Intratumoral microbial heterogeneity in HCC reveals a potential therapeutic target.

Introduction
Hepatocellular carcinoma (HCC) is the fourth leading cause of cancer-related mortality worldwide, accounting for over 700,000 deaths annually.1,2 Owing to the lack of apparent symptoms and reliable predictive biomarkers at early stages, most patients are diagnosed at advanced stages, thereby missing the optimal window for surgical intervention. Moreover, the high heterogeneity of HCC results in significant variations in therapeutic responses and prognoses among patients, even at the same pathological stage.2,3] Therefore, elucidating the mechanisms underlying HCC pathogenesis and identifying novel biomarkers and therapeutic targets remain critical priorities in current research.
The tumor microenvironment (TME) plays a pivotal role in tumor initiation, progression, and treatment response. It comprises various cellular components, including tumor cells, immune cells, endothelial cells, and microbial metabolites, the complex interactions of which directly influence tumor behavior and therapeutic outcomes.4,5 Advances in microbial detection technologies have identified the intratumoral microbiota in multiple cancers, such as breast, lung, ovarian, pancreatic cancers, melanoma, bone tumors, and brain tumors. As critical components of the TME, these tumor-resident microbes modulate immune responses via their metabolites, either by promoting or suppressing tumor immunity.6,7 For example, intratumoral Lactobacillus reuteri metabolizes dietary tryptophan into indole-3-aldehyde, activating the aryl hydrocarbon receptor and inducing the activation of interferon-gamma (IFN-γ)-secreting CD8+ T cells, thereby enhancing antitumor immunity.8 Conversely, metabolites such as deoxycholic acid produced by Clostridium scindens inhibit CD8+ T cell-mediated antitumor responses, thereby promoting colorectal cancer progression.9 These findings underscore that distinct microbial communities and their metabolites exert opposing effects on tumor progression by modulating the immune microenvironment. As an inherently heterogeneous and complex component of tumors, the TME warrants further investigation to elucidate its heterogeneity and internal processes, which may help identify critical regulatory nodes and improve the precision of cancer therapy.10 Microbiota-based TME profiling has attracted increasing attention. For instance, Sun et al.11 classified microbial profiles based on intratumoral microbial heterogeneity and analyzed the interactions among the microbiome, clinical characteristics, and prognosis in HCC. Moreover, a clinical study of HBV-related HCC categorized intratumoral microbiota into two subtypes, bacteria-dominant and virus-dominant, demonstrating correlations between microbial heterogeneity, clinical features, and TME differences.12 Previous studies have reported altered intratumoral microbial diversity and composition in HCC compared with adjacent non-tumor tissues, suggesting a potential link between microbial dysbiosis and hepatocarcinogenesis.13 Despite these advancements, the biological functions of the dominant microbial species contributing to heterogeneity, their molecular mechanisms in reshaping the TME, and the interactions among microorganisms within specific subtypes remain inadequately characterized, necessitating further investigation.
In this study, we performed quantitative analysis of the microbiota in tissue samples from patients with HCC using 2bRad-M sequencing, classifying tumors into Ralstonia-dominant (RT) and non-Ralstonia-dominant (nRT) subtypes. Metabolic profiling of different microbial communities was conducted using ultra-performance liquid chromatography–mass spectrometry (UPLC-MS) and gas chromatography–mass spectrometry (GC-MS), which revealed the metabolic characteristics associated with different microbial subtypes. Further in vitro and in vivo experiments demonstrated that Ralstonia sp. might suppress HCC progression by modulating glycerophospholipid metabolism. Additionally, analysis of the immune cell composition within the tumor microenvironment suggested that Ralstonia sp.-mediated regulation of glycerophospholipid metabolism may alter immune cell infiltration, highlighting its potential antitumor activity. This study advances our understanding of the HCC microbiome and identifies a novel microbial subtype with therapeutic implications. These findings contribute to our understanding of the microbial composition of HCC and suggest a potential antitumor role for certain bacteria, providing preliminary insights into microbiota-based therapeutic strategies for HCC.

Materials and methods

Tissue specimens
This study involved 113 tissue samples, which were all obtained from patients undergoing hepatectomy at the Second Hospital of Nanjing, affiliated with Nanjing University of Chinese Medicine between May 2021 and October 2022. Tumor tissues (T group) and adjacent normal tissues (NA group) were collected from 48 HCC patients, along with normal tissues (N group) from 17 patients with liver hemangioma. Individuals aged ≤18 or ≥80 years, as well as those diagnosed with autoimmune diseases, HIV infection, syphilis, alcoholism, or recent antibiotic treatment, were excluded. Sterile surgical blades were utilized for the collection of tumors and adjacent normal tissues from patients with HCC, as well as normal tissues from patients with hepatic hemangioma. To ensure the prevention of cross-contamination, a new blade was used for each individual specimen. The collected samples were promptly transferred into enzyme-free, sterile cryopreservation tubes under aseptic conditions. Additionally, an empty tube was prepared to serve as a negative control for 2bRAD-M sequencing. This control tube was briefly exposed to ambient air for 30 s before being cryopreserved in liquid nitrogen alongside the tissue samples.

Study approval
The study was conducted in strict compliance with the ethical guidelines outlined in the Declaration of Helsinki. The study was approved by the Ethics Committee of Nanjing Second Hospital (Approval No.: 2022-LS-ky034), and written informed consent was obtained from all participants before sample collection. All animal experiments were conducted in compliance with the guidelines of the Institutional Animal Care and Use Committee of the Nanjing Drum Tower Hospital, Nanjing, China (Approval No.: 2024AE01044).
For further details regarding the materials and methods, please refer to supplementary materials.

Results

Characteristics of study participants
The study cohort included 113 samples, tumor tissues (T group, n = 48), and matched normal adjacent tissues (NA group, n = 48) from patients with HCC and normal liver tissues (N group, n = 17) from patients with hepatic hemangioma who served as the normal control group. The detailed clinical and pathological information on the patients is provided in Table 1 and Fig. S1. The patients with HCC consisted of 37 males (77.1%) and 11 females (22.9%) (59 ± 10.45 years old), and the patients with hepatic hemangioma consisted of 11 (64.7%) males and six (35.3%) females (54 ± 16.34 years old). The majority of patients with HCC (81.25%) were infected with HBV or HCV, compared with 29.4% in the control group, highlighting the strong association between viral hepatitis and liver cancer progression. In addition, alpha-fetoprotein (AFP) levels were markedly elevated in the HCC group, which is consistent with its role as a clinical marker. However, there was no close correlation between AFP level and tumor size or the degree of liver cirrhosis, suggesting that AFP elevation may be influenced by other biological factors beyond tumor burden. Comorbidity patterns further differentiate the cohorts. In the HCC group, 12 patients (25%) had coexisting diabetes and 13 (27.1%) had hypertension. In contrast, among the hepatic hemangioma group, four patients (23.5%) had diabetes and six (35.3%) had hypertension. Notably, liver cirrhosis was common in patients with HCC, affecting 36 of 48 (75%), whereas only two cases (11.8%) were observed in the hemangioma group. These features provide additional insight into the clinical complexity of HCC and highlight the potential interplay between metabolic dysregulation and tumor pathology.

Intratumoral microbial and metabolic landscapes of HCC were heterogeneous
Quality control information for 2bRAD-M sequencing is shown in Table S1, and 1,050,446,365 clean reads were obtained. A total of 344 unique microbial species were detected and the alignment was based on 2b-Tag-DB (Fig. S2A). Rank abundance analysis showed the richness and uniformity of the species in the samples (Fig. S2B). Species richness in the normal liver tissues group was significantly higher than that in the tumor and matched adjacent liver tissues groups (Fig. 1A). Notably, the normal group had the highest biodiversity (which was estimated by the Chao 1 index, Shannon index, and Simpson index), whereas the tumor group had the lowest biodiversity (Fig. 1B) (p <0.0001, p = 0.0031, and p = 0.0018, respectively). A Venn diagram showed that 58 of the 344 total species were shared among the three groups, and 88 species were common in patients with HCC. Moreover, 79 and 57 species were unique to the tumor and normal groups, respectively (Fig. 1C). β-Diversity analysis, measured by principal coordinate analysis (PCoA) and non-metric multidimensional scaling (NMDS) analysis, showed a distinct cluster in tumor tissues compared with non-tumor tissues (NA and N groups) (Fig. 1D).
To identify the key phylotypes in the tumor microenvironment, the bacterial community composition in all samples was analyzed. The dominant bacteria in each group were Proteobacteria, Actinobacteria, Firmicutes, and Bacteroidetes, accounting for more than 98% of the taxonomy (Fig. S2C). The average abundance of Proteobacteria was higher in the HCC group (81.86%) than that in the normal group (70.25%). However, the average abundance of Actinobacteria was lower in the HCC group (13.44%) than that in the normal group (16.91%). Similarly, the abundance of Firmicutes was significantly lower in the HCC group (3.95%) than that in the normal group (10.80%). Additionally, the abundance of Bacteroidetes was not significantly different between the HCC and control groups (0.38% vs. 0.46%). At the genus level, the abundances of Ralstonia, Klebsiella, Acinetobacter, Cutibacterium, Sphingomonas, and Bacillus were highest in the microbial profiles (Fig. S2D). To recognize the high-dimensional biomarkers in patients with HCC, linear discriminant analysis (LDA) effect size (LEfSe) was used to search for the dominant microbiota responsible for differences between different clinical characteristics (Fig. 2A). Firmicutes and Ralstonia genera (including Ralstonia_sp001078575, Ralstonia_sp003851545, and Ralstonia_sp000801955) were enriched in the normal tissues. The random forest algorithm suggested that microbes, including high-abundance Ralstonia sp. and Klebsiella are important biomarkers for group classification (Fig. S2E, Fig. 2B). The presence of the dominant populations plays a crucial role in preserving the stability of the microbial community composition and function within tissue environments. Therefore, we identified 22 key bacterial species by comprehensively considering the factors of species abundance, LEfSe, and random forest analysis, and these species showed significant differences among the three groups (Fig. 2C). In addition, Sankey diagrams showed that the dominance of Proteobacteria and Klebsiella pneumoniae gradually increased, accompanied by changes in other bacterial populations (Fig. 2D). Notably, Ralstonia (Ralstonia_pickettii, Ralstonia_sp007997035, Ralstonia_pickettii_B, and Ralstonia_sp003851545) exhibited a markedly higher abundance in normal tissues than in tumor tissues, underscoring the distinct disparity between the two (Fig. 2E). Consistent with the results of microbial sequencing, qPCR analysis demonstrated that the levels of Ralstonia sp. were significantly higher in normal tissues than in tumor tissue samples (p <0.05) (Fig. 2F). We conjecture that elevated bacterial levels in both normal tissues and those adjacent to tumors may play a role in inhibiting tumor progression.
To investigate the impact of microbial metabolites on tumors, non-targeted metabolomics was conducted using UPLC-MS and GC-MS on the same cohort used for 2bRAD-M sequencing. Principal component analysis showed a clear separation between the T group and both the NA and N groups, with minor overlap between NA and N, indicating distinct biological characteristics (Fig. S3A). A total of 835 and 734 differentially expressed metabolites (DEMs) were identified in the T vs. NA and T vs. N groups, respectively, with a p value <0.05 and variable importance of projection (VIP) >1 (Fig. S3B, Table S2). We performed a correlation analysis between 22 indicator bacterial genera and DEMs. The results revealed distinct microbial–metabolite associations, particularly involving phospholipid-related metabolites (Fig. S3C). Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment of these associated metabolites showed significant enrichment in pathways such as glycerophospholipid metabolism, choline metabolism in cancer, and glycerolipid metabolism (Fig. S3D), indicating that specific microbial signatures may influence lipid metabolic reprogramming in the tumor microenvironment.

Specific microbial population inducing metabolic alterations as potential biomarkers for HCC
Based on differential abundance analysis, Klebsiella and Ralstonia were identified as key taxa associated with tumor-specific microbial signatures. We subsequently explored the relationship between microbial alterations and metabolic changes. Correlation analysis indicated that specific shifts in the microbiota may be linked to distinct metabolic profiles. Notably, metabolites most strongly associated with the high-dimensional microbial markers Ralstonia and Klebsiella included glycerophospholipids, glycerolipids, and several other metabolite classes (Fig. 3A). These findings suggest a potential interplay between microbial dysbiosis and host metabolic reprogramming within the tumor microenvironment. These intratumoral microbe-associated metabolites were significantly enriched in KEGG pathways, such as glycerophospholipid and choline metabolism in cancer (Fig. 3B).
Next, we developed a microbial signature based on Klebsiella pneumoniae and Ralstonia species (including Ralstonia_sp000801955, Ralstonia_sp003851545, and Ralstonia insidiosa), as well as a signature based on glycerophospholipids and choline metabolites (including LysoPC(18:1(11Z)/0:0), LysoPC(16:0/0:0), PC(16:1(9Z)/0:0), glycerophosphocholine, and acetylcholine). A combined signature was created by integrating the two. Receiver operating characteristic (ROC) curve analysis was used to evaluate the potential of these key microbes and metabolite signatures as biomarkers for HCC identification (Fig. 3C). The AUC values indicated high accuracy for both the microbial and metabolite-based signatures in distinguishing between normal and HCC cohorts, with AUC values of 0.84 and 0.87, respectively. The combined signature demonstrated superior classification performance with an AUC value of 0.91. In contrast, traditional tumor biomarkers, including AFP, carcinoembryonic antigen (CEA), and carbohydrate antigen199 (CA199), exhibited relatively low AUC values of 0.71, 0.56, and 0.64, respectively. The microbial, metabolite, and integrated microbial–metabolite signatures developed in this study exhibited promising predictive performance compared with classical tumor markers. However, the discriminatory performance of this microbial–metabolite-based model requires further validation in independent cohorts to establish its reproducibility and generalizability.
To further explore the potential clinical relevance of these microbial signatures, we conducted Spearman correlation analysis between the intratumoral microbiota and clinical parameters. The results indicated that the Chao1 index correlated negatively with alkaline phosphatase (ALP) and gamma-glutamyl transferase (GGT), Shannon index correlated negatively with ALP, and Simpson index correlated positively with C-reactive protein but negatively with the lymphocyte-to-monocyte ratio (p <0.05, Fig. S4). In particular, Ralstonia sp. exhibited a negative association with CA199 and liver cirrhosis, suggesting its potential role in HCC resistance (p <0.05) (Fig. 3D).

Ralstonia-signature microbial subtypes with clinical relevance influence glycerophospholipid metabolism in hepatocellular carcinoma
As an important intratumoral microbial candidate for HCC differentiation, Ralstonia is closely associated with the clinical characteristics of patients with HCC. According to NMDS, the 48 HCC tumor samples were classified into two molecular subtypes based on microbial signatures (Fig. 4A, Table S3). One subtype, termed RT, was characterized by a microbial community structure prominently enriched for Ralstonia species. The other subtype, termed nRT, showed higher relative abundances of other genera, such as Klebsiella and Acinetobacter (Fig. 4B). This molecular classification reflects underlying microbial heterogeneity within the tumor microenvironment. Notably, Ralstonia relative abundance was substantially higher in the RT group (≥0.45), further supporting this subtype distinction. LEfSe analysis (threshold value with LDA score >4) (Fig. 4C) revealed a significant enrichment of Ralstonia in the RT group, whereas Klebsiella was more abundant in the nRT group (with p <0.05). Both the random forest analysis (Fig. S5A) and indicator analysis (Fig. S5B) further support the validity of the classification and confirm the opposite enrichment patterns of Ralstonia and Klebsiella between the two subtypes.
Alpha diversity comparisons using Chao1, Shannon, and Simpson indices revealed higher microbial richness and evenness in the RT group. Although Chao1 showed no statistically significant difference, Shannon and Simpson indices were significantly higher in the RT group (p <0.05), indicating greater microbial diversity (Fig. 4D). Functional predictions based on the KEGG database indicated that bacterial functions were predominantly linked to various metabolic and immune-related pathways, including purine metabolism, pyruvate metabolism, Th17 cell differentiation, glycerophospholipid metabolism, choline metabolism in cancer, glycerolipid metabolism, and other pathways (Fig. 4E, Table S4). Interestingly, significant differences in glycerophospholipid and choline metabolism in cancer were observed in the microbial communities between the RT and nRT groups, as well as between tumor and normal tissues. Moreover, NMDS and PCoA analyses demonstrated considerable similarity between the microbial communities in the RT group and those in normal tissues, suggesting that the Ralstonia-dominated microbiota may be more representative of a healthy microbial balance (Fig. S5C and D).
To investigate the metabolic heterogeneity between the two subtypes, we performed an in-depth analysis of non-targeted metabolomics data (Table S5). Supervised orthogonal partial least-squares discrimination analysis (OPLS-DA) showed a clear separation between the two subtypes (Fig. 4F). DEMs were identified through univariate statistical analysis (p <0.05, and VIP >1), showing that lipids, lipid-like molecules, and organic oxygen compounds were downregulated in the RT microbial subtype compared with nRT (Fig. 4G). KEGG pathway predictions aligned with the T vs. N group comparison, with DEMs significantly enriched in glycerophospholipid and choline metabolism in cancer pathways (Fig. 4H). This suggests that Ralstonia-dominant microbiota (RT subtype) may be associated with a healthier metabolic profile.
We also analyzed clinical and pathological data, including comorbidities (diabetes, hypertension, and HBV infection) and tumor characteristics (size, thrombus, Edmondson-Steiner grade, and Scheuer’s stage) for patients with HCC in the RT and nRT subtypes. Clinical assessment indicated that patients in the RT group had fewer comorbidities and milder tumor characteristics, including smaller tumor size and reduced tumor thrombus formation, as well as lower Scheuer’s classification stages, suggesting a possible association between Ralstonia-enriched microbiota and less severe liver fibrosis (Fig. 4I).
Finally, the clinical tumor markers CA199, AFP, and CEA were analyzed and compared between the two subtypes. Although the AFP and CEA levels were lower in the RT group, no statistically significant differences were observed for AFP (p = 0.9218) or CEA (p = 0.0646). In contrast, CA199 levels were significantly lower in the RT group than in the nRT group (p = 0.0269), suggesting that Ralstonia-dominant microbiota may exhibit antitumor properties (Fig. 4J).

Ralstonia sp. inhibits hepatocellular carcinoma proliferation and metastasis with demonstrated safety in mice
The antitumor effects of Ralstonia sp. on HCC were examined through a series of in vitro and in vivo experiments using the human HCC cell line Huh-7 and the mouse HCC cell line Hepa1-6. Bifidobacterium longum was used as a positive control based on its reported tumor-suppressive activity. In the CCK-8 assay, Ralstonia sp. exhibited dose- and time-dependent inhibition of cell viability, surpassing the effects observed with Bifidobacterium longum (p <0.05, Fig. 5A). Colony formation assays further confirmed these inhibitory effects, as a significant reduction in colony number was observed after Ralstonia sp. treatment (p <0.05, Fig. 5B). Transwell migration and invasion assays demonstrated that Ralstonia sp. reduced the number of migrating and invading cells in a dose-dependent manner, indicating its capacity to suppress these malignant behaviors (Fig. 5C). To further explore whether the observed cytotoxicity of Ralstonia sp. was associated with apoptosis, Hoechst 33258 staining and Annexin V-FITC/PI double revealed an apparent increase in apoptotic cell features in Ralstonia sp.-treated cells (Fig. S6). These findings indicate that Ralstonia sp. can inhibit proliferation, migration, and invasion of HCC cells while promoting apoptosis, highlighting its potential antitumor properties.
To verify the in vivo inhibitory effect of Ralstonia sp. on HCC, we used the Hepa1-6 cell line to establish a transplantation tumor model in 5-week-old C57BL/6J mice (Fig. 5D). The experiment was divided into three groups: mock, Bifidobacterium longum, and Ralstonia sp. treatment groups. Mice received intratumoral injections of PBS, Bifidobacterium longum, or different concentrations of Ralstonia sp. every 3e days. The results demonstrated that Ralstonia sp. treatment significantly inhibited tumor growth, with pronounced effects in the medium- and high-concentration groups (Fig. 5E and F). In contrast, although the Bifidobacterium longum group also showed some inhibitory effects, it was less effective than the Ralstonia sp. treatment group.
Next, we assessed the safety of Ralstonia sp. in localized tumor therapy. Over the 28-day experiment, all mice exhibited a steady increase in body weight, with no statistically significant differences observed between the Bifidobacterium longum, Ralstonia sp., and mock groups (Fig. 5G). Blood biochemical analyses revealed that Ralstonia sp. had no significant effect on liver function markers (alanine aminotransferase [ALT], aspartate aminotransferase [AST], and ALP) or kidney function markers (creatinine [CRE] and blood urea nitrogen [BUN]) (Fig. 5H). Histopathological analysis showed no apparent tissue damage in the liver, kidney, or spleen of the mice treated with Ralstonia sp. (Fig. 5I). In contrast, H&E staining of tumor tissues in the model group revealed disorganized, densely packed tumor cells with prominent nuclear atypia and loss of tissue structure. However, the Ralstonia sp. treatment group exhibited an improvement in the cellular structure. TUNEL (terminal deoxynucleotidyl transferase dUTP nick-end labeling) analysis revealed enhanced apoptotic signal in Ralstonia sp.-treated tumor tissues, suggesting a possible role in apoptosis induction (Fig. S7). To further confirm the specificity of the observed antitumor effects, we performed additional in vivo experiments using Klebsiella pneumoniae (a negatively associated species) and heat-killed Ralstonia sp. (Ral-HK), both administered at 1 × 108 colony-forming units (CFU). Neither Klebsiella pneumoniae nor Ral-HK suppressed tumor growth compared with the mock group, whereas viable Ralstonia sp. retained potent inhibitory effects, suggesting that the antitumor activity is both viability-dependent and specific to Ralstonia sp. (Fig. S8).
Together, these results indicate that viable Ralstonia sp. may exert a specific and effective antitumor response in vivo, without causing systemic toxicity, and may promote structural improvement within tumor tissues.

Transplanted tumors transcriptome reveals Ralstonia sp.-associated gene alterations mainly enriched in immune and metabolic pathways
To characterize tumor transcriptomic profiles associated with Ralstonia sp., we collected transplanted mouse tumor samples for RNA sequencing. A total of 1,018 differentially expressed genes (DEGs) were identified based on the criteria of p value <0.05 and |log2 FC| >1, including 372 downregulated genes and 646 upregulated genes (Fig. 6A, Table S6). Hierarchical clustering highlighted the top 10 upregulated and downregulated genes with the highest fold-change between the two groups (Fig. S9A). Gene Ontology (GO) enrichment analysis revealed that these DEGs were primarily enriched in biological processes related to cholesterol metabolic process, lipid metabolic process, and negative regulation of lipid catabolic process; cellular components such as high-density lipoprotein particles, very-low-density lipoprotein particles, and lipid droplets; and molecular functions, including serine-type endopeptidase activity and ketosteroid monooxygenase activity (Fig. 6B). These results indicate that the DEGs were predominantly enriched in lipid metabolism-related pathways. Moreover, reactome enrichment analysis revealed significant enrichment of DEGs in pathways associated with the innate immune system, chemokine receptor binding, complement cascade, general metabolism, and lipid metabolism, all of which are associated with immune and metabolic functions (Fig. 6C). Subsequently, RNA-sequencing data were analyzed using CIBERSORT to assess immune cell infiltration within the tumor microenvironment. The analysis revealed that macrophages, memory CD8+ T cells, plasma cells, and memory CD4+ T cells were the predominant infiltrating immune cell types across samples (Fig. 6D). A comparative trend was observed between the mock and Ralstonia sp. treatment groups, with higher estimated infiltration levels of M1 macrophages, CD8+ T cells, gamma delta T cells, and Th2 cells in the Ralstonia sp. group (Fig. 6E). This suggests that Ralstonia sp. may inhibit tumor growth by activating or enhancing the host immune response.
To further explore how Ralstonia sp. treatment modulates tumor immunity and metabolism, we performed a correlation analysis between metabolism-related genes and immune cell infiltration. The results showed that macrophages, CD8+ T cells, and gamma delta T cells were strongly associated with several metabolism-related genes (Fig. 6F). Notably, genes involved in lipid metabolism were significantly correlated. Moreover, GO and KEGG enrichment analyses highlighted the pivotal role of lipid metabolism in shaping the tumor microenvironment (Fig. S9B and C).

Transplanted tumors metabolomics reveal Ralstonia sp. affect glycerophospholipid metabolism and its correlation with immune cells infiltration
To explore the alterations in the tumor metabolic microenvironment associated with Ralstonia sp., we conducted UPLC-MS and GC-MS. OPLS-DA demonstrated a clear separation between the Ralstonia sp.-treated group and mock-treated groups, indicating a substantial remodeling effect of Ralstonia sp. on tumor metabolism (Fig. 7A). Univariate statistical analysis adhering to the criteria of p <0.05 and VIP >1, we observed 39 metabolites exhibiting downregulation and 117 upregulation in their expression profiles (Fig. 7B, Table S7). KEGG pathway enrichment analysis of these DEMs (p <0.05) identified pathways primarily associated with glycerophospholipid, glycerolipid, and choline metabolism in cancer (Fig. 7C). Notably, this analysis also revealed pathways related to Th1 and Th2 cell differentiation.
Consistent with findings from human tissue samples, the metabolomic profiles of tumors from Ralstonia sp.-treated mice showed a general trend of glycerophospholipid metabolism downregulation, as observed in the metabolomic profiles (Fig. S10), further supporting the role of Ralstonia sp. in modulating host physiology via this pathway. Furthermore, our correlation analysis between tumor metabolic changes associated with Ralstonia sp. treatment and immune cell infiltration suggested a trend of negative associations between most glycerophospholipid metabolites and M0 macrophages, M2 macrophages, Th1 cells, Th2 cells, CD4+ naïve T cells, and DC-active cells (Fig. 7D). These data suggest that Ralstonia sp.-associated metabolic profiles are potentially linked to distinct immune cell signatures.

Discussion
With advancements in microbial detection technologies, intratumoral microbes have been identified across various tumor types, including HCC.5,6,11,14 Nevertheless, there remains a lack of systematic and comprehensive understanding of the microbial ecosystem within tumor tissues. Previous studies on intratumoral microbes in HCC have predominantly focused on prospective predictive analyses, with comparatively few validation studies. In this study, we examined microbial community data and incorporated metabolomic data from the same clinical cohort to explore the microbial mechanisms in HCC at the molecular level. Based on microbial heterogeneity analyses, we categorized intratumoral microbes in HCC and identified two distinct subtypes with different metabolic characteristics. Further screening revealed a bacterium with antitumor activity, Ralstonia sp., which was subsequently validated by in vitro and in vivo experiments. Additionally, in a mouse hepatocellular carcinoma transplanted model, we characterized tumor metabolic and immune signatures associated with Ralstonia sp. administration in vivo, highlighting its potential as a microbial target for HCC treatment.
Microbial dysbiosis has been demonstrated to be associated with various liver diseases, such as primary biliary cholangitis, cirrhosis, and fatty liver disease.15,16 Our results similarly showed that the microbial diversity in HCC tissues is lower than that in normal tissues.14,17 Altered microbial diversity and composition in HCC relative to adjacent non-tumor tissues have been reported, implicating potential links to hepatocarcinogenesis.13 This may be attributable to the pro-inflammatory milieu and immunosuppressive characteristics of the TME, alongside adverse factors, such as hypoxia and acidity, which further contribute to the reduction in microbial diversity.18,19 Additionally, metabolites produced by the microbiota serve as critical modulators of inflammation, immune responses, and signaling pathways within the TME.20 A clinical study on colorectal cancer established a model integrating eight gut microbiome-associated serum metabolites (GMSM panel), which demonstrated superior diagnostic accuracy compared with the widely utilized biomarker CEA across multiple cohorts.21 In this study, we developed a model associating intratumoral microbiota with glycerophospholipid and choline metabolism, which exhibited enhanced diagnostic accuracy for HCC compared with CEA, AFP, and CA199. However, given the variability of microbiome signatures across populations and platforms, validation in larger, geographically diverse, and prospective cohorts is essential to confirm the robustness and clinical relevance of our findings. Comparative analysis of the microbiota across various pathological liver tissues revealed that Ralstonia sp. was more abundant in adjacent non-tumor and normal tissues than in tumor tissues. Furthermore, the abundance of Ralstonia sp. negatively correlated with serum CA199 levels and cirrhosis-related pathological features, suggesting that this bacterium and its metabolites may exert tumor-suppressive effects. Consistent with this observation, similar patterns of Ralstonia sp. depletion in tumor tissues have also been observed in other HCC studies,14,22] as well as in other malignancies such as breast cancer,23 gastric cancer,24,25 and intrahepatic cholangiocarcinoma,26 supporting the possibility that reduced intratumoral Ralstonia sp. abundance may be a shared feature across malignancies. Huang et al.27 observed that patients with a higher relative abundance of Megasphaera in tumor tissues had prolonged survival. In vitro and vivo experiments have demonstrated that Megasphaera, in combination with anti-PD-1 antibody therapy, enhances tumor suppression in murine tumor models, accompanied by metabolic alterations.27 Based on these findings, it can be inferred that the tumor-specific microbiota may improve patient prognosis by modulating specific metabolic pathways to enhance antitumor immunity. Conversely, evidence suggests that intratumoral injection of Roseburia intestinalis accelerates tumor progression in a subcutaneous xenograft lung cancer model, highlighting the heterogeneous roles of distinct bacteria in tumor progression across different cancer types.28
By conducting comprehensive analyses of tumor genetic characteristics, clinicians can achieve a more precise understanding of the biological behavior of tumors, thereby facilitating the development of tailored treatment strategies for patients. Lee et al.29 identified five clinically and molecularly distinct consensus subtypes based on transcriptomic and clinical features, which demonstrated high reproducibility and offered significant guidance for precision therapy in HCC. As a component of the TME, the intratumoral microbiome influences the initiation, progression, and prognosis of HCC. Recent studies have classified HCC into bacteria-dominant and virus-dominant subtypes based on the microbial expression profiles in tumor tissues, correlating these subtypes with clinical characteristics.12 Certain microbial subtypes have also been utilized to predict postoperative outcomes in patients.11 Given the critical role of microbial metabolites in TME, we performed metabolomic analyses of the two microbial subtypes.20,30 These subtypes exhibited distinct metabolic profiles, with differences that resembled those observed between the tumor and normal liver tissues. Furthermore, the Ralstonia sp.-dominant microbial subtype displayed close similarity to the microbial diversity and composition of normal tissues, suggesting that this microbial community may maintain stability within the TME and potentially exert tumor-suppressive effects on HCC growth.
In vitro experiments revealed that Ralstonia sp. inhibits the proliferation, migration, and invasion of HCC cells. To further validate the antitumor effects of Ralstonia sp. In vivo, we used the Hepa1-6 syngeneic cell line to establish an orthotopic liver cancer model in mice. This approach minimizes severe immune rejection and preserves immune integrity, enabling the investigation of microbial influences on the immune and metabolic landscapes of the TME.31 To replicate its native growth conditions accurately, Ralstonia sp. was directly injected into the tumor tissues. Compared with systemic administration via intravenous injection, intratumoral delivery demonstrated a more pronounced inhibitory effect on tumor growth, with increased production of metabolites and more direct and rapid immune responses.32 The efficacy of systemically delivered therapeutic microbes in reaching the tumor site is considerably reduced, particularly in large animals and human patients, because of the larger blood volume and relatively smaller tumor size, which present significant challenges.33,34 Intratumoral injection overcomes these limitations by directly depositing a substantially higher microbial load within the target tumor. Moreover, the risks of microbial leakage and systemic infection are higher with intravenous administration than with localized intratumoral injection, where the risk is more manageable. In addition, orally delivered therapeutic microbes are often rapidly cleared by phagocytes or metabolic organs, resulting in low bioavailability and a limited therapeutic window.35
In functional mechanism studies, the impact of the microbiome on tumor progression and development is not only mediated by the regulation of cell proliferation and apoptosis but also through reprogramming of the immune system and its responses. Specific bacteria identified in pancreatic cancer tissues, such as Glycomyces, Pseudomonas fluorescens, and Streptomyces, have been shown to enhance the antitumor immune response by promoting the recruitment and activation of CD8+ T cells.5 Microbial metabolites serve as crucial mediators of the interaction between the intratumoral microbiome and immune microenvironment.20,36 Recent studies have shown that under the influence of exogenous IFN-γ and IL-12 released by dendritic cells, the microbial metabolite inosine binds to the adenosine 2A receptor on T cells, consequently promoting Th1 cell differentiation.37 This mechanism significantly enhances the antitumor effects of Th1 cells in various cancers including melanoma, bladder cancer, and colorectal cancer. Additionally, butyrate, a metabolic product of gut bacteria, has been reported to upregulate TLR4 expression and mediate innate immune responses against tumor cells, thereby inhibiting cell proliferation and promoting apoptosis.38 In this study, the metabolic profiling differences between the two microbial subtypes highlighted the glycerophospholipid metabolism. In a mouse xenograft model, the metabolic profile similarly revealed the influence of Ralstonia sp. on glycerophospholipid metabolic pathways. Moreover, an immune microenvironment analysis of paired tumor tissues indicated that macrophages, helper T cells, CD4+ T cells, and dendritic cells were all associated with glycerophospholipids. Therefore, we hypothesized that Ralstonia sp. may influence the tumor immune landscape indirectly via associations with glycerophospholipid metabolic alterations, potentially contributing to the resistance of Ralstonia sp. to HCC progression and underscoring its potential as a therapeutic target for HCC. Our data further indicate that Ralstonia-associated alterations in glycerophospholipid metabolism are closely associated with immune cell infiltration, particularly involving macrophages and CD8+ T cells. However, the underlying molecular mechanisms remain unclear. Previous studies have shown that glycerophospholipid metabolism can influence immune responses by modulating membrane lipid composition, lipid raft signaling, and the production of pro-inflammatory lipid mediators such as lysophosphatidylcholine and phosphatidic acid, which in turn regulate macrophage polarization and T cell activation.[39], [40], [41], [42]
Ralstonia sp. may alter the abundance of local bioactive glycerophospholipid metabolites, thereby influencing the activation of key immune regulatory pathways such as NF-κB, PI3K-Akt, and PPARγ. In addition, bacterial metabolites may affect the expression of host lipid-metabolizing enzymes, such as PLA2G4 and LPCAT, ultimately reshaping the metabolic–immune axis. Although these mechanistic pathways remain to be functionally validated, they are supported by our correlative analyses and are consistent with previously reported microbiota-immune interactions. In future work, mechanistic investigations will be undertaken to further clarify these molecular interactions.
This study has several limitations. First, our research was based on a relatively small cohort of patients with HCC from the East China region, which may limit the generalizability of the results. Larger, multicenter, prospective clinical trials are needed to further validate the clinical significance of microbiome-based HCC subtypes. Second, because of its cross-sectional design, the study cannot capture dynamic changes in the intratumoral microbiome throughout the course of disease progression or treatment. Longitudinal studies incorporating serial sampling and time-resolved analyses will be essential to elucidate how microbial shifts relate to tumor evolution, therapeutic response, and clinical outcomes. Finally, the precise molecular mechanisms by which Ralstonia sp. modulates the immune microenvironment through metabolic regulation remain to be elucidated.
Overall, this study provides a comprehensive characterization of intratumoral microbial heterogeneity in HCC and identifies distinct microbial subtypes with clinical and functional relevance. Ralstonia sp. emerged as a potential antitumor candidate, exerting its effects through modulation of glycerophospholipid metabolism and remodeling of the tumor immune microenvironment. These findings offer novel insights into microbiota-driven tumor biology and support the development of microbiome-informed strategies for patient stratification and therapeutic intervention in HCC.

Abbreviations
AFP, alpha-fetoprotein; ALP, alkaline phosphatase; ALT, alanine aminotransferase; AST, aspartate aminotransferase; BUN, blood urea nitrogen; CA199, carbohydrate antigen199; CEA, carcinoembryonic antigen; CFU, colony-forming unit; CRE, creatinine; DEGs, differentially expressed genes; DEMs, differentially expressed metabolites; GC-MS, gas chromatography–mass spectrometry; GGT, gamma-glutamyl transferase; GGT, gamma-glutamyl transferase; GO, Gene Ontology; HCC, hepatocellular carcinoma; IFN-γ, interferon-gamma; KEGG, Kyoto Encyclopedia of Genes and Genomes; LDA, linear discriminant analysis; LEfSe, linear discriminant analysis effect size; NMDS, non-metric multidimensional scaling; nRT, non-Ralstonia-dominant subtype; OPLS-DA, orthogonal partial least-squares discrimination analysis; PCoA, principal coordinate analysis; Ral-HK, heat-killed Ralstonia sp.; ROC, receiver operating characteristic; RT, Ralstonia-dominant subtype; TME, tumor microenvironment; TUNEL, terminal deoxynucleotidyl transferase dUTP nick-end labeling; UPLC-MS, ultra-performance liquid chromatography–mass spectrometry; VIP, variable importance of projection.

Financial support
This work was supported by the Jiangsu Provincial Municipal Health Commission (M2021017), the Suzhou Municipal Health Commission (QNXM2024077), the Jiangsu Provincial Graduate Practice and Innovation Program (SJCX23_0821, KYCX24_2175), the High-level Talent Research Project of the Second Hospital of Nanjing (0313504), and the Jiangsu Innovation and Entrepreneurship Talent Program –Jiangsu Province Outstanding Medical Expert Talent Project (0313505).

Authors’ contributions
Study conception and design: LL, YY. Patient recruitment and screening: YY, ZJ. Sample collection and clinical data gathering: LL, YY, MW, ZJ. Assistance in the collection of clinical data: MW, CY. Oversight of trial implementation: YY, CY. Data analysis: LL, MW. Manuscript writing: LL, MW. Manuscript revision: YY, CY. Authorship order was determined based on overall contributions to the study conception, sample collection, data analysis, and manuscript preparation.

Data availability
Data supporting the findings of this study are included in this article and the supplementary materials. The 2bRAD-M sequencing data reported in this paper were deposited in the GSA, China National Center for Bioinformation/Beijing Institute of Genomics, Chinese Academy of Sciences (https://ngdc.cncb.ac.cn/gsa: accession no. CRA021380). Non-targeted metabolomics sequencing data were deposited in the OMIX database (https://ngdc.cncb.ac.cn/omix: accession no. OMIX008326). The relevant data are available from the corresponding authors upon request.

Conflicts of interest
The authors declare no conflicts of interest that pertain to this study. Please refer to the accompanying ICMJE disclosure forms for further details.