Exploration of m7G-related immune genes and construction of a prognostic risk model in gastric cancer

Abstract

Gastric cancer (GC) is a lethal digestive malignancy with poor outcomes. Although N7-Methylguanosine (m7G) modification and immune-related genes (IRGs) individually affect GC prognosis and the tumor microenvironment, their mechanistic crosstalk remains unclear. Differentially expressed genes (DEGs) in GC were identified and used alongside m7G-related genes (m7G-RGs) to perform consensus clustering, defining molecular subtypes and yielding additional DEGs. Intersection of these gene sets identified m7G-related immune genes. Prognostic genes were selected via Least Absolute Shrinkage and Selection Operator (LASSO), and Cox regression analysis to construct a validated prognostic risk model. Associations between clinicopathological features and risk scores were assessed, identifying independent prognostic factors and enabling nomogram development. The risk-stratified groups underwent gene set enrichment analysis (GSEA), immune infiltration, tumor immune dysfunction and exclusion (TIDE), tumor purity, drug sensitivity, and tumor mutation burden (TMB) analyses. Using GC and normal samples, this study identified 4,458 DEGs. Intersection with subtype-specific DEGs (2,098) and immune genes (4,172) yielded 193 m7G-associated immune genes. Six prognostic genes (ELANE, ASCL2, APOA1, GRP, CD36, MUC15) were used to construct a prognostic risk model that stratified patients into high- and low-risk groups with significant survival differences and strong predictive accuracy. Age, stage, and risk score were independent prognostic factors, and a nomogram was developed. High-risk patients showed enrichment in tumor-promoting pathways (e.g., calcium signaling), immunosuppressive microenvironments, higher stromal scores, and TP53 mutations. Low-risk patients had activated DNA repair pathways, lower TIDE scores, higher immune infiltration, and TTN mutations. Drug sensitivity analysis identified 130 compounds with differential responses between groups (p < 0.05). This study developed a robust prognostic risk model based on six prognostic genes, which effectively reveals molecular mechanisms and immune features of GC, offering a foundation for individualized therapy.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-47456-z.

Introduction

Gastric cancer (GC) is a significant global health issue, responsible for the deaths of up to 650,000 people annually and severely impacting human health. Characterized by a high degree of molecular and phenotypic heterogeneity, the occurrence and progression of GC are influenced by numerous factors, such as Helicobacter pylori infection, smoking, obesity, a high-salt diet¹. In advanced patients, traditional chemotherapy utilizing platinum, fluorouracil, and paclitaxel was employed, and the median overall survival typically amounted to only about 1 year². Although immune checkpoint inhibitors and targeted therapies are on the rise, they can only extend the survival of some patients. The issues of toxicity and drug resistance remain the greatest concerns³.Therefore, an in-depth exploration of the new molecular mechanisms that drive gastric cancer progression, as well as the establishment of biomarkers capable of accurately predicting patient prognosis and immune treatment response, has become an important research direction for achieving personalized treatment.

The development and progression of cancer are driven by a variety of mechanisms, including genetic instability, epigenetic reprogramming, and interactions with the tumor microenvironment(TME)⁴. TME is an intricate system composed of immune cells, tumor cells, stromal cells, extracellular matrix, etc.Recent single-cell and genomic profiling elucidated GC TME heterogeneity, highlighting functional crosstalk between CAFs and myeloid cells. These interactions drive EMT, metabolic reprogramming, and immune remodeling, shaping the tumor’s evolutionary landscape⁵.Its functional status is regulated by various immune-related genes (IRGs), cytokines, chemokines, and metabolites, which collectively influence the tumor’s immune evasion ability and therapeutic efficacy^6,7.Epigenetic reprogramming, which includes genetic mutations, DNA methylation, and RNA modifications, plays a crucial role in cancer development. Research indicates that RNA modifications, such as N7-methylguanosine (m7G), significantly influence tumor immune regulation. These m7G modifications directly affect RNA stability, transport efficiency, and translation performance, while also regulating the functions and states of immune cells, thereby influencing the progression of cancer^8,9.For example, FAT4 mutations have been linked to tumor mutational burden (TMB) and favorable prognosis in GC¹⁰.M7G-related genes (m7G-RGs) are believed to interact with immune signaling pathways via various mechanisms, including the regulation of immune cell infiltration and antigen presentation processes, thereby shaping the immune microenvironment^11,12.In GC tissues, high m7G expression is positively correlated with immune checkpoint molecule PD-L1, whereas low m7G levels are associated with immune cell infiltration^13,14.Pharmacological evidence confirms that natural compounds and bioactive peptides modulate GC proliferation and TP53-mediated apoptosis, underscoring the interplay between molecular signatures and immune responses¹⁵.However, the specific molecular mechanisms by which m7G contributes to the occurrence of gastric cancer and regulates the immune microenvironment have not been fully elucidated and require further investigation.

This study systematically investigated the prognostic value of m7G-related immune genes in gastric cancer (GC) and established a reliable prognostic risk model. Through multi-dimensional approaches, including gene set enrichment analysis (GSEA), immune cell infiltration analysis, tumor mutational burden (TMB) assessment, and drug sensitivity analysis, we comprehensively revealed the molecular mechanisms and immune microenvironment characteristics of GC. These findings provide crucial theoretical foundations and experimental support for developing personalized treatment strategies.

Graphical Abstract: Study workflow and core findings.

This study integrated GC DEGs, immune-related genes, and subtype-specific genes to identify m7G-related immune signatures. A prognostic risk model was constructed via LASSO and Cox regression, then validated through survival, ROC, and nomogram analyses. Further characterization included GSEA, TME assessment (TIDE, purity, infiltration), drug sensitivity, and TMB profiling. Mechanistically, our findings suggest that m7G modification may modulates immune gene expression and the TME, thereby driving GC progression and clinical outcomes.

Materials and methods

Data sources

In this study, The Cancer Genome Atlas Stomach Adenocarcinoma (TCGA-STAD) dataset was obtained from TCGA (accessed on July 23, 2025; https://portal.gdc.cancer.gov/), which included 383 gastric cancer (GC) samples and 36 normal samples. This dataset comprises gene expression (mRNA) and somatic mutation data, with all 383 GC samples having complete survival information. The data were generated using high-throughput sequencing technology, and the clinical variables utilized included age, sex, Tumor, Node, Metastasis (TNM) stage, overall survival time, and status^16,17. This dataset serves as the training set. Three independent validation sets were acquired from the Gene Expression Omnibus (GEO) database (accessed on July 25, 2025; https://www.ncbi.nlm.nih.gov/geo/), including GSE66229 (GPL570 platform, 300 GC samples), GSE84426 (GPL6947 platform, 76 GC samples), and GSE62254 (GPL570 platform, 300 GC samples). All validation datasets were based on array technology^18–20. A total of 29 N7-Methylguanosine-related genes (m7G-RGs) were obtained from the reference²¹ as listed in Table S1. The immune-related gene set (IRGs) was compiled from two public databases, Innate Immune Database (InnateDB, https://www.innatedb.com/; 1,697 genes) and Immunology Database and Analysis Portal (ImmPort, https://www.immport.org/home; 3,118 genes). After removing duplicates, a total of 4,172 IRGs (Table S2) were retained for subsequent analysis²².

Differential expression analysis

To systematically identify differentially expressed genes (DEGs) between GC tissues and normal tissues, this study utilized the TCGA-STAD training set, which includes all available GC samples and normal samples. Differential expression analysis was performed using the R package “DESeq2” (v 1.38.0)²³. DEGs were screened using the threshold criteria of |log₂FoldChange (FC)| > 1 and adjusted p-value (adj.p) < 0.05. Visualization was carried out using the R package “ggplot2” (v 3.4.1) (https://ggplot2.tidyverse.org) to generate a volcano plot and the R package “ComplexHeatmap” (v 2.16.0)²⁴ to construct a heatmap. The top 10 up- and down-regulated DEGs were labeled and displayed based on ranking by adj.p.

Consistency cluster analysis

To identify molecular subtypes of GC defined by the expression patterns of m7G-RGs, this study employed consensus clustering for molecular subtyping. Using the R package “ConsensusClusterPlus” (v 1.64.0)²⁵, consensus clustering was performed on 383 GC samples from the TCGA-STAD training set with complete survival information. First, Cox regression analysis of m7G-RGs was conducted using the R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival) with screening criteria set at p < 0.05 and HR ≠ 1. Significantly associated genes were visualized via a forest plot generated with the R package “forestplot“ (v 3.1.5) (https://CRAN.R-project.org/package=forestplot). Subsequently, the selected genes were further subjected to proportional hazards (PH) assumption testing (p > 0.05) using the cox.zph function from the R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival). The R package “survminer” (v 0.4.9) (https://CRAN.R-project.org/package=survminer) was used to generate Schoenfeld residual plots with the ggcoxzph function to evaluate the conformity with model assumptions. Based on these results, m7G-RGs were used for consensus clustering analysis. The optimal number of clusters was determined through comprehensive assessment of the cumulative distribution function (CDF) plot, relative change in area under the CDF curve, and cluster consensus heatmaps. To further evaluate prognostic differences between subtypes, patient survival data were integrated. Kaplan-Meier (K-M) survival analysis accompanied by log-rank tests was performed using the R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival). Survival curves were plotted with the R package “survminer” (v 0.4.9) (https://CRAN.R-project.org/package=survminer) to quantitatively compare survival outcomes among different molecular subtypes, with a log-rank p < 0.05 considered statistically significant.

Analysis of molecular subtype differences

To identify DEGs between m7G-mediated molecular subtypes, this study performed transcriptomic differential expression analysis using the R package “DESeq2” (v 1.38.0)²³ on all GC samples from the TCGA-STAD training set. The screening criteria for DEGs were set as |log₂FC| > 1 and adj.p < 0.05. To visually represent the overall distribution of DEGs, a volcano plot was generated using the R package “ggplot2” (v 3.4.1) (https://ggplot2.tidyverse.org), in which the top 10 most significantly up- and down-regulated DEGs were labeled based on adj.p ranking. To further illustrate the expression patterns of the DEGs, the top 10 up-regulated and top 10 down-regulated genes (totaling 20 genes) ranked by adj.p were selected as representatives, and their expression heatmap was constructed using the R package “ComplexHeatmap” (v 2.16.0)²⁴.

Identification of candidate genes and comprehensive functional analysis

To identify m7G immune-related candidate genes in GC, the R package “ggvenn” (v 0.1.9) (https://CRAN.R-project.org/package=ggvenn) was used to intersect DEGs between GC and normal samples, DEGs from molecular subtypes, and IRGs. The resulting overlapping candidate genes were visualized in a Venn diagram. To elucidate the potential functions and pathways associated with these candidate genes, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed using the R package “clusterProfiler” (v 4.15.0.003)²⁶. The GO analysis covered three categories—Biological Process (BP), Cellular Component (CC), and Molecular Function (MF)—to systematically characterize the potential functional and pathway mechanisms of the candidate genes in disease. The significance threshold was set at an adj.p < 0.05. The enrichment results were visualized using the R package “enrichplot” (v 1.20.3) (10.18129/B9.bioc.enrichplot). To reveal the functional and regulatory mechanisms of the candidate genes in biological processes and disease pathogenesis through their encoded protein interactions, a protein-protein interaction (PPI) network was constructed using the STRING online database (https://cn.string-db.org/) with an interaction confidence score threshold greater than 0.7. The interactions were ranked based on node degree values, and the PPI network was visualized using the R package “circlize” (v 0.4.16)²⁷.

Prognosis gene screening and expression level validation

In the TCGA-STAD training set with complete survival information, Cox regression analysis was first performed on candidate genes. The R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival) was used to identify genes significantly associated with GC prognosis (p < 0.05 and HR ≠ 1), and the results were visualized using the R package “forestplot” (v 3.1.5) (https://CRAN.R-project.org/package=forestplot) to generate a forest plot. Subsequently, genes that showed significance in the Cox regression analysis were further subjected to the PH assumption test using the cox.zph function from the R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival) (p > 0.05). The R package “survminer” (v 0.4.9) (https://CRAN.R-project.org/package=survminer) was employed to plot Schoenfeld residuals via the ggcoxzph function to assess whether the model met the PH assumption.To optimize the predictive performance of the prognostic gene set, Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis was conducted on the GC-related prognostic genes obtained from the aforementioned analysis within the TCGA-STAD training set GC samples, using the R package “glmnet” (v 4.1–8.1)²⁸ with ten-fold cross-validation to avoid overfitting. Finally, the genes selected by LASSO regression were incorporated into a Cox regression analysis (p < 0.05 and HR ≠ 1). The R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival) was again used for stepwise regression analysis, and the PH assumption of the final model was re-evaluated using the cox.zph function (p > 0.05), with Schoenfeld residuals visualized via the R package “survminer” (v 0.4.9) (https://CRAN.R-project.org/package=survminer). To examine the expression patterns of the prognostic genes in tissues, a Wald test was applied based on TCGA-STAD training set to compare the expression differences of these genes between GC tissues and adjacent normal tissues (p < 0.05). The results were visualized using the R package “ggplot2” (v 3.4.1) (https://ggplot2.tidyverse.org).

Prognostic risk model construction and validation

To construct and evaluate a prognostic risk model, a prognostic signature was developed based on the prognostic genes identified through Cox analysis. The prognostic risk model was subsequently subjected to validation analysis. Risk scores were calculated for each sample with available survival information in both the TCGA-STAD training set and the external validation sets (GSE66229, GSE84426, and GSE62254) using the following formula:

In the above formula, β denotes the coefficient (coef), with detailed values provided in the supplementary material, and x represents the expression level of the corresponding gene in the sample.

Based on the risk scores, the optimal cutoff value was determined using the surv_cutpoint function to establish the risk stratification threshold, dividing the GC samples in the training set into high- and low-risk groups. Subsequently, the R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival) was employed to compare survival differences between the high- and low-risk groups via K-M survival curves, with a log-rank test p < 0.05 indicating statistically significant survival differences. The R package “survivalROC” (v 1.0.3.1) (https://CRAN.R-project.org/package=survivalROC) was used to plot receiver operating characteristic (ROC) curves for 1-, 3-, and 5-year survival rates, validating the predictive accuracy of the risk scoring model. A higher area under the curve (AUC) value indicates better predictive performance, with an AUC > 0.6 considered acceptable. Finally, risk score distribution plots and survival status distribution plots were generated using the plot function, and the R package “pheatmap” (v 1.0.12) (https://CRAN.R-project.org/package=pheatmap) was utilized to visualize the expression heatmap of prognostic genes in the high- and low-risk groups. To validate the applicability of the prognostic risk model, risk scores for each patient in the validation sets were calculated based on the expression levels of the prognostic genes and their corresponding coefficients, using the same formula. The optimal cutoff value derived from the risk scores was applied to stratify patients in the validation sets into high- and low-risk groups. The same analytical methods—including survival difference comparison, time-dependent ROC analysis, and visualization—were applied to each validation set to ensure consistency and robustness of the evaluation.

Independent prognosis analysis and nomogram construction

In the TCGA-STAD training set comprising GC samples with complete survival information, Cox regression analysis was performed using the R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival) to evaluate various clinicopathological features and the risk score. Statistical significance was defined as HR ≠ 1 with p < 0.05. Factors that showed significance in the Cox regression analysis were further subjected to the PH assumption test using the cox.zph function from the same R package “survival” (v 3.7.0) (https://CRAN.R-project.org/package=survival), with a significance level of p > 0.05. The R package “survminer” (v 0.4.9) (https://CRAN.R-project.org/package=survminer) was utilized to generate Schoenfeld residual plots via the ggcoxzph function to assess compliance with the PH assumption. Subsequently, variables satisfying the PH assumption were included in a Cox regression analysis. The PH assumption was re-evaluated for the model using the cox.zph function (p > 0.05), and corresponding residual plots were generated with the ggcoxzph function. Finally, independent prognostic factors were identified based on the Cox model results using a significance threshold of HR ≠ 1 and p < 0.05.

Based on the identified independent prognostic factors, a prognostic nomogram was constructed using the R package “rms” (v 6.8-1.8) (https://CRAN.R-project.org/package=rms) in the TCGA-STAD training set with available survival information. The nomogram was designed to predict patients’ 1-, 3-, and 5-year overall survival probabilities. Calibration curves at each time point were plotted using the R package “calibrate” (v 1.7.7)²⁹ to evaluate the agreement between predicted and observed probabilities. Time-dependent ROC curves were generated with the R package “timeROC” (v 0.4)³⁰ to assess the discriminative ability of the nomogram at different time points. Additionally, decision curve analysis (DCA) was performed using the R package “ggDCA” (v 1.1) (https://CRAN.R-project.org/package=ggDCA) to examine the clinical utility of the nomogram.

Gene set enrichment analysis (GSEA) in risk groups

To further explore the relevant signaling pathways and potential biological mechanisms in high- and low-risk GC patients, differential expression analysis was performed between the high- and low-risk groups using the R package “DESeq2” (v 1.38.0)²³ based on samples with complete survival information from the TCGA-STAD training set. The log₂FC values were calculated and ranked in descending order. GSEA was then conducted using the R package “clusterProfiler” (v 4.15.0.003)²⁶ with the KEGG gene set “c2.cp.kegg_legacy.v2024.1.Hs.symbols” from the Molecular Signatures Database (MSigDB) (https://www.gsea-msigdb.org/gsea/msigdb) as the reference gene set. Significantly enriched pathways were visualized using the R package “enrichplot” (v 1.20.3) (10.18129/B9.bioc.enrichplot). The significance thresholds were set as |NES| > 1 and adj.p < 0.05, with the top five most significantly enriched pathways displayed based on adj.p ranking.

Tumor immune dysfunction and exclusion (TIDE) analysis and tumor purity analysis

The TIDE online platform (http://tide.dfci.harvard.edu/) was employed to evaluate potential differences in response to immunotherapy between the high- and low-risk groups. Based on expression profile data from samples with survival information in the TCGA-STAD training set, the TIDE score was calculated for each patient. A higher TIDE score indicates a greater likelihood of immune evasion during immunotherapy. Subsequently, the Wilcoxon test was applied to compare the statistical differences in TIDE scores between the high-risk and low-risk groups (p < 0.05), and the R package “ggpubr” (v 0.6.0) (https://CRAN.R-project.org/package=ggpubr) was used to visualize the comparison results.

To further investigate differences in tumor microenvironment scores between the two groups, the Estimation of STromal and Immune cells in MAlignant Tumor tissues using Expression data (ESTIMATE) (https://bioinformatics.mdanderson.org/estimate/) was utilized. The R package “estimate” (v 1.0.13) (https://R-Forge.R-project.org/projects/estimate/) was applied to calculate the Stromal Score, Immune Score, and ESTIMATE Score for all samples in the TCGA-STAD training set. The ESTIMATE Score was used to infer tumor purity. The Wilcoxon rank-sum test was employed to compare the differences in these three scores between the high-risk and low-risk groups, with a significance level set at p < 0.05.

Immune infiltration analysis

To evaluate the immunological relevance of the prognostic model, a systematic analysis of tumor-infiltrating immune cell (TIC) abundance was conducted. In the TCGA-STAD training set, the relative proportions of 22 immune cell types were quantified using the R package “CIBERSORT” (v 0.1.0)³¹. The analysis followed the official recommended workflow and was repeated the algorithm 1,000 permutations to enhance the robustness of the results. Other key parameters were set as follows: the “LM22” signature matrix was used to distinguish 22 immune cell types. Quantile normalization was disabled for RNA-seq data (quantile normalization = FALSE), as recommended for RNA-seq datasets.Samples with a CIBERSORT output p > 0.05 were excluded, and subsequent tumor microenvironment analyses were performed on the remaining samples. To compare differences in immune cell infiltration levels between the high- and low-risk groups, the Wilcoxon test was applied for group-wise comparisons, with a significance threshold set at p < 0.05. Significantly differentially infiltrated immune cells (defined as differential immune cells) were identified, and the results were visualized using the R package “ggplot2” (v 3.4.1) (https://ggplot2.tidyverse.org). Further correlation analysis of these differential immune cells was performed using the R package “psych” (v 2.4.6.26) (https://CRAN.R-project.org/package=psych). Pearson correlation coefficients were calculated, with filtering criteria set at |correlation coefficient (cor)| > 0.3 and p < 0.05. A heatmap was generated using the R package “ggplot2” (v 3.4.1) (https://ggplot2.tidyverse.org) to visualize the correlation patterns. To investigate the association between prognostic genes and differential immune cells, Spearman correlation analysis was conducted based on samples from the high-risk and low-risk groups in the TCGA-STAD training set. The correlation between prognostic gene expression and differential immune cell infiltration levels was assessed, with significance criteria defined as |cor| > 0.3 and p < 0.05.

Drug susceptibility analysis

To evaluate the differences in response to chemotherapeutic agents between high- and low-risk patient groups, common chemotherapy drugs for GC were retrieved from the Genomics of Drug Sensitivity in Cancer (GDSC) database. Using the R package “oncoPredict” (v 1.2) (https://CRAN.R-project.org/package=oncoPredict), drug response prediction was performed. The oncoPredict algorithm employs ridge regression to construct predictive models based on pharmacogenomic data from the GDSC database, and empirical Bayes methods were applied to correct potential batch effects between the training data and TCGA expression data. Subsequently, the half-maximal inhibitory concentration (IC₅₀) values of commonly used chemotherapeutic and molecularly targeted agents were calculated for GC samples with survival information in the TCGA-STAD training set. The Wilcoxon test was employed to compare differences in IC₅₀ values between the high- and low-risk groups for each drug (p < 0.05), and boxplots were generated using the R package “ggplot2” (v 3.4.1) (https://ggplot2.tidyverse.org) to visualize the results. To explore potential therapeutic agents for GC, drug compound prediction was performed based on the identified prognostic genes using the Drug Signatures Database (DsigDB, https://dsigdb.tanlab.org/DSigDBv1.0/). Interactions between predicted drug compounds and prognostic genes were analyzed, and a drug compound–biomarker interaction network was visualized using Cytoscape (v 3.9.1)³².

Tumor mutational burden (TMB) analysis

The Wilcoxon test was applied to compare differences in TMB between the high- and low-risk groups, with a significance threshold of p < 0.05. Additionally, to investigate the gene mutation profiles in high-risk and low-risk groups of GC patients, somatic mutation data from the TCGA-STAD training set were comprehensively summarized, analyzed, and visualized using the R package “maftools” (v 2.16.0)³³. A somatic mutation waterfall plot was generated to display variant information from the Mutation Annotation Format (MAF) file, focusing on the top 15 genes with the highest mutation rates. Distinct color annotations were used to differentiate mutation types. Besides, Kaplan-Meier survival analysis was performed to compare overall survival between high- and low-TMB groups based on the median TMB value. To assess the combined prognostic value of TMB and the risk score, patients were further stratified into four subgroups according to TMB status and risk score. Spearman correlation analysis was conducted to evaluate the relationships among the risk score, TMB, and TIDE scores.

Quantitative real-time PCR (RT-qPCR) and RT-PCR

Collect 10 cases of untreated postoperative gastric cancer tissue from the Fourth Hospital of Hebei Medical University, and these samples were not included in the TCGA or GEO datasets, ensuring independent validation of our prognostic genes. Samples were selected according to the following criteria: patients were diagnosed with primary gastric cancer, had not received preoperative chemotherapy or radiotherapy, and had complete clinical information.The studies involving humans participants were approved by the Ethics Committee of the Fourth Hospital of Hebei Medical University (2024KY234). Informed consent was obtained from all subjects and/or their legal guardians. All procedures were conducted in accordance with local legislation, institutional requirements, and the Declaration of Helsinki.

According to the manufacturer’s instructions, total RNA from tissues was isolated using TRIzol reagent (Tiangen, China) and quantified. mRNAs were reverse transcribed by using PrimeScript™ RT reagent Kit with gDNA Erase and SYBR^® Premix Ex Taq™ II (Tli RNaseH Plus) (TaKaRa) in accordance with the manufacturer’s protocol. GAPDH was used as the internal control.The reactions were performed independently in triplicate. All the primer sequences are listed in Additional file (Table S10).The risk scores of 10 patients were calculated separately.

Statistical analysis

R (v 4.2.2) was used to conduct the bioinformatics analysis. Differences between the two groups were assessed using the Wilcoxon test (p < 0.05).

Results

Identification of DGEs in GC

Through transcriptomic differential expression analysis between GC tissues and normal tissues, a total of 4,458 significantly DEGs were identified, among which 2,195 genes were up-regulated and 2,263 were down-regulated in GC tissues (Fig. 1A-B, Table S3).

Fig. 1.

Identification of DGEs in GC. Figure 1A Volcano plot of differential expression analysis; Fig. 1B Heatmap of differential expression analysis.

Prognostic m7G signature and molecular subtyping

Further univariate Cox regression analysis (p < 0.05 and HR ≠ 1) combined with PH assumption testing (p > 0.05) identified three m7G-RGs significantly associated with prognosis, including NUDT10, NUDT11, and EIF4E1B (Fig. 2A-D). Consensus clustering analysis based on the expression profiles of these genes determined the optimal number of clusters as two, dividing the patients into two molecular subtypes: C1 (n = 96) and C2 (n = 287) (Fig. 2E). Survival analysis revealed that patients in the C2 subtype had significantly better overall survival outcomes than those in the C1 subtype (p < 0.004, Fig. 2F), indicating that this classification can effectively distinguish patient prognosis and providing important clues for investigating the molecular mechanisms of GC and clinical prognostic assessment. Differential expression analysis between C1 and C2 identified 2,098 significantly differentially expressed genes for subsequent analyses, among which 472 genes were up-regulated and 162 were down-regulated in the C2 group compared to the C1 group (Fig. 2G-H, Table S4).

Fig. 2.

Prognostic m7G Signature and Molecular Subtyping. Figure 2A Forest plot of univariate Cox regression analysis for m7G-RGs; Fig. 2B-D Residual plots for testing the PH assumption; Fig. 2E Heatmap of consensus clustering analysis; Fig. 2F KM curve for molecular subtypes; Fig. 2G Volcano plot of DEGs across molecular subtypes; Fig. 2H Heatmap of DEGs among molecular subtypes.

Identification of candidate genes and comprehensive functional analysis in GC

By intersecting the DEGs from GC and normal samples, the DEGs between C1 and C2 subgroups, and IRGs, 193 overlapping genes were identified and defined as candidate genes (Fig. 3A, Table S5). KEGG and GO enrichment analyses were performed on these candidate genes (adj.p < 0.05). KEGG analysis revealed enrichment in 25 pathways, with significant involvement in cytokine-cytokine receptor interaction, lipid and atherosclerosis, and the JAK-STAT signaling pathway, among others (Fig. 3B, Table S6). GO analysis identified 1053 enriched terms, including 963 BP terms, 50 CC terms, and 40 MF terms (Fig. 3C, Table S7). Notable BP terms included humoral immune response, leukocyte migration, and chemotaxis; CC terms included collagen-containing extracellular matrix, secretory granule lumen, and cytoplasmic vesicle lumen; MF terms included cytokine activity, enzyme inhibitor activity, and cytokine receptor binding. These enrichment results indicate that the candidate genes are significantly associated with key pathways such as immune and inflammatory responses and infectious diseases, suggesting that m7G modification may play an important role in GC progression by regulating the tumor immune microenvironment and host defense mechanisms. This provides new theoretical insights into the function of m7G in immune regulation in GC. Furthermore, a PPI network of the candidate genes was constructed, comprising 121 genes and 200 interaction pairs. The top five genes ranked by degree were IL6, CXCL12, CRP, IFNG, and APOA1 (Fig. 3D). The identified candidate genes and signaling pathways may serve as potential therapeutic targets and mechanisms of action in GC.

Fig. 3.

Identification of candidate genes and functional enrichment in GC. Figure 3A Venn diagram showing DEGs between GC and normal tissues, molecular subtype DEGs, and immune-related genes; Fig. 3B KEGG enrichment analysis of candidate genes; Fig. 3C GO enrichment analysis of candidate genes. Figure 3D. PPI network of candidate genes.

Identification and expression level validation of 6 prognostic genes

Univariate Cox regression analysis was performed on the aforementioned 193 candidate genes, identifying 84 genes significantly associated with prognosis (p < 0.05, HR ≠ 1) (Fig. 4A). After PH assumption testing, 83 genes met the criteria (p > 0.05) (Figure S1). Using LASSO regression, 10 important feature genes were selected at lambda.min = 0.041 (Fig. 4B). Subsequently, PH assumption testing and multivariate Cox analysis were applied to these 10 genes, ultimately identifying six prognostic genes: ELANE, ASCL2, APOA1, GRP, CD36, and MUC15 (Figure S2, Fig. 4C). Analysis of the expression levels of these prognostic genes in GC tissue samples and normal tissue samples revealed that GRP, MUC15, CD36, ELANE, and APOA1 were significantly downregulated in GC tissues, while ASCL2 was significantly upregulated (p < 0.05) (Fig. 4D).

Fig. 4.

Screening and expression level validation of prognostic genes. Figure 4A Forest plot of univariate Cox regression analysis; Fig. 4B-C Plots of LASSO regression analysis; Fig. 4D Forest plot of multivariate Cox analysis.

Construction and validation of prognostic risk model in GC

Based on the six prognostic genes identified above, a prognostic risk model for GC was constructed. Using the optimal cutoff value (−1.986047), GC samples in the training set were stratified into a high-risk group (n = 175) and a low-risk group (n = 208). In the training set, the risk score distribution plot clearly demonstrated that the model successfully classified patients into two distinct risk groups. The survival status diagram indicated a higher number of deaths in the high-risk group (Fig. 5A–B). Furthermore, the K-M curve for the training set revealed a significant difference in survival time between the high- and low-risk groups (p < 0.05), with patients in the high-risk group exhibiting poorer prognosis (Fig. 5C). The time-dependent ROC curves showed AUC values greater than 0.6 for 1-, 3-, and 5-year survival, confirming the predictive efficacy of the model (Fig. 5D). The heatmap of prognostic gene expression revealed that GRP, ELANE, CD36, APOA1, and MUC15 were highly expressed in the high-risk group, while ASCL2 was highly expressed in the low-risk group (Fig. 5E).

Fig. 5.

Construction and validation of prognostic risk model in GC. Figure 5A Distribution of risk scores in high- and low-risk groups from the TCGA-STAD training set; Fig. 5B Survival status distribution; Fig. 5C KM survival curve; Fig. 5D ROC curve; Fig. 5E Heatmap of prognostic gene expression.

The predictive accuracy of this prognostic risk model was further validated in three independent datasets: GSE66229, GSE84426, and GSE62254. Using dataset-specific optimal cutoff values for risk scores, the validation sets were divided into high- and low-risk groups as follows: GSE66229 (120 vs. 180, cutoff = 0.915334), GSE84426 (42 vs. 34, cutoff = 4.388249), and GSE62254 (120 vs. 180, cutoff = 0.9172028). In all three validation sets, the risk score distribution plots confirmed successful patient stratification into two risk groups, and the survival status diagrams consistently showed a higher number of deaths in the high-risk groups. The KM curves for all validation sets yielded p < 0.05, and the AUC values for 1-, 3-, and 5-year survival in the ROC curves were all above 0.6. The prognostic gene expression heatmaps consistently demonstrated high expression of GRP, ELANE, CD36, APOA1, and MUC15 in the high-risk groups, and high expression of ASCL2 in the low-risk groups (Figure S3). These results, consistent with those from the training set, collectively indicate that the prognostic model constructed in this study exhibits robust predictive performance.

Identification of independent prognostic factors and construction of a predictive nomogram in GC

An analysis of various clinicopathological characteristics and risk scores identified age, stage, T stage, N stage, and risk score as significant factors meeting the criteria of univariate Cox analysis (HR ≠ 1, p < 0.05) and the PH assumption test (p > 0.05) (Fig. 6A, Figure S4). Further analysis revealed that age, stage, and risk score satisfied both the multivariate PH assumption test (p > 0.05) and the multivariate analysis screening criteria (HR ≠ 1, p < 0.05) (Fig. 6B, Figure S5). Therefore, age, stage, and risk score were selected as independent prognostic factors.

Fig. 6.

Independent prognosis analysis and nomogram construction. Figure 6A Forest plot of univariate Cox analysis; Fig. 6B Forest plot of multivariate Cox analysis; Fig. 6C Nomogram; Fig. 6D-F ROC curve, calibration curve, and decision curve analysis of the nomogram.

Based on these independent prognostic factors, a prognostic nomogram was constructed to predict patients’ 1‑, 3‑, and 5‑year survival probabilities, as shown in Fig. 6C. Using one patient from the training set as an example, the predicted mortality rates at 1, 3, and 5 years were 57.2%, 94.4%, and 98.3%, respectively. Furthermore, the time-dependent ROC curves of the nomogram exhibited AUC values all greater than 0.7 (Fig. 6D), the calibration curves demonstrated slopes approaching 1 (Fig. 6E), and the decision curves showed net benefits above 0 for 1-, 3-, and 5-year predictions (Fig. 6F). These results indicate that the nomogram developed in this study has robust predictive performance.

GSEA reveals distinct pathway enrichment patterns between high- and low-risk GC subgroups

Based on samples with survival information in the training set, GSEA was performed between the high- and low-risk groups, revealing enrichment in 31 pathways (|NES| > 1, adj.p < 0.05). Among these, 15 pathways were significantly enriched in the high-risk group and 16 in the low-risk group (Table S8). As shown in Fig. 7A-B, the high-risk group was primarily enriched in pathways related to tumor metastasis and progression, such as the calcium signaling pathway and focal adhesion. In contrast, the low-risk group showed significant enrichment in functional pathways involved in maintaining genomic stability, including DNA repair and RNA metabolism. These findings suggest that the prognostic model effectively distinguishes GC subtypes with distinct biological characteristics: tumors in the high-risk group may exhibit greater invasiveness, while those in the low-risk group tend to maintain higher genetic stability.

Fig. 7.

GSEA analysis. Figure 7A GSEA plot for the high-risk group; Fig. 7B GSEA plot for the low-risk group.

Association between risk stratification and tumor immune microenvironment in GC

As shown in Fig. 8A, the TIDE score was significantly lower in the low-risk group than in the high-risk group (p < 0.05), suggesting that patients in the low-risk group may exhibit greater sensitivity to immune checkpoint inhibitor therapy. Furthermore, tumor purity analysis revealed that both the stromal score and tumor purity were significantly higher in the high-risk group compared to the low-risk group (p < 0.05) (Fig. 8B), indicating a more abundant stromal component and a higher proportion of tumor cells in the tumor microenvironment of this subtype. These findings imply that the high-risk subgroup may be characterized by a more immunosuppressive microenvironment and a tendency toward poorer clinical outcomes.

Fig. 8.

TIDE analysis and tumor purity analysis. Figure 8A Differences in IDE scores between high- and low-risk groups; Fig. 8B Differences in stromal score, immune score, and ESTIMATE score between high- and low-risk groups.

Distinct immune microenvironment profiles between high- and low-risk GC subgroups

As illustrated in the immune cell distribution profile between high- and low-risk groups (Fig. 9A), patients in the low-risk group exhibited a high infiltration pattern of antitumor immune cells such as CD8⁺ T cells and activated NK cells. Furthermore, six types of immune cells showed significant differences between the high- and low-risk groups (p < 0.05), including activated memory CD4⁺ T cells, M2 macrophages, and monocytes (Fig. 9B). As shown in Fig. 9C, a significant negative correlation was observed between activated memory CD4⁺ T cells and resting memory CD4⁺ T cells (cor = −0.52, p < 0.05), as well as between resting NK cells and resting mast cells (cor = −0.45, p < 0.05). In the correlation analysis between differentially expressed immune cells and prognostic genes (Fig. 9D), ELANE showed significant positive correlations with resting mast cells (cor = 0.41, p < 0.05) and monocytes (cor = 0.34, p < 0.05), but a significant negative correlation with activated memory CD4⁺ T cells (cor = −0.35, p < 0.05). Meanwhile, CD36 was significantly negatively correlated with activated memory CD4⁺ T cells (cor = −0.37, p < 0.05). In summary, these findings suggest that the low-risk group may possess a more active antitumor immune microenvironment, whereas the high-risk group is characterized by a higher abundance of immunosuppressive cell types.

Fig. 9.

Immune infiltration analysis. Figure 9A Distribution of immune cells in high- and low-risk groups; Fig. 9B Differences in immune cell infiltration between high- and low-risk groups; Fig. 9C Correlations among differentially expressed immune cells; Fig. 9D Correlations between differentially expressed immune cells and prognostic genes.

Identification of differential drug sensitivity and potential therapeutic agents in GC risk subgroups

A total of 198 chemotherapy drugs commonly used in GC were obtained from the GDSC database. After excluding drugs with abnormal IC₅₀ values, 180 drugs remained, among which 130 exhibited significant differences in IC₅₀ between the high- and low-risk groups (p < 0.05) (Table S9). As shown in Fig. 10A, the top 10 most significantly associated drugs were selected and displayed based on p-value ranking. Furthermore, potential therapeutic drugs targeting the prognostic genes were predicted: 85 drugs were predicted for ELANE, 9 for ASCL2, 81 for APOA1, 7 for GRP, 57 for CD36, and 7 for MUC15 (Fig. 10B). These findings provide new candidate strategies and a theoretical foundation for personalized chemotherapy and targeted therapy tailored to specific molecular subtypes.

Fig. 10.

Drug sensitivity analysis and drug prediction. Figure 10A Drug sensitivity analysis; Fig. 10B Prognostic gene-drug interaction network.

TMB and mutation landscape in relation to the prognostic risk model

As illustrated in Fig. 11A, the TMB was significantly lower in the high-risk group compared to the low-risk group (p < 0.05), suggesting potential differences in genomic stability and immune therapy response potential between the two subtypes. Furthermore, as shown in Fig. 11B-C, distinct differences in mutation profiles were observed between the high-risk and low-risk groups: the TP53 gene exhibited the highest mutation frequency in the high-risk group, whereas PIK3CA mutations were most prevalent in the low-risk group. Although 12 genes were common to the top 15 most frequently mutated genes in both groups, their ranking and the distribution of genes such as PTEN, NCOR1, and LRP2 differed significantly between the groups.

Fig. 11.

TMB analysis. Figure 11A Differences in TMB between high- and low-risk groups; Fig. 11B Waterfall plot for the high-risk group; Fig. 11C Waterfall plot for the low-risk group.

Survival analysis in the TCGA-STAD cohort showed that patients with high TMB had significantly better prognosis (Fig. 12A). Integrating TMB with the risk score further stratified patients, with the Low-risk/High-TMB group exhibiting the best survival and High-risk/Low-TMB the worst (Fig. 12B). Positive correlations between risk score, TMB, and TIDE scores suggest that higher mutational burdens may indicate lower immune escape potential and better immunotherapy response (Fig. 12C).

Fig. 12.

Association of TMB with the prognostic risk model and immunotherapy response. Figure 12A Kaplan–Meier survival analysis comparing overall survival between high- and low-TMB groups in the TCGA-STAD cohort; Fig. 12B Kaplan–Meier survival analysis of four subgroups stratified by TMB status and risk score; Fig. 12C Correlation analysis among risk score, tumor mutational burden (TMB), and TIDE score.

Clinical characteristics between high- and low-risk GC subtypes

Collect 10 cases of untreated postoperative gastric cancer tissue from the Fourth Hospital of Hebei Medical University. The risk scores of 10 patients were calculated separately.The T stage and tumor stage of high-risk group were worse than those of low-risk group, which suggested that the prognosis of high-risk group was worse(Table 1).

Table 1.

Clinical characteristics of high- and low-risk groups.

Patient characteristic	High-risk group(N = 8)	low-risk group(N = 2)	p
Men, N(%)	7(87.5)	1(50%)	0.104
Age Median, (IQR)
Primary tumor location, N(%)			0.301
Gastro-oesophageal junction	3(37.5%)	0
Stomach	5(62.5%)	2(100%)
Lauren’s type, N(%)			0.301
Intestinal	5(62.5%)	2(100%)
Non-intestinal	3(37.5%)	0
T stage, N(%)			0.016
T1−2	1(12.5%)	2(100%)
T3−4	7(87.5%)	0
N stage, N(%)			0.114
N0−1	3(37.5%)	2(100%)
N2−3	5(62.5%)	0
pTNM, N(%)			0.016
I-II	1(12.5%)	2(100%)
III-IV	7(87.5%)	0
Grade			0.747
G1−2	5(62.5%)	1(50%)
G3	3(37.5%)	1(50%)
Vascular invasion, N(%)	4(50%)	0	0.197
Nerve infiltration, N(%)	4(50%)	0	0.197

Discussion

This study investigates the prognostic value and underlying mechanisms of m7G-related immune genes in GC. We utilized transcriptomic data from gastric cancer (GC) and normal tissues. Through differential expression analysis and multi - omics integration, we identified 193 m7G - related immune genes, pinpointed six key prognostic genes (ELANE, ASCL2, APOA1, GRP, CD36, MUC15), and constructed a prognostic risk model. The model effectively differentiates between high-risk and low-risk patient groups. The high-risk group exhibits poorer overall survival, and the model demonstrates robust predictive performance at 1, 3, and 5 years (AUC > 0.6).Cox analysis identified age, stage, and risk score as independent prognostic factors, and the nomogram demonstrated excellent calibration and clinical applicability.High-risk groups are enriched in tumor progression-related pathways (e.g., calcium signaling pathways), while low-risk groups are predominantly enriched in DNA repair pathways. The lower TIDE scores in low-risk groups suggest greater sensitivity to immunotherapy, whereas high-risk groups exhibit higher stromal scores and tumor purity, indicating more pronounced immune suppression. Furthermore, pharmacogenomic sensitivity and mutation profiling revealed significant differences between the two groups in IC50 and high - frequency mutations (TP53 and TTN) across 130 drugs. These findings provide potential biomarkers and provide theoretical insights based on observed associations, that may guide future studies on the molecular mechanisms and personalized treatment of GC, which require further experimental validation.

The study identified six prognostic genes (ELANE, ASCL2, APOA1, GRP, CD36, MUC15). GRP, MUC15, CD36, ELANE, and APOA1 were downregulated, while ASCL2 was upregulated in GC tissues compared to normal tissues. In the prognostic risk model, GRP, ELANE, CD36, APOA1, and MUC15 are highly expressed in the high-risk group, while ASCL2 is highly expressed in the low-risk group.Current research indicates that ELANE (elastase, neutrophil expressed) is abnormally expressed in GC tissues³⁴.ELANE encodes neutrophil elastase, which may enhance anti-tumor immunity by degrading immune suppressors and boosting T-cell infiltration. Neutrophils rapidly synthesize large amounts of ELANE upon activation.The expression of ELANE in normal tissues was significantly higher than that in gastric cancer tissues.The elevated expression of ELANE is correlated with the progression of GC. ELANE participates in apoptosis and immune infiltration, and it represents a significant prognostic risk factor in GC^35–37.This is consistent with our findings. ELANE is highly expressed in the high - risk group.Notably, in our study, ELANE demonstrates a positive correlation with resting mast cells and monocytes, while displaying a negative correlation with activated memory CD4⁺ T cells. These correlations suggest that ELANE may be associated with GC progression through interactions with the immune microenvironment. MUC15 (Mucin 15, Cell Surface Associated) belongs to the mucin family and is widely distributed on cell surfaces and in secretions. It plays a crucial role in intercellular communication and epithelial barrier protection. MUC15 is abnormally expressed in various tumors, exhibiting both pro - oncogenic and anti - tumor effects³⁸.Research indicates that MUC15 expression is reduced in cervical cancer, while some studies suggest its overexpression in GC^39–41. These discrepancies may arise from molecular and cellular heterogeneity of tumors, differences in patient cohorts (such as clinical characteristics or ethnicity), as well as technical variability among sequencing platforms or detection methods. Acknowledging these inconsistencies highlights the complex regulatory role of MUC15 in tumor initiation and progression and underscores the necessity of large-scale, multi-center studies to further validate its expression patterns and functional mechanisms.This contradicts our findings, and further research is needed to confirm the expression and regulatory mechanisms of MUC15 in GC. Research suggests that MUC15 may regulate specific immune cell populations in the tumor microenvironment⁴². As key regulators of the GC immune microenvironment, ELANE and MUC15 are associated with immune cell infiltration and function in GC, including T cells, mast cells, and monocytes.However, their specific mechanisms and expression patterns require further investigation.Previous studies have shown that APOA1(apolipoprotein A1) and CD36(CD36 molecule)are minimally expressed in GC⁴³, which is consistent with our findings.The protein encoded by the CD36 gene plays a vital role in the cell membrane. It is primarily involved in the uptake and transport of fatty acids and is a key component in cellular energy utilization.The protein encoded by the CD36 gene plays a vital role in the cell membrane. It is primarily involved in the uptake and transport of fatty acids and is a key component in cellular energy utilization⁴⁴.METTL1 (the primary m7G writer) has been shown to directly regulate metabolic reprogramming in tumors^45,46. CD36, a key metabolic gene, is likely a target of this process.Our research demonstrates that CD36 is associated with immune responses, showing a negative correlation with activated memory CD4 + T cells. The gene-mediated lipid uptake not only promotes M2 macrophage polarization but may also directly inhibit T cell activation and function^47,48.The high expression of CD36 in high-risk group suggests that lipid metabolism reprogramming may contribute to immune escape.The APOA1 gene encodes apolipoprotein A1, the primary component of high - density lipoprotein (HDL). In vitro studies demonstrate that APOA1 promotes the proliferation, migration, and invasion of gastric cancer cells⁴⁹.In conclusion, CD36 and APOA1 may jointly promote GC progression and immune escape by regulating lipid metabolism reprogramming and interacting with the immune microenvironment. The protein encoded by the ASCL2 (achaete - scute family bHLH transcription factor 2) gene belongs to the basic helix - loop - helix (BHLH) family and plays a key role in cell development and differentiation.Previous studies have demonstrated that ASCL2 expression is upregulated in GC tissues⁵⁰, which is consistent with our findings. ASCL2 expression levels are inversely correlated with survival rates in GC. Its ectopic overexpression promotes cell proliferation and induces resistance to 5-fluorouracil in GC⁵¹. M7G modifications may enhance the translation efficiency of ASCL2 mRNA, thereby increasing the tumorigenicity and drug resistance of GC cells. GRP(gastrin-releasing peptide)is a neuropeptide that stimulates gastrin secretion. It is overexpressed in various cancers, including lung and breast cancers^52–54. In our study, GRP expression is lower in GC tissues, but higher in high-risk group compared to low-risk group. GRP regulates angiogenesis, cell adhesion, and migration, while m7G serves as the core mechanism driving oncogene translation upregulation^55,56. In our study, the prognostic genes ASCL2 and GRP may play a pivotal role in GC progression by promoting cell proliferation and differentiation.

Furthermore, we conducted a GSEA analysis on the high- and low-risk groups, which revealed significant enrichment in 31 pathways. Notably, the high-risk group exhibited marked enrichment in calcium signaling pathways, adhesion plaques, and neuroactive ligand-receptor interactions. Calcium signaling plays a key role in regulating cell proliferation, migration, and apoptosis^57,58, and its dysregulation can promote GC proliferation and invasion⁵⁹. The adhesion plaque pathway is closely associated with cytoskeletal reorganization and motility, serving as a key driver of tumor metastasis^60,61. It has been identified as a risk factor for GC chemotherapy efficacy and prognosis⁶². Furthermore, neural ligand-receptor interactions have been reported to correlate with tumor immunotherapy responses⁶³. The enrichment of these pathways collectively suggests that the high - risk group has stronger proliferative, migratory, and immunomodulatory capacities, which is consistent with their poor prognosis.Conversely, the low - risk group was significantly enriched in pathways such as DNA replication, aminoacyl - tRNA biosynthesis, and base excision repair.Acquired DNA damage is a key early driver of tumor development. Unrepaired or persistent DNA damage compromises genomic stability and promotes cancer⁶⁴.

Activating the DNA repair pathway may be associated with a lower mutation load, potentially enhancing the response to conventional chemotherapy. The biosynthesis of aminoacyl-tRNA is directly linked to translation efficiency, and its proper functioning forms the basis of cellular metabolic homeostasis⁶⁵.These findings suggest that the low-risk group may be associated with higher genetic stability and more intact cellular functions, resulting in better prognoses. Not only do these results support the reliability of the prognostic risk model, but they also potentially inform the development of personalized treatment strategies tailored to different risk groups.

Differences in infiltrating immune cells were observed between the high-risk and low-risk groups. The significant enrichment of M2 macrophages and monocytes in the high-risk group may be associated with an immunosuppressive tumor microenvironment. M2 macrophages can suppress tumor immunity by secreting inhibitory cytokines such as IL-10 and TGF-β,and promote tumor angiogenesis and tissue remodeling^66,67. Monocytes serve as a bridge between innate and adaptive immune responses and can influence the tumor microenvironment through multiple mechanisms⁶⁸. Resting mast cells are elevated In high-risk group. Recent studies indicate that mast cells may promote tumor progression by modulating the stromal microenvironment and secreting fibrotic factors⁶⁹. These cells contribute to an inhibitory tumor immune microenvironment, which could be associated with poorer prognosis in the high-risk group. The low-risk group exhibits a higher proportion of activated memory CD4⁺ T cells, which play a regulatory role in anti-tumor immunity by facilitating cytotoxic T cell responses and B cell antibody production⁷⁰. Resting NK cells are more abundant in the low-risk group. When activated by antigens, they rapidly transform into cytotoxic NK cells and directly kill tumor cells⁷¹. These findings suggest that the immune microenvironment of the low - risk group may be associated with enhanced anti-tumor potential, consistent with its better clinical outcome. The results suggest that m7G - related genes may be associated with tumor progression and treatment efficacy by regulating immune cell infiltration and function in the GC immune microenvironment. The significantly higher TMB (tumor mutational burden) in the low-risk group compared to the high-risk group suggests that the low-risk group may be more sensitive to immune checkpoint inhibitor therapy. As a predictive biomarker for ICI therapy, higher TMB enhances immune recognition via neoantigen generation. Our study demonstrates a significant association between TMB and the prognostic risk score. Furthermore, the negative correlation between TMB and TIDE scores implies reduced immune evasion potential in high-TMB tumors. Collectively, these findings suggest the risk model reflects tumor immunogenicity and may identify patients likely to benefit from immunotherapy.

In this study, the screened prognostic genes were validated by RT-qPCR using 10 independent clinical gastric cancer samples. Given the relatively small sample size, the statistical power and generalizability of the results may be limited. Additionally, functional validation in gastric cancer cell lines or animal models was not performed, and therefore the precise molecular mechanisms by which these genes influence gastric cancer progression remain unclear. Future studies will further investigate the functions and regulatory roles of these genes through in vitro and in vivo experiments, as well as expand the sample size and perform multi-center cohort validation, in order to more comprehensively assess their prognostic value and potential clinical applications in gastric cancer. This study systematically analyzed key genes in GC associated with m7G modifications and the immune microenvironment using bioinformatics methods, established a reliable prognostic model, and revealed significant differences in molecular pathways and immune infiltration among different risk groups. However, it should be noted that the findings of this study primarily rely on the retrospective analysis of public databases and bioinformatics predictions. Although these results are statistically significant and demonstrate a certain level of reliability, they still require further validation through subsequent experiments. Only through multi - level, multi - center verification can we better promote the clinical translation of these findings, ultimately benefiting GC patients.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

NL and RZ wrote the main manuscript text; JY and YZ and YF proofread the text; JW and ZG sorted the data.

Funding

Partial financial support was received from Hebei Natural Science Foundation(H2025206271).

Data availability

All data supporting the findings of this study are available within the paper and its Supplementary Information.

Declarations

Competing interests

The authors declare no competing interests.

Ethics approval

The studies involving humans were approved by Ethics Committee of the Fourth Hospital of Hebei Medical University. The studies were conducted in accordance with the local legislation and institutional requirements (2024KY234).