Gastric cancer (GC) is a major global health challenge, with pronounced geographic heterogeneity in incidence and mortality, and a particularly heavy burden in East Asia [1, 2]. Although overall GC incidence and mortality have declined worldwide in recent years, GC remains among the leading causes of cancer morbidity and mortality in certain regions such as China, accounting for nearly half of newly diagnosed cases globally [2]. The etiologies of GC are multifactorial, involving Helicobacter pylori (H. pylori) infection, dietary exposures, host genetic background, and environmental factors [3, 4]. Moreover, GC is highly heterogeneous in its clinicopathological presentation, encompassing distinct histological types and molecular features that jointly shape disease progression, therapeutic response, and patient outcomes [5, 6].

GC development follows a multistep pathological process known as the Correa sequence (CS). This classical model describes the gradual transition from normal gastric mucosa through chronic gastritis, atrophic gastritis, intestinal metaplasia, and dysplasia, ultimately culminating in intestinal-type GC [7, 8, 9]. H. pylori infection is considered a principal driver of this cascade, promoting lesion advancement through persistent inflammation and a series of molecular alterations [10, 11]. Recent progress in molecular biology and immunology has deepened our understanding of the CS, revealing stage-specific genomic alterations, epigenetic modifications, immune-microenvironment remodeling, and stem-cell-like properties [12]. These advances validate the traditional pathological framework while uncovering more complex mechanisms and heterogeneous trajectories, thereby providing a theoretical basis for early prevention, precision screening, and targeted therapy.

Multiple studies have demonstrated that the gut microbiota (GM) of patients with GC differs markedly from that of healthy individuals, typically characterized by reduced $\alpha$-diversity, depletion of beneficial taxa, and enrichment of conditionally pathogenic bacteria [13, 14]. Such dysbiosis may promote gastric carcinogenesis through several mechanisms, including induction of chronic inflammation, disruption of the mucosal barrier, production of genotoxins, and modulation of the immune microenvironment [15]. For example, H. pylori can disturb the gastric microbiome, further disseminate to the intestine, and trigger systemic immune responses [16]. Mendelian randomization analyses have suggested that certain taxa (e.g., Clostridium sensu stricto 1 and Rikenellaceae) are associated with reduced GC risk, whereas Roseburia and the Eubacterium brachy group may confer increased risk. Compositional signatures—such as the ratio of Streptococcus to Bacteroides—can differentiate GC patients from healthy controls, with diagnostic performance (area under the curve, AUC, 0.81–0.86) that may surpass some traditional tumor markers [17]. Microbiota features also track with tumor stage, age, and molecular markers; for instance, alterations are often more pronounced in early GC than in advanced disease, and microbial composition can differ between younger and older patients [18]. Collectively, these findings support the potential of microbiome profiling as a noninvasive diagnostic tool and prognostic indicator. Recent multi-omics studies have begun to unravel the complex microbial and metabolic ecosystem underlying gastric carcinogenesis. For example, a large-scale multi-omics integration [19] demonstrated that dysbiosis-associated depletion of short-chain fatty acid (SCFA) producers coincides with metabolic signatures of inflammation and oxidative stress in early gastric lesions. Another comprehensive atlas combining metagenomics and metabolomics [20] revealed that specific microbial guilds—particularly Streptococcaceae and Lactobacillaceae—shape amino acid and nucleotide metabolic pathways that are progressively rewired along the Correa cascade. Moreover, metabolome-resolved microbiome profiling [21] identified distinct microbial–metabolite co-modules that differentiate gastritis, intestinal metaplasia, and early gastric cancer with high discriminatory accuracy. A recent integrative study [22] further highlighted that microbial shifts in Actinobacteria and Firmicutes strongly correlate with perturbations in bile acid, lipid, and tryptophan metabolism, emphasizing the value of multi-layered omics for mechanistic insight.

Metabolomics studies leveraging plasma, serum, saliva, and other biofluids have identified numerous GC-associated metabolites. For example, a plasma model combining trimethylamine-N-oxide (TMAO) with rhamnose achieved an AUC of 0.961 for discriminating GC [23], while a serum panel composed of acylcarnitines (e.g., C6DC, C16OH) and arginine yielded an AUC of 0.99778 [24]. Salivary metabolomics identified cytosine and 2-oxoglutaric acid as biomarkers with sensitivities exceeding those of some classical serum markers [25]. Moreover, metabolomics of extracellular-vesicle cargo combined with a nanocapture workflow and machine learning achieved highly accurate classification of early GC (AUC = 1.0) [26]. These observations underscore the promise of metabolomics for noninvasive diagnosis.

Despite these advances, several critical knowledge gaps remain. Most previous studies have focused on either the gut microbiota or metabolomics in isolation, without systematically integrating both approaches to delineate stage-specific alterations during GC progression. In particular, the dynamic interplay between microbial dysbiosis and host metabolic reprogramming across the Correa sequence has not been comprehensively characterized. Furthermore, the diagnostic and prognostic utility of combined microbiota–metabolite signatures for early detection and disease stratification remains largely unexplored. Therefore, in this study we aimed to (i) comprehensively characterize gut microbiota composition and metabolic profiles in fecal and gastric tissue samples across different stages of GC; (ii) integrate microbiome and metabolomic data to elucidate key microbe–metabolite–host interaction networks; and (iii) evaluate the diagnostic potential of combined microbial and metabolic features in distinguishing early GC and precancerous lesions from controls. This integrative approach is expected to provide novel insights into the microbiota–metabolism axis in gastric carcinogenesis and may facilitate the development of non-invasive biomarkers for early detection and precision management.

This study systematically delineates dynamic changes in the gut microbiota and metabolites during gastric carcinogenesis, highlighting coordinated and associated changes between specific bacterial taxa and perturbations in related metabolic pathways. We confirm the gradual dysbiosis and metabolic remodeling predicted by the Correa sequence, with microbial and metabolic alterations emerging most prominently at the atrophic stage (G3). In the G1–G3 (vs2) comparison, richness indices (Chao1, Observed OTUs) showed significant reductions (p 0.05), indicating selective loss of low-abundance taxa, while evenness-based indices (Shannon, Simpson, Pielou’s) remained stable. This apparent discrepancy between decreased richness and unchanged Shannon/Simpson indices is biologically plausible. Early microbial shifts along the Correa sequence tend to involve the selective loss of low-abundance or conditionally beneficial taxa rather than a redistribution of dominant taxa, resulting in reduced richness but preserved community evenness. Because Shannon and Simpson metrics are more strongly influenced by high-abundance taxa, their stability suggests that the major bacterial groups remain relatively unchanged while rare taxa diminish. Moreover, the continuous, stepwise nature of our cohort—representing adjacent stages of gastric lesion progression rather than binary healthy–cancer contrasts—further explains why subtle richness-focused alterations emerge earlier than changes in overall community uniformity. Beta-diversity analysis by PERMANOVA (Bray–Curtis, R² = 0.052, p = 0.031) further supported compositional divergence from controls beginning at G3. In parallel, both fecal and tissue metabolomics revealed the first consistent appearance of amino acid– and lipid-related alterations at this stage, which persisted or amplified through erosion (G4) and GC (G5). These convergent microbial and metabolic signals suggest that atrophy represents an emerging inflection point—rather than an abrupt breakpoint—marking the onset of sustained ecological and metabolic remodeling along gastric carcinogenesis. Erosion (G4) appears to act as a transitional phenotype, bridging atrophic dysbiosis and tumor-associated reprogramming. We further demonstrate that combined microbiota–metabolite signatures have potential value for early diagnosis and disease stratification. Our findings are consistent with and extend recent multi-omics studies that have characterized the microbial–metabolic landscape of gastric cancer. A multi-center analysis [19] reported early depletion of SCFA-producing bacteria accompanied by inflammatory metabolic signatures, aligning with our observation that richness declines at the atrophic stage while SCFA-related metabolites decrease. Another integrative framework [20] highlighted microbial networks involving Streptococcaceae and Lactobacillaceae as key drivers of amino acid reprogramming; similarly, we observed enrichment of Streptococcus-related taxa in erosive and cancer stages together with amino acid and lipid pathway disruptions. Furthermore, stage-specific metabolite–microbe modules identified in a multi-layered cohort [21] mirror our finding that atrophy represents a pivotal metabolic inflection point, with erosion functioning as a transitional phenotype. A recent metagenomic–metabolomic coupling study [22] reported strong associations between Actinobacteria expansion and dysregulated bile acid and tryptophan metabolism, which is consistent with our ROC-identified role of Actinobacteria and the observed perturbations in these metabolic pathways.

We acknowledge that several clinical covariates may influence microbiome and metabolome variation, and have therefore expanded our analysis to better account for potential confounding. Platelet counts, systemic inflammatory markers, BMI, H. pylori infection status, smoking and alcohol consumption, and dietary salt intake were incorporated into Table 1 and examined for differences across groups. Although none of these variables fully explained the observed multi-omic patterns, their potential contributions—particularly inflammation-related indices—should be considered when interpreting the results. Given the modest sample size and cross-sectional design, residual confounding cannot be excluded, and future longitudinal studies with stratified or multivariable modeling will be needed to more precisely quantify the effects of these clinical factors on microbiome–metabolome interactions.

Beyond biological interpretation, the combined microbiota–metabolite signatures identified in this study have meaningful clinical translational potential. Such integrative markers could be particularly useful for screening individuals at high risk for gastric cancer—such as those with chronic gastritis, atrophic gastritis, or long-term H. pylori infection—by enabling non-invasive detection of early mucosal alterations before endoscopically visible lesions emerge. In addition, these signatures may serve as dynamic indicators for monitoring the evolution of precancerous lesions, providing a complementary tool to endoscopy for risk stratification and longitudinal surveillance. From a feasibility perspective, fecal sample collection is simple, low-cost, and well accepted by patients, and the workflow for 16S rRNA sequencing and standardized metabolomics is becoming increasingly accessible in clinical laboratories. As sequencing and mass spectrometry costs continue to decline, these combined markers could be incorporated into routine population-level screening programs or follow-up algorithms for high-risk cohorts. Thus, the translational potential of microbiota–metabolite profiling extends beyond mechanistic insights and may inform real-world strategies for early detection, disease monitoring, and precision prevention.

Microbiome analyses revealed a decline in $\alpha$-diversity and progressive restructuring of community composition. Controls were enriched for SCFA-producing genera (Bacteroides, Faecalibacterium), whereas disease stages featured increased Firmicutes, Actinobacteria, and inflammation-related taxa. These patterns align with prior work suggesting that dysbiosis can disrupt the mucosal barrier, sustain chronic inflammation, and accumulate genotoxic metabolites, thereby promoting gastric carcinogenesis [27, 28]. Interestingly, the enrichment of Bifidobacterium observed in advanced stages (G5) warrants a nuanced interpretation. Although Bifidobacterium is traditionally regarded as a probiotic genus with anti-inflammatory and immunostimulatory properties, recent studies highlight its strain- and context-dependent roles in tumor biology. Certain strains can enhance antitumor immunity and improve immune checkpoint inhibitor efficacy through STING and interferon signaling pathways, whereas others have been identified within tumor tissues as colonizers of hypoxic or necrotic niches, likely reflecting ecological adaptation rather than causation. Reports in gastric cancer cohorts are also mixed—some showing enrichment, others depletion—depending on sampling site (stool vs. mucosa), tumor stage, H. pylori infection, and medication exposure. Therefore, the observed increase in Bifidobacterium in G5 should be interpreted as an ecological signal rather than a uniformly deleterious event. Notably, the atrophic stage (G3) concentrated the most salient microbial shifts, supporting its role as a critical node where “microbiota–host interactions” transition from homeostasis to carcinogenesis.

At the metabolomic level, both fecal and tissue profiles exhibited stage-wise perturbations. Fecal data suggested reductions in anti-inflammatory SCFAs and increases in pro-inflammatory or pro-carcinogenic metabolites (e.g., kynurenine pathway intermediates, secondary bile acids) emerging already at G3. Tissue profiles revealed broader reprogramming in GC, including energy, nucleotide, and lipid metabolism, consistent with proliferative demands and immune evasion. Importantly, all metabolomic analyses were corrected for multiple testing using the Benjamini–Hochberg FDR procedure, and biological interpretations were based primarily on metabolites passing the FDR 0.05 threshold. Nominally significant findings were described only as exploratory observations pending further validation. In line with earlier studies in serum, saliva, and extracellular vesicles [29, 30], our findings reinforce the value of multi-omics integration for pinpointing metabolic inflection points and candidate biomarkers.

Correlative analyses linked depletion of SCFA producers to reduced mucosal immunity and altered signaling patterns that were associated with lesion advancement. Within the Correa framework, H. pylori-driven chronic inflammation likely perturbs gastric and intestinal communities, with metabolite shifts subsequently modulating local immunity and signaling to accelerate lesion advancement [31, 32].

Importantly, ROC analyses indicated that specific taxa (e.g., class Actinobacteria, family Oscillospiraceae) and metabolite combinations (e.g., butyrate + kynurenine + secondary bile acids) achieved good discrimination among controls, precancerous lesions, and GC (AUCs $>$0.75). The class Actinobacteria reached an AUC of 0.935 for control versus GC, underscoring its potential as a non-invasive biomarker. Compared with single-modality approaches, combined microbiota–metabolite signatures may improve early screening accuracy and stratification robustness.

The integrative correlation analysis of the G3 vs G1 comparison revealed biologically coherent microbe–metabolite modules that reflect functional remodeling during the early atrophic stage. The strong positive associations between Alistipes, Akkermansia, Actinomyces, and Bacteroides with lipid-related metabolites—including monoacylglycerols and arachidonic-acid derivatives—suggest enhanced lipid turnover, altered membrane remodeling, and possible activation of inflammation-linked lipid pathways in G3. The parallel enrichment of amino-acid metabolites and their correlations with G3-enriched taxa further imply increased proteolytic activity and nitrogen metabolism. Conversely, the negative associations observed for Muribaculaceae and Nitrosphaera with several lipid metabolites are consistent with the decline of saccharolytic functionalities reported in early atrophic gastritis. Although the PERMANOVA effect sizes were modest, which is common in microbiome studies, the coordinated shifts observed at G3 suggest that this stage may represent a potential transitional or acceleration point rather than a discrete inflection boundary.

This study has limitations. First, the sample size is modest; some findings did not remain significant after multiple-testing correction and require validation in larger cohorts. Second, the cross-sectional design precludes causal inference; longitudinal studies and mechanistic models are needed. In addition, although we incorporated available data on H. pylori infection and basic dietary habits into exploratory analyses, residual confounding cannot be fully excluded. Future large-scale longitudinal studies should incorporate comprehensive dietary assessment and H. pylori stratification to refine the microbiota–metabolome interaction models. Third, metabolite annotation is inherently limited; some features lack definitive pathway mapping. Moreover, because the present study is exploratory and cross-sectional with a modest sample size, advanced causal inference analyses such as co-abundance network modeling and mediation effect testing were not performed, and will be addressed in future longitudinal studies with expanded cohorts. Additionally, the functional roles of key microbial biomarkers identified by LEfSe (such as Bifidobacterium and Streptococcaceae) were not experimentally validated through in vitro assays or animal models, and future studies will incorporate bacterial culture supernatant experiments and gnotobiotic or microbiota-transplant models to strengthen causal inference. To overcome these limitations, future work will employ a longitudinal cohort design with repeated sampling across the Correa sequence to capture temporal microbial and metabolic dynamics. Integration of metagenomics, metatranscriptomics, and targeted metabolomics will enable higher-resolution functional mapping of microbial metabolism and host signaling.

In parallel, gnotobiotic mouse and antibiotic-treated models will be used to verify causality between specific taxa (e.g., Actinobacteria, SCFA-producing bacteria) and gastric mucosal inflammation or metabolic reprogramming through immune and signaling pathways such as NF-$\kappa$B, AMPK/mTOR, and cytokine networks. Finally, expansion of cohort size and implementation of machine-learning-based integrative frameworks will improve the robustness and predictive value of microbiota–metabolite signatures for early GC detection and risk stratification. Furthermore, while our findings reveal clear associations between gut microbial shifts, SCFA depletion, and metabolic remodeling along the Correa sequence, the causal mechanisms remain to be elucidated. In particular, whether decreased SCFAs exacerbate mucosal inflammation via NF-$\kappa$B or cytokine signaling, and how Actinobacteria enrichment may shape metabolic dependencies through AMPK/mTOR or immune-modulatory pathways, warrants further mechanistic validation. Future studies integrating metatranscriptomics, targeted metabolomics, and immune assays in cellular and animal models are planned to elucidate these regulatory networks. Future work should combine targeted validation, deeper multi-omics (transcriptomics/proteomics), and functional assays to elucidate causal “microbiota–metabolism–host” mechanisms in GC.

In summary, by integrating fecal and tissue untargeted metabolomics with 16S profiling, we chart a stage-resolved “microbiota–metabolite–host” interaction landscape across GC progression. Our results suggest that atrophy (G3) may represent a potential inflection stage, whereas erosion (G4) exhibits transitional features. Combined microbial and metabolic features show promise for early detection and stratified management, while offering mechanistic insights into metabolic dependencies that could inform precision prevention and therapy.

The datasets used and analyzed during the current study are available from the corresponding author on reasonable request.

JY and XS conceived and designed the research. BW and YL performed the experiments. YY, HZ, XZ and GW analyzed the results and data. JY and BW wrote, YL and XS revised the manuscript. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

This study was approved by the ethics committee of The People’s Hospital of Chizhou City (approved No. 2023-KY-17). We certify that the study was performed in accordance with the 1964 declaration of HELSINKI and later amendments. Written informed consent to participate in this study was provided by the patients or their families/legal guardian.

We thank Zhao Zhang and Yan Ma (Center of Human Microecology Engineering and Technology of Guangdong Province, Guangdong Longsee Biomedical Corporation, Guangzhou, China) for statistical consultation and technical support provided by the Longseek high-throughput zebrafish screening platform for drug and probiotic evaluation. Also we acknowledge Dr. Shihao Huang from Hainan University for language polishing.

This study was funded by 2023 Chizhou City major science and technology special project.

The authors declare no conflict of interest.

During the preparation of this work the authors used Deepseek-R1 in order to check spell and grammar. After using this tool, the authors reviewed and edited the content as needed and takes full responsibility for the content of the publication.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/FBL46553.