Research advances in the identification of gut keystone bacteria and formulation of suitable dietary intervention strategies for restoring gut homeostasis Authors: Qiang Meng, Huimin Zhang, Jianming Ye, Ting Liu, Yihua Li, Wei Song, Yane Luo, Tianli Yue Published: 2026 ABSTRACT Methodological inconsistencies hinder the identification of gut keystone bacteria and their functional mechanisms. This work critically compares identification approaches, elucidates their interactions within the microbiota and host, and systematically details targeted dietary interventions. By explicitly linking keystone identification to diet-driven modulation, it provides a practical and precise framework for developing gut health strategies and managing microbiota-related diseases. Introduction A diverse community of commensal bacteria, commonly known as “microbiota”, colonize the small and large intestines of humans as well as animals and aid in nutrient metabolism, local and systemic immunomodulation, protection of intestinal barrier, and maintenance of gut health. Although the gut microbiota is rich and diverse (with approximately 1011–1012 microorganisms per cm3 in the human colon), its structure and functions are greatly affected by specific species, which constitute the “keystone taxa1”. The following definition of keystone taxa proposed by Banerjee et al. is well accepted: “Microbial keystone taxa are highly connected taxa that individually or in a guild exert a considerable influence on microbiome structure and functioning irrespective of their abundance across space and time. These taxa have a unique and crucial role in microbial communities, and their removal can cause a dramatic shift in microbiome structure and functioning2”. These keystone taxa are not only central to maintain gut ecosystem stability but are also critically implicated in the pathogenesis and progression of various diseases, supporting a promising target for therapeutic intervention3. Although the gut microbes include organisms from Bacteria, Archaea, Eukarya, and viruses, Bacteria and Archaea contribute to more than 99% of the total biomass and unique gene repertoire of gut microbiota4. This underscores the urgent necessity for identifying keystone bacteria from specific gut ecosystems, particularly those associated with certain diseases or therapy for specific diseases. A universally accepted standard for measuring parameters or experimental procedures for determining “keystone taxa” among the gut microbiota are currently lacking; however, the keystone bacteria may be identified via direct experimental evidence (referred to as keystone taxa) or inferred following computational analysis (known as keystone OTUs)5. Interestingly, the original intent of proposing keystone OTUs was to reduce the high costs associated with experimental validation and allow the rapid identification of potential keystone taxa from a gut microbial community5. Certain network-based methods centered on the calculation of co-occurrence or coabundance of OTUs have been proposed, which appear promising for inferring keystone taxa from microbial communities. For these, the combined score differentiating high-closeness and low-betweenness centrality has been frequently utilized as a threshold for defining keystone OTUs2. Unfortunately, the microbial network interactions inferred from these approaches frequently do not reflect the real ecological interactions6, prompting researchers to strive for more efficient methods of modeling microbial interactions. On the one hand, certain methods involving multidimensional calculations (such as omics, network analysis, and correlation analysis) have been established for improving the accuracy of predicting keystone bacteria. On the other hand, certain synthetic gut bacterial communities have been developed as tools for supplementing and validating the results attained via network analysis1,6. The irreplaceable roles of keystone bacteria in the gut are evident from the discovery of an ever-increasing number of special roles and influences of these bacteria7,8; these have been instrumental in the formulation of effective intervention programs, with a series of breakthroughs achieved in health- and diet-related programs9,10. However, the keystone species are highly influenced by environmental factors such as nutritional conditions1, pH11, and oxygen levels12,13, which shape their abundance and functions. Moreover, the interactions among microbes and with the host are crucial for efficient nutrient utilization and the expression of functional genes involved in metabolism, colonization, and host-microbe crosstalk14–16. Therefore, the interactions among the intestinal microbiota and their underlying molecular mechanisms have been the subject of ongoing research, with the goal of aiding dietary interventions. A comprehensive understanding and efficient utilization of keystone bacteria in the gut necessitate in-depth research on their identification, evaluation of dynamic changes resulting from microbial interactions, and elucidation of the underlying mechanisms. However, the techniques for identifying keystone bacteria have hitherto been reported sporadically. Moreover, the factors that impact the keystone microbes in the gut and their underlying mechanisms remain to be well characterized and analyzed. The scientific and technological developments in the identification and application of keystone bacteria from the gut microbiota are presented herein; moreover, the regulation of intestinal microbiota by the keystone species as well as the underlying molecular mechanisms are elucidated in a comprehensive manner, including the roles of daily intervention factors such as diets. These efforts will benefit the development of dietary intervention strategies to promote health via the regulation of gut microbiota. Techniques facilitating the identification of keystone bacteria from the gut microbiome As mentioned above, keystone taxa are defined as highly connected taxa that significantly influence microbiome structure and functioning, regardless of their abundance2. The identification of these pivotal entities is fundamental to gut microbiome research, while the involved methodologies vary widely in calculating or analytical principles, data requirements, and outputs. To provide a clear comparative overview, the main computational and experimental approaches for keystone identification are summarized in Table 1, encompassing their fundamental approach, main strengths, inherent limitations, and translational potential.Table 1Comparative overview of key methodologies for identifying gut keystone bacteriaIdentification approachTypeOutlineKey indicatorsStrengthsLimitationsTranslational potentialReferencesTop-down identification of keystone taxaPrediction processCalculating EPI values (D1, D2, Q); only the top N = 1000 OTUs with the top mean abundance were kept; only OTUs with a relative frequency between 0.25 and 0.75 were then calculatedDistance based separation (D1, D2) and the modularity of network associated with presence/absence of a certain species (Q)• Does not require longitudinal data;•Suitable for both perturbation experiments and cross-sectional surveys;•Identifies keystones based on systemic impact.• Difficult to differentiate true ecological interactions from environmentally driven associations;•Biological interpretation remains ambiguous without validation.Moderate for screening. It is useful for analyzing large human cohort data to flag high-impact taxa. Like network analysis, it serves as a starting point for downstream experimental and clinical verification.3Metagenome-wide association analysisPrediction processDesigning the corhots in a MWAS study; sampling according to the design; gene prediction and functional annotation; gene clustering using their abundances; identification of disease associations with individual microbial genes, taxa or functions; validation via additional metagenomic datasets or further investigation by animal modelsN.A.• Directly links microbes to clinical phenotypes;•Can identify recovery-associated patterns via cross-cohort validation.• Requires large, well-characterized cohorts (resource-intensive);•Reveals association, not causation;•Results may be cohort-specific.High, but requires validation. It provides candidate targets for diagnostics and therapies. Machine learning models built on such data can predict outcomes, but clinical deployment needs mechanistic validation and interventional trials.17Experimental validation using synthetic and gnotobiotic modelsEmpirical studySystematically tests candidate keystones in controlled settings. Includes single-species dropout in defined synthetic communities (e.g., OMM12), mono-/co-colonization in gnotobiotic animals, and functional assays (gene knockout, metabolomics)N.A.• Provides causal evidence and mechanistic insight;•The gold standard for confirming keystone function and host effects.• Time-consuming, laborious, and costly;•Animal models may not fully recapitulate human physiology;•Simplified models may not reflect native community complexity.Essential for translation. All computational predictions require experimental validation to confirm ecological role and safety before clinical application (e.g., developing biotherapeutics or dietary plans).1,120Data-driven keystone species identification frameworkPrediction processComputing structural and functional keystoneness values (Ks(i, s), Kf(i, s))Structural and functional keystoneness• Quantifies impact independent of abundance;•Captures low-abundance but high-impact keystones;•Evaluates impact on specific functions.• Requires large, high-quality datasets for training;•Limited model interpretability hampers biological insight;•Performance depends on training data diversity.Emerging and promising. It is potentially powerful for personalized microbiome diagnostics by identifying individual-specific keystones. Clinical integration necessitates robust validation in diverse populations and transparent model interpretability.21Co-occurrence networksPrediction processProducing an interaction matrix and directed network for a metacommunity; simulating population dynamics individual communities until reach steady state abundances; constructing a co-occurrence network; assessing the alignment between the co-occurrence and interaction networks and the ecological relevance of network topologyMean degree; betweenness centrality; closeness centrality; transitivity• Efficient screening from high-throughput sequencing data;•Applicable to large, cross-sectional datasets.• Correlations may be spurious, reflecting shared environment rather than true ecological interaction;•Often biased towards detecting high-abundance taxa.Primarily for hypothesis generation. It identifies potential keystone candidates and interaction patterns. It must be combined with experimental validation (e.g., synthetic communities) to confirm ecological role before clinical translation.18Learning interactions from microbial time seriesPrediction processGenerating species interaction network by employing sparse linear regression with bootstrap aggregation; cross-sectional and time-series data as inputsOutgoing interactions• Models dynamic species interactions (competition/cooperation);•Can predict keystone nature with high accuracy using network metrics.• Susceptible to spurious correlations from indirect effects;•Often limited to analyzing species with sufficient abundance.Moderate, depends on data quality. It is useful for modeling community dynamics in controlled settings. Clinical application is limited by the need for dense, longitudinal sampling.19Metage2MetaboPrediction processInputting a collection of annotated genomes; reconstructing metabolic networks; identifying individual and collective metabolic capabilities; calculating the added-value of cooperation; computing a minimal community and key speciesN.A.• Provides functional mechanism (metabolic cross-feeding);•Identifies minimal communities and key metabolic contributors.• Requires high-quality genomic data (reference or MAGs);•Computationally intensive;•Predictions are limited to known metabolic pathways.High for mechanism-driven design. It is excellent for engineering defined consortia for probiotics or biotherapeutics. Direct clinical translation is challenging for complex native microbiota but guides targeted dietary interventions.20N.A. not applicable. Table 1Comparative overview of key methodologies for identifying gut keystone bacteriaIdentification approachTypeOutlineKey indicatorsStrengthsLimitationsTranslational potentialReferencesTop-down identification of keystone taxaPrediction processCalculating EPI values (D1, D2, Q); only the top N = 1000 OTUs with the top mean abundance were kept; only OTUs with a relative frequency between 0.25 and 0.75 were then calculatedDistance based separation (D1, D2) and the modularity of network associated with presence/absence of a certain species (Q)• Does not require longitudinal data;•Suitable for both perturbation experiments and cross-sectional surveys;•Identifies keystones based on systemic impact.• Difficult to differentiate true ecological interactions from environmentally driven associations;•Biological interpretation remains ambiguous without validation.Moderate for screening. It is useful for analyzing large human cohort data to flag high-impact taxa. Like network analysis, it serves as a starting point for downstream experimental and clinical verification.3Metagenome-wide association analysisPrediction processDesigning the corhots in a MWAS study; sampling according to the design; gene prediction and functional annotation; gene clustering using their abundances; identification of disease associations with individual microbial genes, taxa or functions; validation via additional metagenomic datasets or further investigation by animal modelsN.A.• Directly links microbes to clinical phenotypes;•Can identify recovery-associated patterns via cross-cohort validation.• Requires large, well-characterized cohorts (resource-intensive);•Reveals association, not causation;•Results may be cohort-specific.High, but requires validation. It provides candidate targets for diagnostics and therapies. Machine learning models built on such data can predict outcomes, but clinical deployment needs mechanistic validation and interventional trials.17Experimental validation using synthetic and gnotobiotic modelsEmpirical studySystematically tests candidate keystones in controlled settings. Includes single-species dropout in defined synthetic communities (e.g., OMM12), mono-/co-colonization in gnotobiotic animals, and functional assays (gene knockout, metabolomics)N.A.• Provides causal evidence and mechanistic insight;•The gold standard for confirming keystone function and host effects.• Time-consuming, laborious, and costly;•Animal models may not fully recapitulate human physiology;•Simplified models may not reflect native community complexity.Essential for translation. All computational predictions require experimental validation to confirm ecological role and safety before clinical application (e.g., developing biotherapeutics or dietary plans).1,120Data-driven keystone species identification frameworkPrediction processComputing structural and functional keystoneness values (Ks(i, s), Kf(i, s))Structural and functional keystoneness• Quantifies impact independent of abundance;•Captures low-abundance but high-impact keystones;•Evaluates impact on specific functions.• Requires large, high-quality datasets for training;•Limited model interpretability hampers biological insight;•Performance depends on training data diversity.Emerging and promising. It is potentially powerful for personalized microbiome diagnostics by identifying individual-specific keystones. Clinical integration necessitates robust validation in diverse populations and transparent model interpretability.21Co-occurrence networksPrediction processProducing an interaction matrix and directed network for a metacommunity; simulating population dynamics individual communities until reach steady state abundances; constructing a co-occurrence network; assessing the alignment between the co-occurrence and interaction networks and the ecological relevance of network topologyMean degree; betweenness centrality; closeness centrality; transitivity• Efficient screening from high-throughput sequencing data;•Applicable to large, cross-sectional datasets.• Correlations may be spurious, reflecting shared environment rather than true ecological interaction;•Often biased towards detecting high-abundance taxa.Primarily for hypothesis generation. It identifies potential keystone candidates and interaction patterns. It must be combined with experimental validation (e.g., synthetic communities) to confirm ecological role before clinical translation.18Learning interactions from microbial time seriesPrediction processGenerating species interaction network by employing sparse linear regression with bootstrap aggregation; cross-sectional and time-series data as inputsOutgoing interactions• Models dynamic species interactions (competition/cooperation);•Can predict keystone nature with high accuracy using network metrics.• Susceptible to spurious correlations from indirect effects;•Often limited to analyzing species with sufficient abundance.Moderate, depends on data quality. It is useful for modeling community dynamics in controlled settings. Clinical application is limited by the need for dense, longitudinal sampling.19Metage2MetaboPrediction processInputting a collection of annotated genomes; reconstructing metabolic networks; identifying individual and collective metabolic capabilities; calculating the added-value of cooperation; computing a minimal community and key speciesN.A.• Provides functional mechanism (metabolic cross-feeding);•Identifies minimal communities and key metabolic contributors.• Requires high-quality genomic data (reference or MAGs);•Computationally intensive;•Predictions are limited to known metabolic pathways.High for mechanism-driven design. It is excellent for engineering defined consortia for probiotics or biotherapeutics. Direct clinical translation is challenging for complex native microbiota but guides targeted dietary interventions.20N.A. not applicable. Comparative overview of key methodologies for identifying gut keystone bacteria • Does not require longitudinal data; •Suitable for both perturbation experiments and cross-sectional surveys; •Identifies keystones based on systemic impact. • Difficult to differentiate true ecological interactions from environmentally driven associations; •Biological interpretation remains ambiguous without validation. • Directly links microbes to clinical phenotypes; •Can identify recovery-associated patterns via cross-cohort validation. • Requires large, well-characterized cohorts (resource-intensive); •Reveals association, not causation; •Results may be cohort-specific. • Provides causal evidence and mechanistic insight; •The gold standard for confirming keystone function and host effects. • Time-consuming, laborious, and costly; •Animal models may not fully recapitulate human physiology; •Simplified models may not reflect native community complexity. • Quantifies impact independent of abundance; •Captures low-abundance but high-impact keystones; •Evaluates impact on specific functions. • Requires large, high-quality datasets for training; •Limited model interpretability hampers biological insight; •Performance depends on training data diversity. • Efficient screening from high-throughput sequencing data; •Applicable to large, cross-sectional datasets. • Correlations may be spurious, reflecting shared environment rather than true ecological interaction; •Often biased towards detecting high-abundance taxa. • Models dynamic species interactions (competition/cooperation); •Can predict keystone nature with high accuracy using network metrics. • Susceptible to spurious correlations from indirect effects; •Often limited to analyzing species with sufficient abundance. • Provides functional mechanism (metabolic cross-feeding); •Identifies minimal communities and key metabolic contributors. • Requires high-quality genomic data (reference or MAGs); •Computationally intensive; •Predictions are limited to known metabolic pathways. N.A. not applicable. Critical evaluation of methodologies for inferring keystone bacteria The inference of keystone taxa is highly sensitive to a range of technical and analytical variables, including amplicon sequencing biases, bioinformatics pipelines, cohort characteristics (e.g., diet, geography), and the specific parameters of analytical models6. This sensitivity underscores the necessity of a critical evaluation of each major approach in view of reproducibility, biases, and clinical applicability. Disease-association studies, such as Metagenome-Wide Association Studies (MWAS), directly link microbial features to host phenotypes and can identify recovery-associated bacteria through cross-cohort validation17. However, they fundamentally struggle to distinguish correlation from causation, and their reliance on large-scale, resource-intensive surveys can limit reproducibility across smaller, independent cohorts17. Co-occurrence network analysis offers an efficient mean to infer potential interactions and identify highly connected “hub” taxa from high-throughput data1,18. A major, well-recognized limitation is its proneness to spurious correlations, as statistical associations in abundance data may reflect shared environmental responses rather than direct biological interactions such as competition or cross-feeding6. This method also exhibits a bias towards detecting high-abundance taxa, potentially overlooking low-abundance keystones. Dynamic models, like those based on the Lotka-Volterra equations, advance beyond static correlation by modeling population dynamics and inferring interaction strengths18,19. Despite this, they face challenges in parameter inference from metagenomic time-series data and remain susceptible to spurious correlations arising from indirect interactions. Furthermore, their focus on population dynamics often restricts the detection of species with sufficient abundance, making them unlikely to identify low-abundance but high-impact keystone OTUs19. Mechanistic models, such as Metage2Metabo, provide valuable functional insights by reconstructing metabolic networks to identify species essential for community-level metabolic functions20. Their predictions, however, are constrained by the completeness of genomic annotations and reference databases, potentially overlooking strain-specific or novel functions. Additionally, their computational resource requirements—such as the need for high-performance computing clusters and long processing times—can pose a significant practical barrier to widespread application20. Emerging machine learning and deep learning frameworks, including the Data-Driven Keystone Identification (DKI) model, demonstrate power in detecting complex patterns and evaluating species impact independent of abundance21. Their “black-box” nature can obscure biological interpretation, and they require large, high-quality training datasets21. A universal challenge across all in silico and the simplified in vitro systems (such as synthetic communities like OMM12) is arising from the functional redundancy inherent in complex native gut ecosystems1,21. The species identified as a keystone in a synthetic model system may not hold the same pivotal role in vivo, where its function could be compensated by other taxa1. A fundamental disconnect exists between statistically identifying a “keystone OTU” and validating it as a therapeutic target despite that computational approaches efficiently generate hypotheses, leading to a critical translational gap. Thereby, independent experimental validation is indispensable to clarify the biological meaning of a high-centrality node in a network1,3. The roles of a candidate keystone must be causally established through in vitro strategies (such as gene knockout, coculture experiments, etc.) and in vivo validation in gnotobiotic animal models or via fecal microbiota transplantation1,17,22–25. True clinical translation requires moving further, from observational association to demonstrating that targeted modulation of the keystone (via diet, prebiotics, or probiotics) yields predictable and beneficial outcomes in human intervention trials. This necessitates a shift towards longitudinal, hypothesis-driven research that integrates computation, experimental validation, and clinical investigation to bridge the current translational bottleneck. Methodologies for experimental validation of keystone bacteria The experimental validation of these candidate bacteria typically follows an integrated, two-pronged strategy employing complementary in vitro and in vivo methodologies. The in vitro experiments are designed to simulate the human gut environment or specific metabolic conditions, enabling an investigation of the metabolic properties of the microbes, interspecies interactions, and effects of keystone bacteria on host physiology. The specific approaches employed in these studies include validation of gene functions (utilizing gene knockout or overexpression strategies)1,22,23, profiling of enzyme activities (via protein purification and functional assays)26, quantification of metabolites (via mass spectrometry or high-performance liquid chromatography analyses)1,17,22,24,27–29, and in vitro assessments of metabolic interactions (utilizing bacterial coculture experiments)22–24,27,28. These in vitro approaches provide rapid and precise insights into the metabolic characteristics and ecological functions of keystone species, thereby establishing a robust foundation for subsequent in vivo investigations. in vivo experiments are indispensable for validating the functional roles of keystone species within the complex gut environment and primarily utilize animal models or human cohort studies for evaluating their impact on microbial community dynamics and host health outcomes. Commonly employed approaches include the use of mouse models (involving monocolonization or cocolonization of mice with one or more bacterial strains, respectively)1,17,22,29–32, fecal microbiota transplantation (involving the transfer of human-derived microbiota containing keystone species into germ-free or antibiotic-treated mice)23–25, histopathological analysis (involving staining with hematoxylin and eosin or immunofluorescence techniques)22,24,30, and longitudinal human cohort studies (tracking changes in the abundance of keystone species in response to antibiotic interventions or disease progression). These in vivo approaches offer a comprehensive understanding of the ecological significance and host–microbe interactions of the keystone species, consolidating their role as critical regulators of the function and stability of the gut ecosystem. A representative example is Methanobrevibacter smithii, which has been rigorously characterized through multi-dimensional experimental validation. Firstly, M. smithii was initially identified as a predominant anaerobe in the gut microbiota of healthy individuals33,34. Subsequent colonization experiments in germ-free mice demonstrated that M. smithii can promote obesity by enhancing the digestion of carbohydrates35. Furthermore, genetic and metabolic evidences for the polysaccharide-degrading capabilities of M. smithii were obtained via genome sequencing and metabolite analyses, respectively36. These corroborative experimental observations supported the consideration of M. smithii as a low-abundance keystone pathogen with significant functional contributions7. Table 2 lists other identified keystone species in the gut microbiota based on experimental evidence, providing a validated target repository for the dietary and therapeutic interventions discussed in the following sections.Table 2Gut keystones reported in recent years (listed partly)Gut keystone bacteriaIdentification methodsaKeystone OTUs or taxaEstablished mechanistic pathways / Link to processesbReferencesBacteroides caecimuris; Blautia coccoidesIn vitro (metagenomics, coculture experiment, 16S rRNA gene amplicon sequencing, biochemical assays) and in vivo (colonization assays, biochemical assays) experimentsKeystone taxaNot mentioned1Bacteroides thetaiotaomicronIn vitro (transcriptomics, coculture experiment, biochemical assays) and in vivo (colonization assays, biochemical assays) experimentsKeystone taxonB. thetaiotaomicron exacerbates enteric infection by producing succinate, which is sensed by the pathogen’s transcription factor Cra, leading to enhanced virulence gene expression.22Ruminococcus bromiiIn vitro (coculture experiment, 16S rRNA gene amplicon sequencing, biochemical assays)Keystone taxonR. bromii serves as a keystone degrader of resistant starch in the colon by secreting potent amylases that efficiently break down insoluble starch particles into fermentable sugars.27,28Akkermansia muciniphilaIn vitro (metagenomics, transcriptomics, 16S rRNA gene amplicon sequencing) and in vivo (biochemical assays) experimentsKeystone taxonA. muciniphila plays a protective role in ulcerative colitis by degrading mucin to produce anti-inflammatory short-chain fatty acids and maintaining mucosal integrity.29Klebsiella pneumoniae; Proteus mirabilisIn vivo (colonization assays, 16S rRNA gene amplicon sequencing, biochemical assays) experimentsKeystone taxaK. pneumoniae and P. mirabilis promote colitis by synergizing with the endogenous gut microbiota to induce mucosal inflammation and TNF-α production30Bacteroides strain CL1-UCIn vivo (metagenomics, 16S rRNA gene amplicon sequencing, biochemical assays) experimentsKeystone taxonEarly-life engraftment of Bacteroides strain CL1-UC restores gut microbiome development, promotes immune tolerance, and prevents colitis in genetically susceptible IL-10-deficient mice, whereas late-life engraftment does not confer protection.31Bacteroides thetaiotaomicronIn vitro (biochemical assays) and in vivo (colonization assays, genomics, 16S rRNA gene amplicon sequencing, biochemical assays) experimentsKeystone taxonB. thetaiotaomicron undergoes rapid diet-driven evolution, acquiring mutations that enhance mucin glycan degradation under Western-style diet, thereby adapting to nutrient shifts and serving as a genetic biomarker of dietary history.32Bacteroides thetaiotaomicron; Methanobrevibacter smithiiHuman cohort studies (MWAS, 16S rRNA gene amplicon sequencing); In vivo (biochemical assays) experimentsKeystone taxaB. thetaiotaomicron enhances polysaccharide fermentation, producing hydrogen, formate, and ethanol, which are utilized by M. smithii to drive methanogenesis, thereby optimizing metabolic efficiency and supporting symbiotic stability in the gut ecosystem.36Bacteroides fragilisIn vivo (biochemical assays) experimentsKeystone taxonEnterotoxigenic B. fragilis promotes rapid colitis and colon tumorigenesis by selectively activating Stat3-dependent TH17 cell responses, which drive IL-17-mediated inflammation and tumor formation.121Christensenella minutaIn vivo (metagenomics, biochemical assays) experimentsKeystone taxonC. minuta enhances host energy homeostasis by increasing gut microbial biomass, which promotes energy excretion in feces and elevates voluntary physical activity.122FirmicutesMetagenomicsKeystone OTUCrohn’s disease123Prevotella; KlebsiellaCo-occurrence network analysisKeystone OTUsThe development of hypertension124Collinsella16S rRNA gene amplicon sequencing and correlation analysisKeystone OTUNonalcoholic steatohepatitis125Corynebacterium glutamicum; Akkermansia muciniphila16S rRNA gene amplicon sequencing and Linear discriminant analysis Effect SizeKeystone OTUsType 2 diabetes126Helicobacter pyloriNon-metric multidimensional scaling clustering analysisKeystone OTUGastric cancer127Lachnospiraceae UCG-001Co-occurrence network analysisKeystone OTUAlzheimer’s disease progression128Blautia; Faecalibacterium; Roseburia; Bacteroides; Enterococcus; Escherichia-Shigella; EnterobacterCorrelation network analysisKeystone OTUsChildren mortality in pediatric allogeneic hematopoietic stem cell transplantation129Marvinbryantia; Lachnospiraceae UCG-006; Pseudoflavonifractor; [Eubacterium] coprostanoligenes group; Defluviitaleaceae UCG-011; Lachnospiraceae ND3007 group; UCG-004Co-occurrence network analysisKeystone OTUsNon-communicable diseases130Bacteroides spp.; Lachnospiraceae family; Bifidobacterium spp.; Lactobacillus reuteriIn vitro (16S rRNA gene amplicon sequencing, metatranscriptomics, coculture experiment, biochemical assays) experimentsKeystone OTUsPropionate synthesis and gut microbiota metabolic homeostasis131OscillibacterIn vitro (Biochemical analyses; histochemical hnalysis; 16S rRNA gene amplicon sequencing and RT-qPCR) experimentsKeystone OTUOxidative stress; obesity132aThe identification method denotes the specific experimental approaches employed in the original literature.b“Keystone taxa” corresponds to established mechanistic pathways, while “keystone OTUs” refers to the link to disease-related processes. Table 2Gut keystones reported in recent years (listed partly)Gut keystone bacteriaIdentification methodsaKeystone OTUs or taxaEstablished mechanistic pathways / Link to processesbReferencesBacteroides caecimuris; Blautia coccoidesIn vitro (metagenomics, coculture experiment, 16S rRNA gene amplicon sequencing, biochemical assays) and in vivo (colonization assays, biochemical assays) experimentsKeystone taxaNot mentioned1Bacteroides thetaiotaomicronIn vitro (transcriptomics, coculture experiment, biochemical assays) and in vivo (colonization assays, biochemical assays) experimentsKeystone taxonB. thetaiotaomicron exacerbates enteric infection by producing succinate, which is sensed by the pathogen’s transcription factor Cra, leading to enhanced virulence gene expression.22Ruminococcus bromiiIn vitro (coculture experiment, 16S rRNA gene amplicon sequencing, biochemical assays)Keystone taxonR. bromii serves as a keystone degrader of resistant starch in the colon by secreting potent amylases that efficiently break down insoluble starch particles into fermentable sugars.27,28Akkermansia muciniphilaIn vitro (metagenomics, transcriptomics, 16S rRNA gene amplicon sequencing) and in vivo (biochemical assays) experimentsKeystone taxonA. muciniphila plays a protective role in ulcerative colitis by degrading mucin to produce anti-inflammatory short-chain fatty acids and maintaining mucosal integrity.29Klebsiella pneumoniae; Proteus mirabilisIn vivo (colonization assays, 16S rRNA gene amplicon sequencing, biochemical assays) experimentsKeystone taxaK. pneumoniae and P. mirabilis promote colitis by synergizing with the endogenous gut microbiota to induce mucosal inflammation and TNF-α production30Bacteroides strain CL1-UCIn vivo (metagenomics, 16S rRNA gene amplicon sequencing, biochemical assays) experimentsKeystone taxonEarly-life engraftment of Bacteroides strain CL1-UC restores gut microbiome development, promotes immune tolerance, and prevents colitis in genetically susceptible IL-10-deficient mice, whereas late-life engraftment does not confer protection.31Bacteroides thetaiotaomicronIn vitro (biochemical assays) and in vivo (colonization assays, genomics, 16S rRNA gene amplicon sequencing, biochemical assays) experimentsKeystone taxonB. thetaiotaomicron undergoes rapid diet-driven evolution, acquiring mutations that enhance mucin glycan degradation under Western-style diet, thereby adapting to nutrient shifts and serving as a genetic biomarker of dietary history.32Bacteroides thetaiotaomicron; Methanobrevibacter smithiiHuman cohort studies (MWAS, 16S rRNA gene amplicon sequencing); In vivo (biochemical assays) experimentsKeystone taxaB. thetaiotaomicron enhances polysaccharide fermentation, producing hydrogen, formate, and ethanol, which are utilized by M. smithii to drive methanogenesis, thereby optimizing metabolic efficiency and supporting symbiotic stability in the gut ecosystem.36Bacteroides fragilisIn vivo (biochemical assays) experimentsKeystone taxonEnterotoxigenic B. fragilis promotes rapid colitis and colon tumorigenesis by selectively activating Stat3-dependent TH17 cell responses, which drive IL-17-mediated inflammation and tumor formation.121Christensenella minutaIn vivo (metagenomics, biochemical assays) experimentsKeystone taxonC. minuta enhances host energy homeostasis by increasing gut microbial biomass, which promotes energy excretion in feces and elevates voluntary physical activity.122FirmicutesMetagenomicsKeystone OTUCrohn’s disease123Prevotella; KlebsiellaCo-occurrence network analysisKeystone OTUsThe development of hypertension124Collinsella16S rRNA gene amplicon sequencing and correlation analysisKeystone OTUNonalcoholic steatohepatitis125Corynebacterium glutamicum; Akkermansia muciniphila16S rRNA gene amplicon sequencing and Linear discriminant analysis Effect SizeKeystone OTUsType 2 diabetes126Helicobacter pyloriNon-metric multidimensional scaling clustering analysisKeystone OTUGastric cancer127Lachnospiraceae UCG-001Co-occurrence network analysisKeystone OTUAlzheimer’s disease progression128Blautia; Faecalibacterium; Roseburia; Bacteroides; Enterococcus; Escherichia-Shigella; EnterobacterCorrelation network analysisKeystone OTUsChildren mortality in pediatric allogeneic hematopoietic stem cell transplantation129Marvinbryantia; Lachnospiraceae UCG-006; Pseudoflavonifractor; [Eubacterium] coprostanoligenes group; Defluviitaleaceae UCG-011; Lachnospiraceae ND3007 group; UCG-004Co-occurrence network analysisKeystone OTUsNon-communicable diseases130Bacteroides spp.; Lachnospiraceae family; Bifidobacterium spp.; Lactobacillus reuteriIn vitro (16S rRNA gene amplicon sequencing, metatranscriptomics, coculture experiment, biochemical assays) experimentsKeystone OTUsPropionate synthesis and gut microbiota metabolic homeostasis131OscillibacterIn vitro (Biochemical analyses; histochemical hnalysis; 16S rRNA gene amplicon sequencing and RT-qPCR) experimentsKeystone OTUOxidative stress; obesity132aThe identification method denotes the specific experimental approaches employed in the original literature.b“Keystone taxa” corresponds to established mechanistic pathways, while “keystone OTUs” refers to the link to disease-related processes. Gut keystones reported in recent years (listed partly) aThe identification method denotes the specific experimental approaches employed in the original literature. b“Keystone taxa” corresponds to established mechanistic pathways, while “keystone OTUs” refers to the link to disease-related processes. Mechanisms underlying the interspecies–interactions of keystone bacteria Understanding the interactions among keystone bacteria and gut microbes is the first step toward dietary intervention to promote intestinal microecological balance. Although interactions among the members of microbiota typically vary with the environmental conditions, space, and time, they still can be classified into positive, neutral, and negative interactions, with negative interactions—both direct and indirect—outnumbering the positive ones1,8. Positive interactions Positive interactions between microbes are crucial for maintaining stability in diverse communities via mutual regulation. However, modeling and empirical data shows that positive interactions between microbial species range from commensalism to true cooperation, with cross-feeding or syntrophy typically encountered in the microbiome. Cross-feeding refers to the sharing of energy and nutrients in the form of metabolites between various species or strains of the microbial community. Bacteroides represents a crucial keystone genus in the human gut microbiome that aids in the digestion of ingredients in food and produces vital nutrients and energy needed for the growth of other bacterial species, and thereby, in preserving gut homeostasis. For instance, Bacteroides uniformis and Bacteroides plebeius are capable of utilizing agarose and alginate polysaccharides, respectively, and cooperate to exploit the native alginate and agarose polysaccharides derived from Porphyra haitanensis26 (seaweed; Fig. 1A). Furthermore, Bacteroides ovatus secretes the exogenous inulin–degrading hydrolases Bo04502 and Bo04503; the activities of these enzymes release oligosaccharides or monosaccharides that can be assimilated by Bacteroides vulgatus37 (Fig. 1B). Intriguingly, the fitness of B. ovatus is enhanced via the extracellular digestion of inulin due to reciprocal benefits received from B. vulgatus (Fig. 1B). Additionally, B. ovatus hydrolyzes konjac glucomannan into β-mannans owing to the activity of its endogenous mannanase BoMan26B and facilitates the cogrowth of Faecalibacterium prausnitzii on the polysaccharide konjac glucomannan14 (Fig. 1B). The presence of the transmembrane protein FpMOBP (a component of a specific ABC transporter that binds exogenous β-mannans) and cellular proteins FpGH113, FpCE17, and FpCE2 in F. prausnitzii facilitates the degradation of intracellular β-mannans to glucose and galactose14 (Fig. 1B).Fig. 1Positive interactions between gut keystones and the key metabolites-producing pathways.A Bacteroides plebeius and Bacteroides uniformis collaboratively utilize polysaccharides from Porphyra haitanensis. B Bacteroides ovatus degrades inulin to support Bacteroides vulgatus, while also breaking down konjac glucomannan to nourish Faecalibacterium prausnitzii. C Nutritional interactions between Bacteroides thetaiotaomicron and Desulfovibrio piger, Collinsella aerofaciens, Faecalibacterium prausnitzii, Clostridium difficile, Escherichia coli, and Salmonella typhimurium. D The metabolic pathways of Bacteroides thetaiotaomicron producing succinate and acetate (pyruvate metabolism) and Collinsella aerofaciens producing lactate (glycolytic pathway) and hydrogen (NADH oxidation pathway). Solid black arrows indicate direct production effects; dashed black arrows denote reactions for which the required genes are unknown in the corresponding model strains. Solid red arrows represent known utilization or promoting effects of key metabolites, while dashed red arrows indicate unknown utilization or promoting effects. Solid green arrows illustrate cellular proliferation processes. Light green highlighted text denotes key metabolites involved in cross-feeding. Text within rectangular boxes indicates enzymes (upright) or genes (italicized). Red pentagons signify that the production of these metabolites involves multiple steps. Positive interactions between gut keystones and the key metabolites-producing pathways. A Bacteroides plebeius and Bacteroides uniformis collaboratively utilize polysaccharides from Porphyra haitanensis. B Bacteroides ovatus degrades inulin to support Bacteroides vulgatus, while also breaking down konjac glucomannan to nourish Faecalibacterium prausnitzii. C Nutritional interactions between Bacteroides thetaiotaomicron and Desulfovibrio piger, Collinsella aerofaciens, Faecalibacterium prausnitzii, Clostridium difficile, Escherichia coli, and Salmonella typhimurium. D The metabolic pathways of Bacteroides thetaiotaomicron producing succinate and acetate (pyruvate metabolism) and Collinsella aerofaciens producing lactate (glycolytic pathway) and hydrogen (NADH oxidation pathway). Solid black arrows indicate direct production effects; dashed black arrows denote reactions for which the required genes are unknown in the corresponding model strains. Solid red arrows represent known utilization or promoting effects of key metabolites, while dashed red arrows indicate unknown utilization or promoting effects. Solid green arrows illustrate cellular proliferation processes. Light green highlighted text denotes key metabolites involved in cross-feeding. Text within rectangular boxes indicates enzymes (upright) or genes (italicized). Red pentagons signify that the production of these metabolites involves multiple steps. Blautia coccoides plays a crucial role in the model community OMM12 by participating in higher-order interactions with the other members, utilizing hydrogen, and altering the energy balance of hydrogen-producing reactions1. B. thetaiotaomicron and B. vulgatus can hydrolyze glycoconjugates at the mucosal surface into N-acetylneuraminic acid via the activity of the endogenously expressed sialidases Bt0455 and BVU_4143, respectively (Fig. 1C). In addition, B. thetaiotaomicron can convert dietary sugars into succinate; in turn, this enables the expansion of Clostridioides difficile, which harbors a gene cd2344 encoding a putative succinate transporter, in the gut via the utilization of succinate38 (Fig. 1C). Moreover, B. thetaiotaomicron employs the Bacteroides-encoded sulfatase Bt0238 to convert chondroitin sulfate (a glycosaminoglycan found in the host intestinal mucosa) into inorganic sulfate, which supports the growth and colonization of Desulfovibrio piger; in turn, D. piger significantly increases the abundance of Collinsella aerofaciens but decreases that of Escherichia coli. The lactic acid and hydrogen produced by C. aerofaciens is utilized by D. piger39 (Fig. 1C). B. thetaiotaomicron is also known to produce acetic acid and may increase goblet cell differentiation, mucus-related gene expression (for instance, that of the gene st3gal4 encoding sialyltransferase), and the ratio of sialylated to sulfated mucus, thereby contributing to mucus production40. F. prausnitzii utilizes acetic acid and produces butyrate, accompanied by absence of goblet cell differentiation and mucus glycosylation, which helps to maintain the proper ratio of various cell types in the epithelium. Consequently, the intestinal diseases caused by C. difficile and Salmonella typhimurium may be prevented by interfering with the colonization of the concerned microorganisms by regulating the abundance of B. vulgatus and B. thetaiotaomicron; notably, the enzymes encoded by operons involved in sialic acid metabolism in C. difficile (CdnanE, CdnanA, and CdnanT) and S. typhimurium (StnanA and StfucI) allow these microorganisms to utilize sialic acid to enhance their own abundance in the gut38,41. As mentioned above, cross-feeding appears to establish four trophic levels. The breakdown of complex polysaccharides by primary degraders such as bacteria of the genus Bacteroides to release oligo- and monosaccharides, which are then available for utilization by the other species, constitutes the first trophic level. The next level involves the breakdown of sugars by primary fermenters such as B. thetaiotaomicron and C. aerofaciens, which produce phosphoenolpyruvate (PEP) via the glycolytic pathway. PEP is then utilized for substrate level phosphorylation, resulting in the production of organic acids such as formate, acetate, and succinate (Fig. 1D) as well as alcohols, including 1,2-propanediol. At the third trophic level, secondary fermenters such as F. prausnitzii utilize these by-products via several fermentative or respiratory pathways to generate short-chain fatty acids (SCFAs) such as acetate, butyrate, and propionate. At the fourth level, both primary and secondary fermenters utilize molecular hydrogen as an electron donor for sulfate reducing bacteria, methanogens, and acetogens42. Negative interactions Negative interactions chiefly comprise competition for nutrients and space in the gut as well as interference. Interestingly, bacteria of the phylum Bacteroidetes generally have a detrimental effect on the community, while certain members of Actinobacteria and Firmicutes show multiple beneficial interactions with others42. The relative abundance of Ruminococcus obeum was found to consistently increase when mice were infected with Vibrio cholerae, effectively limiting the colonization of the pathogen in mice43. The underlying mechanisms were investigated via RNA-sequencing of fecal samples; an obvious decrease in the expression of the genes tcpA (encoding a major factor for colonization of the human intestine), rtxA and hlyA (encoding accessory toxins), and VC1447-VC1448 (encoding RtxA transporter) was observed in R. obeum43 (Fig. 2A). The gene luxS encoding S-ribosylhomocysteine lyase is a key regulator in R. obeum and involved in the synthesis of autoinducer-2 (AI-2), which directly impacts the abundance of V. cholerae. The expression of LuxS and production of AI-2 in R. obeum are greatly enhanced upon invasion by V. cholerae; in turn, the AI-2-mediated quorum sensing leads to the suppression of multiple colonization factors in V. cholerae, including toxin coregulated pilus, which are essential for V. cholerae infection in humans44,45 (Fig. 2A). Furthermore, cocolonization experiments with V. cholerae mutants have shown that R. obeum AI-2 can reduce the colonization and pathogenicity of the pathogen via a hitherto uncharacterized pathway that is independent of the V. cholerae AI-2 sensor, LuxP43. Additionally, the quorum-sensing transcriptional regulator VqmA of V. cholerae influences the early colonization of gut via an unidentified mechanism43 (Fig. 2A). These interactions therefore serve to limit the colonization of V. cholerae or other enteropathogens43.Fig. 2Negative interactions between gut keystones.A Ruminococcus obeum synthesizes the signaling molecule autoinducer-2 (AI-2) to inhibit the colonization of Vibrio cholerae in the gut. B Collinsella scindens produces secondary bile acids (SBA) that suppress Clostridium difficile. C Enterococcus faecalis gains a competitive advantage for colonization by transferring the pPD1 plasmid to surrounding bacteria through conjugation and by synthesizing the bacteriocin bac-21. Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased expression levels, respectively. pPD1 the conjugative plasmid pPD1, BAC bacteriocin, bac-21 bacteriocin-21. Negative interactions between gut keystones. A Ruminococcus obeum synthesizes the signaling molecule autoinducer-2 (AI-2) to inhibit the colonization of Vibrio cholerae in the gut. B Collinsella scindens produces secondary bile acids (SBA) that suppress Clostridium difficile. C Enterococcus faecalis gains a competitive advantage for colonization by transferring the pPD1 plasmid to surrounding bacteria through conjugation and by synthesizing the bacteriocin bac-21. Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased expression levels, respectively. pPD1 the conjugative plasmid pPD1, BAC bacteriocin, bac-21 bacteriocin-21. The metabolites produced by certain keystone species in the microbiome can impact other members of the community indirectly or without providing any clear benefits. These keystone bacteria may also release bacteriocins to selectively alter the composition of the microbiota, which may range from type VI secretion systems and bacteriocins to feedback inhibition via the induction of host immune responses46. A majority of species in the human gut microbiome, including Lactobacillus spp., E. coli, Bacillus subtilis, Enterococcus spp., and Bacteroides spp., have been observed to secrete one or more types of antagonistic compounds such as bacteriocins47. For instance, the competition in the gastrointestinal tract caused by the bacteriocin secreted by Enterococcus faecalis results in a shift in the composition of the microbiota. This shift may be a result of displacement of the aboriginal microflora containing E. faecalis, competing out of other bacteria such as Porphyromonas gingivalis, or stimulation of the growth of B. thetaiotaomicron2. The presence of E. faecalis decreased the levels of antioxidants such as those produced via the ergothioneine biosynthesis pathway48. A recent study reported that the sex hormone–responsive conjugative plasmid pPD1 encodes enterococcal bacteriocin-21 (bac-21) (Fig. 2B), which shares the same nucleotide sequence with AS-48, another well-researched enterococcal bacteriocin. The operon bac on the plasmid pPD1 comprises nine genes (bacA to bacI); the gene bacA encodes the precursor of bac-21, while the genes bacB to bacI encode proteins necessary for the processing, modification, and secretion of a mature bacteriocin composed of 70 amino acids. In general, bac-21 facilitates the competitive colonization of E. faecalis as well as immune evasion49 (Fig. 2B). Clostridium scindens is another bacterium found in the intestine and is capable of dehydroxylating secondary bile acids, significantly inhibiting the growth of C. difficile, and improving resistance to infection by promoting the production of secondary bile acids50 (Fig. 2C). Butyrate is secreted by butyrate-producing bacteria such as Blautia and is known to enhance the ability of mouse macrophages to neutralize various intestinal pathogens, including S. typhimurium and adherent-invasive enterotoxigenic E. coli, by increasing the production of antimicrobial peptides and improving the ability of the macrophages to engulf pathogens51. Additionally, butyrate reduces oxygen levels in the epithelial cells, restricting the aerobic growth of Salmonella52. Propionate, an SCFA, is also capable of inhibiting the growth of S. typhimurium by disrupting its intracellular pH homeostasis53. Dietary modulation on keystone bacteria As mentioned above, the composition of the gut microbiome is frequently impacted by a range of environmental and biological factors, including temperature, pH, dietary habits, immune system of the host, metabolites, as well as interspecies microbial interactions. Among these, the influence of diet on gut microbiota is paramount and cannot be ignored, especially given that daily dietary intake has been reported to reshape the gut microbiota by impacting keystone bacteria. The regulatory mechanisms underlying the selective influence of various dietary patterns or substrates on the growth and metabolic activities of keystone bacteria are presented herein. Firstly, dietary compositions directly influence luminal pH, creating selective intestinal environments for keystone bacteria. High-protein diets promote an alkaline shift via microbial urease, favoring the proliferation of Proteobacteria (such as E. coli) and alkali-tolerant Bacteroides54–56. Besides, high-protein diets activate the gene gad encoding glutamic acid decarboxylase, which reduces the intracellular concentration of H+, thereby suppressing the growth of acidophilic bacteria such as Lactobacillus and Streptococcus thermophiles57,58 (Fig. 3). By contrast, diets rich in fermentable fibers promote the production of SCFAs, lowering pH and fostering the growth of acidophilic bacteria such as Bifidobacterium and Lactobacillus but inhibiting that of pathogens such as C. difficile and Salmonella spp.59–61 (Fig. 4).Fig. 3The functioning mechanisms of nitrogen metabolites produced by keystone bacteria from diets.A The active ingredients of dietary biomarkers such as dietary protein and alkaloids can modulate the abundance of key bacterial populations within the intestinal microbiome, leading to variations in nitrogen metabolites levels. B The intestinal nitrogen metabolites from diets are closely related to a variety of diseases (such as gout, nephritis, and neurological diseases). Black arrows indicate promoting effects; gray arrows illustrate macro-level influences. Red and green arrows represent increased and decreased levels, respectively. The black dashed boxes indicate regulated gut microbiota. DC dendritic cell, ILC3 type 3 innate lymphoid cells, GLP-1 glucagon-like peptide-1, PXR pregnane X receptor, IECs intestinal epithelial cells, UA uric acid, SLC2A9 soluble carrier protein 2 family member 9, ABCG2 adenosine triphosphate binding cassette subfamily G2.Fig. 4The functioning mechanisms of short-chain fatty acids (SCFAs) produced by keystone bacteria from diets.A The active ingredients of dietary biomarkers can modulate the abundance of key bacterial populations within the intestinal microbiome, leading to variations in SCFAs levels. B Effect of SCFAs on the four-layer intestinal barrier in the host. Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased levels, respectively. Gray arrows illustrate macro-level influences. The black dashed boxes indicate regulated gut microbiota or key metabolites. AD Alzheimer’s disease, T2DM diabetes mellitus type 2, CVD cardiovascular disease, NAFLD nonalcoholic fatty liver disease, ALD alcoholic liver disease, AS atherosclerosis, IBD inflammatory bowel disease, CRC colorectal cancer, AMPK AMP-activated protein kinase, HIF hypoxia-inducible factor, MUC2 Mucin-2, HDAC8 histone deacetylase 8, SYNPO synaptopodin, TLR4 toll-like receptor 4, GLP-1 glucagon-like peptide-1, PYY peptide YY, HDAC histone deacetylase. The functioning mechanisms of nitrogen metabolites produced by keystone bacteria from diets. A The active ingredients of dietary biomarkers such as dietary protein and alkaloids can modulate the abundance of key bacterial populations within the intestinal microbiome, leading to variations in nitrogen metabolites levels. B The intestinal nitrogen metabolites from diets are closely related to a variety of diseases (such as gout, nephritis, and neurological diseases). Black arrows indicate promoting effects; gray arrows illustrate macro-level influences. Red and green arrows represent increased and decreased levels, respectively. The black dashed boxes indicate regulated gut microbiota. DC dendritic cell, ILC3 type 3 innate lymphoid cells, GLP-1 glucagon-like peptide-1, PXR pregnane X receptor, IECs intestinal epithelial cells, UA uric acid, SLC2A9 soluble carrier protein 2 family member 9, ABCG2 adenosine triphosphate binding cassette subfamily G2. The functioning mechanisms of short-chain fatty acids (SCFAs) produced by keystone bacteria from diets. A The active ingredients of dietary biomarkers can modulate the abundance of key bacterial populations within the intestinal microbiome, leading to variations in SCFAs levels. B Effect of SCFAs on the four-layer intestinal barrier in the host. Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased levels, respectively. Gray arrows illustrate macro-level influences. The black dashed boxes indicate regulated gut microbiota or key metabolites. AD Alzheimer’s disease, T2DM diabetes mellitus type 2, CVD cardiovascular disease, NAFLD nonalcoholic fatty liver disease, ALD alcoholic liver disease, AS atherosclerosis, IBD inflammatory bowel disease, CRC colorectal cancer, AMPK AMP-activated protein kinase, HIF hypoxia-inducible factor, MUC2 Mucin-2, HDAC8 histone deacetylase 8, SYNPO synaptopodin, TLR4 toll-like receptor 4, GLP-1 glucagon-like peptide-1, PYY peptide YY, HDAC histone deacetylase. Secondly, dietary compositions can significantly alter the level of dissolved oxygen within the gut. High-fat diets, including the typical American (“Western”) diet containing approximately 36–40% fat, compromise epithelial integrity and lead to elevated local levels of dissolved oxygen62. This oxygen-rich environment promotes the proliferation of facultative anaerobes such as E. coli, but suppresses the growth of strict anaerobes like Bacteroides (Fig. 5). Conversely, the fermentation of high-fiber diets consumes substantial quantities of oxygen, generating an anaerobic milieu that favors the growth of strict anaerobes such as Bifidobacterium and Prevotella (Fig. 4).Fig. 5The functioning mechanisms of bile acids (BAs) produced by keystone bacteria from diets.A The active ingredients of dietary biomarkers can modulate the abundance of key bacterial populations within the intestinal microbiome, leading to variations in BAs levels. B Effect of BAs on the gut-liver axis by regulating key receptors, including FXR, TGR5, PXR, and CAR. Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased levels, respectively. Gray arrows illustrate macro-level influences. The black dashed boxes indicate regulated gut microbiota or key metabolites, while the red dashed boxes represent specific biological processes. FGF15/19 fibroblast growth factor 15/19, GLP-1 glucagon-like peptide-1, mTOR mammalian target of rapamycin, FoxO1 forkhead box protein O1, IBD inflammatory bowel disease, SHP small heterodimer partner, SERBP-1C sterol regulatory element-binding protein-1C, MRP2/3/4 multidrug resistance proteins 2/3/4, UGTIA1 UDP glucuronosyltransferase family 1 member Al, MDR1 multidrug resistance-1, CYP7A1 cholesterol 7 alpha-hydroxylase. The functioning mechanisms of bile acids (BAs) produced by keystone bacteria from diets. A The active ingredients of dietary biomarkers can modulate the abundance of key bacterial populations within the intestinal microbiome, leading to variations in BAs levels. B Effect of BAs on the gut-liver axis by regulating key receptors, including FXR, TGR5, PXR, and CAR. Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased levels, respectively. Gray arrows illustrate macro-level influences. The black dashed boxes indicate regulated gut microbiota or key metabolites, while the red dashed boxes represent specific biological processes. FGF15/19 fibroblast growth factor 15/19, GLP-1 glucagon-like peptide-1, mTOR mammalian target of rapamycin, FoxO1 forkhead box protein O1, IBD inflammatory bowel disease, SHP small heterodimer partner, SERBP-1C sterol regulatory element-binding protein-1C, MRP2/3/4 multidrug resistance proteins 2/3/4, UGTIA1 UDP glucuronosyltransferase family 1 member Al, MDR1 multidrug resistance-1, CYP7A1 cholesterol 7 alpha-hydroxylase. Thirdly, although gut temperature is generally stable, dietary patterns can indirectly influence keystone bacteria by modulating body temperature or local microenvironments. High-calorie diets that are rich in fats and sugars generate significant metabolic heat, elevating local intestinal temperatures63. This thermal stress activates the expression of the gene hsp and clp (encoding heat shock proteins, and caseinolytic protease, respectively) in thermophilic bacteria such as Bacillus stearothermophilus and Clostridium thermocellum, enabling them to maintain cellular functions under conditions of high temperature64. Additionally, the thermophiles upregulate the expression of the kdp operon to enhance the uptake of potassium ions and mitigate heat-induced cellular damage. By contrast, temperature-sensitive keystone bacteria such as Lactobacillus and Bifidobacterium downregulate the expression of ackA under conditions of high temperature, which reduces acetate synthesis and impairs their proliferation. Fourthly, the water content and water activity (Aw) of dietary components are critical determinants of the growth and metabolism of keystone bacteria. Diets rich in high-water-content foods (such as fruits and soups) maintain a high luminal Aw, promoting the growth of Bacteroides, Bifidobacterium, and Lactobacillus65. In the case of Bacteroides, the diets with high water content upregulate genes within polysaccharide utilization loci (PULs), enhancing the expression of glycoside hydrolases and outer membrane lipoproteins (SusD, SusE, and SusF) to improve polysaccharide degradation and mucosal colonization of Bacteroides66,67 In Bifidobacterium, high Aw promotes the expression of ABC transporters for carbohydrate uptake and key metabolic enzymes like fructose-6-phosphate phosphoketolase, driving the fermentation of oligosaccharides to acetate and lactate68–70. Similarly, the diets with high water content upregulate opp (oligopeptide permease) gene family for uptake of amino acids and stress-response genes sod, kat, and gpx (encoding superoxide dismutase, lysine acetyltransferase, and glutathione peroxidase respectively) to mitigate oxidative damage, supporting its survival and metabolic activity of Lactobacillus71. Conversely, diets with low water content reduce intestinal Aw, inhibiting hydration-sensitive pathogens like V. cholerae but favoring desiccation-tolerant Staphylococcus aureus and Streptomyces72,73. Additionally, dietary components directly affect luminal osmolarity. High-salt diets elevate the concentrations of intestinal Sodium (Na+) and chloride (Cl−) ions, enhancing osmotic stress that can inhibit salt-sensitive keystone bacteria like Bifidobacterium, Lactobacillus and Bacteroides74. Concurrently, dietary nutrients exert a profound and far-reaching impact on the functions of keystone bacteria by regulating their metabolism. In fact, the production of key metabolites such as SCFAs, bile acids, and nitrogenous compounds by keystone bacteria is highly dependent on dietary intake (Figs. 3–5). For instance, soluble dietary fibers such as pectin and β-glucan are usually served as preferred carbon sources for saccharolytic keystones like Bacteroides and Bifidobacterium to yield health-associated SCFAs75 (Fig. 4). A diet rich in polyphenols can act as a potent stimulant for the proliferation of key bacteria such as A. muciniphila that metabolized polyphenols into urolithin A, playing a crucial role in bile-acid homeostasis, cholesterol efflux, and mucin synthesis76,77 (Fig. 4). Polyunsaturated fatty acids can be readily assimilated by A. muciniphila as essential nutrients for its growth with forming primary and secondary bile acids78 (Fig. 5). Animal-derived proteins, typified by red meat being copiously rich in aromatic amino acids such as tryptophan and tyrosine, can be metabolized by bacteria such as E. coli and Bacteroides into nitrogen-containing compounds, including indole and p-cresol79 (Fig. 3). In general, the composition and physicochemical properties of dietary nutrients cooperatively modulate the gut microenvironment and metabolic pathways of keystone bacteria, thereby shaping the gut microbial community. Formulation of dietary intervention strategies The influences of diets on gut microbiota are both profound and complex. Thus, a meticulous compilation of dietary regulation strategies is presented from the perspective of diverse dietary patterns, dietary conditions, and feeding patterns. Specific dietary patterns Ketogenic diet (KD) is high in fats and low in carbohydrates. This diet has been shown to modulate gut microbiota by increasing the abundance of A. muciniphila and subsequently, the production of SCFAs, particularly butyrate; in turn, this is effective for a subset of patients with drug-resistant seizure80. Concurrently, KD stimulates the synthesis of ketone bodies, which can directly influence the metabolic activities of gut microbes. This leads to a reduction in proinflammatory bacterial species and an increase in anti-inflammatory bacteria, thereby promoting gut health and regulating inflammation81 (Fig. 6).Fig. 6Impact of different dietary patterns and feeding patterns via keystone bacteria on the host’s health.Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased levels, respectively. CRC colorectal cancer, HTN hypertension, PD Parkinson’s disease, AS atherosclerosis, MS metabolic syndrome, CD castleman disease, ALD alcoholic liver disease, MDD major depressive disorder, NAFLD nonalcoholic fatty liver disease. Impact of different dietary patterns and feeding patterns via keystone bacteria on the host’s health. Black arrows indicate promoting effects; T-shaped arrows represent inhibitory effects. Red and green arrows represent increased and decreased levels, respectively. CRC colorectal cancer, HTN hypertension, PD Parkinson’s disease, AS atherosclerosis, MS metabolic syndrome, CD castleman disease, ALD alcoholic liver disease, MDD major depressive disorder, NAFLD nonalcoholic fatty liver disease. Dietary approaches to stop hypertension (DASH) diet is enriched with polyphenols and has been associated with an increased diversity of the gut microbiota and enrichment of A. muciniphila (Fig. 6). Additionally, the DASH diet diminishes the abundance of Allobaculum mucolyticum and Shigella, in turn alleviating alcoholic liver disease (ALD)82. A gluten-free diet (GFD) is predominantly composed of polysaccharides and can alter the nutritional milieu in the intestine, impeding the growth of harmful bacteria. However, it may lead to a decrease in the abundance of beneficial bacteria such as Bifidobacterium longum and Lactobacillus, while increasing that of E. coli and Enterobacteriaceae. Even under these conditions, Bifidobacterium and other beneficial bacteria can maintain intestinal health through multiple mechanisms, including the production of SCFAs, enhancement of intestinal barrier integrity, and regulation of the immune response, which contribute to the alleviation of celiac disease83. The low FODMAP (low fermentable oligosaccharides, disaccharides, monosaccharides, and polyols) diet is efficacious for the treatment of metabolic syndrome (MS). It upregulates beneficial gut microbiota such as Bifidobacterium and Akkermansia, reduces potentially harmful bacteria of the family Enterobacteriaceae, improves intestinal barrier function, mitigates inflammation, and regulates metabolism-related signaling pathways such as the gut–brain axis and insulin-related signaling84. Similarly, the plant-based diet (PBD) abundant in dietary fiber, polyphenols, and lean proteins fosters the growth of beneficial gut microbiota, including Akkermansia and Gordonibacter, while reducing the abundance of proinflammatory bacteria such as those belonging to Enterobacteriaceae. PBD facilitates the production of SCFAs to enhance the integrity of the intestinal barrier, reduce systemic inflammation, and improve lipid metabolism, thereby alleviating atherosclerosis85 (Fig. 6). The Chinese dietary pattern is rich in whole grains, legumes, and dietary fiber; it increases the abundance of beneficial gut microbiota, including Bifidobacterium, Lactobacillus, Faecalibacterium, and Ruminococcus, while decreasing that of pathogenic bacteria such as Prevotella, Clostridium, Klebsiella, Streptococcus, and Oscillibacter. It promotes the production of SCFAs and modulates the gut–brain axis, thereby alleviating major depressive disorder (MDD)86. The Mediterranean diet (MeDi) abundant in dietary fiber, polyphenols, and unsaturated fats has been associated with an increase in the diversity of gut microbiota and enrichment of Prevotella, F. prausnitzii, and Muribaculaceae. It has also been shown to cause a decrease in the F/B ratio. These keystone bacteria ferment dietary fibers to generate SCFAs, which exert anti-obesity effects by activating GPCRs such as GPR43 and GPR109A87 (Fig. 6). Special dietary conditions A high-fat diet (HFD) rich in saturated lipids and cholesterol increases the abundance of gut microbes belonging to Verrucomicrobiota, Clostridia, Firmicutes, and Proteobacteria, while reducing beneficial bacteria such as Bacteroidetes. It contributes to the progression of NAFLD by increasing intestinal permeability, inducing systemic inflammation, and promoting hepatic lipid accumulation, thereby having a detrimental impact on health88. A high-salt diet (HSD) is rich in sodium chloride and can inhibit enterotoxigenic B. fragilis–induced tumorigenesis in colorectal cancer through the suppression of IL-17A and inducible nitric oxide synthase, thereby influencing the development of intestinal tumors to a certain extent89 (Fig. 6). Feeding patterns The feeding patterns may be chiefly divisible into caloric restriction (CR) and intermittent fasting (IF). The biomarker for the CR pattern is chiefly vitamins, and it may cause an increase in the F/B ratio but decrease the abundance of Enterobacteriaceae, thereby reducing hypertension (HTN)90,91. In comparison, the IF pattern involves supplementation chiefly with lipid-soluble vitamins and may cause an increase in the abundance of Firmicutes, Tenericutes, and Opisthokonta but decrease that of Proteobacteria, thus preventing Parkinson’s disease (PD)92,93 (Fig. 6). Overall, a rational selection and combination of the aforementioned dietary and feeding patterns based on their regulatory effects on keystone microbes may allow the formulation of specific diet plans; these may help restore the balance of the gut microbial ecosystem and effectively reduce the risk of various diseases, including IBD, hypertension, type 2 diabetes, coronary heart disease, and diseases of the nervous system94. In general, the effects of diets on host health are mediated via the modulation of the composition and activities of keystone bacteria in the gut. These keystone bacteria, in turn, produce metabolites that influence various physiological processes, including immune regulation, metabolic homeostasis, and intestinal barrier function. Thus, keystone bacteria serve as critical mediators between diet and host health, highlighting the importance of dietary interventions in managing gut microbiota–related diseases. Developing therapeutic keystone bacteria into next-generation probiotics (NGPs) Probiotics are defined as “live microorganisms that, when administered in adequate amounts, confer a health benefit on the host”, NGPs refer to those microorganisms that have not been used as agents to promote health to date, and which are more likely to be delivered under a drug regulatory framework95,96. Thus, certain keystone bacteria with therapeutic functions may be developed as NGPs, such as A. muciniphila, F. prausnitzii, Christensenella minuta, etc. Direct supplementation of therapeutic keystone bacteria as NGPs C. minuta was found to have a causal relationship with the host phenotype after large-scale human studies on individuals with obesity, metabolic syndrome, or a family history of these conditions. And the subsequent mono-colonization experiments in gnotobiotic models confirmed its direct role in modulating host adiposity and microbial community structure97. As a bacterium associated with leanness, it influences the interspecies interactions among gut microbes and alters metabolite compositions, resulting in the restructure of microbial community and increase in SCFAs (Fig. 4). This highlights its potential as NGPs for treating metabolic disorders, particularly in individuals with a low-abundance Christensenellaceae enterotype25,97. However, its strict anaerobic nature is a challenge for large-scale cultivation and vitality maintenance during gastrointestinal transportation, which usually requires complex and expensive manufacturing processes. For instance, the targeted oral delivery has been created to supplement this therapeutic keystone bacterium. This evolution reflects a paradigm shift from broad ecological modulation to precision microbiome engineering, involving mechanism-driven interventions, sophisticated formulation science, and an emerging focus on personalized delivery97,98. In models of diet-induced metabolic dysfunction, such as HFD feeding, gut barrier impairment and reduced A. muciniphila abundance are hallmarks99. Oral supplementation with live A. muciniphila can partially reverse HFD-induced metabolic disturbances in mice, including alleviating adipose tissue inflammation, improving insulin resistance, and enhancing intestinal tight junction protein expression100. Thus, a strategy of “A. muciniphila supplementation coupled with reduced dietary saturated fat” constitutes a synergistic approach to counter the pathological consequences of an HFD101. With the development of A. muciniphila as a therapeutic agent, a significant conceptual and practical progress appears. The pasteurized (heat-killed) A. muciniphila can also improve metabolic parameters in mice99. A randomized, double-blind, placebo-controlled trial demonstrated that daily supplementation with 1010 inactivated A. muciniphila cells for three months significantly improved insulin sensitivity, reduced plasma total cholesterol, and attenuated systemic inflammation in overweight/obese individuals with metabolic syndrome102. Moreover, a purified membrane protein, Amuc_1100®, has been identified as a key active component mediating host immune-metabolism, especially bile acid metabolism and immune modulation103 (Fig. 5). The pasteurized bacteria or purified functional components can circumvent the hurdle of colonization resistance from the gut inhabitants and enhance product stability and safety, representing a major technological upgrade in the field of probiotics100. Direct supplementation of NGPs has evolved along a trajectory marked by increasing mechanistic understanding and technological sophistication, from live, challenging-to-culture strains toward development of stabilized, formulation-optimized, and sometimes non-viable derivatives. This advancement is guided by deeper insights into keystone functions involving critical microbial metabolites. Formulation of combined approaches An expanding body of evidence indicates that specific dietary patterns can act synergistically with probiotic interventions to ameliorate various human diseases. For instance, clinical studies demonstrate that a multi-strain probiotic administered alongside an energy-restricted diet significantly improves glycated hemoglobin (HbA1c) levels in overweight or obese individuals with hypertension, underscoring its beneficial role in regulating glucose homeostasis104. Recent research suggests that integrating probiotic supplementation within a “recovery diet” framework can partially restore gut microbiota composition and enhance cardiometabolic health105. As to therapeutic F. prausnitzii, a dominant butyrate producer severely depleted in inflammatory bowel disease (IBD), is required to consider specific cultivation and application conditions when serving as NGPs106,107. In a gnotobiotic mouse model, F. prausnitzii exerted mucus-promoting and anti-inflammatory effects only when co-colonized with bacteria capable of degrading complex polysaccharides such as B. thetaiotaomicron, and fed a diet containing the corresponding fiber40,108. To ensure its therapeutic efficacy, advanced encapsulation technology and synbiotic strategy are developed to co-deliver its preferred prebiotic substrates such as inulin and fructose, and a “dietary fiber–F. prausnitzii” combination approach is expected to design in future109–111. Fortunately, the PREDICT study (Personalized Responses to Dietary Composition Trial) is developed for integrating deep phenotyping with standardized nutritional challenges112,113. This has propelled the combination of NGPs with nutritional support, to design a precision dietary regimen assisting NGPs to colonize and function. The rational design of such combinations is increasingly guided by computational models that not only predict functional outcomes but can also prospectively identify and validate candidate NGPs114,115. For instance, the metabolic network of F. prausnitzii was reconstructed in silico with the tools like the Assembly of Gut Organisms through Reconstruction and Analysis (AGORA) framework116. Almost all the known metabolic reactions of F. prausnitzii have been taken into account, this model allows researchers to simulate its growth and metabolite production under various nutritional conditions117. Based on Metage2Metabo or MICOM (Microbial Community Modeling), F. prausnitzii fed with adequate complex fiber to fuel the entire metabolic network can ameliorate colitis in gnotobiotic mouse models40,118. Besides, the MeDi being rich in diverse fibers and polyphenols, provides an ideal nutritional landscape for NGPs like A. muciniphila and F. prausnitzii, thus supplementation with A. muciniphila alongside this polyphenol-abundant diet can yield synergistic benefits119. In general, the therapeutic efficacy of NGPs is fundamentally constrained by the intestinal ecological niche, which is predominantly sculpted by host diet. These findings highlight a promising future to investigate dietary-NGP synergy and point toward its potential clinical translation. Future perspectives and research directions In this review, current identification approaches—ranging from disease-association studies and co-occurrence network analyses to dynamic and metabolic modeling—have been assessed in view of development of technology. Although each identification method has its advantages and disadvantages, combined schemes may be proposed according to their characteristics. As to identified keystone bacteria, their interactions as well as intrinsic mechanisms have been elucidated from gene level to host level, i.e., positive cross-feeding among gut microbes is essential for maintaining community diversity and stability, and their negative interactions such as competition and inhibition dominate microbial ecology. In order to modulate keystone bacteria, dietary pattern and composition have been taken into consideration for formulating dietary intervention strategies, especially dietary-NGP formula with clinical transformation potential. Future research should focus on developing more physiologically relevant models and high-resolution functional assays to unravel the intricate relationships among diet, keystone bacteria, and host physiology. There are at least three footholds. First, the deep integration of multiomics functional analysis and validation are required, metagenomic, metatranscriptomic, metabolomic, and culturomic data will evolve from correlation analysis to mechanistic understanding. The metabolic functions and dietary responsiveness of keystone bacteria will be clarified. Second, the tailored dietary or probiotic regimens can be designed by integrating individual microbiome data with machine learning predictions, paving the way for precision nutrition and disease management based on the modulation of keystone bacteria. Personalized approaches that account for host genetics, microbiome baseline, and lifestyle are needed to optimize interventions. Third, emerging sequencing technologies (such as single-cell metagenomics, metatranscriptomics, and metaproteomics) combined with integrated multi-omics and in vivo validation approaches will facilitate a deeper understanding of the metabolic potential, interactions, and beneficial effects of keystone bacteria. These technologies will help resolve strain-level functions and real-time activity in complex communities. By leveraging systems-biology frameworks and translational validation, we can design effective, evidence-based dietary NGPs therapies targeting keystone taxa, offering promising strategies for preventing and managing gut microbiota-related diseases.