A tumor-resident Staphylococcus epidermidis strain exhibits genomic and metabolic traits driving hepatocellular carcinoma progression Authors: Jinhua Yang, Yang Liu, Bowen Li, Tianqi Zhang, Liying Yan, Kunyu Zhang, Fangqi Xu, Zhiyue Wang, Ying Liu, Baiqiang Lin, Jianming Wang, Yunwei Wei Published: 2026 ABSTRACT SummaryAlthough bacteria have been detected in hepatocellular carcinoma (HCC), the low microbial biomass has hindered isolation of viable intratumoral strains and evaluation of their functional relevance. Intratumoral microbiota profiling by 5R 16S rRNA sequencing, together with a stringent contamination-controlled workflow, identified enrichment of Staphylococcus in advanced-stage (Barcelona Clinic Liver Cancer [BCLC] C) HCC and enabled isolation of a viable S. epidermidis strain, GX3-2, from HCC tissue. GX3-2 promoted HCC cell proliferation and migration in vitro and was associated with accelerated xenograft growth in a pilot intratumoral inoculation model. Multi-omics analyses linked GX3-2 to distinct metabolic features, including adenosine and 2-phenylethylamine (2-PEA), together with host transcriptional programs related to cell cycle and epithelial-mesenchymal transition. These findings provide proof-of-concept evidence that a tumor-derived S. epidermidis isolate may modulate malignant phenotypes in HCC.Subject areas: oncology, precision medicine, molecular biology, microbiome, cell biology, omics Introduction The gut microbiota is well established as a critical regulator of tumor progression and therapeutic response.1,2 However, increasing evidence has demonstrated that intratumoral microbiota components are present in various tumor tissues, challenging the long-held doctrine of tissue sterility.3 These intratumoral microbes are not passive bystanders but are increasingly implicated as active players, contributing to tumor development by activating carcinogenic pathways and promoting tumor progression.4,5 Although large pan-cancer surveys and spatial imaging studies have mapped bacterial DNA and its intracellular localization within tumors, a critical knowledge gap remains regarding whether living, tumor-adapted strains directly drive oncogenic programs.6,7 Recent studies combining culturomics and metagenomics have provided direct experimental evidence for the translocation of viable microbes from the gut to distal organs, including the liver, via mesenteric lymph nodes.8 This validation of a microbial axis between the gut and the liver fundamentally shifts the research focus from merely detecting bacterial DNA to understanding the functional roles of these living, resident microorganisms. Addressing this functional question necessitates a departure from reliance on commercial reference strains, which are often attenuated and fail to recapitulate the complex adaptations acquired within the tumor microenvironment (TME). In contrast, bacterial isolates obtained directly from patient tumor tissues, such as hepatocellular carcinoma (HCC) tissues, are poised to reveal authentic, clinically relevant phenotypes.9 Therefore, the successful isolation and functional characterization of these bacteria derived from patients represent critical prerequisites for mechanistically dissecting their potential tumor-promoting or therapeutic contributions to cancer biology.10,11 HCC is one of the most prevalent and lethal solid tumors worldwide and is characterized by a profoundly immunosuppressive TME as well as its unique anatomical connection with the portal venous system.12 This distinctive anatomical and immunological context facilitates gut-derived bacterial translocation to the liver, positioning HCC as a particularly informative model for investigating interactions along the gut-liver axis and the intratumoral microbiota. Although recent integrative proteogenomic and transcriptomic studies have identified prognostic molecular signatures in HCC,13 the contribution of intratumoral bacteria to tumor progression and molecular heterogeneity remains insufficiently characterized.14 Unlike colorectal and other mucosa-associated cancers, HCC harbors very low bacterial biomass. Hence, successful cultivation offers a direct route to dissect their functions in HCC. Recent studies have increasingly linked HCC to the presence of intratumoral bacteria,15 with bacterial taxa such as Fusobacterium and Enterococcus being detected in HCC tissues.4,16 Some emerging studies suggest that bacterial translocation and dysbiosis may contribute to liver inflammation and carcinogenesis,17 but direct evidence of specific intratumoral bacterial profiles driving HCC remains scarce. In this study, we present a proof-of-concept (PoC) investigation to address this gap by combining multi-omics with functional validation, while recognizing that multi-omics approaches offer promising prospects for the precise diagnosis and classification of HCC.18 5R 16S rRNA profiling of HCC tissues initially revealed a positive association between Staphylococcus abundance and advanced Barcelona Clinic Liver Cancer (BCLC) staging, along with the regulation of tumor-associated transcriptomic pathways. We then successfully isolated a viable tumor-resident strain, S. epidermidis GX3-2, which was used as a representative model to explore the functional impact of this species within the TME, where S. epidermidis was frequently detected. By integrating bacterial genomics and metabolomics with host cell transcriptomics, we demonstrate that GX3-2 metabolic capabilities, particularly its enrichment in purine (e.g., adenosine) and phenylethylamine metabolism, directly align with the pro-tumorigenic transcriptional reprogramming of HCC cells, thereby defining a potential microbe-host oncogenic axis. As a foundational validation, preliminary functional assays suggest that GX3-2 alone enhances oncogenic phenotypes. Together, these findings identify HCC-derived S. epidermidis GX3-2 as a candidate microbial contributor to HCC progression and underscore the importance of studying tumor-derived strains to dissect microbe-host interactions in cancers. Results Distinct intratumoral microbiota signatures characterize early- and advanced-stage HCC To define the relationship between the intratumoral microbiota and HCC progression, we performed 5R 16S rRNA sequencing on tumor tissues from HCC patients with either early-stage (BCLC A) or advanced-stage (BCLC C) disease. Tumor tissues from the advanced-stage C cohort displayed significantly greater alpha diversity, as measured by both the Sobs and Chao indices (Figures 1A and 1B). This was complemented by beta diversity analysis, where the principal-component analysis (PCA) revealed distinct clustering of the microbial communities at the genus level, separating stage A tumors from stage C tumors (Figure 1C). Microbial signature analysis in HCC (A and B) Alpha diversity indices (Sobs and Chao) of tumor tissues stratified by BCLC stage. On the basis of the clinical records, 31 patients with early-stage (BCLC A) HCC and 20 patients with advanced-stage (BCLC C) HCC were included in this analysis (51 patients in total). Differences in alpha diversity were assessed using the Wilcoxon rank-sum test; ∗, p < 0.05. (C) PCA of microbial communities at the genus level. Beta diversity differences were evaluated by ANOSIM based on Bray-Curtis distances with 999 permutations. (D) Genus-level differential abundance analysis based on the Wilcoxon rank-sum test. (E) CCA of microbial composition with clinical parameters. (F) KEGG pathway enrichment analysis of DEGs in Staphylococcus-positive HCC patients. Differential genus analysis and KEGG enrichment analysis were corrected for multiple testing using the Benjamini-Hochberg method. A two-sided p value < 0.05 or adjusted p value < 0.05 was considered statistically significant. Differential abundance analysis subsequently identified specific taxa driving this separation. Notably, Staphylococcus genus was markedly enriched in stage C tumors, alongside Providencia and Lactobacillus, whereas genera such as Serratia and Comamonas were more abundant in stage A tumors (Figure 1D). To contextualize these microbial shifts, canonical correspondence analysis (CCA) was performed, which established a strong correlation between the microbial composition and key clinical features of tumor progression, including BCLC stage, tumor size, and microvascular invasion (Figure 1E). Moreover, we characterized the general bacterial community profile in both tumor (T) and paratumor (PT) tissues, and the results revealed that bacterial richness and diversity did not significantly differ between the T and PT tissues (Figure S1). To further investigate the relationship between Staphylococcus and tumor progression in HCC tissues, transcriptomic sequencing was performed on tumor samples from HCC patients to identify the molecular pathways associated with Staphylococcus positivity. Notably, significant alterations, including activation of the MAPK signaling pathway, which has been linked to cancer cell proliferation, survival, and metastasis, were observed in Staphylococcus-positive patients (Figure 1F). Collectively, these data show that the HCC microbiota undergoes significant remodeling during progression and highlight Staphylococcus as a key genus enriched in advanced tumors, where it is associated with a pro-proliferative host transcriptional signature. Isolation of active bacteria from HCC reveals S. epidermidis as a key functional strain Building on the 5R 16S sequencing data that identified Staphylococcus as an enriched genus in advanced HCC, we next sought to isolate viable bacteria directly from the resected tumor tissues. A rigorous culture protocol incorporating multiple environmental and procedural controls was applied, which confirmed that all the isolated colonies were authentic tissue residents and not contaminants. Comprehensive characterization via 16S rRNA sequencing and MALDI-TOF mass spectrometry identified several bacterial species, most notably S. epidermidis, alongside Enterobacter cloacae, Klebsiella pneumoniae, and various Bacillus species (Figure 2A). Isolation, characterization, and phenotypic analysis of HCC tumor-resident S. epidermidis (A) Summary of viable bacterial species isolated from the T and PT tissues of four HCC patients. Species identification was confirmed by 16S/18S rRNA sequencing and MALDI-TOF mass spectrometry. (B) FISH images showing the spatial localization of S. epidermidis within HCC tumor tissue. Green signals correspond to the S. epidermidis-specific or E. cloacae-specific probe, red signals correspond to the K. pneumoniae-specific probe, while blue signals represent the cell nuclei stained with DAPI. Scale bars, 20 μm. (C) Comparison of detection methods for S. epidermidis across the four patient samples. Positive detection by either culture-based isolation or FISH is indicated. (D) Colony morphologies of GX-1 and GX3-2 compared with those of the reference strain ATCC 12228 on Columbia blood agar plates after 24 h of incubation. (E) SEM images revealing the ultrastructures of GX-1, GX3-2, and ATCC 12228 strains. The arrows highlight the raised granular structures on the surface of the clinical isolates. Scale bars, 500 nm. (F) Growth curves of the GX-1, GX3-2, and ATCC 12228 strains cultured in BHI medium. The optical density (OD) was measured at 600 nm over 24 h. (G) Representative growth inhibition curves for S. epidermidis GX3-2. Bacterial growth, measured by OD600, is plotted against serially diluted concentrations of various antibiotics. The data are presented as the mean ± SEM from three independent experiments. To visualize the spatial distribution of these bacteria within the tumor architecture, we performed fluorescence in situ hybridization (FISH). Probes specific to S. epidermidis revealed distinct fluorescence signals within the tumor tissues from all 4 patients analyzed (Figure 2B). To further substantiate the clinical prevalence of S. epidermidis in a larger cohort, we expanded our analysis to an additional 20 archived HCC tissue sections. FISH signals were identified in 12 of these 20 samples (60%; Figure S2), resulting in a cumulative detection rate of 66.7% (16/24) across the total clinical cohort. Interestingly, while FISH confirmed the widespread presence of S. epidermidis, culturable colonies of this species were successfully recovered from fresh tissue from 2 of the 4 patients (Figure 2C). The inability to isolate viable strains from all FISH-positive cases, including the expanded cohort where archived samples were unsuitable for culturing, likely reflects the extremely low bacterial biomass in liver tumors and the stringent requirements for maintaining bacterial viability. Given its prevalence, we selected two S. epidermidis isolates derived from HCC patients, designated GX-1 (from patient 4) and GX3-2 (from patient 1), for further functional characterization. The reference strain S. epidermidis ATCC 12228 was used as a negative control. Compared with the reference strain, the tumor-resident isolates presented distinct colony morphologies, while the ATCC 12228 colonies were white and non-hemolytic. Both GX-1 and GX3-2 formed grayish colonies and displayed clear zones of hemolysis on Columbia blood agar plates (Figure 2D). Scanning electron microscopy (SEM) was employed to examine the ultrastructures of the strains. All three strains presented as typical spherical bacteria approximately 0.7 μm in diameter, forming grape-like clusters with smooth surfaces. However, a striking distinction emerged upon closer inspection. Numerous raised granular structures were observed on the surfaces of the HCC tumor-resident isolate GX-1, particularly GX3-2, whereas such features were absent on the ATCC 12228 reference strain (Figure 2E). This unique surface morphology suggests potential phenotypic adaptations of GX3-2 within the TME. Growth curve analysis revealed similar growth profiles for all three strains, with no significant differences in maximum biomass, although GX3-2 displayed a marginally shorter lag phase (Figure 2F). GX3-2 was identified as a multidrug-resistant isolate characterized by high-level resistance (minimum inhibitory concentration [MIC] > 256 μg mL−1) to oxacillin, fusidic acid, fosfomycin, and erythromycin, while remaining susceptible or showing intermediate resistance to gentamicin, vancomycin, chloramphenicol, and tetracycline (Figures 2G and S3A–S3C). S. epidermidis drives the oncogenic behaviors of HCC cells To investigate whether these intratumoral S. epidermidis strains with distinct phenotypic traits promote oncogenic transformation in HCC cells, we performed functional assays in which S. epidermidis strains were co-cultured with Huh7 and HepG2 cells. Proliferation assays, which were carefully controlled to distinguish host cell viability from potential bacterial artifacts, revealed marked differences in the pro-proliferative capacities of each strain across two HCC cell lines (Figures 3A and 3B). In Huh7 cells, the reference strain ATCC 12228 had a negligible impact on viability, whereas the intratumoral isolate GX1 significantly increased cell viability. Strikingly, the second intratumoral isolate, GX3-2, drove the most substantial increase in proliferation, significantly outperforming both GX1 and the reference strain. A similar pattern of GX3-2 superiority was observed in HepG2 cells. Although the specific effects of the other two strains varied, with ATCC 12228 showing a minor increase and GX1 having no discernible impact, GX3-2 once again proved to be the most potent driver of proliferation, robustly enhancing cell viability far beyond all other conditions. Consistent with this potent pro-proliferative phenotype, immunofluorescence analysis of the proliferation marker Ki67 revealed a parallel molecular signature (Figures 3C and 3D). GX3-2 induced a dramatic increase in Ki67 mean fluorescence intensity, an effect that far exceeded the minimal-to-moderate impact observed with both the reference strain and the other intratumoral isolate GX1. HCC tumor-resident S. epidermidis promotes cell proliferation and migration (A and B) Proliferation of Huh7 (A) and HepG2 (B) cells infected with the S. epidermidis reference strain ATCC 12228 and HCC isolates (GX1 and GX3-2), measured via the CCK-8 assay. Briefly, the cells were co-cultured with the indicated bacterial strains for 4 h. The medium was then replaced with fresh medium containing CCK-8 reagent and 1% antibiotics to inhibit residual bacterial metabolism. Absorbance at 450 nm was measured after a 2-h incubation. (C and D) Quantitative analysis of Ki67 immunofluorescence of Huh7 (C) and HepG2 (D) cells. Scale bars, 50 μm. (E and F) Transwell migration assay showing bacterial-induced invasive capacity of Huh7 (E) and HepG2 (F) cells (crystal violet staining; scale bars, 100 μm). (G and H) Western blot analysis and quantification of N-cadherin and E-cadherin expression of Huh7 (G) and HepG2 (H) cells. (I) Volcano plot of DEGs. Red and green dots denote significantly upregulated and downregulated genes, respectively, filtered by absolute log2(fold change) > 1 and p value < 0.05. (J) Bubble chart of the KEGG enrichment results. The size of the bubble represents the number of DEGs enriched in the pathway. (K) Western blot analysis and quantification of p38/p-p38 and ERK/p-ERK expression. Data are represented as the mean ± SEM. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple-comparisons test; ∗∗∗∗, p < 0.001; ∗∗∗, p < 0.005; ∗∗, p < 0.05; ∗, p < 0.01; and ns, non-significant differences. Given the potent pro-proliferative effect of GX3-2, we next investigated whether it also enhanced cell migration (Figures 3E and 3F). In Huh7 cells, neither the reference strain ATCC 12228 nor the GX1 had any significant effect on cell migration compared with the control group. In stark contrast, GX3-2 triggered a robust increase in the number of migrated cells. While the other two strains again failed to elicit a significant migratory response, GX3-2 once more proved to be a powerful inducer, substantially increasing the number of transmigrated cells compared to all other groups. Western blot analysis was performed to assess epithelial-mesenchymal transition (EMT)-related protein expression following co-culture treatment (Figures 3G and 3H). In Huh7 cells, GX3-2 treatment markedly enhanced the expression of the mesenchymal marker N-cadherin, leading to a significant increase in the N-cadherin/E-cadherin ratio. Similarly, in HepG2 cells, GX3-2 co-culture decreased E-cadherin expression while increasing vimentin and N-cadherin expressions, resulting in a higher N-cadherin/E-cadherin ratio compared with the control and ATCC 12228 groups. Overall, GX3-2 consistently induced EMT-associated protein expression in vitro. To elucidate the molecular pathways affected by GX3-2, we performed RNA sequencing (RNA-seq) on Huh7 cells following co-culture with GX3-2 or intervention with PBS. The analysis identified a total of 1,260 differentially expressed genes (DEGs), with 992 genes being significantly upregulated and 268 genes downregulated in the GX3-2-infected cells (Figure 3I). Pathway enrichment analysis of these DEGs revealed a significant overrepresentation of genes associated with several key signaling pathways, including the HIF-1 and MAPK signaling pathways (Figure 3J). To assess the activation status of the MAPK signaling pathway, the phosphorylation levels of key proteins in the p38 and ERK pathways were examined by western blot analysis. As shown in Figure 3K, GX3-2 markedly increased p38 phosphorylation compared with the control group, while total p38 protein levels remained largely unchanged, resulting in a significant elevation of the p-p38/p38 ratio. The ATCC 12228 also exhibited an increased p-p38/p38 ratio, although to a lesser extent than that observed in the GX3-2. For the ERK pathway, the total ERK protein levels were comparable among all groups. Notably, GX3-2 displayed the highest p-ERK levels and p-ERK/ERK ratio, indicating an enhanced trend toward ERK phosphorylation, although this difference did not reach statistical significance. Integrated multi-omics for GX3-2 microbe-host interactions Given that HCC tumor-resident S. epidermidis GX3-2 promotes Huh7 tumorigenesis in vitro, we conducted comprehensive genomic and metabolomic analyses to investigate the molecular mechanisms underlying this phenotype. The complete genome of GX3-2 consists of a 2.44 Mb circular chromosome, with a GC content of 32.16% and five plasmids, collectively encoding 2,274 coding sequences (CDSs) (Figure 4A). The functional annotation of the genome highlighted a predominance of genes involved in energy production and conversion, amino acid transport and metabolism, coenzyme transport and metabolism, and transcription. Notably, clusters of genes related to cell wall biogenesis and signal transduction reflected potential adaptations for survival within the host (Figure 4A). This genomic analysis was complemented by the phenotypic testing of antibiotic susceptibility. Comparative genomics and metabolomics analysis of GX3-2 and its pro-oncogenic role in HCC (A) Circular plot of the complete genome of strain GX3-2, with genes annotated according to COG functional classification. (B) KEGG enrichment analysis of genes in strain GX3-2 versus ATCC 12228. Strain-specific genes are associated primarily with amino acid biosynthesis, carbohydrate metabolism, transport systems, and stress response pathways. (C) Principal-component analysis of the metabolome revealed distinct separation between the two strains. Beta diversity differences were evaluated by ANOSIM based on Bray-Curtis distances with 999 permutations. (D) Volcano plot highlighting significantly altered metabolites, including the compounds 2-phenylethylamine, adenosine, and tyramine. (E) KEGG enrichment analysis of the differentially altered metabolites. Altered compounds are enriched in core energy metabolism pathways such as amino acid metabolism, nucleotide metabolism, glycolysis, and the tricarboxylic acid cycle. (F) Association between metabolites and tumor-related gene expression in HCC cells infected with S. epidermidis GX3-2. The confidence score and expression trends were used to quantify the strength and direction of these correlations. (G) 5 μM adenosine significantly promoted the proliferation of Huh7 cells. ∗∗∗∗, p < 0.001; ∗, p < 0.01. (H) Treatment with 1 μM 2-phenylethylamine significantly promoted the proliferation of Huh7 cells. ∗∗∗∗, p < 0.001; ns, non-significant differences. (I) Human cancer pathway finder PCR array of Huh7 cells. (J) Human cancer inflammation and immunity cross-PCR of Huh7 cells. In (I) and (J), n = 3 independent experiments with similar results. In (G)–(J), data are represented as the mean ± SEM. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple-comparisons test. ∗∗∗∗, p < 0.001; ∗∗∗, p < 0.005; ∗∗, p < 0.05; and ∗, p < 0.01; “ns” denotes non-significant differences. Next, comparative genomic analysis was conducted between GX3-2 and the reference strain ATCC 12228. The two strains shared 2,009 genes, whereas GX3-2 presented 175 unique genes. Pathway enrichment analysis of these GX3-2-specific genes revealed a significant overrepresentation of functions related to amino acid biosynthesis, carbohydrate metabolism, various transport systems, and stress responses (Figure 4B). These unique genetic features suggest that GX3-2 has acquired specific adaptations that may confer enhanced metabolic flexibility and resilience within the TME. To further explore the oncogenic mechanisms by which GX3-2 contributes to tumor progression, we performed a comparative metabolomic analysis on the entire culture broths of GX3-2 and ATCC 12228. PCA of the resulting metabolomic profiles revealed a clear and distinct separation between the two strains (Figures 4C, 4D, and S4). The Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis further revealed that these differentially abundant metabolites were associated with amino acid metabolism, nucleotide metabolism, and central energy pathways. Significant alterations were observed in several metabolites, including 2-phenylethylamine (2-PEA) and adenosine (Figure 4E), highlighting them as lead candidates for mediating the observed pro-tumorigenic effects. To establish a mechanistic link between the metabolic profile of GX3-2 and the transcriptional changes it induces in Huh7 cells, we utilized the Search Tool for Interactions of Chemicals (STITCH) database. This database provides an evidence-based framework by mapping known chemical-protein interactions. Our analysis systematically filtered a pool of 241 differential bacterial metabolites against 1,260 host DEGs, which yielded a network of 17 high-confidence interactions between 10 metabolites and 12 host genes. Further analysis revealed that GX3-2 metabolites, including adenosine, 2-PEA, and cyclic AMP, were associated with proteins encoded by altered host genes implicated in oncogenesis, such as VEGFA, CXCR4, and IL1B (Figure 4F and Table S1). To further functionally validate the role of bacterial-derived metabolites in regulating tumor cell phenotypes, we conducted intervention experiments in which the metabolites identified from GX3-2 were used to treat HCC cells and observed their effects on tumor-promoting phenotypes. Notably, 0.1–10 μM adenosine significantly promoted tumor cell proliferation, with a concentration-dependent increase in optical density (OD), suggesting its role in tumor progression (Figure 4G). In contrast, 2-PEA influenced proliferation at 0.1 and 1 μM, but no significant effect was observed at higher concentrations, indicating its complex interaction with the TME (Figure 4H). To highlight the specific roles of these metabolites, we also tested two other abundant metabolites, uridine and tyramine; the results showed they exerted only a marginal pro-proliferative effect at low concentrations (Figure S8), reinforcing that adenosine and 2-PEA are the key functional metabolites. These findings underscore the role of bacterial metabolites in modulating the TME and driving oncogenic processes. To dissect the molecular pathways underlying these proliferative effects, we performed a quantitative reverse-transcription PCR (RT-qPCR) array analysis on Huh7 cells treated with either adenosine (5 μM) or 2-PEA (1 μM), which elicit maximal pro-proliferative responses. The two metabolites triggered divergent transcriptional programs. Adenosine induced a broad signature related to angiogenesis and inflammation, marked by an approximately 8-fold increase in ANGPT2 and a 6-fold increase in CCL2 and IL13. In contrast, 2-PEA activity was dominated by a striking 12-fold upregulation of FLT3, with a more limited impact on immune factors (Figures 4I and 4J). Importantly, the upregulation of CXCR4 by adenosine provided direct experimental corroboration for a key interaction identified in our initial network analysis, thus completing a cycle of discovery from bioinformatic prediction to mechanistic validation. S. epidermidis GX3-2 promotes HCC progression and resides in tumor tissue in vivo On the basis of the prominent carcinogenesis phenotype of S. epidermidis GX3-2, we aimed to verify whether HCC-resident GX3-2 can act as a single pathogen that promotes HCC progression in an in vivo setting. As a PoC validation, Huh7 xenografts were established in BALB/c nude mice, and intervention via intratumoral injection was initiated (Figure 5A); no significant differences were initially observed between the groups (Figure S5A). Strikingly, tumors injected with GX3-2 exhibited a significant acceleration in growth relative to those injected with either PBS or the reference strain ATCC 12228 (Figures 5B and 5C). The observed statistical significance, notwithstanding the pilot nature of the study, identifies a discernible biological signal. The growth kinetics of the ATCC 12228-injected tumors were indistinguishable from those of the PBS vehicle control group. Furthermore, to isolate the effects of the injected strain from gut-derived influences, we administered dual antibiotic (DA) orally. This regimen did not significantly alter the accelerated growth kinetics, suggesting that the oncogenic drive of GX3-2 is independent of endogenous microbial translocation (Figure S5B). S. epidermidis GX3-2 promotes HCC progression in mice (A) Representative images comparing tumors from PBS-injected controls and mice infected with S. epidermidis. (B and C) Quantitative comparison of tumor weights between the control and infected groups; n = 4 per group. The data analysis was performed using the Kruskal-Wallis test. ∗, p < 0.01; ns, non-significant differences. (D and E) Immunohistochemical staining illustrating the expression of E-cadherin, N-cadherin, VEGFA, Ki67, and PCNA in HCC tissues; n = 4 per group. Scale bars, 50 μm and 310 μm (H&E). Data are represented as the mean ± SEM. Statistical significance was determined by one-way ANOVA followed by Tukey’s multiple-comparisons test. ∗∗∗∗, p < 0.001; ∗∗∗, p < 0.005; ∗∗, p < 0.05; ∗, p < 0.01; ns, non-significant differences. (F) The overall survival (OS) and disease-free internal (DFI) analysis of Staphylococcus-positive cases compared with Staphylococcus-negative tumors. Survival curves for OS and DFI were generated using the Kaplan-Meier method. Statistical differences between the positive and negative groups were evaluated by the two-sided log-rank test. Censored cases are indicated by tick marks on the curves, and 95% confidence intervals are displayed as shaded bands. A p value < 0.05 was considered statistically significant. (G) Confirmation of S. epidermidis colonization and density in tumor tissues by FISH. Scale bars, 20 μm. (H) High bacterial loads of S. epidermidis GX3-2 isolated from the resected HCC tissues. (I) A single deletion was used to compare the GX3-2 genome with the GX3-2-G2 tumor isolate from HCC xenograft tissue. To assess the molecular correlates of GX3-2-mediated tumor promotion, we performed immunohistochemistry (IHC) analysis on xenograft tumors. Tumors from the GX3-2-injected group presented significantly upregulated expression of N-cadherin and VEGFA, along with concomitant downregulation of E-cadherin, which is indicative of an EMT phenotype. Ki67 expression was significantly greater in GX3-2-injected tumors than in ATCC 12228-injected tumors; however, no significant differences in proliferating cell nuclear antigen (PCNA) expression were observed among the groups in this model (Figures 5D and 5E). Moreover, among patients who underwent R0 resection, those with Staphylococcus-positive tumors exhibited significantly poorer OS than those with Staphylococcus-negative tumors; disease-free interval (DFI) also showed a trend toward shorter survival in Staphylococcus-positive patients, although the difference did not reach statistical significance (Figure 5F). Besides, Staphylococcus-positive tumors presented significantly greater expression of the proliferation markers PCNA and Ki67 than Staphylococcus-negative tumors (Figure S6), supporting a potential link between intratumoral Staphylococcus colonization, aggressive tumor biology, and adverse clinical outcomes. We next sought to confirm the colonization and viability of these injected S. epidermidis strains within the TME. FISH analysis using S. epidermidis and the EUB338 probe confirmed the presence of GX3-2 bacteria within the tumor tissue (Figures 5G and S7). To assess viability, homogenates from excised tumors were cultured. This procedure recovered a significantly greater bacterial load from viable S. epidermidis GX3-2-injected tumors than from the ATCC 12228 group (Figure 5H). We then wondered the efficacy of oral double-antibiotic water (gentamicin and vancomycin) against intratumoral GX3-2 and found that it reduced the intratumoral bacterial load by approximately 20%, likely representing the attenuated gut-derived fraction (Figures S5C and S5D). Finally, we assessed the genetic stability of the bacteria after in vivo passage. From GX3-2-injected tumors, we isolated a representative colony, designated GX3-2-G2, for comparative genomic analysis. The results revealed that GX3-2 and ATCC 12228 are completely distinct strains and that GX3-2-G2 shares an average nucleotide identity (ANI) of >99.99% with the parental GX3-2 strain, differing only by a single deletion (Figure 5I). Discussion This study explored the intratumoral bacterial composition in HCC, observing a pronounced shift in the intratumoral microbial composition in advanced HCC. Additionally, we found that Staphylococcus-positive patients presented activation of tumor-associated signaling pathways. We successfully isolated several intratumoral bacterial strains, with Staphylococcus species consistently identified, in line with the 16S rRNA sequencing data. By expanding our clinical validation to a total of 24 patients, FISH analysis identified S. epidermidis in 66.7% (16/24) of the tissues analyzed. This frequent detection in a low-biomass organ such as the liver suggests that S. epidermidis is recurrently present in a substantial subset of HCC tumors and provides a clinically relevant context for the functional characterization of GX3-2 in this study. We established the authentic intratumoral presence of S. epidermidis through a multi-layered validation framework designed to rigorously exclude contamination. First, our proactive contamination controls, including concurrently processed operating room air and instrument blanks, consistently tested negative for any viable microorganism. Second, moving beyond insights from 5R 16S rRNA sequencing, we used in situ visualization to provide direct evidence of S. epidermidis forming microcolonies deep within the HCC tumor parenchyma, minimizing the likelihood of surface artifacts. Finally, our findings are supported by external literature establishing biological plausibility, with recent studies confirming both the presence of viable S. epidermidis in supposedly “sterile” organs9,19 and a physiological gut-to-organ translocation route for such microbes.8 However, S. epidermidis was not isolated in all cases, suggesting variability in bacterial viability or cultivability among different tumor environments. This variability is likely influenced by multiple factors inherent to the TME, which is complex and heterogeneous in nature.20 The TME consists of not only cancer cells but also stromal components such as fibroblasts, immune cells, and the extracellular matrix (ECM), all of which can create a dynamic and often hostile environment for bacterial colonization.21,22 A key functional observation of this study is that tumor-derived S. epidermidis isolates may differ from the reference strain in their tumor-modulatory potential. For the co-culture experiments, we infected HCC cell lines with patient-derived S. epidermidis isolates to evaluate their relative functional effects. The results showed that these intratumoral isolates, particularly GX3-2, significantly enhanced HCC cell proliferation and migration compared with the other strains, including the reference strain ATCC 12228. This observation suggests that tumor-derived isolates may harbor distinct pathogenic or tumor-modulatory traits and highlights the importance of studying authentic patient-derived strains when investigating tumor-microbe interactions. We conducted integrative analyses combining whole-genome sequencing, metabolomics, and host transcriptomics to establish a multi-omics association linking GX3-2 genomic traits, metabolic profiles, and host transcriptional responses. This approach revealed that GX3-2 harbors distinct metabolic and regulatory pathways, particularly those related to purine metabolism, amino acid biosynthesis, and membrane transport, that were strongly correlated with altered metabolite profiles and downstream transcriptional changes in infected HCC cells. Notably, metabolites such as adenosine and 2-PEA were elevated in the entire culture broth of GX3-2 and were functionally validated to enhance tumor cell proliferation in vitro, which is consistent with the transcriptomic enrichment of the cell cycle- and EMT-related pathways. While adenosine and 2-PEA emerged as prominent metabolites associated with GX3-2 and were sufficient to nominate a genomic-metabolic-transcriptional axis linked to cell cycle- and EMT-related programs, the precise molecular cascade connecting these metabolites to specific signaling nodes and transcription factors in HCC cells remains incompletely defined. Importantly, the aim of this work is not to exhaustively map every downstream step of metabolite signaling, but rather to provide initial functional evidence that a viable, tumor-resident strain can act as an active agent promoting malignant phenotypes and that its metabolic outputs align with coherent host transcriptional reprogramming. This framework provides a tractable experimental entry point for mechanistic dissection of intratumoral bacteria-host interactions in oncology. These findings support a genomic-metabolic-transcriptional axis that may contribute to GX3-2-associated tumor-promoting phenotypes and TME reprogramming. Notably, GX3-2 was re-isolated from tumors, which were intratumorally injected with the original strain. The load of GX3-2 in the tumor far exceeded that of the reference strain, despite identical inoculation doses. To isolate its specific functional impact, we utilized oral antibiotics to minimize confounding from gut-derived bacterial translocation. While this regimen reduced the total intratumoral bacterial load by ∼20%, tumor growth kinetics remained unchanged. This finding suggests that the growth-promoting signal associated with GX3-2 in this model was not obviously dependent on continuous microbial influx from the intestinal reservoir. Nevertheless, the current in vivo findings should be interpreted cautiously because the subcutaneous xenograft experiment was designed as a pilot PoC model rather than a physiological model of spontaneous bacterial colonization. In particular, the use of BALB/c nude mice and intratumoral bacterial injection provided a tractable first-pass functional system, but it also introduced important limitations with respect to immune clearance, host-microbe interactions, and physiological relevance. Future work will, therefore, need to extend these observations into orthotopic and immunocompetent models to better recapitulate the native hepatic niche and more accurately define the biological impact of GX3-2 in vivo. In conclusion, this study provides initial PoC evidence that viable intratumoral bacteria may contribute to malignant phenotypes in HCC. By demonstrating recurrent detection of S. epidermidis in a sizable clinical cohort and functionally characterizing the patient-derived isolate GX3-2, we move beyond descriptive association toward a strain-resolved framework for investigating microbe-host interactions in HCC. Our findings support an emerging genomic-metabolic-transcriptional axis, illustrating how a viable, tumor-resident strain can act as an active agent in promoting malignancy. While the in vivo results should be interpreted cautiously given the pilot-scale design, they provide preliminary evidence that GX3-2 may accelerate xenograft growth in a simplified intratumoral inoculation model. This work also establishes a standardized, contamination-controlled workflow for investigating live intratumoral bacteria and provides a practical basis for future high-resolution, multicenter, and mechanistically refined studies. Limitations of the study Several limitations should be acknowledged in this study. First, although S. epidermidis was detected in approximately 60% of HCC tumors in our cohort, this species-level detection does not establish the prevalence of the GX3-2 strain itself. In this study, the influence of sex and age on the results was not specifically addressed. Although we neither conducted detailed analyses stratified by sex or age nor examined them as independent variables, it must be acknowledged that sex and age could potentially influence tumor biology, treatment response, and prognosis. Future studies should consider incorporating these factors as variables for further investigation of their potential impact and association with the outcomes. Determining whether GX3-2 represents a recurrent lineage or a functionally informative example of tumor-associated S. epidermidis will require genome-resolved strain tracking across larger prospective cohorts. Approaches such as multilocus sequence typing (MLST), isolate-level comparative genomics, and strain-specific molecular assays will be important for defining the broader relevance of GX3-2 and related tumor-associated S. epidermidis strains. In addition, despite our contamination-free workflow, the discrepancy between 5R 16S rRNA sequencing and culture recovery highlights an inherent bias toward culturable species. This gap likely reflects the presence of viable but non-culturable states and the challenge of replicating TME-specific nutritional requirements ex vivo.23 Future work using optimized media is needed to capture functional diversity of the uncultured fraction. Second, we prioritized functional characterization of GX3-2 over determining its precise spatial localization within the tumor architecture. While prior studies indicate that distinct micro-niches drive different immune outcomes,24 high-resolution spatial mapping of GX3-2 remains a subject for future investigation. Third, our in vivo findings should be interpreted within the framework of a foundational PoC study. While GX3-2 exhibited superior tumor colonization and was associated with accelerated growth relative to controls, these data do not constitute definitive evidence of physiological tumor regulation in vivo. In addition, the small cohort size, the use of an immunodeficient host, and the artificial nature of intratumoral bacterial inoculation limit the extent to which these findings can be generalized to spontaneous tumor colonization in human HCC. Finally, although we linked bacterial metabolites (adenosine and 2-PEA) to host proliferative and EMT signatures, the specific receptor-mediated signaling cascades remain to be fully delineated. Thus, the proposed pathway should be regarded as a mechanistic framework supported by the current data, rather than a comprehensively resolved signaling map. Resource availability Lead contact Requests for further information, resources, and reagents should be directed to and will be fulfilled by the lead contact, Yunwei Wei (hydwyw11@hotmail.com). Materials availability The accession number has been included in the manuscript. The S. epidermidis GX3-2 strain has been deposited in the Guangdong Microbial Culture Collection Center (GDMCC) under accession number GDMCC 65927. Due to ethical regulations and institutional policies, the availability of strain samples and related materials generated in this study may be subject to restrictions. Requests for materials should be directed to the lead contact and reviewed by the Ethics Committee of Ningbo No. 2 Hospital. Data and code availability Data The data of 5R 16S rRNA sequence and eukaryotic transcriptome have been uploaded to the NCBI database, with ID: PRJNA1345378 (https://www.ncbi.nlm.nih.gov/Traces/study/?acc=PRJNA1345378). The data of GX3-2 WGS have been deposited at the NCBI database, with ID: PRJNA1278975 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1278975). The metabolomics data from two S. epidermidis strains, GX3-2 and ATCC 12228, have been uploaded to the CNSA database, project number CNP0008289 (https://db.cngb.org/search/project/CNP0008289/). Code This paper does not report original code. All other items Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Acknowledgments This research was supported by the National Natural Science Foundation of China (grant nos. U23A20458 and 82300631), the Noncommunicable Chronic Diseases-National Science and Technology Major Project (2024ZD0525400), the Zhejiang Province Medical and Health Science and Technology Plan Project (2025KY1391), Key Program of the Zhejiang Provincial Natural Science Foundation (LHDMZ25H160004), and the Ningbo Top Medical and Health Research Program (2022010101). Author contributions Conceptualization, J.Y. and Yang Liu; methodology, J.Y., B. Li, K.Z., and Z.W.; investigation, J.Y., Yang Liu, T.Z., L.Y., and K.Z.; data curation, B. Li, T.Z., L.Y., F.X., and B. Lin; formal analysis, J.Y. and Yang Liu; visualization, J.Y., Yang Liu, and Ying Liu; validation, J.Y., Yang Liu, J.W., and Y.W.; writing – original draft, J.Y. and Yang Liu; writing – review & editing, J.Y. and Yang Liu; resources, B. Li, J.W., and Y.W.; supervision, Yang Liu, J.W., and Y.W.; funding acquisition, J.Y., Yang Liu, and Y.W.; project administration, Y.W. All authors have read and approved the final manuscript. Declaration of interests All authors declare no conflict of interest. Declaration of generative AI and AI-assisted technologies in the writing process During the preparation of this work, the author used Curie (Springer) and Genimi (Google) to polish the language and enhance readability. After using this tool/service, the author reviewed and edited the content as needed and takes full responsibility for the content of the published article. STAR★Methods Key resources table REAGENT or RESOURCESOURCEIDENTIFIERAntibodiesN-cadherin antibodyAbcamCat# ab18203; RRID:AB_444317E-cadherin antibodyAbcamCat# ab40772; RRID:AB_731493Vimentin antibodyAbcamCat# ab137321; RRID:AB_2921312p38 antibodyProteintechCat# 14064-1-AP; RRID:AB_2878007phospho-p38 antibodyProteintechCat# 28796-1-AP; RRID:AB_2918205ERK antibodyProteintechCat# 11257-1-AP; RRID:AB_2139822phospho-ERK antibodyProteintechCat# 28733-1-AP; RRID:AB_2881202β-actin antibodyAbcamCat# ab6276; RRID:AB_2223210GAPDH antibodyAbcamCat# ab181602; RRID:AB_2630358Bacterial and virus strainsStaphylococcus epidermidis GX3-2Isolated from patientGuangdong Microbial Culture Collection Center (GDMCC) under accession number GDMCC 65927Staphylococcus epidermidis ATCC 12228ATCCATCC 12228Biological samplesHuman HCC tumor tissuesNingbo No.2 HospitalN/AHuman Paired paratumor tissuesNingbo No.2 HospitalN/AMice xenograft tumor tissuesNingbo No.2 HospitalN/AChemicals, peptides, and recombinant proteinsOxacillinBBICat# A460096Fusidic acidSigma-AldrichCat# F0881TetracyclineSangon Biotech, Shanghai, ChinaCat# A600504GentamicinSangon Biotech, Shanghai, ChinaCat# B540724KanamycinSangon Biotech, Shanghai, ChinaCat# A506636VancomycinSangon Biotech, Shanghai, ChinaCat# A460243Brain Heart Infusion (BHI) brothThermo Fisher ScientificCat# CM1135BGlutaraldehydeBBICat# A500484Proteinase KThermo Fisher ScientificCat# 25530015ParaformaldehydeBBICat# E672002Crystal violetBBICat# A600331Tris-buffered saline with Tween-20Sigma-AldrichCat# T9039AdenosineSigma-AldrichCat# A92512-phenylethylamine (2-PEA)Sigma-AldrichCat# 8.07334UridineBBICat# A424857TyramineSigma-AldrichCat# T90344TRIzol™ ReagentThermo Fisher ScientificCat# 15596026RT reagent kit with gDNA eraserTakaraCat# RR047aDulbecco's Modified Eagle Medium (DMEM)Thermo Fisher ScientificCat# 11965118Fetal Bovine Serum (FBS)BBICat# E600051PVDF MembraneSigma-AldrichCat# ISEQ00010DAPIAbcamCat# ab104139Enhanced chemiluminescence (ECL)TanonCat# 180-5001Critical commercial assaysExKine™ Pro Total Protein Extraction Kit for Animal Cultured Cells/TissuesAbbkineCat# KTP3007Rapid PAGE gel preparation kitEpizyme Biomedical Technology Co., Ltd., Shanghai, ChinaCat# PG112BCA protein assay kitBeyotime Biotech, Inc., Shanghai, ChinaCat# PK10026Cell Counting Kit-8 (CCK-8)BBICat# E606335Cancer inflammation and immunity crosstalk qPCR arrayWcgene BiotechnologyCat# wc-MRNA0021-HCancer pathway finder qPCR arrayWcgene BiotechnologyCat# wc-MRNA0020-HDeposited data5R 16S rRNA sequencing dataThis paperNCBI database with ID PRJNA1345378 (https://www.ncbi.nlm.nih.gov/Traces/study/?acc=PRJNA1345378).Eukaryotic transcriptome sequencing dataThis paperNCBI database with ID PRJNA1345378 (https://www.ncbi.nlm.nih.gov/Traces/study/?acc=PRJNA1345378).Whole-genome sequencing data of bacterial strainsThis paperNCBI database with ID PRJNA1278975 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1278975)LC-MS untargeted metabolomics dataThis paperCNSA database, project number CNP0008289 (https://db.cngb.org/search/project/CNP0008289/)TCGA-LIHC microbial, clinical, and transcriptomic dataThe Cancer Genome Atlashttps://portal.gdc.cancer.gov/Experimental models: Cell linesHuh7ATCCRRID: CVCL_U336HepG2ATCCRRID: CVCL_0027Experimental models: Organisms/strainsMouse: BALB/c nude, male, 6–8 weeks oldMuhao Biotech, Hangzhou, ChinaN/AOligonucleotidesPrimers 27F: 5'-AGAGTTTGATCCTGGCTCAG-3;Tianlin Biotech, Wuxi, ChinaN/APrimers 1492R: 5'-GGTTACCTTGTTACGACTT-3'Tianlin Biotech, Wuxi, ChinaN/AStaphylococcus epidermidis FISH probe: 5'-TTCAATATATTATCCGGT-3'FOCOBIO, Guangzhou, ChinaN/AKlebsiella pneumoniae FISH probe: 5'-CCTACACACCAGCGTGCC-3'FOCOBIO, Guangzhou, ChinaN/AEnterobacter cloacae FISH probe: 5'-AACGATAGTGTGTCGGAACACA-3'FOCOBIO, Guangzhou, ChinaN/ASoftware and algorithmsQIIME2QIIME2 developersQIIME2SILVASILVA database v138SILVAGoatoolsGoatools v1.4.4GoatoolsDESeq2DESeq2 v1.42.0https://bioconductor.org/packages/release/bioc/html/DESeq2.htmlCGViewCGView Softwarehttp://wishart.biology.ualberta.ca/cgview/download.htmlResfinderResfinderhttps://bitbucket.org/genomicepidemiology/resfinder/src/master/Bowtie2Bowtie2 v2.5.4Bowtie2GraphPad PrismGraphPad Software 9.5.0https://www.graphpad.com/Majorbio Cloud PlatformShanghai Majorbio Bio-pharm Biotechnologyhttps://cloud.majorbio.com/ImageJNIHhttps://imagej.nih.gov/ij/OtherPacBio Sequel platformPacific BiosciencesSequelNanoDrop spectrophotometerThermo Fisher ScientificND-2000Agilent BioanalyzerAgilent Technologies5300Tissue homogenizerRWDDSC-410Fluorescence microscopeOlympusBX63FColumbia blood agar platesHuankai Microbial, Guangdong, ChinaCat# CP0910 Experimental model and study participant details Clinical sample collection Written informed consent was obtained from all human samples, and this study was approved by the Ethics Committee of Ningbo No.2 Hospital Ethics Committee Approval Letter of Biomedical Research Involving Humans (ID: PJ-NBEY-KY-2024-189-01). Corresponding compensation was provided to the participants. Fresh tumor tissues were collected from HCC patients who underwent surgical resection at Ningbo No.2 Hospital between 2023 and 2024. All cases were clinically staged according to the Barcelona Clinic Liver Cancer (BCLC) system. On the basis of the clinical records, 31 patients with early-stage (BCLC A) HCC and 20 patients with advanced-stage (BCLC C) HCC were included in this analysis (51 patients in total). All patients were Han Chinese men and women. The inclusion criteria for HCC patients were as follows: pathologically confirmed HCC (per guideline standards), clear clinical staging (TNM/BCLC), complete baseline data (clinical, imaging, laboratory), normal major organ function, aged 18-80 years, signed informed consent with good compliance, and meeting study-specific requirements (e.g., no prior systemic treatment for interventional studies). The exclusion criteria were non-HCC pathology or concurrent malignant tumors (excluding curable tumors), uncontrollable complications (severe decompensated cirrhosis, liver failure), severe organ insufficiency, mental illness or cognitive impairment, expected survival < 3 months, and uncontrolled active hepatitis virus infection. Animal experiments model Six- to eight-week-old male BALB/c nude mice were obtained from Muhao Biotech, Hangzhou, China. All mice were maintained at standard housing conditions, the humidity range is 40% to 60%, and the room temperature is maintained between 24°C and 26°C. They can freely obtain food and water. All mice experiments were approved by Ningbo No.2 Hospital, approval of animal use protocol, IACUC (GK-2024-XM-1060), which requires a maximum tumor volume not exceeding 2000 mm3, a maximum diameter not exceeding 15 mm and a tumor weight not exceeding 5% of the body weight. Microbe strains Bacteria were cultured in Columbia broth medium (pH 7.0). Monoclonal isolates were identified as Staphylococcus epidermidis by 16S rRNA gene sequencing or MALDI-TOF MS. Confirmed S. epidermidis isolates were inoculated into Columbia broth medium (pH 7.0) and incubated at 37°C with shaking until they reached the logarithmic growth phase. Bacterial cultures were serially passaged and discarded after the fifth passage. For long-term preservation, isolates were stored at -80°C in culture medium supplemented with 20% glycerol. All procedures involving human tissues and bacterial culture were conducted under biosafety level 2 conditions, with strict aseptic techniques applied to minimize contamination from environmental or skin-derived microbiota. Cell lines Huh7 and HepG2 cells were were purchased from American Type Culture Collection (ATCC) and cultured in medium containing 10% fetal bovine serum (FBS) and 1% dual antibody at 37°C in a 5% CO2 cell culture incubator. Cells were passaged twice or three times weekly at a density of 3×105-5×105 cells/mL by centrifugation at 1200 rpm for 3 minutes. Experiment was initiated after two passages. Method details 5R 16S rRNA sequencing and analysis Genomic DNA was extracted from freshly resected tumor tissues via a standard microbial DNA isolation kit under stringent sterile conditions. The full-length bacterial 16S rRNA genes were amplified via universal primers targeting the V1-V9 regions and sequenced via the PacBio Sequel platform (5R strategy). The sequencing reads were quality-controlled, demultiplexed, and clustered into operational taxonomic units (OTUs) at 97% similarity via the QIIME2 pipeline. Taxonomic assignment was performed against the SILVA 138 database. The downstream analyses included alpha diversity (Sobs, Chao), principal component analysis (PCA), differential abundance testing (Wilcoxon rank-sum test), and canonical correspondence analysis (CCA). Eukaryotic transcriptome To investigate the host gene expression profiles, eukaryotic transcriptomic analysis was performed on two sets of samples: fresh-frozen tumor tissues corresponding to those used for 5R 16S rRNA sequencing and an in vitro model using the Huh7 cell line. Total RNA was extracted from the cell pellets via TRIzol® Reagent following the manufacturer's protocol. The concentration and integrity of the extracted RNA were quantified via a NanoDrop ND-2000 spectrophotometer and an Agilent 5300 Bioanalyzer, respectively. The construction of sequencing libraries, including RNA purification and reverse transcription, as well as the final sequencing, was performed by Shanghai Majorbio Bio-pharm Biotechnology Co., Ltd. (Shanghai, China). Differential expression analysis was conducted via the DESeq2 package to identify genes whose expression was altered between the infected and control groups. Genes were classified as significantly differentially expressed genes (DEGs) on the basis of the criteria of an absolute log2(fold change) ≥ 1 and a false discovery rate (FDR)-adjusted P value < 0.05. To elucidate the biological functions of these DEGs, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed via Goatools and the SciPy library in Python, respectively. Functional terms and pathways with a Bonferroni-corrected P value < 0.05 were considered to be significantly enriched. HCC tumor-resident strains isolation and identification To isolate viable bacteria, resected tumor (T) and paired paratumor (PT) tissues were processed under a strict contamination-control protocol. Tissues were minced and homogenized (DSC-410, RWD), and 100 μL of tissue homogenate was spread directly onto Columbia blood agar plates and incubated at 37°C under anaerobic and aerobic conditions. To exclude possible contamination during isolation of intratumoral bacteria, operating room air was collected as blank control 1, and the instruments and air involved in tissue shearing and grinding and coating processes were collected as blank control 2.25 Full-length 16S rRNA sequencing or microbial mass spectrometry (Shimadzu Biotech Axima Assurance) was used for strain identification. Scanning electron microscopy (SEM) of S. epidermidis cells Bacteria or bacterial precipitates were collected and fixed with 2.5% glutaraldehyde at 4°C overnight and then with 1% osmium for 1 h to enhance the degree of membrane structural contrast. Gradient ethanol dehydration was performed for 15 min per stage; anhydrous ethanol dehydration was repeated three times to completely remove the water. After dehydration, the samples were displaced with isoamyl acetate and injected with liquid CO2 for critical point drying. Whole-genome sequence and comparative genomics of strains Total DNA was extracted from the bacterial culture, ensuring that the DNA quality and concentration met the sequencing requirements. The assembled genome was subsequently functionally annotated, and genes, promoters, and other functional elements were identified via Shanghai Majorbio Bio-pharm Biotechnology Co., Ltd. (Shanghai, China). Comparative genomics methods, such as variant analysis and gene function prediction, were used to analyze the strain genome characteristics and perform biological analyses. Determination of minimum inhibitory concentrations (MICs) The minimum inhibitory concentrations (MICs) for the S. epidermidis isolate GX3-2 were determined via the broth microdilution method in 96-well microtiter plates. The procedure was performed in accordance with the guidelines established by the Clinical and Laboratory Standards Institute (CLSI).26 Stock solutions of antibiotics such as oxacillin and fusidic acid were prepared. Two-fold serial dilutions were performed in cation-adjusted Mueller-Hinton broth (CAMHB) via a 96-well microplate. A suspension of GX3-2 cultured to the stationary phase was collected and adjusted to a turbidity equivalent to 0.5 McFarland standard using sterile PBS. This suspension was subsequently diluted to a final inoculum density of approximately 5×105 CFU·mL-1 per well and incubated at 37°C for 24 h. The MIC was defined as the lowest antibiotic concentration that completely inhibited visible bacterial growth. Fluorescence in situ hybridization (FISH) FISH was performed on paraffin-embedded liver tumor tissue and paratumor tissue sections. After deparaffinization and rehydration, each step was treated with 0.2 N HCl and proteinase K for 10 min. After incubation with blocking buffer for 2 h at 55°C, the sections were analyzed via a Staphylococcus epidermidis probe: 5'-TTCAATATATTATCCGGT-3', labeled with FITC; a Klebsiella pneumoniae probe: 5'-CCTACACACCAGCGTGCC-3', labeled with Cy3; and an Enterobacter cloacae probe: 5'-AACGATAGTGTGTCGGAACACA-3', hybridized with 0.25 μM hybridization buffer and mounted with DAPI (ab104139, Abcam). A fluorescence microscope (BX63F, Olympus) was used for microscopic observation. Immunohistochemistry (IHC) staining Immunohistochemistry for N-cadherin, E-cadherin, VEGFA, Ki67 and PCNA was performed on paraffin-embedded xenograft tumor sections. For IHC quantification, tissue images were captured under uniform conditions. Five random areas of each section were selected for analysis. The percentage of positive cells relative to the total number of cells was calculated, and the average score from the five fields was used to represent the overall protein expression level. Cell coculture and microbial infection The number of bacteria was calculated according to the multiplicity of infection (MOI), and cells in the logarithmic growth phase were taken and added to the bacteria for coculture according to the set MOI (without antibiotic, pH∼7.4). Determination of cell proliferation via the CCK-8 assay Huh7 and HepG2 cells were digested with trypsin for 3 min and centrifuged at 1500 rpm for 5 min. The cells were subsequently resuspended in complete culture medium and seeded into 96-well plates at a density of 10,000 cells per well. After overnight incubation to allow cell attachment, bacteria were added at an MOI of 100:1 for coculture. After 4 h, the culture medium was removed, and antibiotics were used to eliminate extracellular bacteria and minimize interference with the absorbance measurements. Subsequently, 10% CCK-8 solution (v·v-1) in fresh medium was added to each well, and the cells were incubated at 37 °C for 2 h. Absorbance was measured at 450 nm via a microplate reader. Transwell migration assay Cell migration was assessed via transwell chambers to establish an in vitro invasion model. Huh7 and HepG2 cells (2×104) were resuspended in serum-free medium and added to the upper chambers. Bacteria were added at an MOI of 1:50. The lower chambers were filled with medium containing 10% FBS to serve as a chemoattractant. After 24 h of incubation, the nonmigrating cells were removed from the upper surface of the membrane. The migrating cells on the lower surface were washed three times with PBS, fixed with 4% paraformaldehyde, and stained with 0.1% crystal violet. Excess stain was removed by rinsing with PBS until the solution was colorless. Western blot The cells from the different groups were collected after the coculture described above. Total protein was extracted via the ExKine™ Pro Total Protein Extraction Kit for Animal Cultured Cells/Tissues. Protein levels were measured via a BCA protein assay kit (Shanghai Beyotime Biotech, Inc.). A 10% PAGE gel was prepared via a rapid PAGE gel preparation kit (Shanghai Epizyme Biomedical Technology Co., Ltd.). A total of 20 μg of protein per well was loaded and run at a constant voltage of 120 V for 80 min. The samples were transferred at a constant current of 200 mA for 90 min. The samples were blocked with 5% milk at room temperature for 1 h and incubated with the following primary antibodies: N-cadherin (Abcam, ab18203), E-cadherin (Abcam, ab40772), vimentin (Abcam, ab137321), p38 (Proteintech, 14064-1-AP), p-p38 (Proteintech, 28796-1-AP), ERK (Proteintech, 11257-1-AP), p-ERK (Proteintech, 28733-1-AP), β-actin (Abcam, ab6276), and GAPDH (Abcam, ab181602). The samples were incubated overnight at 4°C with shaking. The next day, the samples were washed three times with TBST buffer for 10 min each; after incubation with secondary antibody at room temperature for 1 h, the samples were washed again three times with Tris-buffered saline with Tween-20 (TBST) for 10 min each. The membranes were incubated with enhanced chemiluminescence solution, after which images were captured via a chemiluminescence imager. LC-MS based untargeted metabolomics For metabolite extraction, 100 mg samples of ATCC 12228 and GX3-2 were weighed and placed into a 2 mL centrifuge tube containing a 6 mm grinding bead. The extraction, quality control and analysis of metabolites were subsequently performed by Shanghai Majorbio Bio-pharm Biotechnology (Shanghai, China). The data were analyzed through the free online platform of the majorbio cloud platform (cloud.majorbio.com). RNA extraction and quantitative real-time PCR array Huh7 cells were seeded at a density of 3×106 cells/well in 6-well plates and subsequently treated with metabolites for 24 h, with the PBS-treated group serving as the control. Following intervention, the cells were harvested, and total RNA was isolated from the cell pellets via TRIzol® Reagent following the manufacturer's instructions. Gene expression profiles were then assessed via a cancer inflammation and immunity crosstalk qPCR array (Wcgene Biotechnology, wc-MRNA0021-H) and a cancer pathway finder qPCR array (Wcgene Biotechnology, wc-MRNA0020-H). Analysis was performed according to the manufacturer's guidelines, and genes exhibiting a fold change of ≥2.0 or ≤0.5 were considered to have significant changes in expression. Xenograft tumor model establishment and intervention Six- to eight-week-old male BALB/c nude mice were subcutaneously inoculated in the right axilla with 1 × 106 Huh7 cells, which were subsequently suspended in sterile PBS buffer. One week post-inoculation, the mice were randomized into four experimental groups (n = 4 per group, and no significant differences were initially observed between groups). Tumor volumes were monitored via caliper measurements and calculated via the formula V = (L × W2)/2. To assess therapeutic efficacy, the relative tumor volume for each mouse was determined by normalizing the final tumor volume measured at the study endpoint to its corresponding initial volume. Interventions were administered via multi-point intratumoral injections twice a week for a total duration of two weeks. The experimental groups were defined as follows: Group A, which was injected with a volume of sterile PBS equivalent to the intervention groups; Group B, which was injected with ATCC 12228; Group C, which was injected with GX3-2; and Group D, which was injected with GX3-2 and concurrently provided with drinking water containing a dual-antibiotic (DA) solution ad libitum. For the bacterial intervention groups (B, C, and D), the inoculum was administered at MOI of 100:1, which was calculated on the basis of the initial number of injected tumor cells. To minimize potential confounding from gut-derived bacterial translocation, mice in Group D were administered gentamicin (0.5 mg·mL-1) and vancomycin (1.0 mg·mL-1) via drinking water to attenuate the intestinal microbiota prior to intratumoral inoculation.5 At the end of the three-week intervention period, the mice were euthanized under anesthesia, and the subcutaneous tumors were surgically excised. Re-isolation of S. epidermidis from xenograft tumor tissue in vivo To confirm the intratumoral presence of the administered S. epidermidis, a fresh piece of tissue was aseptically excised from each harvested tumor. The tissue was homogenized and ground in sterile PBS. The resulting homogenate was then serially diluted and plated onto Columbia blood agar plates. Following incubation at 37°C under aerobic conditions, colonies were isolated for subsequent identification to verify the presence of the inoculated S. epidermidis strains. TCGA cohort analysis and clinical relevance of intratumoral Staphylococcus Microbial profiles and matched clinical data for HCC were obtained from The Cancer Genome Atlas (TCGA) cohort. Tumor samples were stratified into Staphylococcus-positive and Staphylococcus-negative groups according to the relative abundance of Staphylococcus; specifically, we applied an upper-quartile (75th percentile) relative abundance cutoff-aligned with the threshold defined in our primary cohort-to classify tumors as Staphylococcus-positive (≥75th percentile) or Staphylococcus-negative (<75th percentile).27 Corresponding host transcriptomic profiles were used to compare the expression levels of proliferation markers (e.g., MKI67 and PCNA) between the two groups. For survival analysis, overall survival (OS) and disease-free interval (DFI) was estimated using the Kaplan-Meier method and compared by the log-rank test. Quantification and statistical analysis Statistical analyses were performed using GraphPad Prism 9.5.0 unless otherwise specified. For comparisons between two groups, unpaired two-tailed Student’s t tests were used when data satisfied assumptions of normality and homogeneity of variance. For comparisons among multiple groups, one-way ANOVA was used, followed by appropriate post hoc multiple-comparison tests as indicated in the corresponding figure legends. When data did not satisfy assumptions of normality or equal variance, nonparametric tests were applied, including the Mann-Whitney U test for two-group comparisons and the Kruskal-Wallis test for multiple-group comparisons. Differences were considered statistically significant at P < 0.05 unless otherwise specified. For microbial community analyses, alpha-diversity indices were compared using the Wilcoxon rank-sum test. Beta-diversity differences were evaluated by analysis of similarities (ANOSIM) based on Bray-Curtis distances with 999 permutations. Differential genus abundance analysis was performed using the Wilcoxon rank-sum test. Canonical correspondence analysis (CCA) was used to assess associations between microbial composition and clinical parameters. For host transcriptomic analyses, differential expression was evaluated using DESeq2, and genes with an absolute log2(fold change) ≥ 1 and a false discovery rate (FDR)-adjusted P value < 0.05 were considered significantly differentially expressed. Functional enrichment analyses, including KEGG pathway enrichment, were performed as described in the method details. For differential genus analysis and enrichment analysis, multiple-testing correction was performed using the Benjamini-Hochberg method, and adjusted P < 0.05 was considered statistically significant. Exact statistical tests, exact n values, and the definition of biological units are provided in the corresponding figure legends. Asterisks indicating statistical significance are defined in each relevant figure legend. Where applicable, exact P values and relevant test statistics are reported in the figure legends. In xenograft experiments, n refers to the number of mice per group. In cell-based assays, n refers to the number of biologically independent experiments unless otherwise indicated. Footnotes Supplemental information can be found online at https://doi.org/10.1016/j.isci.2026.116084. Contributor Information Jianming Wang, Email: wjm18jgm@aliyun.com. Yunwei Wei, Email: hydwyw11@hotmail.com. Supplemental information Document S1. Figures S1–S8 and Table S1 References Associated Data Supplementary Materials Document S1. Figures S1–S8 and Table S1 Data Availability Statement Data The data of 5R 16S rRNA sequence and eukaryotic transcriptome have been uploaded to the NCBI database, with ID: PRJNA1345378 (https://www.ncbi.nlm.nih.gov/Traces/study/?acc=PRJNA1345378). The data of GX3-2 WGS have been deposited at the NCBI database, with ID: PRJNA1278975 (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA1278975). The metabolomics data from two S. epidermidis strains, GX3-2 and ATCC 12228, have been uploaded to the CNSA database, project number CNP0008289 (https://db.cngb.org/search/project/CNP0008289/). Code This paper does not report original code. All other items Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.