The Gut Microbiota and Diabetes: Research, Translation, and Clinical Applications—2023 Diabetes, Diabetes Care, and Diabetologia Expert Forum

In October 2023, a closed-door, day-long forum organized under the auspices of the journals Diabetes, Diabetes Care, and Diabetologia took place during the European Association for the Study of Diabetes 2023 Annual Meeting in Hamburg, Germany. The express goal of the forum was to create consensus and perform gap analyses to advance research into the role of the gut microbiota (GM) in diabetes. Discussions fell under four main headings: epidemiology; physiology and pathophysiology; technology and methodology; and clinical applications.

The group acknowledged that many of the gaps in understanding of the GM’s role in metabolic diseases are not unique to the diabetes field, but rather reflect broader needs to 1) conduct more well-controlled prospective and retrospective human studies that are followed up mechanistically with model systems studies and 2) refine computational tools and welcome a return of microbiology and molecular biology to our experimental toolkit. Nonetheless, there was agreement that the current reproduced microbiome data represent compelling target areas for future diabetes research. This article presents a distillation of the evidence and recommendations on important microbiome focus areas that would benefit from the attention of young and established diabetes researchers alike. Key knowledge gaps and challenges discussed in this article are summarized in Table 1.

The GM is the largest and most complex microbial community of the human body, connecting our external and internal milieu (Fig. 1). The motivation for epidemiological studies of the GM in obesity and cardiometabolic diseases, including type 2 diabetes, emerged from rodent studies demonstrating links among the GM, adiposity, and glucose tolerance (1,2). In humans, epidemiological studies have observed decreased microbial diversity in obesity, but no generalizable obesity-associated gut microbial signature has emerged from meta-analyses of small cohorts profiled by 16S rRNA gene sequencing (3,4) or whole-genome metagenomics (5). However, a large GM study using deep-sequencing whole-genome metagenomics in 34,057 individuals from Israel and the U.S. demonstrated consistent GM-phenotype associations and the predictive accuracy of machine learning models trained on microbiome data for BMI and glycated hemoglobin (A1C) that could be replicated across the cohorts (6). By subsampling the training cohort, these authors showed increased predictive accuracy with increased cohort size, with ∼7,500–10,000 individuals optimal for replicable results. This finding highlights the necessity of using large cohorts with hundreds of individuals and deep-sequencing whole-genome metagenomics that adequately represent the embedded interindividual heterogeneity and regional and demographic variation in human GM cross-sectional studies.

Several observational studies have reported associations between the GM and type 2 diabetes. Consistent features of altered GM composition in type 2 diabetes and impaired glucose tolerance/fasting glucose, found in epidemiological studies worldwide and also occurring in the metabolic syndrome, are reduced diversity and decreased abundance of bacteria that produce the short-chain fatty acid (SCFA) butyrate (Figs. 1 and 2) (7–14). Some studies have also observed an increase of opportunistic pathogens (7,8,11), some of which have been linked to subclinical coronary atherosclerosis (15). The mucus-degrading bacterium Ruminococcus gnavus, which has been linked to inflammatory bowel diseases, recently has also been identified as a predictor of several features of the metabolic syndrome, including low-grade inflammation, large waist circumference, elevated serum triglycerides, elevated A1C, and decreased HDL cholesterol (16). However, as indicated by meta-analyses of GM alterations across different diseases, including gastrointestinal (GI) and metabolic diseases, several of these features are not disease specific and might characterize a general GM dysbiosis (4,5). Therefore, to disentangle disease-specific microbial signatures beyond differences in race/ethnicity, lifestyle, and other demographic characteristics, it will be important to perform studies in large populations and to include healthy individuals/control participants from different studies as references; these approaches have been shown to increase disease prediction accuracy (5,6).

In addition to a decreased capacity to produce butyrate, GM functional capabilities that are altered in type 2 diabetes are involved in the production of branched-chain amino acids (BCAAs) and the metabolism of B vitamins and simple sugars (7,12,17,18). Increased levels of circulating BCAAs have been described in insulin-resistant individuals and linked to a higher risk of type 2 diabetes (19). In line with this, an increased potential to synthesize BCAAs and decreased microbial BCAA uptake and catabolism have been described in the GM of insulin-resistant individuals with normoglycemia (18). However, the analyses of GM functions only show an altered potential. Quantification of metabolites has recently been performed to validate these findings. Figure 2 depicts the links between GM metabolites and signaling molecules that have been observed for glucose metabolism and type 2 diabetes.

Evidence supporting the role of the GM in type 2 diabetes has been strengthened by observational and interventional studies demonstrating changes in the relative abundance of multiple bacterial species in the GM of metformin users (10,20–22). A higher relative abundance of Escherichia coli and a decreased abundance of Intestinibacter bartlettii (10,20,21) have been described in multiple cohorts involving individuals being treated with metformin. Additionally, an increase in Escherichia marmotae and a decrease in Romboutsia timonensis have been found in metformin-treated individuals in a recent large metagenomic study (23).

Support for the causal effects of these GM differences in type 2 diabetes has been provided by randomized trials and studies in drug-naive individuals demonstrating that the GM compositional changes translate to enhanced production of propionate and butyrate (20,21) and modulation of bile acid pools (21), which may mediate some of the glucose-lowering effects of metformin (Fig. 3) (20–22,24). However, the GM might also be responsible for the transient or persistent intestinal discomfort experienced by ∼30% of individuals who take metformin (e.g., through increased gas production by some Escherichia species) (22,25).

With regard to other oral glucose-lowering drugs, studies have shown effects of dipeptidyl peptidase 4 (DPP-4) inhibitors and α-glucosidase inhibitors on the GM and microbial metabolites but less clear effects of sodium–glucose cotransporter 2 (SGLT2) inhibitors, thiazolidinediones, and glucagon-like peptide 1 (GLP-1) receptor agonists (26–28). The majority of studies to date involving SGLT2 inhibitors have been conducted in mouse models, and the few existing human studies have provided contradictory results and were unable to clearly discriminate the effects of the SGLT2 inhibitor from those of previous or concomitant metformin treatment or concurrent lifestyle modifications (29). GLP-1 receptor agonists may exert anti-inflammatory effects (e.g., through activation of the intraepithelial lymphocyte GLP-1 receptors), which in turn could contribute to modulating the gut microbiota (30,31). Although much more research is needed, existing evidence suggests that the GM may mediate some of the benefits of glucose-lowering treatments (26), and certain probiotics or prebiotics might further improve the glucose-lowering effects of these drugs through their effects on the GM or its functions (32). Further interventional and translational studies are needed to determine whether drug-induced GM changes are causally involved in mediating health effects and to uncover the underlying mechanisms.

Importantly, the GM might also influence the efficacy of glucose-lowering drugs, for example, by expressing homologs of human DPP-4, which can decrease the activity of GLP-1 and affect glucose metabolism (Fig. 3) (33,34). Because bacterial DPP-4 homologs seem resistant to some drugs targeting human DPP-4 (33), inhibition of bacterial isozymes might be required to improve metabolic responses to current medications.

The GM ferments plant-based dietary carbohydrates and fiber, as well as peptides that reach the large intestine, to produce SCFAs—mainly acetate, propionate, and butyrate. After hepatic metabolism, ∼70% of colonic acetate, but only small amounts of propionate and butyrate (<2% for butyrate), reach the circulation (35). As described in more detail in physiological and pathophysiological perspectives, SCFAs regulate several processes, including intestinal motility and pH, gut barrier immune responses and systemic metabolism through pathways affecting gluconeogenesis, insulin sensitivity, and insulin secretion (Fig. 2) (36). However, human studies show extensive variation in the levels of different SCFAs in the stools and/or blood of individuals with type 2 diabetes, which is likely because of methodological limitations (36). The strongest support for the role of SCFAs in the regulation of glucose metabolism is provided by animal studies and one recent human study using the Mendelian randomization (MR) statistical method (36,37) (see role of mr in elucidating causal effects, below).

Bile acids are amphipathic molecules that mediate the absorption of dietary fats and lipid-soluble vitamins. These molecules are also recognized as major players in regulating lipid, glucose, and energy metabolism. Consequently, alterations in bile acid pools have been found in type 2 diabetes and other obesity-related diseases and identified as possible contributors to the pathophysiology of type 2 diabetes (Fig. 2) (24,38–41).

Increased levels of 12α-hydroxylated bile acids (41) and decreased levels of 6α-hydroxylated bile acids (42,43) are linked to insulin resistance and occur in people with type 2 diabetes. Increased levels of 6α-hydroxylated bile acids are observed after gastric bypass surgery and can predict type 2 diabetes remission (42). The GM can deconjugate and transform bile acids, thus contributing to a highly variable but important portion of human bile acid pools (Fig. 1) (44). For example, circulating levels of 6α-hydroxylated bile acids are found to co-vary with levels of specific Clostridia species in the gut (43).

Intervention studies have also investigated the potential importance of bile acids in human metabolism. Elevated systemic bile acid levels and intestinal signaling to stimulate the release of GLP-1 have been demonstrated after bariatric surgery, with postprandial increases found to be particularly important (45). However, exaggerated bile acid responses have been found in some individuals with cholecystectomy and are associated with further increased GLP-1 and insulin responses (46). In people with type 2 diabetes, metformin has been shown to improve glucose metabolism via a decreased abundance of Bacteroides fragilis, which has been linked to increased levels of glycoursodeoxycholic acid in the gut and inhibition of the farnesoid X receptor (FXR) (24). However, our understanding of direct interactions between the GM and bile acids and their associations with the development and treatment of type 2 diabetes and related diseases is still limited, and more human studies are warranted.

GM-produced amino acid metabolites have also been linked to type 2 diabetes (Figs. 1 and 2). Increased circulating levels of 3-indolepropionic acid, a tryptophan catabolite, have been associated with improved insulin secretion and sensitivity and a decreased risk of type 2 diabetes (47). Furthermore, increased plasma levels of imidazole propionate, a bacterial product of histidine metabolism, have been reported in individuals with insulin resistance and type 2 diabetes (48). These metabolites and several others derived from GM catabolism of aromatic amino acids have also been associated with incident cardiovascular risk and mortality in independent cohorts from Europe and the U.S. (49,50). Finally, GM ethanol production has been associated with fatty liver disease (51–53) and might be linked to insulin resistance (Fig. 2).

The GM can affect and interact with host health in numerous ways, and the arrow of causality is often bidirectional or even multidirectional. GM features at different levels (e.g., community, species, pathway, gene, and metabolite) can affect a host phenotype (e.g., altering the risk of obesity), while the development of a phenotype (e.g., obesity) can, in turn, change the GM.

MR is a statistical method that uses human genetic variants related to exposures to discriminate causal effects on disease outcomes from associations that result from confounding, reverse causation, or something else. To apply MR to investigate the connection between the GM and type 2 diabetes, the GM feature in question needs to be affected by a human genetic variant or multiple variants strongly enough to allow their use as instruments in instrumental variant analysis.

Although several genome-wide association studies have been performed on different GM features such as gut bacterial taxa relative abundances and human fecal microbial metabolites, large MR studies investigating their causal role in type 2 diabetes have been limited and are not yet confirmed in replication studies. Sanna et al. (37) identified human genetic variants that associate with fecal SCFA levels and reported evidence for a potential causal connection between the GM’s butyrate production potential (i.e., genes responsible for GM butyrate production) and improved response to insulin during an oral glucose tolerance test. These authors also found a causal link between abnormal fecal propionate levels and increased type 2 diabetes risk (37). Another MR study reported that type 2 diabetes and kidney disease increased plasma levels of the GM-dependent metabolite trimethylamine oxide (TMAO) and proposed that the earlier observational evidence of elevated risk of cardiovascular diseases with higher TMAO levels might have been the result of confounding or reverse causality rather than a causal effect (54). Another recent study suggested that certain bacterial genera could have a causal link to type 2 diabetes (55). Considering the limitations of both MR (e.g., pleiotropy and problems related to weak instrumental variables) and GM research (e.g., methodological differences, interindividual heterogeneity, and intraindividual variability), large, high-quality studies are needed to assess the ability of host genetic variants using MR to mimic specific GM features—whether specific bacterial species, genera, or metabolite products—to understand causal connections with type 2 diabetes pathogenesis.

Decreased dietary fiber intake has been associated repeatedly with increased risk of type 2 diabetes; accordingly, new dietary recommendations for diabetes management encourage high consumption of minimally processed plant foods such as whole grains, vegetables, whole fruit, legumes, nuts, and seeds (56). Diet is a driver of the GM ecosystem, and microbially accessible carbohydrates promote GM diversity and SCFA generation, which decreases inflammation and supports the maintenance of the gut barrier (57).

In relation to the GM and glucose metabolism, increased fiber intake has been associated with increased levels of distinct species, for example, Prevotella copri (58) (now renamed Segatella copri). Studies have also shown that the beneficial effects of fibers on A1C may be mediated by the specific baseline GM composition and diversity of fiber-promoted SCFA-producing bacteria (59). However, variable effects are observed even in well-controlled dietary interventions (60), and given the high interindividual variability of the GM, dietary responses of the GM are highly individualized (61). Precision, or “personalized,” nutrition is an evolving field based on identifying individual-specific response-predictive features that can be used to design dietary interventions (62). Using personal data on GM composition and other information such as blood biomarkers and dietary habits, machine-learning approaches have been applied to predict postprandial glycemic responses to standardized meals with greater accuracy than other predictive methods (63,64). These studies have revealed that the specific composition of the GM contributes to the specific response of its host (i.e., the response to the diet differs in the presence of different bacteria). Hence, the GM determines, at least in part, metabolic heterogeneity among humans. Being modifiable and highly metabolically active, the GM offers possibilities for more precise lifestyle interventions and novel treatments.

Several large, high-quality reference genome catalogs now exist (65–67) and greatly facilitate taxonomic assignment and functional characterization of the GM in human studies. However, these databases are not without limitations (Table 1). For epidemiological analyses, GM data are fraught with challenges, including great inter- and intraindividual variability, high dimensionality (i.e., the number of observed GM features may be larger than the number of samples and subjects), and sparsity (i.e., GM features such as species are only detected in part of the samples) (68). At the population level, the GM is composed of thousands of interacting species, each harboring genetic diversity across hosts and within a host over time; yet, commonly performed analyses often ignore such nonindependence, the complex additive and interaction effects among the microbes, and the modifiability and fluctuations of the GM. However, some recent analyses have revealed different patterns of intraindividual variation and adaptation to host physiology for different bacterial species (14,69,70).

Other challenges relate to the remarkable number of phenotypic and environmental factors that the GM may influence and to which it may respond. The requirement of large cohorts has unquestionably been demonstrated in human genetics; most polygenic traits are known to be affected by many genetic variants with small effect sizes, which nonetheless can be summed to powerful polygenic risk scores of clinical importance (71,72). Similarly, as evidenced by the findings of the large metagenomic study from Israel and the U.S. (6), single bacterial species might have associations of low effect size with human phenotypes or be present in low abundance. Thus, large sample sizes for adequate statistical power and coverage of interindividual variability are necessary to obtain replicable results and high prediction accuracy.

To better understand the long-term influence of GM variation and dynamics on type 2 diabetes, prospective studies are crucial. In the few prospective studies published to date, GM features have been associated with incident type 2 diabetes in a geographically diverse Chinese population (73) and a subset of a Spanish clinical trial (74), both studies using 16S rRNA gene sequencing. GM features were also linked to type 2 diabetes in a large population-based Finnish cohort with 18 years of follow-up using shallow metagenomic sequencing (75). However, in these studies, the number of incident cases was restricted, and the analyses had limited resolution (i.e., were restricted to the most dominant GM taxa), as none of the studies used deep-sequencing whole-genome metagenomics.

The importance of subspecies- and strain-level resolution in metagenomic studies may have been undervalued and is an important limitation to harnessing the GM for human health. For example, Faecalibacterium prausnitzii is among the most promising candidates for next-generation probiotics, but there are also other promising candidates, such as Akkermansia muciniphila and P. copri (76). With regard to F. prausnitzii, several potential subspecies have been found in the human gut, harboring different functional potential for the use of complex polysaccharides (77). In line with this observation, several F. prausnitzii subspecies were also identified in the large metagenomic study from Israel and the U.S., and negative associations with BMI were observed only for a subset of them (6). In the case of P. copri, both positive and negative associations have been found with the host metabolic phenotype (e.g., visceral fat and glucose responses).

These inconsistent findings could be partially explained by intra- and interspecies diversity (78,79). The current code of nomenclature defines bacterial species based on genome similarity, with conspecific genomes having ≥70% similarity by DNA-DNA hybridization and an average nucleotide identity ≥94% in the core genome and ≥96% in universal marker genes (80). However, these genomic variations can translate into important phenotypic differences. For example, these differences may define a strain within the same species as commensal or pathogenic, as in the cases of B. fragilis and Clostridium difficile, depending on whether the strain encodes virulence factors (80). Overall, the studies mentioned above demonstrate that differences at the strain or even substrain level are highly meaningful, and low-resolution analyses (such as 16S rRNA gene sequencing) miss key information.

Another knowledge gap concerns the viral component of the GM, predominantly comprising viruses that infect bacteria, known as bacteriophages (or, more simply, “phages”). Although these phages have not been well studied in the epidemiological setting, they may be important for understanding the bacterial dynamics of the GM that may affect its interactions with the host. To date, only a few epidemiological studies have reported associations between the gut phageome and type 2 diabetes (81,82) or the metabolic syndrome (83). Although initially promising, the conclusiveness of the results is limited because of the restricted sample sizes. Future studies of the role of phages as regulators of the GM and cardiometabolic health are warranted but will face challenges related to, among other issues, virome isolation and the limitations of current databases (84,85).

Integrative multiomics studies might be needed to investigate the intricate connections among environmental factors, the GM, the virome/phageome, and cardiometabolic phenotypes. Some pioneering examples of reasonably large studies have recently demonstrated the power of such approaches (14,86–88). The interactions are multifactorial and multidirectional and demand untargeted, large-sized, multiomics and longitudinal approaches of high depth and resolution.

During their evolution, mammals had to adapt to a world rich in microbes, viruses, and fungi (89). During and immediately after birth from a sterile intrauterine environment, mammals are exposed to potentially harmful microbes (90–92). Evolution has created substantial barriers, including the GI transit process (93), immunoglobulin A (IgA) (94–96), mucus (97), the epithelial layer (98), the endothelial barrier (99), lymph nodes (100), and the liver (101), all of which prevent microbial translocation into the body but create an optimal reservoir for the microbial ecosystem (102). Low microbe numbers are present in the upper GI tract. At the same time, high microbial density and richness are observed in the large intestine, along with physiological changes in the pH and aerobic/anaerobic conditions from the small to the large intestine, with anaerobic conditions in the large intestine (Fig. 1) (103).

Besides being a potential deleterious threat for mammals, gut microbes also provide essential functions for mammals, including the education of the immune system, protection from pathogens (i.e., colonization resistance) (104), metabolic functions, and the supply of nutrients (e.g., vitamins [105]), gut motility, and detoxification of xenobiotics (106). At the same time, there is competition between microbes and the host for nutrients in the small intestine, and microbially produced macronutrient byproducts are provided to the host. Nutrients (i.e., fibers) and mammalian metabolites such as glucuronides, mucus polysaccharides, and bile acids are fermented or transformed by microbial metabolism (Fig. 1) (107). Microbial metabolism and microbial cell death and turnover contribute to pools of microbial metabolites in the peripheral blood, where ∼30% of all peripheral blood metabolites show associations with the GM and its genes (108,109). These microbial metabolites are recognized by receptors such as G-protein–coupled receptors (GPCRs) (110) or the aryl hydrocarbon receptor (AHR) (111) or are further processed by mammalian enzymes such as TMAO to regulate mammalian gene expression by epigenetic modifications, with implications for cardiometabolic health (112–115).

Nondigestible carbohydrates are an energy source for specific bacteria in the large intestine that contain enzymes, lacking in the host, that metabolize these fibers and promote the production of SCFAs. Numerous studies have demonstrated that exogenous administration of SCFAs, particularly propionate and butyrate, is beneficial in rodent models of diabetes-like phenotypes (116–118). However, the evidence from clinical trials in both type 1 and type 2 diabetes is less clear (119–123).

In the colon, SCFAs activate enteroendocrine cells (EECs) via binding to GPCRs and free fatty acid receptors 2 and 3 to induce the release of gut peptides, mainly GLP-1 and peptide YY (Fig. 1) (124). In support of this finding, supplementation with prebiotics in rodents and humans, which can improve glucose tolerance and insulin resistance, has been associated with increased levels of gut peptides (125). In one study, a high-fiber diet improved glucose tolerance in individuals with type 2 diabetes, an effect that was associated with increased fecal butyrate levels and circulating GLP-1 (59). GLP-1 regulates glucose homeostasis by increasing insulin secretion, promoting insulin sensitivity, and reducing hepatic glucose production (Fig. 1).

Additionally, SCFAs are crucial for maintaining overall gut health and the gut barrier, as butyrate is the primary fuel source for colonocytes. In contrast, reduced butyrate drives colonocytes toward anaerobic glycolysis, which increases epithelial oxygenation, disrupting the anaerobic environment of the colon (126). Although SCFAs can act to increase gut peptide release or improve the gut barrier, additional work has highlighted a glucoregulatory role via their action in intestinal gluconeogenesis and on energy expenditure via brown adipose tissue, as well as direct action at the liver, pancreas, and even brain, all of which requires further exploration (127–131).

The GM produces a plethora of metabolites in addition to SCFAs, which likely play a crucial role in host glucose homeostasis (Fig. 2) (132). For example, bile acids are known glucoregulatory signaling molecules, and their affinity for both FXR and Takeda GPCR 5 (TGR5) is significantly affected by deconjugation and metabolism into secondary bile acids coordinated by gut microbes (133–136). Additionally, the GM converts tryptophan and other nutrients into indoles that act via the AHR to reduce inflammation in the metabolic syndrome, especially at the gut level (137,138). Furthermore, other gut-derived molecules such as TMAO and imidazole propionate have been implicated in the development of diabetes (139,140).

The GM plays a vital role in gut barrier functioning. Impairment in the gut barrier leads to a leaky gut, contributing to low-grade systemic inflammation, a characteristic of obesity and diabetes (141,142). Although the mechanisms have been studied mostly in experimental models, one potential mechanism contributing to systemic inflammation is an increase in circulating lipopolysaccharide (LPS) endotoxins derived from the cell envelope of Gram-negative bacteria, also known as metabolic endotoxemia (Fig. 1). LPSs can act on a specific pathogen-associated molecular pattern (PAMP)—toll-like receptor 4 (TLR4)—throughout the body to elicit a proinflammatory immune response that negatively affects glucose homeostasis. A series of studies have suggested a potential role for A. muciniphila in mediating some of the effects of alterations in the GM on systemic inflammation through actions on TLR4 and the gut barrier; however, less evidence is available on its role in mediating effects on glucose metabolism in metabolic disease (143–146). However, much more research is needed to determine whether metabolite sensing by PAMPs other than TLR4 is implicated in regulating host-microbe cross talk and gut barrier integrity (147) in humans.

In parallel, the accumulation of proinflammatory macrophages (Fig. 1), CD8αβ T-cell infiltration, and reduced IgA+ immune cells are observed in the intestines of individuals with obesity (148–150), contributing to insulin resistance (149,150). GM modulation strategies could mitigate the adverse gut immune effects of hypercaloric diets. For example, reducing the proportion of proinflammatory macrophages and increasing type 3 innate lymphoid cells and regulatory T cells are associated with improved glucose metabolism. Nonetheless, understanding the precise molecular mechanisms driving microbe-immune interactions in the gut and their translation to humans will also require extensive future research.

Reductionist approaches are required to progress from correlations of bacterial phylotypes/strains with metabolic phenotype (e.g., diabetes) to mechanistic and causal relationships. However, many of the earlier analyses performed in epidemiological studies do not have sufficient resolution (i.e., 16S rRNA gene short amplicon sequencing only provides accurate identification at the genus level but not at the species, strain, and functional levels) (151), and deep-sequencing whole-genome metagenomics, coupled with strain-level analyses, are needed to identify bacteria and their functions that are linked to diseases and thereby to correctly downsize further causality and mechanistic studies (152).

The complexity of the GM and the multitude of possible phylotypes and microbial networks associated with specific phenotypes in population studies also make it difficult to test all possible hypotheses experimentally. Advanced statistics (e.g., MR and mediation analysis) and machine learning methods are helping to establish causal inferences in human studies, but further validation stages are still needed to provide direct evidence of causality. For this purpose, the use of rodent models in which confounding variables (e.g., genetic background, microbial communities, and diet) can be controlled is key to clearly identifying the effects caused by the precise microbe investigated (namely, a bacterial strain or metabolite).

The use of whole fecal microbiota transplantation (FMT) or co-housing experiments in which the microbiota of rodents with different phenotypes are interexchanged have substantially helped to unravel cause and effect in whole communities of microorganisms (2,153). Nonetheless, the results of those trials could be biased by limitations in study design. For example, bias could arise in relation to the limited number of donors selected and the possible variations in rodent response to microbes from a different donor species (e.g., lack of colonization or persistency of part of the microbial population or failure in replicating host-microbe interactions of the original host in which the microbiota co-evolved) and consequent differences in physiological effects (154).

Moreover, identifying the key microbial actors driving a health or disease phenotype and the nonactive players within the community is a rather complex undertaking. Specific components of the GM could play a role per se or in coordination with other community members (155). Furthermore, correlations and causal relationships between GM components and disease could also depend on the specific host (e.g., disease predisposition) and environmental context (e.g., dietary patterns), which can vary in different geographical locations and population groups (11).

One reductionist approach that could aid in establishing the causal role of microbes in metabolic disease is the use of defined microbial communities in gnotobiotic mouse models of diabetes (156). This approach consists of assembling a defined community of well-characterized and genetically tractable microbes, which is then used to colonize gnotobiotic mice. This so-called defined microbiota approach allows for the manipulation of a specific microbial feature in the background of a complex, yet manageable, microbial community to determine whether specific microbial functions play a causative role in the pathogenesis of diabetes and related complications.

Intervention studies with specific bacteria in the ecosystem, as well as more sophisticated strategies that deplete specific microbial components or functions, are ideal for providing evidence of causality (157,158). However, even if a particular bacterial genus and species has been correlated with a specific disease phenotype in a relatively reproducible manner in epidemiological studies (159) and proven to be causally involved in an intervention trial, differences between species, and even between strains, may also lead to different outcomes. As explained above, small genomic and phenotypic differences between strains belonging to the same species can translate into functional differences affecting the host phenotype (e.g., with regard to their immunomodulatory effects) (160). Therefore, the results of efficacy and mechanistic analyses deduced from studies performed with a specific strain cannot be systematically generalized to all strains of the given species.

Historically, the study of the impact of GM on human disease has been focused on the large intestine microbiota because the human colon is the site in the body with the highest abundance of microbes and the most accessible intestinal section. The contribution of the large intestine microbiota to the pathogenesis of metabolic disease has been demonstrated by FMT studies in mouse models of obesity and related complications (2,153). However, it is important to note that the small intestine overshadows the large intestine with regard to metabolic regulation. The small intestine is home to EECs that produce GLP-1 and other incretin hormones, which are key glucose metabolism regulators (161). The small intestine epithelium also plays an essential role in glucose and fat uptake and metabolism, protecting the host from features of metabolic dysfunction (162,163), and the small intestine microbiota is a regulator of EEC function and nutrient absorption, metabolism, and secretion (164,165). Thus, microbiota-host interactions in the small intestine would be expected to contribute to diabetes pathogenesis. Recent work in mouse models has determined that specific members of the small intestine microbiota can inhibit lipid secretion by enterocytes and limit serum triglyceride concentrations during the consumption of a Western-style obesogenic diet (166). Additionally, the small intestine microbiota in rodents has been demonstrated to affect nutrient-induced gut-brain signaling, which regulates glucose homeostasis (167,168).

Despite their importance, host-microbiota interactions in the small intestine and their relevance to diabetes are understudied because of limitations in the process of acquiring small intestine microbiota samples from humans and the reduced abundance of microbes in this portion of the digestive tract. Novel technologies to overcome technical limitations in the study of the small intestinal microbiota are discussed further in technological and methodological advancements.

GM fluctuation, especially related to dietary intake, should be considered when establishing what are normal and what are dysfunctional microbiota changes for metabolic health. The GM mirrors individuals’ habitual diet and daily choices. Therefore, longitudinally considering dietary history and GM variations over multiple days could help to fine-tune associations and infer causal relationships regarding the metabolic health of individuals (169). Moreover, daily oscillations of the GM related to eating patterns also affect its functional roles, such as appetite regulation and postprandial responses to food intake, with potential long-term effects on metabolic health and diabetes risk. For example, mouse studies have shown that daily oscillations in GM composition are required to maintain the circadian release of GLP-1, which in turn is required to achieve appropriate circadian control of metabolic homeostasis (170). In humans, type 2 diabetes and obesity are correlated with alterations of GM circadian rhythms (171), suggesting that daily oscillations are relevant to understanding the role of the GM in controlling energy homeostasis.

Sex also seems to contribute to GM variations, although its relevance for predicting health associations and their underlying mechanisms are underinvestigated. The epidemiology and pathophysiology of obesity and associated cardiometabolic disorders such as type 2 diabetes have a sex dimorphism that may be related to not only the role of sex hormones in fat distribution, metabolism, and immunity but also differences in the GM (172). Studies in mouse models suggest reciprocal interactions between sex hormones and the GM. On one hand, the GM may regulate the production and/or metabolism of sex hormones (i.e., testosterone and estrogens), as proven in a nonobese diabetic mouse model of type 1 diabetes (173). On the other hand, physiological effects of sex hormones (e.g., on immunity and intestinal transit) may affect the GM (174). Therefore, sex should be considered a confounding variable in epidemiological studies and in the design of mechanistic studies using mouse models because, to date, most preclinical studies have been carried out exclusively in males.

Separating phenomenology from actual biology in the microbiome field requires tools and approaches to identify mechanisms that deconvolute whether the microbiome may be a driver of, or offer therapeutic opportunities for, metabolic diseases. Here, we discuss the most promising technological developments for advancing the field.

When comparing model systems for studying the relationship between the GM and metabolic diseases, it is essential to consider both traditional models (e.g., germ-free [GF] and gnotobiotic mice) and emerging technologies (e.g., organs-on-a-chip and nonmurine GF models such as zebrafish and pigs).

GF animals have been used widely to investigate the role of the human GM in obesity and diabetes (1,173,175). These animals, which are born without any microbiota, allow for the interrogation of interventions in the absence of a microbiome. As a result, we can gain insight into whether the microbiome is necessary for a given biological process.

Gnotobiotic disease models are established by colonizing GF mice with either an entire GM via donor stool or specific isolated bacterial strains (176). Studies have demonstrated that GF animals, when inoculated with fecal microbiota from individuals with obesity and type 2 diabetes, successfully replicated disease phenotypes, providing evidence for the involvement of the GM in metabolic diseases (2,59,177). Additionally, an overgrowing endotoxin-producing bacterium, Enterobacter cloacae B29, isolated from the gut of a person with morbid obesity and diabetes, induced obesity, fatty liver, and insulin resistance in GF C57BL/6J mice that were otherwise resistant to high-fat diet–induced metabolic defects. Knocking out the endotoxin-producing gene in the B29 bacterial strain or the Tlr4 gene in C57BL/6J mice prevented the metabolic defects, underscoring the causal relationship between specific gut bacteria and host responses in the initiation and progression of metabolic disease (178–180).

However, certain concepts have been perpetuated about GF mice that are the result of studying only one genotype. For example, GF C57BL/6J mice are resistant to diet-induced obesity (175), whereas GF Swiss Webster mice are not (181); therefore, because the majority of GF mouse studies use C57BL/6J mice, it has been stated as fact that GF mice in general have to eat more than conventional mice to maintain weight. The divergent responses of these models to high-fat diets underscore the importance of genetic background in research outcomes (182).

The availability of additional GF models, such as pigs and zebrafish, complement the use of GF mice. GF pigs and piglets offer more human-relevant insights than do mice when developing human microbiota–associated gnotobiotic models (183), although the space required to house them is prohibitive for many institutions or limits studies to using just a few animals. GF zebrafish, on the other hand, have proven useful for studies of the GM and distinct host cellular developmental stages (184). The transparency of the fish body and the ability to fluorescently tag and image different cell types in the presence of different bacteria, as well as the ease of housing and propagating zebrafish, is advantageous for investigating specific questions (185). These models do not fully replicate human physiology, but they allow longitudinal and invasive sampling in tightly controlled conditions, which is important when asking mechanistic questions.

Organs-on-a-chip, such as the gut-on-a-chip, offer more human-relevant systems because they can be derived directly from human tissue or blood-derived induced pluripotent stem cells, which retain the genetic signature of the host; thus, they enable the study of complex human tissues and cellular interactions in a controlled environment (186). Recent efforts have demonstrated the ability to seed the gut-on-a-chip with microbiota in a semi-anaerobic environment (187), and many groups are now testing the efficiency of seeding increasingly complex communities on these chips. Although the gut-on-a-chip model lacks some key cell types such as immune cells, major advances include the ability to connect different organ chips such as the gut-chip and neuron-chip (188) to model gut-brain interactions. Creative uses of organs-on-a-chip to study the microbiome will continue to emerge and are likely to fill important gaps to complement animal models.

The ability to sequence and assemble whole genomes of bacteria is an enormously powerful approach for identifying lineages and the relatedness of bacterial strains and for identifying putative pathways involved in a given bacterial phenotype that may have relevance in human health or disease. If we think about the mechanisms of human disease that have been elucidated from the study of genetically manipulated mice, it is not hard to imagine the wealth of information to be gained from doing the same in bacteria. The ability to knock out and manipulate bacterial genes is not new. Nearly 80 years of bacterial genetics have clarified how pathogens colonize the gut epithelium and secrete toxins, leading to diseases such as cholera, how they share information with each other to adapt to different environments, and how nutritional selection drives their composition in a host. E. coli can be considered the bacterial version of the C57BL/6 mouse; its genetics are well-defined and easily engineered (189–191), and it has become the workhorse for testing the effect of modifications in a given environment. However, the commensal GM consists of far more diversity than just E. coli; thus, researchers are actively seeking a deeper understanding of GM genetics, using, for example, Bacteroides and Clostridium as representative organisms (192,193), as numerous human and mouse studies have demonstrated the important roles of these organisms in health and disease.

Advanced computational tools, including artificial intelligence, have shed new light on unannotated parts of a bacterial genome by predicting the three-dimensional structures of proteins, a task greatly advanced by technologies such as AlphaFold2 (194,195). By analyzing these structures, researchers can infer possible functions based on their shapes and binding sites. These potential roles can be confirmed by experimental validation in biochemical and microbiological studies (196). This knowledge, especially regarding how proteins influence metabolic pathways, is crucial for linking microbial activity to health conditions such as diabetes, offering insights into disease mechanisms and potential therapeutic targets.

The most critical issue with current database-dependent approaches in microbiome sequencing analysis is their limitation in detecting novel or understudied microbes (197). When microbial community samples are analyzed using databases based on reference genomes from well-characterized bacteria, sequences that do not match are overlooked or misclassified. This process results in a biased view of the microbial ecosystem, potentially missing crucial components that could have significant roles in health and disease, including diabetes. Therefore, advancing microbiome research necessitates the development and use of methods that can uncover and characterize these underrepresented microbial entities.

Assembling genomes de novo from metagenomic sequencing data is a powerful approach in microbiome research that involves constructing genomes directly from sequencing reads without relying on reference databases (197). This method uses advanced computational algorithms to piece together DNA fragments from a sample, allowing the identification of genetic material from a wide range of organisms, including those not previously sequenced or cataloged. By assembling these genomes, researchers can discover novel species and uncover new gene functions, significantly expanding our understanding of microbial diversity and its potential roles in various environments, including the human body. This approach is particularly useful in revealing the full spectrum of microbial life, including rare or unknown species that might play crucial roles in health and disease.

The small intestine is the primary site of nutrient uptake, enterohepatic recycling, and intestinal hormone stimulation; thus, it is essential to gain a deep understanding of microbial function in this region of the body. However, most of our knowledge of human microbiomes has been based on stool samples and the colonic microbiota because accessing the small intestine microbiota is challenging, even with modern endoscopy methods.

Recent advances use innovative methods such as ingestible capsules that sample intestinal material throughout the GI tract (198). Because each capsule is triggered by a different pH along the gut, this method can provide a microbial atlas of intestinal communities. These tools are being further refined and commercially developed for use in both diagnostics and research. One caveat, however, is that there is potential for microbes to continue growing after the sample has been collected within a capsule, thus giving an inaccurate representation of the native microbiome community. Additionally, these and other capsules have been developed for sampling in the fasting state, leaving the study of postprandial responses still limited, although these responses are likely important to reach a complete understanding of microbial contributions to the regulation of glucose metabolism. Addressing these issues is crucial for ensuring the reliability and accuracy of microbiome studies with such devices.

Isozyme and small molecule screens in microbiome research are crucial for identifying specific bacterial products that can be targeted therapeutically. Microbial isozymes are enzymes that have different molecular structures but catalyze the same reaction as the host enzymes. Screening these products can reveal variations in microbial metabolism that might influence health and potentially interfere with medications, as in the case of bacterial DPP-4 isozymes (33). Small molecule screens focus on identifying bioactive compounds produced by microbes (199). These compounds can have significant effects on host pathophysiology (140,200). By identifying specific isozymes and small molecules, researchers can target them for degradation or enhancement, offering potential therapeutic strategies for diseases such as diabetes.

As described above, no diagnostic and generalized fecal microbiota taxonomic signature has been found for type 1 or type 2 diabetes (86,201). Future research should therefore move toward strain-level studies in large prospective populations and, when possible, focus on functional profiling of intestinal microbes along the GI tract (198), with special attention to stable isotope precursors to study production and substrate fluxes of important microbially produced metabolites in different GI regions (202).

With regard to GM-based therapies for diabetes, high-fiber diets have been shown to be effective in controlling blood glucose levels and reducing insulin resistance in both type 1 and type 2 diabetes (203,204). Although the direct mode of action of dietary fiber via the GM remains to be shown, these trials underscore the potential importance of including GM modulation strategies as part of diabetes intervention trials, especially for the production of beneficial metabolites such as SCFAs (205). However, as noted above, intervention trials of oral SCFA butyrate supplementation have shown no effect on glycemic control or other markers of diabetes regulation in either type 1 or type 2 diabetes (116,122,123,206), probably because the site of delivery does not mimic endogenous production. For other SCFAs, including propionate and acetate, data are too scarce to draw any conclusions regarding possible effects on metabolic regulation.

Probiotic therapies for diabetes can be divided into conventional probiotics, particularly Lactobacillus and Bifidobacterium strains, which have a history of use for human consumption in fermented foods or supplements to promote health, and next-generation probiotics, which are strains of new bacterial species recently identified as indigenous members of the human GM. These strains are associated with health, and their presence is diminished in disease settings (76,207). With regard to the conventional probiotics, prospective randomized controlled trials (RCTs) in new-onset type 1 diabetes are ongoing (NCT03961854, NCT03961347, NCT04769037, and NCT05767450), and a smaller trial has shown only moderate effects in longstanding type 1 diabetes (208). However, an open-label trial of probiotics (strains of Bifidobacterium, Lactobacillus, and Streptococcus salivarius) found beneficial effects on susceptibility to and progression of type 1 diabetes in siblings of people with type 1 diabetes (209). In type 2 diabetes, a recent meta-analysis described some efficacy of these probiotic strains in metabolic control and reduced insulin resistance (210).

With regard to next-generation probiotics, fewer data have been generated in humans. For example, despite specific strains (e.g., Akkermansia) having been associated with a healthy metabolic phenotype (211), an RCT intervention with A. muciniphila did not identify strong metabolic effects (143). This finding could be the result of a lack of a causal role of these tested strains in the metabolic syndrome, reduced viability upon passage through the stomach (212,213), inadequate dosages, or a lack of engraftment when introduced in the human gut (214). Because the small intestine is important for the pathophysiology of both type 1 and type 2 diabetes, further analyses of small intestinal microbiota from individuals with type 1 (215) and type 2 (216,217) diabetes are needed, with defined combinations of next-generation probiotic strains studied as possible interventions for diabetes. However, this effort should consider conditions of bacterial strain engraftment, ecological or functional dependencies on other bacterial members, and potential redundancies in functionality, as shown by a meta-analysis of FMT (218).

Until such investigations with defined combinations of strains are completed, donor FMT might provide insight into the magnitude of effect of modulating the GM and the effect of such modulation on the pathophysiology and potential reversibility of diabetes. de Groot et al. (219) recently published research on the efficacy of fresh FMT in maintaining residual β-cell function and dampening autoimmunity in new-onset type 1 diabetes. Other studies have been performed for type 2 diabetes and insulin resistance, showing a modest effect of FMT on insulin resistance and nonalcoholic fatty liver disease (216,217,220–223), whereas one study showed no effect on these parameters (224). Additionally, a combined intervention of encapsulated donor FMT and fiber supplementation showed beneficial effects on glucose metabolism, suggesting the possible need to design interventions not only with synthetic bacterial strain consortia but also with dietary support (e.g., fiber to nourish the bacterial strains) (221). Finally, studies evaluating whether autologous FMT after lifestyle intervention could help prevent weight regain have suggested that diet-induced changes in low-abundance bacteria might be responsible for weight loss maintenance, which could guide more precise interventions with less ethical burden and lower risks of transmitting diseases (225).

Overall, donor FMT is a more diffuse approach than interventions with targeted strains or metabolites (226). Additionally, there are differences in mode of fecal material administration (capsules vs. fresh FMT), intestinal pH (e.g., due to antacids), and colonic transit time in existing data sets, and the amount of fecal microbiota administered also seems to affect engraftment of donor bacterial strains (218,227). These factors preclude generalization of the results of studies to date. We therefore advocate further standardization of intestinal microbiota composition measurements (228), with strict dietary monitoring. Also, better standardization is needed in human studies of FMT-based interventions. In this context, production of lyophilized capsules for FMT must follow Good Manufacturing Practices to maintain viability and ensure adequate shelf life (229).

Nevertheless, based on its wide availability and general safety (provided that donors are adequately screened [230]), FMT could provide clinicians with new treatment modalities for diabetes until interventions with defined combinations of strains are available, especially if next-generation probiotics can be spiked in donor fecal microbiota to boost therapeutic efficacy (231). However, these interventions should adhere to the international Nagoya Protocol on Access to Genetic Resources and the Fair and Equitable Sharing of Benefits Arising From Their Utilization to the Convention on Biological Diversity (232), which seeks to prevent researchers or their institutions from financially capitalizing (at the expense of vulnerable individuals or populations) on identified bacterial strains as next-generation probiotics. With regard to trial outcomes for diabetes and GM-based therapies, using dynamic measurements of glucose metabolism over time (e.g., mixed-meal tests or continuous glucose monitoring) could provide better insights into the interactions between the GM, diet, and glucose homeostasis during both FMT and administration of defined strain combinations.

New insights into the GM are increasingly associating it with diabetes in humans, although the microbiome of the small intestine remains understudied. Intervention studies with FMT in humans have been able to dissect associations from causality and have indeed shown some clinical benefit, although the contrast between, on average, relatively small therapeutic effects and ethical concerns (233) preclude widespread practical use of this treatment option in diabetes clinical care. Additional studies are thus needed of prospective associations between the GM and diabetes in multiethnic cohorts. Alongside this effort, the therapeutic potential of synthetic GM-derived bacterial strains and/or communities and engineered systems for targeting intestinal delivery of identified metabolites in diabetes should be explored.

Over the past two decades, alterations in the GM have been associated with aberrant glucose metabolism and steatosis in individuals with diabetes. Larger sample sizes in epidemiological studies have now started to show the magnitude and possible consistency of correlations between the GM and human metabolic traits of relevance to obesity and/or type 2 diabetes; however, for type 1 diabetes, the picture is much less clear.

Interaction with diabetes medications in relation to ethnicity and dietary intake should be taken into account more rigorously in future studies. Moreover, in recent years, more insights have been gained into the function of the GM beyond just its composition, and this information nicely dovetails with earlier reports of links between specific metabolites, including SCFAs, BCAAs and bile acids, and obesity and diabetes.

With regard to GM composition, only a few studies have addressed the role of phages and fungi and the interactions between these inhabitants and bacterial strains in diabetes. It is clear that future studies also need to focus on small intestine microbiota function as well as developing adequate bioinformatic pipelines and correctly assembling genomes (Table 1).

We must also take into account that most data to date have been generated in mouse studies whose relevance to human diabetes needs further confirmation because of the large differences between mice and humans in diet, genetics, and life span. Nevertheless, human intervention studies of single strains and FMT in the setting of human diabetes have shown a range of clinical metabolic effects (compared with the more consistent effects of medications) but without serious side effects. In conclusion, after almost two decades of study, we must still look to future efforts to illuminate the clinical diagnostic and therapeutic applicability of GM research to human diabetes.

Acknowledgments. Editorial support services for this article were provided by Debbie Kendall of Kendall Editorial in Richmond, VA. The authors also thank Robert Caesar and Anna Hallen of Wallenberg Laboratory, Department of Molecular and Clinical Medicine, Institute of Medicine, University of Gothenburg, Gothenburg, Sweden, for their help with the figures.

Funding. The 2023 Diabetes, Diabetes Care, and Diabetologia expert forum “The Microbiome and Diabetes: Research, Translation and Clinical Applications,” including a subsequent symposium presented on this topic and the preparation of this article, was supported by a grant from the Novo Nordisk Foundation (NNF23SA0085453). M.B. is a Howard Hughes Medical Institute Freeman Hrabowski Scholar and is supported by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grant NIDDK R01DK131104, the Pew Charitable Trusts (2022-A-19568), and the Burroughs Wellcome Fund Investigators in the Pathogenesis of Infectious Disease (022792). S.D. is supported by NIDDK grants NIDDK R01DK123446 and 1DP1 DK130687. J.H.N. is supported by Swiss National Science Foundation (SNSF) grant 310030_219210. M.N. is supported by a personal ZONMW-VICI grant 2020 (09150182010020) and a DFN-DON grant 2020 (2020.10.002). Y.S. is supported by grants from the Spanish Ministry of Science, Innovation and Universities MICIU/AEI (PID2020-119536RB-I00) and Severo Ochoa Center of Excellence (Ref. CEX2021-001189-S/10.13039/501100011033). V.T. is a co-principal investigator of a research project funded by the European Union’s Innovative Medicines Initiative (IMI grant 875534).

Duality of Interest. J.H.N. has served as an advisor for Abbvie, Bristol Myers Squibb, Falk, Janssen, Roche, and Takeda. M.N. is the founder and a scientific advisory board member of Caelus Pharmaceuticals and Advanced Microbiome Interventions in Amsterdam, the Netherlands. Y.S. is co-author of patents based on the use of intestinal bacteria for obesity and metabolic comorbidities. V.T. holds a patent for the use of microbiota in type 2 diabetes. L.Z. is a co-founder of Notitia Biotechnologies, a microbiome company providing products related to diabetes. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. All authors contributed to writing, reviewing, and revising the manuscript and approved the final version for submission. M.O.-M. is the guarantor of this work and, as such, takes responsibility for the integrity of the review.

Open Access. This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.