Identification of a novel human gut microbes and microbial metabolites related genes signature for prognostic implication in head and neck squamous carcinomas Authors: Xuan Yang, Sai Liang, Piao Wang, Wei Huang, Kai Song, Ming Yu, Zhengpeng Gong Published: 2026 ABSTRACT AbstractBackgroundThe gut microbiota acts as a critical driver influencing the pathogenesis, therapeutic response, and clinical outcomes across various cancer types. This study aimed to investigate the prognostic value of human gut microbes and microbial metabolites related genes (HGMMMRGs) in head and neck squamous cell carcinoma (HNSCC).MethodsWe constructed a prognostic risk model comprising 19 core HGMMMRGs using LASSO penalized regression and a multivariate Cox proportional hazards model. The predictive performance of the model was evaluated through Kaplan-Meier analysis, receiver operating characteristic (ROC) curves, nomograms, and concordance index. In addition, functional enrichment analysis was performed on the differentially expressed risk genes. Furthermore, the relationship between the immune microenvironment of HNSCC and the risk diagnostic model was examined. Western blot analysis was used to assess the expression levels of IL10 in both HNSCC tissues and adjacent normal tissues. Finally, the correlation between IL10 and the gut microbiota was explored.ResultsThis study developed a risk score model integrating 19 HGMMMRG genes, which can serve as a tool to guide prognosis and immune microenvironment assessment in HNSCC patients. Survival analysis showed that patients in the high-risk group had significantly worse outcomes (P < 0.05). Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis revealed significant enrichment of differentially expressed genes (DRLs) and immune-related pathways. Western blot analysis further confirmed that IL10 was highly expressed in HNSCC, and the abundance of Faecalibacterium prausnitzii and Enterococcus durans colonies was correlated with IL10 expression.ConclusionWe developed a prognostic model for HGMMMRGs that can be effectively used to predict OS in patients with HNSCC. Second, Faecalibacterium prausnitzii and Enterococcus durans can influence the prognosis of patients with HNSCC by mediating the expression IL10 and thereby affecting the prognosis of HNSCC patients. Thus, human gut microbes and microbial metabolite-related genes may be another promising strategy for the treatment of patients with HNSCC.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1007/s12672-026-05442-x.Keywords: Human gut microbes and microbial metabolites, Head and neck squamous cell carcinoma, HNSCC, Immune microenvironment, Prognostic model Introduction As a highly malignant tumor, head and neck squamous cell carcinoma (HNSCC) has a strong capacity for local invasion and metastasis, posing a serious threat to patients’ lives. The development of HNSCC is associated with multiple risk factors, ranging from lifestyle habits (tobacco use, alcohol consumption, betel nut chewing) and chronic local irritation to infection with human papillomavirus [1, 2]. Treatment strategies for HNSCC patients have evolved into comprehensive approaches involving surgical resection, radiotherapy, chemotherapy, targeted therapy, and immunotherapy. Despite these advancements, the 5-year overall survival (OS) rate for HNSCC patients remains at 50%, with little progress over the past decade. Therefore, identifying novel biomarkers for HNSCC and improving prevention, screening, and treatment strategies are critical [3, 4]. Human gut microorganisms and microbial metabolites are highly complex, and vary widely among individuals. The gut microbiota has a major impact on host homeostasis and disease, playing a crucial role in maintaining immune and metabolic homeostasis, as well as defense against pathogens [5, 6]. Studies have shown that the human bacterial microbiota is associated with several cancers and that gut microbes can modulate the immune system and intestinal epithelial cells through direct interactions with the host organismal system, which is essential for the prevention of tumorigenesis, especially colorectal, breast, and prostate cancers [7, 8]. There is clear evidence that Clostridium difficile is strongly associated with colorectal cancer and Helicobacter pylori infection in gastric cancer [6]. The gut microbiota can regulate circulating estrogen levels, thereby influencing the risk of hormone-driven cancers such as endometrial cancer, ovarian cancer, prostate cancer, and breast cancer [9–11]. Tumorigenesis is closely related to the immune system, and thus the significant systemic effects of the gut microbiota on tumor development and treatment are mediated through immune system regulation [12]. Several studies have explored the molecular mechanisms of human gut microbes and microbial metabolites, which have contributed to the understanding of the link between gut microbes and their hosts. Thus, understanding the complex role of gut microbiota in cancer is a rapidly evolving field that could pave new pathways for anticancer therapies in cancer prevention and treatment [8]. This research was designed to build a prognostic model based on the human gut microbiome and its associated metabolic genes (HGMMMRGs) to assess clinical outcomes in patients with HNSCC and provide guidance for therapeutic interventions. Although the genomic and immune landscapes of head and neck squamous cell carcinoma (HNSCC) have been extensively explored, the prognostic potential of the gut-tumor axis remains largely elusive [13, 14]. Specifically, a comprehensive prognostic signature based on Human Gut Microbes and Microbial Metabolites Related Genes (HGMMMRGs) has not yet been established. To address this gap, we integrated transcriptomic data from TCGA, GEO, and gutMGene databases. By employing LASSO penalized and multivariate Cox regression algorithms, we constructed a novel HGMMMRG-based prognostic model. This approach robustly captures the microbe-host transcriptomic cross-talk, aiming to provide a reliable tool for predicting clinical outcomes and immune responses in HNSCC. Materials and methods Data collection and study design Transcriptomic RNA-seq data and corresponding clinical information for Head and Neck Squamous Cell Carcinoma (HNSCC) were downloaded from The Cancer Genome Atlas (TCGA) database. To ensure the reliability of the survival analysis, patients with an overall survival (OS) time of less than 30 days were excluded to minimize the impact of perioperative mortality or extremely short follow-up periods. Samples with missing survival status or missing key clinical variables (age, stage) were also removed. Additionally, the microarray data profile and corresponding clinical details for the GSE41613 cohort were sourced from the Gene Expression Omnibus (GEO) database as an independent validation cohort. Human gut microbes and microbial metabolites related genes (HGMMMRGs) were obtained from the gutMGene database (http://bioannotation.cn/gutmgene/resource). Simple nucleotide variant (SNV) data were also downloaded from the TCGA website. In terms of study design, this research adopted a retrospective consecutive sampling approach. The dependent variable was overall survival (OS), defined as the time interval from the date of initial diagnosis to the date of death or the last follow-up. The independent variables primarily included the prognostic risk score calculated from the 19 HGMMMRGs, along with essential clinical covariates such as age, sex, tumor grade, and pathological stage. The total sample size consisted of 511 patients in the TCGA training cohort and 97 patients in the GEO validation cohort. Survival data and follow-up Overall survival (OS) was utilized as the primary clinical endpoint in this study. The survival data extracted from both the TCGA and GEO databases were right-censored. The median follow-up time was 2.90 years for the TCGA training cohort and 5.47 years for the GEO validation cohort. Furthermore, given the nature of the public repositories utilized in this study, OS represents all-cause mortality, which intrinsically encompasses the effects of potential comorbid conditions. Due to the unavailability of specific cause-of-death records, competing risk models were not employed, which is acknowledged as a common constraint in retrospective database studies. Development and validation of a prognostic HGMMMRGs signature Based on TCGA (training set) and GEO (validation set) data, we performed differential expression analysis of target genes between tumor and normal tissues using the R package “limma”. To identify prognostically significant genes (P < 0.05), a univariate Cox regression analysis of OS was conducted. Subsequently, the “maftools” R package was used to analyze gene mutations and their associations with clinical characteristics in the TCGA HNSCC cohort. A prognostic prediction model was constructed based on the lasso-Cox regression method (R package glmnet), with the parameter λ optimized through 10-fold cross-validation. The risk score for each sample was calculated using the formula: risk score = gene expression Σ coefficient. K-M survival analysis was conducted to compare OS in risk-stratified groups, which were defined by the median training-set-based risk score. The predictive accuracy of the model was evaluated using the ‘timeROC’ R package. Finally, the model’s reliability and effectiveness were validated using the testing set. Principal component analysis (PCA) To visualize the high-dimensional spatial distribution and assess the discriminative ability of the constructed signature, Principal Component Analysis (PCA) was performed based on the expression profiles of the core HGMMMRGs. Prior to PCA, the gene expression data were standardized (scaled to unit variance) to ensure comparability across different variables. To clarify the representativeness of the dimensionality reduction, the proportion of total variance explained by the primary principal components (PC1 and PC2) was calculated and reported. Furthermore, rather than relying solely on visual interpretation, a Permutational Multivariate Analysis of Variance (PERMANOVA) based on Euclidean distance was conducted to quantitatively evaluate the statistical significance of the spatial separation between the high-risk and low-risk groups. A P-value < 0.05 was considered to indicate a statistically significant separation. Development of a nomogram to predict OS This study aimed to develop a nomogram for predicting OS in HNSCC patients. Utilizing the “rms” R package, the nomogram was constructed by integrating transcription level and clinical features from TCGA cohort. To validate the model, calibration curves and time-dependent ROC curves (AUC) were generated. We conducted single-predictor and multi-predictor Cox regression models to determine if the prognostic scoring model was an independent predictor of OS in HNSCC patients. Exploring the correlation between the immune microenvironment and the risk score model in HNSCC Single nucleotide variation (SNV) data for HNSCC patients were obtained from the TCGA database, and the tumor mutation burden (TMB) was calculated for each individual. Subsequently, the correlation between the risk score and TMB was evaluated by Spearman analysis, and TMB levels were compared between the high-risk and low-risk groups were compared. Additionally, using the “cibersort” deconvolution method (1000 permutations), we quantified 22 tumor-infiltrating lymphocyte subsets in both groups’ microenvironment [15]. A P-value threshold of < 0.05 was defined for statistical significance. TMB calculation TMB is an indicator that measures the number of genetic alterations present in cancer [16, 17]. We used the R package “maftools” to analyze the mutation data of HNSCC samples obtained from TCGA. Based on the TMB data of each sample [15] and combined with survival status of patients, we first divided all samples into high- and low-TMB groups based on individual TMB data and survival information. Subsequently, K-M survival analysis was employed to assess the four resulting subgroups: high-TMB/high-risk, high-TMB/low-risk, low-TMB/high-risk, and low-TMB/low-risk. This analysis aims to evaluate the impact of the risk score and TMB on the survival rate of HNSCC patients and explore the potential synergistic effect of TMB and differential risk levels (DRLs) on patient survival. Functional enrichment analysis Differentially expressed genes (DEGs) were filtered using a dual-threshold screen: an absolute log2-fold change greater than 1 and a statistical significance of P < 0.05. These DEGs were then analyzed to explore their biological roles and underlying mechanisms that differentiate the two groups. The “org.Hs.eg.db” R package was used to convert the symbol IDs of each DEG (|logFC| > 1, P < 0.05) into Entrez gene IDs. Finally, for functional annotation, the identified DEGs were subjected to Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses using the R package “clusterProfiler”. Prediction of drug sensitivity The treatment response to established chemotherapy drugs was assessed using the “Oncopredict” package [18]. Based on the Cancer Drug Sensitivity Genomics (GDSC) database (www.cancerRxgene.org) and transcriptomic data, a ridge regression model was constructed to calculate the half-maximal inhibitory concentration (IC50) for each drug, thereby estimating the sensitivity of HNSCC samples to various chemotherapy drugs. PPI network We first constructed a protein–protein interaction (PPI) network for the 19 core genes using the STRING database [19] (version 11.5) with a medium-confidence interaction score threshold (> 0.40). The resulting network was then visualized and analyzed using Cytoscape software (version 3.9.1; Cytoscape Consortium). Finally, key hub genes were identified from the network using the CytoHubba plugin [20] (version 0.1) in Cytoscape. Expression levels of HGMMMRGs Acquisition of tissue specimens Eight pairs of fresh frozen HNSCC tumor tissues and matched adjacent normal tissues (distance > 2 cm from tumor margin) were collected from patients who underwent curative surgery at the Department of Otorhinolaryngology Head and Neck Surgery, Affiliated Hospital of Guizhou Medical University between January 2022 and June 2023. The inclusion criteria were: Pathologically confirmed primary HNSCC (no prior history of HNSCC or other malignancies).No neoadjuvant radiotherapy, chemotherapy, or immunotherapy before surgery.Complete clinical and follow‑up data available (minimum follow‑up 6 months).The exclusion criteria were: Recurrent HNSCC.History of other malignancies within the past 5 years.Severe comorbidities (e.g., uncontrolled cardiovascular disease, autoimmune disorders) that might affect survival.Follow‑up time less than 6 months or loss to follow‑up.All participants signed written informed consent. Western blotting analysis Collected pairs of HNSCC and adjacent normal tissues (n = 8) were stored at − 80 °C. After sectioning, weighing, and mincing on ice, tissues were lysed on ice using RIPA buffer supplemented with protease and phosphatase inhibitors. Total protein was extracted from centrifuged lysates, quantified by BCA assay, and subjected to SDS-PAGE. Proteins were then transferred to a PVDF membrane and probed with specific primary antibodies (anti-IL-10, 1:1000; anti-Tubulin, 1:5000) and corresponding secondary antibodies. Signal detection was performed using ECL, and bands were analyzed with ImageJ. Statistical analysis All statistical analyses were performed using R software (version 4.2.2). Categorical variables were compared using Pearson’s chi-square test, while comparisons of continuous variables between normal and HNSCC samples were conducted via one-way analysis of variance (ANOVA). Kaplan-Meier (K-M) survival curves were generated to compare unadjusted overall survival (OS) across different patient subgroups, with statistical significance determined by the two-tailed log-rank test. To account for the potential confounding influence of comorbidities and clinical covariates, multivariate Cox proportional hazards regression models were subsequently employed to adjust for these factors and identify independent prognostic factors. The results were reported as hazard ratios (HRs) with 95% confidence intervals (CIs). The proportional hazards (PH) assumption for the Cox models was rigorously validated using the Schoenfeld residuals test (P > 0.05). For Principal Component Analysis (PCA), the gene expression data were scaled to unit variance prior to dimensionality reduction, and the statistical significance of the spatial separation between risk groups was quantified using Permutational Multivariate Analysis of Variance (PERMANOVA). Unless otherwise specified, the threshold for statistical significance was set at a two-sided P < 0.05. Results Identification of prognostic HGMMMRGs in the TCGA cohort We obtained 223 HGMMMRGs from the GutMGene database [21] (http://bioannotation.cn/gutmgene/resource). dhtml) (Supplementary Table 1). We integrated these 223 genes with gene expression data from HNSCC patients downloaded from the TCGA database and identified 111 HGMMMRGs that were differentially expressed between HNSCC tumor tissues and normal tissues (Fig. 1A) (Supplementary Table 2). Univariate Cox regression analysis showed that 29 genes were associated with prognosis (P < 0.05) (Fig. 1B). Identification of the prognostic HGMMMRGs in the TCGA cohort. A Heatmap of expression profiles of 50 prognosis-related genes. B Forest plot of one-way Cox regression analysis of HGMMMRGs and HNSCC prognosis. C Adjusted parameter selection in the Lasso model with tenfold cross-validation. D Lasso coefficient curves Development of a prognostic risk score model based on the TCGA cohort To develop a risk score model for prognosis prediction, using Lasso Cox regression, we identified a core set of genes with prognostic value. Based on the optimal λ value, 19 gene features were selected (Fig. 2C, D). The risk score for each sample was defined by the following formula: Risk Score = Gene1 expression × Gene2 expression + … + Gene(N) expression. Patients in the TCGA cohort were stratified into high-risk (n = 255) and low-risk (n = 256) groups based on the median cutoff value. The model demonstrated significant efficacy in distinguishing high-risk from low-risk HNSCC samples. Consistent with this, the GEO cohort was successfully stratified into high-risk (n = 56) and low-risk (n = 41) groups. Based on the median risk score, all individuals in the training group were further categorized into high-risk and low-risk subgroups as required for survival analysis (Fig. 2A, B). Elevated risk scores were significantly associated with reduced survival duration and increased mortality. The differential expression patterns of the 19 prognostic DRGs between high- and low-risk groups were visualized in a heatmap (Fig. 2E, F) (Table 1). Kaplan-Meier survival curves revealed that the high-risk group exhibited a significantly poorer OS than the low-risk group (P < 0.05; Fig. 2G). A significant OS difference was also observed in the validation group (P < 0.05; Fig. 2H). To evaluate the discriminative power of the HGMMMRGs, Principal Component Analysis (PCA) was performed. As shown in Fig. 2I, the high-risk and low-risk groups were distributed in distinct directions based on the overall expression profiles of HGMMMRGs (PC1 = 10.55%, PC2 = 7.84%; p = 0.026). Furthermore, when focusing on the 19 core genes involved in the prognostic model, a more significant spatial separation between the two risk strata was observed (Fig. 2J), with PC1 and PC2 accounting for 17.28% and 16.86% of the total variance, respectively (p = 0.001). These findings indicate that the constructed 19-gene signature can effectively capture the transcriptomic differences and classify patients into high- and low-risk populations. Prognostic risk score model established in the TCGA cohort. A–D Distribution of overall survival status and risk scores in the training, validation groups. E–F Heatmaps of the expression of six model genes in the training, validation groups. G, H K-M survival curves of OS in both groups of patients in the training, validation groups. I PCA based on all HGMMMRGs in HNSCC. J PCA based on the genes involved in the prognostic risk score model in HNSCC Table 1The coefficients of nineteen HGMMMRGs measured by the multivariate coxGeneCoefCXCL80.026966CXCR6− 0.04423CYP2D6− 0.36848DEFB1− 0.03318DUSP100.129887EGFR0.009588FADS30.079445G6PD0.02313IL10− 0.31693IL12B− 0.04567IL1A0.006674LGR5− 0.10195MGAT4B0.101008ONECUT2− 0.21458SMURF20.113303SOD10.265766SSTR3− 0.03019TGFB10.00168TJP3− 0.05142 The coefficients of nineteen HGMMMRGs measured by the multivariate cox The relationship between risk score and clinical characteristics Univariate and multivariate Cox regression analyses were performed to identify independent prognostic factors for overall survival (OS). In the multivariate model, the risk score (HR = 3.365, 95% CI 2.358–4.801, P < 0.001) and pathological stage (HR = 1.453, 95% CI 1.199–1.761, P < 0.001) were independently associated with OS, whereas age, sex, and tumor grade were not (all P > 0.05). Compared with pathological stage, the risk score exhibited a higher hazard ratio (Fig. 3A, B). The forest plots in Fig. 3A and B display the HR and 95% CI for each variable. Application of the risk-scoring system revealed that a lower risk score was correlated with improved progression-free survival and better prognosis (Fig. 3C). The researchers created a nomogram based on age, sex, tumor grade, pathological staging, and risk scores to predict OS in patients with HNSCC in the TCGA cohort. The AUC demonstrated the reliability of prognostic prediction using the nomogram, reaching 0.407 at one year, 0.508 at three years, and 0.800 at five years (Fig. 3D). The nomograms outperformed other prognostic variables (e.g., risk score, age, gender, grade, and stage) in accuracy, offering greater clinical utility for predicting outcomes. The calibration graphs showed strong agreement between the predicted and actual results of the nomograms (Fig. 3E, F). Furthermore, Cox regression analyses (both univariate and multivariate) demonstrated the independent predictive ability and staging. (Fig. 3G, H). Development of a nomogram for OS prediction. A, B The forest plots show the results of univariate (A) and multivariate (B) Cox regression analysis in the TCGA cohort. C Kaplan–Meier curves for progression-free survival (PFS). D Nomogram models for prediction of overall survival. E The ROC risk score and clinicopathologic characteristic curves. F The calibration curve for evaluating the accuracy of the nomogram model. G, H Univariate and multivariate Cox regression analysis for the nomogram and clinicopathological variables Analysis of immune-related characteristics between low- and high-risk score groups To investigate the association between risk scores and immune profiles, we assessed immune infiltration and activity via single-sample gene set enrichment analysis (ssGSEA). The analysis revealed that the low-risk group was characterized by higher abundances of adaptive immune cells (e.g., plasma cells, activated CD4/CD8 T cells, Tregs) and resting mast cells, whereas the high-risk group was enriched for resting CD4 memory T cells, M0/M2 macrophages, and activated mast cells (Fig. 4A). Further immune function scoring analysis indicated that the high-risk group exhibited significantly lower scores in multiple immune dimensions (e.g., antigen presentation, cytokine signaling, and T‑cell regulation), as detailed in Fig. 4B. Immune-related characteristics of low-risk and high-risk scoring groups. A Immune cell infiltration in low-risk and high-risk scoring groups. B Immune cell function score in low-risk and high-risk scoring groups. (* p < 0.05, ** p < 0.01, and *** p < 0.001) Genomic alteration profiling of the prognostic model Correlation analysis revealed a negative association between TMB and risk scores, indicating distinct molecular characteristics across risk-based subtypes. First, we analyzed the distribution of gene mutations in the high-risk and low-risk score subtypes (Fig. 5A, B). Analysis of the cumulative incidence plot showed that the high-risk group had a higher frequency of somatic mutations (95.95%) compared to the low-risk group (88.11%). The top five most frequently mutated genes in both groups were TP53, TTN, FAT1, CDKN2A, and MUC16, though their mutation rates differed: in the high-risk group, the rates were 75%, 36%, 23%, 17%, and 15%, respectively, whereas in the low-risk group, they were 55%, 39%, 17%, 20%, and 19%, respectively (Fig. 5A, B). A, B Gene mutation analysis. C Tumor mutational burden (TMB) scores. D Tumor immune dysfunction and exclusion (TIDE) scores. E Kaplan–Meier survival analysis of the TMB in predicting the survival of HNSCC patients. F Kaplan–Meier survival analysis showed the prognostic values of TMB combined risk score. (* p < 0.05, ** p < 0.01, and *** p < 0.001) Association Between Risk Scores and TMB scores revealed a marked positive correlation between these two variables (Fig. 5C). Analysis of the tumor immune dysfunction and exclusion (TIDE) algorithm further revealed that patients in the high-risk group had lower TIDE scores (Fig. 5D). HNSCC patients were stratified into high and low TMB groups using the median TMB value as the cutoff. Subsequent Kaplan-Meier survival analysis with the log-rank test demonstrated that the low TMB group had significantly superior OS compared to the high TMB group (P = 0.007; Fig. 5E). Subsequently, we integrated risk scores with TMB to predict survival outcomes for HNSCC patients. As shown in Fig. 5F, patient prognosis ranged from most favorable in the low TMB/low-risk group to least favorable in the high TMB/high-risk group (P < 0.001). Comparative enrichment analysis between low- and high-risk groups We performed GO and KEGG enrichment analyses on the 780 DEGs derived from the risk group comparison (Supplementary Table 3) in order to identify biological functions and pathways linked to the risk score. The DEGs were found to be enriched in various immunobiological processes. Specifically, GO analysis revealed that the DEGs are involved in the BCR-mediated signaling, adaptive humoral immunity mediated by circulating immunoglobulins, complement cascade initiation via the classical pathway, immunoglobulin production, antigen receptor-mediated signaling, immune response-activating cell surface receptor signaling, immune response-regulating receptor signaling, immunoglobulin-mediated immune responses, and several others related to immune regulation and antigen binding (Fig. 6A, B) (Supplementary Table 4). These findings suggest potential differences in immune responses to cancer and treatment between the high- and low-risk groups. KEGG analysis further revealed significant enrichment of the DEGs in critical signaling and metabolic pathways, notably those related to cytokine–cytokine receptor interaction, hematopoietic cell lineage, primary immunodeficiency, cell adhesion molecules, and virus–host cytokine interplay (Fig. 6C, D) (Supplementary Table 5). Enrichment analysis of the DEGs. A The top 10 enriched GO analysis results. BP biological process, CC cellular compartment, MF molecular function. B The top 10 enriched GO analysis results. The size of the spots indicates the number of genes, and the color indicates the p-value. C KEGG pathway analysis for DRGs. D KEGG pathway analysis for DRGs. The size of the spots indicates the number of genes, and the color indicates the p-value Response to chemotherapy drugs Figure 7A-L shows the results of a comparative study of the estimated IC50 levels of 12 representative chemotherapeutic drugs or inhibitors used to investigate the differences in responsiveness between the low- and high-risk groups. In AZD6738, BI-2536, Dasatinib, ERK_6604, SB505124, and SCH772984, the half-maximal inhibitory concentration (IC50) values in the high-risk group were significantly lower than those in the low-risk group. These results suggest that these drugs may have potential therapeutic value for patients classified as high-risk. However, the IC50 values for Afatinib, Apatinib, Axitinib, AZD8186, Dolastatin, and Oxaliplatin were significantly higher in the high-risk group compared to the low-risk group (P < 0.001). This indicates that patients classified as high-risk may not benefit from these drugs. The role of prognostic risk score model in chemotherapy. A–F Comparison of estimated IC50 value of AZD6738 (A), BI-2536 (B), Dasatinib (C) ERK_6604 (D), SB505124 (E) and SCH772984 (F) between low- and high-risk score groups. G–L The correlation between risk scores and estimated IC50 value of Afuresertib (G), Alpelisib (H), Axitinib (I), AZD8186 (J), Doramapimod (K) and Oxaliplatin (L) PPI network of prognostic HGMMMRGs The PPIs of the 19 model genes were identified using the STRING database, and the interactions among the 19 model genes are shown in Fig. 8A. We constructed a PPI network in Cytoscape and utilized the degree algorithm within its cytoHubba plugin to pinpoint hub genes. A total of IL10 was identified as a hub gene, which was further analyzed the hub genes (Fig. 8B). Time-to-event analysis showed that the higher the mRNA expression of IL10, the better the prognosis of patients with HNSC (Fig. 8D). Further analysis of hub genes. A PPI network of 20 model genes. B Hub genes identified by cyto-Hubba. C Human gut flora associated with IL10. D Comparison of IL10 expression with patients’ OS. E Differences in IL10 expression and immune cell infiltration. (* p < 0.05, ** p < 0.01, and *** p < 0.001) This study analyzed the gut microbiota associated with IL10, including Faecalibacterium prausnitzii, Akkermansia muciniphila, Faecalibacterium prausnitzii A2-165, Faecalibacterium prausnitzii CNCM I-4543, Faecalibacterium prausnitzii L2-6, Faecalibacterium prausnitzii CNCM I-4573, Faecalibacterium prausnitzii, Bacteroides fragilis, Faecalibacterium prausnitzii, Enterococcus faecium, Faecalibacterium prausnitzii A2-165, Lactobacillus fermentum ACA-DC 179, Lactobacillus plantarum ACA-DC 287, Enterococcus durans EP1, Enterococcus durans EP2, Enterococcus durans EP3, Enterococcus durans M4-5, Faecalibacterium prausnitzii, Escherichia coli, Bacteroides fragilis ATCC 23,745 (Table 2) (http://bioannotation.cn/gutmgene/resource.dhtml) [21]. Employing the median expression of IL10 as a cutoff, we compared the immune cell infiltration profiles between the high- and low-expression groups (Fig. 8C). The results showed that, compared to tumors with low IL10 expression, tumors with high IL10 expression had significantly increased T cell CD4 memory activation phenotypes and T follicular helper T cell phenotypes (Fig. 8E). Table 2Human gut flora associated with IL10HumanGene (ID)Gut Microbiota (ID)AlterationThroughputPMIDHumanIL10Akkermansia muciniphila (239935, gm0030)ElevatedLow-throughput30765132HumanIL10Bacteroides fragilis (817, gm0082)ElevatedLow-throughput23828891HumanIL10Bacteroides fragilis ATCC 23745 (gm1123)ElevatedLow-throughput30507089HumanIL10Enterococcus durans EP1 (gm1082)ElevatedLow-throughput28239378HumanIL10Enterococcus durans EP2 (gm1083)ElevatedLow-throughput28239378HumanIL10Enterococcus durans EP3 (gm1084)ElevatedLow-throughput28239378HumanIL10Enterococcus durans M4-5 (gm1006)ElevatedLow-throughput21178864HumanIL10Enterococcus faecalis (1351, gm0288)ElevatedLow-throughput18234854HumanIL10Escherichia coli (562, gm0305)ElevatedLow-throughput19966295HumanIL10Faecalibacterium prausnitzii (853, gm0327)ElevatedLow-throughput30765132HumanIL10Faecalibacterium prausnitzii (853, gm0327)ElevatedLow-throughput23643066HumanIL10Faecalibacterium prausnitzii (853, gm0327)ElevatedLow-throughput20876719HumanIL10Faecalibacterium prausnitzii (853, gm0327)ElevatedLow-throughput18936492HumanIL10Faecalibacterium prausnitzii A2-165 (411483, gm0932)ElevatedLow-throughput28713353HumanIL10Faecalibacterium prausnitzii A2-165 (411483, gm0932)ElevatedLow-throughput25002542HumanIL10Faecalibacterium prausnitzii CNCM I-4543 (gm0979)ElevatedLow-throughput28713353HumanIL10Faecalibacterium prausnitzii CNCM I-4573 (gm0980)ElevatedLow-throughput28713353HumanIL10Faecalibacterium prausnitzii L2-6 (718252, gm0934)ElevatedLow-throughput28713353HumanIL10Lactobacillus fermentum ACA-DC 179 (gm1059)ElevatedLow-throughput18077037HumanIL10Lactobacillus plantarum ACA-DC 287 (gm1060)ElevatedLow-throughput18077037 Human gut flora associated with IL10 Experimental validation of IL10 expression in HNSCC and adjacent normal tissues Based on the TCGA database, we initially validated the IL10 transcript levels in HNSCC and paired normal tissues, revealing a significant difference (P < 0.001) (Fig. 9A). To further confirm these findings, we performed Western Blot analysis to assess IL10 protein expression in HNSCC and adjacent normal tissues. The results showed that IL10 protein expression was significantly higher in HNSCC tissues compared to paired normal tissues (P < 0.001) (Fig. 9B). A Expression levels of IL10 in HNSCC and adjacent normal tissues, B Western Blotting validates the expression levels of IL10 in HNSCC and adjacent normal tissues (T: tumor tissues, N: normal tissues). (* p < 0.05, ** p < 0.01, *** p < 0.001 and ****p < 0.0001) Discussion According to previous studies [22], the 5-year survival rate of patients diagnosed with HNSCC is extremely low. Predicting prognosis and guiding treatment are crucial and necessary. However, there are no reliable biomarkers or predictive models for predicting HNSCC progression. Gut flora plays a crucial and complex role in the body’s central system to prevent infection and maintain health, and the relationship between the gut microbiota and cancer is a relatively new aspect of anticancer research [8, 23]. The gut microbiota may promote tumorigenesis through several mechanisms: (a) direct oncogenic effects of microorganisms and their metabolites; (b) contributing to the conversion of some circulating metabolites into carcinogens; (c) regulating the expression and mutation of oncogenes and oncogenes; and (d) disrupting the immune surveillance function by inducing immunosuppressive pathways [24, 25]. Studies have shown that gut microbes influence the nutrition, metabolism, and immunity of the gut. In populations receiving oncology treatment. In patients undergoing oncology treatment, the gut microbiota determines carcinogenesis and progression and is significantly implicated in the prognosis and clinical management of colorectal cancer [26, 27], non-small lung [28, 29], pancreatic [30, 31], and prostate cancers [32]. However, the relationship between HGMMMRGs and HNSCC is unclear and there is a lack of relevant prognostic prediction models for HNSCC patients. Therefore, we performed this analysis constructed a new HGMMMRGs prognostic model to predict the prognosis and immune status of HNSCC. This study performed a comprehensive analysis of the transcriptional dynamics of the 111 HGMMMRGs in both HNSCC and paired normal tissues. Next, 29 differential genes closely related to Gut Microbes and Microbial Metabolites were screened by correlation analysis, and a risk model containing 19 differential genes was developed using Cox univariate analysis and LASSO Cox regression analysis to assess the prognosis of HNSCC patients. Our analysis indicated that patients in the low-risk group experienced significantly better overall survival. Furthermore, the risk model showed greater predictive accuracy for 5-year outcomes compared to individual clinical factors, based on AUC evaluation. This demonstrates the robustness and validity of the prognostic model for predicting the prognosis of HNSCC. To evaluate the performance of the model further, we tested it using the GEO dataset. In addition, functional analysis revealed that DEGs between low- and high-risk clusters were closely associated with cellular immunity. Significant differences in immune cell infiltration levels were observed across multiple immune cell types between the high-risk and low-risk clusters. Our study found that IL-10 can predict the prognosis of HNSCC patients. IL-10 acts as a cytokine with potent immunosuppressive activity that can stimulate tumor cell proliferation and metastasis [33]. However, the role of IL-10 in the carcinogenesis process remains controversial and contradictory [34]. IL-10 on the one hand induces immunosuppression and contributes to the tumor immune escape, and on the other hand it exerts antitumor effects by stimulating immune cells, among other things [35]. Gunaydi et al. [36] found that differential expression of IL-10 between the cancer and normal control groups were significantly elevated, and IL-10 increased with tumor stage and lymph node infiltration. Jebreel et al. [37] found that IL-10 was measured more frequently in the serum of laryngeal cancer patients than in normal subjects, and that tumor stages T3 and T4 were associated with high levels of IL-10. Several studies have shown that IL-10 levels in the serum and tumor tissue are positively correlated with poor prognosis in melanoma, lung cancer, and HNSCC [38, 39]. IL-10 production has been demonstrated in many solid tumors, including HNSCC [37, 40], and IL-10 may be a potential biomarker for IL-10-targeted therapies [41, 42]. IL10 is closely associated with immune regulation and inflammation, although its exact role has not been determined. Several studies have shown that IL10 promotes angiogenesis through inhibition of antigen-presenting cells [43] and down-regulation of IL12 and IFNc production, becoming a potent inhibitor of the immune system [44]. It is hypothesized that genetic polymorphisms may modulate IL-10’s transcriptional activity or protein structure, thereby influencing individual susceptibility to tumors including HNSCC through its role in anti-tumor immunity [45]. Thus, IL-10 antagonists may be a new therapeutic approach for patients with cancer. The microbiome contributes to tumorigenesis through its interactions with the immune system, thereby promoting chronic inflammation, altering the balance of cell proliferation and death, and eliciting specific immune responses that favor cancer progression. Microorganisms produce metabolically active molecules that alter homeosta sis or exert oncogenic effects in the body. Imbalances in proliferation and apoptotic pathways can be driven by microbiota via mechanisms including PRR signaling, chronic inflammation, genotoxic metabolite production, and loss of metabolically protective factors. Microbiome-immune interactions are significantly involved in tumor development, with the microbiota promoting oncogenesis by driving chronic inflammation, altering cell fate decisions, and shaping anti-tumor immunity [46]. Microorganisms generate metabolically active molecules that can alter homeostasis or exert oncogenic effects within the body [47]. Via its influence on host signaling and metabolism, the microbiota can disrupt proliferation-apoptosis homeostasis. This includes triggering PRR-mediated inflammation, generating genotoxic metabolites, and reducing the availability of protective metabolites [48]. Fusobacterium prausnitzii is an anaerobic, non-bacteriophilic, non-motile, comprising more than 5% (5–15%) of the total adult fecal microbiota [49–51]. Studies have shown that E. faecalis Przewalski enhances the immune system, reduces lipopolysaccharide-associated signaling, improves the activity of the gut microbiota, and prevents leaky gut by maintaining the integrity of the intestinal barrier [52]. E. faecalis Przewalski contributes to the differentiation of CD4 + T cells into CD4CD8αα T cells (Tregs) in the colonic mucosa and prevents excessive inflammatory responses. In CRC patients, Faecalibacterium prausnitzii and CD4CD8αα T cells are simultaneously depleted [53]. Other studies have shown that Faecalibacterium prausnitzii has anti-inflammatory effects in the colon and that oral administration of Faecalibacterium prausnitzii has immunomodulatory effects in inflammatory colon disease and CRC mouse models [54]. In one study, fecal levels of E. faecalis were reduced in patients with breast cancer [55]. When E. faecalis and MCF-7 breast cancer cells are co-cultured in vitro, the proliferation of the latter is impeded by the inhibition of IL-6 production and phosphorylation of JAK2/signal transduction factor and STAT3 activator [55]. Beyond IL10, other interleukins such as IL-6 also play pivotal roles in bridging gut microbiota dysbiosis and HNSCC progression. Recent systematic studies have begun to elucidate the complex regulatory networks through which gut microbial metabolites influence interleukin signaling cascades, thereby modulating the tumor immune microenvironment [56, 57]. For instance, microbial-derived short-chain fatty acids and secondary bile acids can shape the differentiation of T helper cells and regulatory T cells, subsequently altering the balance of IL-6 and IL-10 production. These insights underscore the necessity of integrating multi-omics approaches to fully decipher the gut‒immune‒cancer axis in HNSCC. From a clinical policy and healthcare management perspective, our findings hold significant implications for the personalized treatment of HNSCC. The HGMMMRG-based risk scoring system can serve as a highly cost-effective, transcriptomic-based triage tool. At the policy level, multidisciplinary tumor boards could utilize this signature to optimize medical resource allocation: high-risk patients may be prioritized for more aggressive targeted therapies or closer clinical surveillance, whereas low-risk patients might be spared from the severe toxicities and financial burden of overtreatment. Integrating such gut microbial metabolite-related markers into standard evaluations could profoundly inform future precision medicine guidelines. Furthermore, it is essential to acknowledge the statistical uncertainties inherent in prognostic modeling. As our core signature was derived from LASSO regression, the coefficients lack traditional 95% confidence intervals (CIs) due to the nature of L1 regularization. However, the subsequent Cox survival analyses reflect real-world uncertainties. While narrow CIs in our large cohorts indicate high precision in mortality risk estimation, wider intervals observed in certain subgroup analyses highlight the inherent heterogeneity of HNSCC. Acknowledging these uncertainties is critical to avoid over-interpreting point estimates and ensures a more objective application of our model in clinical practice. This study has several notable limitations in its design and implementation. First, the prognostic assessment model constructed relies heavily on gene expression profile data, which may not fully capture the spatiotemporal heterogeneity of tumors and their evolutionary dynamics. Additionally, as the model is based on retrospective data, it is susceptible to selection bias and information gaps, and its robustness as well as clinical translational value require further validation through well-designed prospective studies. Second, the current model has not yet incorporated key confounding variables such as HPV infection status, tobacco use, alcohol consumption, differences in treatment regimens, patients‘ comorbid conditions, or individual lifestyle factors. Moreover, the HNSCC sample database used exhibits relatively homogeneous population structures and limited coverage of gene expression information, which may constrain the model’s generalizability. Finally, the specific regulatory networks of HGMMMRGs in the development and progression of HNSCC, as well as their feasibility as therapeutic targets, still require systematic elucidation through multi-omics integration analysis and functional experiments. To address these issues, subsequent research will focus on expanding sample diversity, developing dynamic monitoring models, and conducting mechanistic validation experiments for further exploration and refinement. Conclusion Our study elucidated the association between HGMMMRGs and prognosis, immune microenvironment, and tumor mutational load in HNSCC. HGMMMRGs can be used as independent factors to better predict the prognosis of HNSCC, and this exploration may contribute to the discovery of new biomarkers that can help in the prediction of prognosis, clinical diagnosis, and treatment of patients with HNSCC, and may potentially become a new HNSCC option for clinical treatment.Despite these promising computational findings, vital implications for future research remain. Moving forward, it is imperative to validate this prognostic signature in large-scale, prospective, and multicenter clinical trials. Moreover, comprehensive in vivo and in vitro experimental investigations are urgently warranted to unravel the precise molecular mechanisms by which specific gut microbial metabolites interact with the host tumor microenvironment, ultimately paving the way for novel microbiota-targeted therapeutic interventions in HNSCC. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Acknowledgements We would like to thank the researchers and staff who contributed to the collection and analysis of data, including those at The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) databases. Abbreviations Human gut microbes and microbial metabolites related genes Head and neck squamous cell carcinoma Receptor operating characteristic curves Gene Ontology Kyoto Encyclopedia of Genes and Genomes Overall survival Gene Expression Omnibus Author contributions Data curation, Xuan Yang; Formal analysis, Sai Liang, Visualization: Piao Wang; Writing - original draft: Xuan Yang and Wei Huang; Writing - review & editing: Kai Song, Ming Yu and Zhengpeng Gong. Funding This study was supported by the Science and Technology Foundation of Guizhou Province (Grant No. Guizhou Science and Technology Cooperation ZK [2022]–General 411) and the Scientific Research Fund of the Affiliated Hospital of Guizhou Medical University (Grant No. gyfynsfc [2023]–57). Data availability The Cancer Genome Atlas (TCGA) (https://www.cancer.gov/ccg/research/genome-sequencing/tcga) and Gene Expression Omnibus (GEO) databases (https://www.ncbi.nlm.nih.gov/geo/) belong to public databases. The patients involved in the databases have received ethical approval. Users can download the data for free to conduct research and publish related articles. Declarations Ethics approval and consent to participate The data analyzed in the bioinformatic part of this study were obtained from public databases (TCGA and GEO), which were collected in accordance with relevant ethical guidelines. The collection and experimental use of the human clinical tissue specimens (eight pairs of HNSCC and adjacent normal tissues) were approved by the Affiliated Hospital of Guizhou Medical University (Approval ID: 2023 − 436). All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. All participants provided written informed consent to participate in this study. The study was conducted in accordance with the ethical guidelines and approved by the appropriate institutional review board/ethics committee. Participants were informed about the objectives of the study, the procedures involved, and the potential risks and benefits. They were also assured that their participation was voluntary and that they could withdraw at any time without consequence. Consent for publication The authors confirm that all participants provided written informed consent for their data to be published. Identifying information has been omitted to ensure participant anonymity. The data in this study is published with the consent of all participants, and they have agreed to the publication of any anonymized data in scientific journals and other public forums. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1. Supplementary Material 2. Supplementary Material 3. Data Availability Statement The Cancer Genome Atlas (TCGA) (https://www.cancer.gov/ccg/research/genome-sequencing/tcga) and Gene Expression Omnibus (GEO) databases (https://www.ncbi.nlm.nih.gov/geo/) belong to public databases. The patients involved in the databases have received ethical approval. Users can download the data for free to conduct research and publish related articles.