data Figure 6i). Taken together, our findings reveal four distinct groups with unique taxonomic features and carbohydrate metabolism that are characterized by the utilization and degradation of carbohydrates, associated with IR and its associated markers.
Figure 2IR-associated fecal metabolites are associated with alterations in the gut microbiota and microbial genetic function. a, co-abundance clustering of genus level bacteria and their abundance (n=282). Participants were divided into four clusters A to D based on their taxonomic characteristics. The proportion of patients with IR is shown. B, Homa-IR, BMI, triacylglycerol (TG), and high-density lipoprotein cholesterol (HDL-C) levels in participant clusters. c, Co-emergence of bacteria-metabolite networks between microbiota and fecal metabolites (n=282). Includes all fecal hydrophilic and bacterial-associated lipid metabolites. Only positive and significant (padj < 005) Spearman correlation. Metabolites in CAGS related to carbohydrates in Figure 1b are highlighted in red. unclust.Indicates unclustered. d, the number of positive and negative correlations between bacteria and fecal carbohydrates. The top five genera in each correlation are shown. e, the KEGG pathway associated with carbohydrate metabolism and membrane transport, fecal carbohydrates, the top three genera positively or negatively correlated with fecal carbohydrates, and participant clusters. F, abundance of representative kegg orthologues involved in glycosidases in the participant cluster (n=266). The abundance is converted by square root arcsine.
Consistent with previous reports, the cytokine, metabolomic, and transcriptomic signatures of the host are highly correlated with IR (Supplementary Table 19-21). In addition, these PBMC genes are functionally involved in inflammation (Extended Data Figure 7A) and may be derived from monocytes (Supplementary Table 21). Some studies have shown that microbial components such as lipopolysaccharides play a role in promoting inflammation in metabolic diseases. However, it remains unclear whether microbial metabolism is involved in low-grade inflammation. Therefore, we attempted to infer a possible association between the inflammatory profile of host IR and fecal carbohydrates. First, based on the cross-omics correlation network of individual metabolites, bacteria, transcripts, and cytokines associated with IR, fecal carbohydrates showed a strong association with bacterial and host IR-related features, particularly cytokines, suggesting that these metabolites are the hubs of the host-microbial network in IR (Figure 3A, Extended Data Figure 7B, C and Supplementary Table 22). Differential abundances, calculated as their abundance ratios in IR and IS, are most evident in the association between fecal carbohydrates and cytokines. Notably, IL-10, a plasma cytokine, showed the most prominent association with fecal carbohydrates, while the moderate association with transcripts of PBMC** supported recent studies showing its paradoxical effect of promoting IR. Fecal carbohydrates moderately explain the variance of IL-10 and, to a lesser extent, the variance of adiponectin, leptin, and serpin E1, suggesting that fecal carbohydrates are specifically associated with these cytokines (Figure 3B). Although the proportion of variance explained by fecal carbohydrates is lower than that of plasma metabolites, the abundance is much higher than that of the genus level, highlighting the role of fecal metabolites in bridging gut microbes and host inflammatory responses. We next attempt to infer whether these cytokines mediate the effects of fecal carbohydrates on host metabolism through causal mediation. We found that IL-10, Serpin E1, adiponectin, and leptin mediated most of the in vitro causal relationships between fecal carbohydrates and host IR markers such as Homa-IR (Figure 3C, Extended Data Figure 7D, and Supplementary Table 23). It is important to note that there is a unique correspondence between metabolites and cytokines;For example, IL-10 mediates the effects of fructose, mannose, xylose, and rhamnose, but not the effects of other metabolites. While the biological significance of these unique correspondences remains to be studied, combined analysis of fecal microbiome, metabolome, and host inflammatory phenotypes in IR reveals a previously unrecognized interaction in which excess monosaccharides may affect host cytokine expression.
Figure 3Fecal carbohydrate metabolites correlate with cytokine levels in IR. A, based on IS (insulin sensitive), moderate (i.e., HoMA-IR >1.)6 and < 25) and all omics information (N and 275) of IR (Insulin Resistant) samples to construct a network between fecal carbohydrate metabolites (purple), fecal bacteria (green), plasma hydrophilic metabolites (pink), cytokines (yellow), and PBMC genes (red). The analysis included host-derived markers significantly associated with IR (see Supplementary Table 19-21), 15 fecal carbohydrates and 20 genera identified in Figure 1b and Figure 5f of the extended data, respectively. To construct the omics network, paired PSCs adjusted for age, sex, BMI, and FBG were calculated and a PADJ < 005 interactions. The width of the line represents the absolute value of the coefficient, and the red and gray lines represent positive and negative correlations, respectively. b, Explanatory variance of 10 plasma cytokines using a random forest classifier **per omics dataset**. C, alluvial plot showing plasma cytokines significantly mediating the silico effect of fecal carbohydrates on host metabolic markers. Lines show mediating effects, and colors indicate associations mediated by a single cytokine.
The above findings of human multiomic analysis reveal the association between carbohydrate metabolites and IR pathology. To determine the causal relationship between gut microbiota, fecal carbohydrates, and metabolic diseases, we first analyzed metabolites in bacterial cultures of 22 human fecal IS- and IR-associated bacteria. These bacteria were selected based on the findings of genus-level symbiosis (Figure 2a, b) and species-level (extended data, Figure 5i) characteristics. Principal component analysis plots of 198 metabolites show that bacteroidales, a representative order of IS-associated bacteria, show different metabolic profiles on PC1 (Extended Data Fig. 8a,b and Supplementary Table 24). The top 10 metabolites leading to separation between groups include several amino acids and fermentation products such as succinic acid and fumaric acid, most of which are preferentially produced by bacteroidales (Extended Data Figure 8B,C). We detected 13 of the 15 carbohydrates associated with IR in bacterial cultures (Figure 1B) (Extended Data Figure 8B). Most of these carbohydrates are arranged negatively along PC1, suggesting that these metabolites are inversely correlated with bacteroidales. Glucose, mannose, and glucosamine are preferentially consumed by bacteroidales compared to other mela, while lactulose is mainly produced by eubacteriales (Extended Data Figure 8D). Alistipes Indistinctus is the bacteria that most consume a variety of carbohydrates (Extended Data Figure 8E,F). These findings suggest that bacteroidales species are potent consumers of several carbohydrates, driving the production of their fermentation products.
We next tested the potential effects of seven candidate bacteria associated with IS found in the human cohort. In mice on a high-fat diet, AMice of Indistinctus, Alistipes Finegoldii, and Bacteroides Thetaiotaomicron had particularly reduced postprandial blood glucose levels (Figure 4A). Insulin resistance tests have also shown that these strains improve IR, most notably AAdministration of indistinctus (Figure 4B, C). a.Administration of Indistinctus improved weight gain, hepatic ectopic triglyceride accumulation, and glucose tolerance (Extended Data Fig. 9A-D). a.Serum high-density lipoprotein cholesterol (HDL-C), adiponectin levels, and triglyceride levels were also reduced in mice with Indistinctus** (Extended Data Figure 9E-G). The results of hyperinsulinemic glycemic constant-clamp analysis showed that aAdministration of Indistinctus significantly improved IR, particularly systemic glucose clearance (Extended Data Figure 9H-J). a.indistinctus and aIncreased Akt phosphorylation in liver and epididymal fat in mice of FineGoldII** (Extended Data Figure 9K,L), indicating an improvement in insulin signaling in liver and adipose tissue. These findings reveal aThe potential of Indistinctus administration in ameliorating diet-induced obesity and IR.
Mechanistically, metabolic measurements show, aMice of indistinctus** showed a significant reduction in carbohydrate oxidation, indicating that carbohydrate utilization was limited (Extended Data Figure 9M and Supplementary Table 25). Since dietary intake and exercise activity remain unchanged (Extended Data Figure 9n,o), we infer aIndistinctus** reduces the amount of carbohydrates available to the host in the gut. In this regard, aAdministration of Indistinctus significantly alters cecal metabolites, which are characterized by a decrease in several carbohydrates, including fructose, a monosaccharide synthesized by fat (Extended Data Figures 10a-c and Supplementary Table 26). A similar decrease in fructose was observed in serum (Extended Data Figure 10d). Importantly, AUC for insulin resistance testing was positively correlated with cecal monosaccharides, glucose, and mannose (Figure 4D). Taken together, these findings suggest that aIndistinctus improved IR in mice and affected gut carbohydrate metabolites, supporting our observations in human cohorts.
Figure 4In experimental models, IS-associated bacteria improved IR. A, Postprandial blood glucose at 4 weeks after bacterial administration in mice fed a high-fat diet. n=12 (control group), n=10 (a.).indistinctus and afinegoldii group), n=5 (other groups) mice. B and C, blood glucose levels during the insulin tolerance test (B) and AUC (C) (n=5 per group). D, AUC and A. of insulin tolerance testCorrelations between fructose, glucose, and mannose levels in the cecum of the indistinctus (blue) or control (gray) groups.
To deepen our understanding of the host-microbe relationship in IR, we conducted a comprehensive and extensive study using multimodal techniques to investigate the interactions between the human gut microbiome and metabolic disease. Although the metabolism of carbohydrates by gut microbes has been thought to influence the pathogenesis of obesity and prediabetes, the actual mechanistic link in humans has been elusive due to the lack of detailed metabolomic information. In this regard, the main advantage of our approach is that we combine fecal metabolomics, cataloging more than 2800 annotated metabolites, as well as microbiome and host pathology. This metabolome-based approach allowed us to identify fecal metabolites associated with IR, to discover associations between fecal carbohydrates and low-grade inflammation of IR, and to efficiently select candidate strains for functional validation under experimental conditions (Extended Data Figure 10E). Taken together, our study highlights the advantages of a comprehensive omics strategy in exploring the role of microbial metabolism and its products in the pathogenesis of IR. Excess monosaccharides have the potential to promote ectopic lipid accumulation while activating immune cells, leading to low-grade inflammation and IR. Fructose is a well-known risk factor for inflammation and IR because of its role in lipid accumulation, while galactose has been shown to be involved in energy metabolism in activated immune cells. Our in vivo study confirms that aAdministration of Indistinctus improved lipid accumulation and IR while reducing intestinal monosaccharide levels (Figure 4D). However, we realized that further mechanistic studies were needed to examine the kinetics of absorption and its effects on host metabolism. In particular, how the Alistipes strain inhibits carbohydrate metabolism is an interesting question (e.g., whether these bacteria themselves inhibit carbohydrate metabolism, or whether they interact with other commensal bacteria), as it will directly open up the possibility of a new strategy. Considering aIndistinctus improved systemic IS (Extended Data Figure 9i), and it is important to study the involvement of insulin signaling in the liver and surrounding tissues, including skeletal muscle and adipose tissue, as well as the accumulation of specific lipid molecules such as ceramides and diacylglycerols in these tissues. These studies have the potential to reveal what led to aUnderlying mechanisms of indistinctus-mediated IR improvement. Finally, two participants in the human study were unable to collect their feces in the morning, which could affect the results due to a lack of strict control over the time of day and fasting conditions. Therefore, we believe that longitudinal studies are needed to record dietary Xi in a timely manner to dissect the complex effects of microbial metabolism on diabetes and its complications while considering potential confounders.
This article is translated from: Takeuchi T, Kubota T, Nakanishi y et al gut microbial carbohydrate metabolism contributes to insulin resistance. nature. 2023 sep; 621(7978):389-395.