The value of immunogenic cell death-related gene model in the prognosis of gastric cancer

Abstract

Immunogenic cell death (ICD) is a form of regulatory cell death that has been obtained growing attention for its role in the treatment and prognosis of tumors. This study aims to further investigate the value of immunogenic cell death-related genes (ICDRGs) in prognostic of GC. We obtained the stomach adenocarcinoma (STAD) dataset from the Gene Expression Omnibus (GEO) database and the Cancer Genome Atlas (TCGA) database, got ICDRGs from the GeneCards database, and utilized LASSO regression to construct a prognostic model. Based on the median risk score, patients were separated into high-risk and low-risk groups and differential gene expression analysis related to prognosis was obtained. We further studied the possible mechanisms, biological characteristics, and pathways of genes in different risk groups. Prognostic analysis, chromosomal localization, and ROC curve plotting were performed. Immunohistochemical analysis was conducted using the Human Protein Atlas (HPA) database. We constructed a prognostic model based on 22 ICDRGs and conducted enrichment analysis to obtain relevant biological characteristics and pathways. An 8-hub gene PPI network model was obtained. ROC curves showed that the occurrence of STAD is associated with the expression of 8 hub genes. Immunohistochemical analysis revealed higher expression levels of genes HSP90AA1, HMGB1, IFNGR1, PDIA3 in STAD tumor tissues compared to normal tissues. This research developed a prognostic model for STAD using ICDRGs and investigated the potential influence of these genes in patients with GC, providing a new direction for evaluating GC prognosis and guiding individualized treatment.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-025-04296-z.

Introduction

Gastric cancer (GC) is the third leading cause of cancer-related deaths worldwide[1]. The prognosis is usually poor due to its advanced stage diagnosis[2]. While systemic treatments could improve the situation of patients with advanced or metastatic disease, raise survival rates and improve quality of life, the 5-year survival rate remains low[3].

Immunogenic cell death (ICD) refers to a kind of cancer cell death that can be induced by certain chemotherapeutic agents, oncolytic viruses, physical–chemical therapies, photodynamic therapy and radiotherapy in tumors[4, 5]. ICD is accompanied by the exposure and release of numerous damage-associated molecular patterns (DAMPs), including surface-exposed calreticulin (CALR), as well as secreted ATP, annexin A1 (ANXA1), type I interferons, and high-mobility group box 1 (HMGB1). These molecules are conducive to the recruitment and activation of antigen-presenting cells [4]. Existing research indicates that ICD plays a crucial role in tumor immunotherapy. Immunogenic chemotherapeutic drugs can induce ICD, and exert anti-tumor effects by mediating the release of DAMPs [6]. Moreover, the hallmarks of ICD have been analyzed as biomarkers for predicting the prognosis and survival rates of cancer patients[7]. Previous studies have confirmed that models based on ICD characteristics are instrumental in clarifying the tumor microenvironment (TME) features of stomach adenocarcinoma (STAD), holding significant clinical value for assessing the prognosis and immunotherapy response of STAD patients. Additionally, TGM2 plays a vital role in the progression of STAD[8]. However, there are certain limitations in terms of gene coverage and dataset validation. The primary objective of this study is to conduct a more comprehensive exploration of the impact of ICD on the treatment and prognosis of gastric cancer.

This research is designed to analyze the influence of immunogenic cell death-related genes (ICDRGs) on GC and to establish a predictive model. We obtained ICDRGs from the GeneCards database, applied LASSO regression to establish a model based on 22 ICD prognosis-related genes significantly associated with the gastric cancer patients. We then further explored the biological characteristics and potential mechanisms of these genes. Furthermore, we using the Human Protein Atlas (HPA) database to perform immunohistochemical analysis to validate key findings to prove their clinical relevance. This study provides the foundation for further understanding the prognosis of gastric cancer and potential therapeutic.

Material and methods

Data download

We downloaded the expression matrix of the STAD dataset (TCGA-STAD) from The Cancer Genome Atlas (TCGA) (https://portal.gdc.cancer.gov/) using the R package TCGAbiolinks[9]. The dataset included 375 STAD samples (Cancer group) and 32 adjacent normal samples (Normal group). Every sample was accompanied by its corresponding clinical information and was adjusted to the Fragments Per Kilobase per Million (FPKM) standard. All samples were included in this study, and the corresponding clinical data were obtained from the UCSC Xena database [10] (http://genome.ucsc.edu). The count sequencing data of the TCGA-STAD dataset were normalized using the R package limma [11].

Additionally, we used the R package GEOquery [12] to download STAD-related datasets GSE62254 [13] and GSE84437 [14] from the Gene Expression Omnibus (GEO) database [15]. GSE62254, derived from Homo Sapiens, uses the GPL570 (HG-U133_Plus_2) Affymetrix Human Genome U133 Plus 2.0 Array data platform and contains microarray gene expression profiles of 300 STAD patient samples, all of which were included in this study. GSE84437, also derived from Homo Sapiens, uses the data platform GPL6947 (Illumina HumanHT-12 V3.0 expression beadchip) and contains microarray gene expression profiles of 433 stomach adenocarcinoma patient samples, all of which were included in this study. Probe name annotations in the datasets were accorded to the GPL platform file for the chip.

The GeneCards database[16] provides extensive details about human genes. We collected ICDRGs from the GeneCards database. Using "immunogenic cell death" as a search keyword, we identified a total of 57 ICDRGs. Additionally, we retrieved 34 ICDRGs from literature searches on the PubMed website. After merging and removing duplicates, and excluding genes missing during the probe conversion process, we obtained a total of 77 ICDRGs (Table S1).

Construction of prognostic model for immunogenic cell death-related genes

To establish a prognostic model for ICDRGs in STAD, we performed least absolute shrinkage and selection operator (LASSO) regression with tenfold cross-validation with a significance level of p-value 0.05 and seed value of 2021. In order to prevent overfitting, we ran 1000 cycles for each cycle. LASSO regression, used in constructing prognostic models, bases on linear regression and through adding a penalty term (lambda × absolute value of the slope) to reduce overfitting and enhance the model's generalizability. The outcomes of the LASSO regression were graphically represented. We then further elucidated each sample in the LASSO regression prognostic model through risk factor plots according to the categorization derived from risk scores, the survival outcomes and the expression of immunogenic cell death prognosis-related genes (ICD prognosis-related genes) within each group.The risk factor diagram consists of three parts: risk grouping, survival outcomes, and a heatmap. In the risk grouping, samples from the dataset are categorized based on the median of the Risk Score, which is predicted by the LASSO regression prognostic model. The survival outcomes are illustrated by a dot plot, showing the survival time and outcomes of clinical samples from the TCGA-STAD dataset. The heatmap visualizes the expression of the ICD prognosis-related genes in the samples of our LASSO regression prognostic model.

Analysis of differentially expressed genes related to prognosis in high and low-risk groups of STAD

To identify the potential mechanisms, biological features and pathways of ICD prognosis-related genes in high and low-risk groups of STAD, we used the R package limma to standardize and group the TCGA-STAD, GSE62254 and GSE84437 datasets firstly. We divided the samples into STAD high-risk group (Group: High) and STAD low-risk group (Group: Low) based on the Risk score from the LASSO model constructed with ICD prognosis-related genes, using the median. We then obtained differentially expressed genes (DEGs) among the three STAD datasets by conduct differential analysis on the processed expression profile data. Genes with logFC > 0 and P < 0.05 were considered up-regulated DEGs, while genes with logFC < 0 and P < 0.05 were considered down-regulated DEGs. We utilized the ggplot2 package to draw volcano plots and used the pheatmap package in R for the creation of heatmaps to display the results of the differential analysis. Additionally, we also plotted group comparison diagram to illustrate the expression differences of the ICD prognosis-related genes across different groups within the three STAD datasets.

GO and KEGG analysis

Gene Ontology (GO) analysis [17] is a commonly used method for large-scale functional enrichment studies, encompassing biological processes (BP), molecular functions (MF), and cellular components (CC). The Kyoto Encyclopedia of Genes and Genomes (KEGG) [18] is a widely used database storing information about genomes, biological pathways, diseases, and drugs, etc. We utilized the R package clusterProfiler [19] to perform GO annotation analysis on ICD prognosis-related genes, with P.adj < 0.05 and false discovery rate (FDR) values (q.value) < 0.05 being considered statistically significant. The P-value correction method used was Benjamini-Hochberg (BH).

Gene set enrichment analysis (GSEA)

Gene Set Enrichment Analysis (GSEA) [20] is a tool employed to evaluate the distribution trend of genes within a predefined gene set, which are ranked based on their correlation with a phenotype, in order to enable the assessment of their contribution to that phenotype. In this research, we initially divided the genes from the TCGA-STAD, GSE62254, and GSE84437 datasets into two groups based on high and low phenotypic correlation, using phenotypic correlation ranking. Subsequently, we conducted an enrichment analysis on all genes in the two groups with high and low phenotypic correlations using the R package clusterProfiler. The parameters used in this GSEA were as follows: seed number 2020, number of permutations 1000, minimum gene set size was set to 10, maximum gene set size was set to 500, and Benjamini-Hochberg (BH) was the P-value adjustment method.We obtained the c2.cp.v7.2.symbols gene set from the Molecular Signatures Database (MSigDB) [21], the screening criteria of significant enrichment set as P.adj < 0.05 and FDR value (q.value) < 0.25.

Gene set variation analysis (GSVA)

Gene Set Variation Analysis (GSVA) [22] is a non-parametric, unsupervised analysis method that evaluates the results of gene set enrichment analysis for the chip-based nuclear transcriptome, which helps assess whether different pathways are enriched in different samples. Utilizing the gene set 'h.all.v7.4.symbols.gmt' acquired from the MSigDB database, we performed GSVA at the gene expression level to calculate the functional enrichment differences between the two tissues.

Protein–protein Interaction (PPI) Network

Protein–protein interaction (PPI) networks are constituted by single proteins interacting with each other. The STRING database [23] is used to search for interactions between known and predicted proteins. We constructed a protein–protein interaction network related to differentially expressed genes (DEGs) in this study, utilizing the STRING database. This network was built using ICD prognosis-related genes that were obtained through screening, with the required minimum interaction score is 0.400. Tightly connected local areas within the PPI network may represent molecular complexes, possessing specific biological functions. The PPI network model was visualized using the Cytoscape software (version 3.9.1) [24]. Additionally, we utilized 5 algorithms from the CytoHubba [25] plugin: the closeness, the degree, edge percolated component (EPC), maximal clique centrality (MCC), and maximum neighborhood component (MNC). In the PPI network, we first calculated the scores of the ICD prognosis-related genes, and then arranged the ICD prognosis-related genes in order based on these scores. Finally, the intersection of the top 10 genes from the 5 algorithms was taken, and a Venn diagram was plotted. This process identified hub genes related to immunogenic cell death in STAD.

Prognostic clinical correlation analysis

In order to investigate the clinical prognostic value of hub genes for STAD, we conducted a univariate Cox regression analysis on the expression of hub genes in the TCGA-STAD dataset and clinical variables. We selected the factors with P < 0.1 into a multivariate Cox regression analysis. Based on the outcomes of the multivariate Cox regression analysis, we further constructed a nomogram to predict the 1-year, 3-year, and 5-year survival rates of patients with STAD. A nomogram is used to graphically represent the functional relationship between multiple independent variables on a Cartesian coordinate system, using a cluster of non-intersecting lines. Based on the multivariate Cox regression, it represents the situation of each variable in a multivariate regression model by setting a certain scale to score the variables, finally calculating a total score to estimate the likelihood of an event occurring. Ultimately, we used a calibration curve to assess the accuracy and discriminative capability of the nomogram. The calibration curve assesses the fitting of actual probabilities to model-predicted probabilities under different conditions, to evaluate the predictive effect of the model on actual outcomes. It is primarily used for the fit analysis of models established by the Cox regression method in relation to actual situations. The calibration curve has the model-predicted survival probability on the horizontal axis and the actual survival probability shown by the data on the vertical axis. Different colors of lines and points represent the model's predictions at different time points. The closer a colored line is to the ideal grey line, the better the prediction at that time point. The "rms" package in R is used to construct the nomogram and calibration curve. Decision curve analysis (DCA) is used to assess the utility of clinical prediction models, diagnostic tests and molecular markers. We exploited the ggDCA package [26] in R to construct DCA curves, which facilitated our assessment of the nomogram model's predictive capacity for 1-year, 3-year, and 5-year survival outcomes in STAD patients. The effectiveness of the model could be evaluated by observing the range of x-values for which the model's line consistently remains above both the 'All positive' and 'All negative' lines. A larger x-value range indicates superior performance of the model. Prognosis-associated Kaplan–Meier (KM) curves analysis is a method utilized to analyze and judge patient survival time based on data, investigating the relationship and its degree between survival time, outcome and numerous influencing factors. This method is also known as survival rate analysis or simply survival analysis. As proposed by Kaplan and Meier, this method is commonly referred to as the Kaplan–Meier method or KM method. The KM method calculates the probability of patients surviving a certain period and then surviving the next period (i.e., survival probability). Then, these individual survival probabilities are multiplied to estimate the survival rate for the corresponding period to estimates survival curves. We individually plotted KM curves for the hub genes related to the subtypes of clinical variables, using P = 0.05 as the threshold to identify genes and corresponding clinical variable subtypes with statistical significance.

ROC curve and chromosome localization analysis

Receiver Operating Characteristic curve (ROC) [27] serves as a type of analytical tool in the form of a coordinate diagram, employed to determine the best model, abandon suboptimal models, or identify the ideal cutoff value. The ROC curve is a comprehensive indicator reflecting the sensitivity and specificity of continuous variables. It demonstrates the interrelationship between sensitivity and specificity through graphic, with the area under the ROC curve generally ranging between 0.5 and 1. The closer the Area Under Curve (AUC) is to 1, the better the diagnostic performance. AUC values closer to 1 indicate better diagnostic efficacy. AUC values ranging from 0.5 to 0.7 imply a low level of accuracy, those locating between 0.7 and 0.9 suggest moderate accuracy, while values surpassing 0.9 are indicative of high accuracy. We utilized the survivalROC package in R to generate the ROC curves of the hub genes in the TCGA-STAD dataset. The AUC was calculated to evaluate the diagnostic effectiveness of the hub genes' expression on the survival of STAD patients. To analyze the chromosomal localization of hub genes in 24 pairs of chromosomes, we first determined the start and end sequences of hub genes using the UCSC database (http://genome.ucsc.edu/). Subsequently, we used the RCircos package [28] in R to plot the chromosome localization map.

Immunohistochemistry analysis

Utilizing the Human Protein Atlas (HPA) database [29] (www.proteinatlas.org/), we performed immunohistochemistry analyses to investigate the expression of the identified hub genes, and showcasing the immunohistochemistry analysis results from human cell samples in the database.

Statistical analysis

R software (Version 4.1.2) was utilized to process and analysis the data in this study. Continuous variables showed as mean ± standard deviation. The Wilcoxon rank sum test was employed to analyze continuous variables between two groups, and statistical significance of normally distributed variables was estimated by independent Student's t-test. The Kruskal–Wallis test method was utilized to compare involving three groups or more. Chi-square test or Fisher's exact test was applied to compare and analyze the statistical significance between two groups of categorical variables. The survival analysis, as well as univariate and multivariate Cox analyses, were performed using the survival package in R. LASSO regression analysis was performed utilizing the glmnet package [30], Kaplan–Meier survival curves were applied to display survival differences, and the log-rank test was utilized to evaluate the significance of differences in survival time between two groups of patients. Unless otherwise specified, we using Spearman correlation analysis to determine the correlation coefficients between different molecules, and all P-values were two-tailed with a significance level set at P < 0.05.

Results

Technical roadmap

The study flow diagram is shown in Fig. 1.

Fig. 1.

Technical Roadmap: the flowchart of this study. TCGA, The Cancer Genome Atlas; STAD, Stomach Adenocarcinoma; ICD-related genes, Immunogenic cell death-related genes; LASSO, least absolute shrinkage and selection operator; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; GSVA, Gene Set Variation Analysis; GSEA, gene set enrichment analysis; PPI network, Protein–protein interaction network; KM, Kaplan–Meier; ROC, Receiver Operating Characteristic curve; IHC, Immunohistochemistry

Construct a prognostic model based on immunogenic cell death-related genes

We utilized LASSO regression analysis for the creation of a prognostic model (Fig. 2A) comprised of 22 genes (ABCA1, ABCB1, BAK1, CD274, CD4, CDK4, CFLAR, EGR1, HMGB1, HSP90AA1, ICAM3, IFNA2, IFNB1, IFNGR1, IL17RA, ITGA4, LY96, MLKL, NT5E, PDCD1, PDIA3, SP1) (see Tables 1, 2), aiming to ascertain the prognostic value of 77 ICDRGs (see Table S1) within the TCGA-STAD dataset. We also visualized the results of the LASSO regression and obtained the LASSO variable trajectory plot (Fig. 2B). Subsequently, we visualized the grouping of samples in the constructed LASSO regression prognostic model using a risk factor diagram (Fig. 2C).

Fig. 2.

Construction of the prognostic model for ICD-related genes. A. LASSO regression prognostic model plot for ICD-related genes. B-C. LASSO regression model variable trajectory plot (B) and risk factor plot (C). In the LASSO regression prognostic model plot (A), the vertical axis represents the likelihood deviation of LASSO regression, with the x-axis below showing log(λ) values representing the situation after taking the logarithm of the penalty term’s lambda coefficient in the LASSO regression; the numbers on the upper x-axis represent the number of variables with non-zero coefficients corresponding to each lambda

Table 1.

List of gene symbol of prognosis-related genes

Gene symbol
ABCA1	ABCB1	BAK1	CD274	CD4
CDK4	CFLAR	EGR1	HMGB1	HSP90AA1
ICAM3	IFNA2	IFNB1	IFNGR1	IL17RA
ITGA4	LY96	MLKL	NT5E	PDCD1
PDIA3	SP1

Table 2.

List of description and expression difference of ICD prognosis-related genes

Gene	Description	log2FoldChange	pvalue	padj
ABCA1	ATP Binding Cassette Subfamily A Member 1	0.367701476	7.07231E-05	0.000557243
ABCB1	ATP Binding Cassette Subfamily B Member 1	-0.2294169	0.07765969	0.157492
BAK1	BCL2 Antagonist/Killer 1	0.114065	0.08813879	0.1733198
CD274	CD274 Molecule	0.605293437	2.10965E-05	0.000203902
CD4	CD4 Molecule	-0.01443891	0.9014487	0.9422749
CDK4	Cyclin Dependent Kinase 4	0.383731317	3.83849E-08	1.00036E-06
CFLAR	CASP8 And FADD Like Apoptosis Regulator	-0.112302	0.05499674	0.1211734
EGR1	Early Growth Response 1	-0.516358009	0.00015899	0.001079716
HMGB1	High Mobility Group Box 1	0.154594	0.002909168	0.01191426
HSP90AA1	Heat Shock Protein 90 Alpha Family Class A Member 1	0.757977486	3.93419E-33	7.7E-29
ICAM3	Intercellular Adhesion Molecule 3	-0.4928291	0.001694845	0.007695399
IFNA2	Interferon Alpha 2	-0.2150533	0.2316951	0.362402
IFNB1	Interferon Beta 1	-0.4550724	0.06939922	0.1445678
IFNGR1	Interferon Gamma Receptor 1	0.411534454	1.10186E-08	3.5387E-07
IL17RA	Interleukin 17 Receptor A	0.0419878	0.4139671	0.5545095
ITGA4	Integrin Subunit Alpha 4	0.2153897	0.06891411	0.1438553
LY96	Lymphocyte Antigen 96	-0.1683696	0.1456656	0.2554172
MLKL	Mixed Lineage Kinase Domain Like Pseudokinase	0.2081422	0.002391128	0.01020034
NT5E	5'-Nucleotidase Ecto	-0.06261504	0.6435999	0.7528411
PDCD1	Programmed Cell Death 1	0.01019752	0.9461557	0.9706043
PDIA3	Protein Disulfide Isomerase Family A Member 3	0.441254558	4.47028E-14	9.11378E-12
SP1	Sp1 Transcription Factor	0.185461317	5.80502E-05	0.000470849

ICD, immunogenic cell death

Analysis of differentially expressed genes related to prognosis in high and low-risk groups of STAD

In the TCGA-STAD dataset, a total of 19,572 differentially expressed genes were obtained, with 6,940 genes satisfying the criteria of |logFC|> 0 and P.adj < 0.05. Among these 4,595 genes were high expression in the high-risk group of gastric cancer (logFC positive, low expression in the low-risk, indicating upregulated genes), while 2,345 genes were low expression in the high-risk group (logFC negative, high expression in the low-risk, indicating downregulated genes). A volcano plot was generated to illustrate the differential expression analysis results (Fig. 3A). In the GSE62254 dataset, we got 21,655 differentially expressed genes, with 6,711 genes satisfying the criteria of |logFC|> 0 and P.adj < 0.05. Within these, 2,986 genes were logFC positive (upregulated genes), while 3,725 genes were logFC negative (indicating downregulated genes). A volcano plot was generated to illustrate the differential expression analysis results (Fig. 3B). In the GSE84437 dataset, a total of 25,124 differentially expressed genes were identified. Among them, 10,613 genes met the threshold of |logFC|> 0 and P.adj < 0.05. From these, 5,587 genes were logFC positive (upregulated genes), while 5,026 genes were logFC negative (indicating downregulated genes). A volcano plot was generated to illustrate the differential expression analysis results (Fig. 3C).

Fig. 3.

Differential Gene Analysis in High and Low-Risk Groups of STAD. A-C. Volcano plots of differentially expressed genes between the high-risk (group: High) and low-risk (group: Low) groups of stomach adenocarcinoma in datasets TCGA-STAD (A), GSE62254 (B) and GSE84437 (C). D-F. Complex numerical heatmap of immunogenic cell death-related genes in datasets TCGA-STAD (D), GSE62254 (E) and GSE84437 (F)

Furthermore, the differential expression profiles of the 22 ICD prognosis-related genes were analyzed across different groups in the TCGA-STAD dataset (Fig. 3D), GSE62254 (Fig. 3E) and GSE84437 (Fig. 3F). Heatmaps were generated using the pheatmap package in R to visualize the results, and group comparison plots were created (Fig. 4A-C) to demonstrate the differential expression analysis results of the 22 ICD prognosis-related genes. Notably, in the TCGA-STAD dataset (Fig. 4A), the expression differences of genes EGR1, HMGB1, HSP90AA1, ICAM3, IFNGR1, and PDIA3 between high and low risk groups in STAD are statistically extremely significant (P value < 0.001); the expression differences of genes CD274, IFNA2, LY96, and SP1 between high and low risk groups in STAD are highly statistically significant (P value < 0.01); the expression difference of gene CFLAR between high and low risk groups in STAD is statistically significant (P value < 0.05); the expression differences of genes ABCA1, ABCB1, BAK1, CD4, CDK4, IFNB1, IL17RA, ITGA4, MLKL, NT5E, and PDCD1 between high and low risk groups in STAD are not statistically significant (P value ≥ 0.05).

Fig. 4.

Group Comparison of ICD Prognosis-Related Genes in High and Low-Risk Groups of STAD. A-C. Group comparison plots of ICD prognosis-related genes in the high and low-risk groups of datasets TCGA-STAD (A), GSE62254 (B) and GSE84437 (C). "ns" represents P value ≥ 0.05, indicating no statistical significance; "*" represents P value < 0.05, indicating statistical significance; "**" represents P value < 0.01, indicating high statistical significance; "***" represents P value < 0.001, indicating extreme statistical significance

In the dataset GSE62254 (Fig. 4B), the expression differences of genes ABCA1, CD4, IFNA2, MLKL and PDIA3 in the high and low-risk groups of STAD are extremely statistically significant (P value < 0.001); genes CFLAR and LY96 exhibit highly significant expression differences in the different risk groups of STAD (P value < 0.01); expression differences of genes ABCB1, BAK1 and CD274 in different risk groups show statistically significant in STAD (P value < 0.05); while genes CDK4, EGR1, HMGB1, HSP90AA1, ICAM3, IFNB1, IFNGR1, IL17RA, ITGA4, NT5E, PDCD1 and SP1 do not show statistically significant expression differences in different groups of STAD (P value ≥ 0.05).

In the dataset GSE84437 (Fig. 4C), genes BAK1, ICAM3, NT5E and SP1 exhibit extremely statistically notable expression disparities in the different groups of STAD (P value < 0.001); gene LY96 shows highly significant expression differences in different groups of STAD (P value < 0.01); genes ABCA1, CD274, CD4, HSP90AA1, IL17RA, ITGA4, MLKL and PDCD1 demonstrate statistically significant expression differences in different groups of STAD (P value < 0.05); while genes ABCB1, CDK4, CFLAR, EGR1, HMGB1, IFNA2, IFNB1, IFNGR1 and PDIA3 do not show statistically significant expression differences in different groups of STAD (P value ≥ 0.05).

GO and KEGG

To analyze and the relationship between the biological processes, molecular functions, cellular components, biological pathways and STAD of the 22 ICD prognosis-related genes (see Tables 1 and 2), we first conducted a Gene Ontology (GO) gene function enrichment analysis on the ICD prognosis-related genes (Table 3), with the screening criteria for enrichment being P.adj < 0.05 and FDR value (q.value) < 0.05. The results indicate that the 22 ICD prognosis-related genes (see Table 1) are primarily enriched in biological processes such as lymphocyte differentiation and cytokine secretion, as well as molecular functions like cytokine binding and coreceptor activity in STAD. We have presented the results of the GO function enrichment analysis through a bar graph (Fig. 5A). Subsequently, we applied KEGG enrichment analysis on the 22 ICD prognosis-related genes (Table 4), revealing significant enrichment in 30 KEGG pathways including necroptosis and human cytomegalovirus infection. A bubble graph (Fig. 5B) is used to illustrate the findings from the KEGG pathway enrichment analysis. Additionally, we used network graphs to show the GO and KEGG enrichment analysis results (Fig. 5C-D). Furthermore, a combined logFC GO and KEGG enrichment analysis was performed on the 22 ICD prognosis-related genes. Based on the enrichment analysis, we calculated the z-score for each molecule by providing the logFC values of these 22 ICD prognosis-related genes from the differential analysis results between the high-risk and low-risk groups in the TCGA-STAD dataset for STAD. The results of the combined logFC GO enrichment analysis are revealed in a chord diagram (Fig. 5E), while the results of the combined logFC KEGG enrichment analysis are shown in a circle plot (Fig. 5F), which shows necroptosis (ID, KEGG: hsa04217) as a significantly upregulated biological process.

Table 3.

GO enrichment analysis results of ICD prognosis-related genes

Ontology	ID	Description	GeneRatio	BgRatio	pvalue	p.adjust	qvalue
BP	GO:0007159	leukocyte cell–cell adhesion	8/22	337/18670	2.66e-09	3.45e-06	1.77e-06
BP	GO:0030098	lymphocyte differentiation	7/22	353/18670	1.09e-07	4.70e-05	2.42e-05
BP	GO:0050663	cytokine secretion	6/22	240/18670	2.66e-07	6.94e-05	3.56e-05
BP	GO:0071222	cellular response to lipopolysaccharide	5/22	205/18670	3.44e-06	3.19e-04	1.64e-04
MF	GO:0019955	cytokine binding	4/22	128/17697	1.73e-05	0.001	6.18e-04
MF	GO:0015026	coreceptor activity	3/22	44/17697	2.14e-05	0.001	6.18e-04
MF	GO:0004896	cytokine receptor activity	3/22	96/17697	2.21e-04	0.005	0.002
MF	GO:0005125	cytokine activity	3/22	220/17697	0.002	0.028	0.013

GO, Gene Ontology. ICD, immunogenic cell death. BP, biological process. MF, molecular function

Fig. 5.

GO and KEGG. A. Bar graph illustrating the GO enrichment analysis results of ICD prognosis-related genes. B. Bubble graph illustrating the KEGG enrichment analysis results of ICD prognosis-related genes. C. Circular network diagram showing the GO enrichment analysis results of ICD prognosis-related genes. D. Divergent network diagram displaying the KEGG enrichment analysis results of ICD prognosis-related genes. E. Chord diagram presenting the combined logFC GO enrichment analysis results of ICD prognosis-related genes. F. Circle plot showing the combined logFC KEGG enrichment analysis results of ICD prognosis-related genes. In the bubble graph (B), the y-axis represents GO terms, with bubble colors indicating activation or inhibition of the GO terms (red for activation, blue for inhibition). In the network graphs (C-D), red dots represent specific genes, while blue circles represent specific pathways. In the circle plot (F), red dots denote upregulated genes (logFC > 0), and blue dots represent downregulated genes (logFC < 0)

Table 4.

KEGG enrichment analysis results of ICD prognosis-related genes

Ontology	ID	Description	GeneRatio	BgRatio	pvalue	p.adjust	qvalue
KEGG	hsa04217	Necroptosis	7/22	159/8076	1.34e-07	1.39e-05	7.88e-06
KEGG	hsa05163	Human cytomegalovirus infection	6/22	225/8076	2.25e-05	0.001	6.62e-04
KEGG	hsa04514	Cell adhesion molecules	5/22	149/8076	4.09e-05	0.001	7.74e-04
KEGG	hsa05160	Hepatitis C	5/22	157/8076	5.25e-05	0.001	7.74e-04
KEGG	hsa05164	Influenza A	5/22	171/8076	7.89e-05	0.002	8.51e-04
KEGG	hsa04060	Cytokine-cytokine receptor interaction	5/22	295/8076	9.92e-04	0.009	0.005
KEGG	hsa05165	Human papillomavirus infection	5/22	331/8076	0.002	0.013	0.008
KEGG	hsa04151	PI3K-Akt signaling pathway	5/22	354/8076	0.002	0.016	0.009

KEGG, Kyoto Encyclopedia of Genes and Genomes. ICD, immunogenic cell death

GSEA

To determine the impact of gene expression levels on the differences between high and low risk groups in STAD, we separately analyzed the relationships among all gene expressions and the biological processes involved, the affected cellular components, and the molecular functions exerted in the TCGA-STAD, GSE62254, and GSE84437 datasets through GSEA enrichment analysis. A P.adj < 0.05 and FDR value (q.value) < 0.25 were used as the criteria for significant enrichment selection. We found that in the TCGA-STAD dataset, genes are obviously enriched in pathways such as transcriptional regulation by TP53, gene and protein expression by JAK STAT signaling after interleukin 12 stimulation, signaling by NOTCH4 and dectin 1 mediated noncanonical NF-кB signaling (Table 5). In the GSE62254 dataset, there is a significant enrichment of genes in certain pathways, like the IL23 pathway, TNFR2 non-canonical NF-кB pathway, JAK STAT signaling pathway and NFAT TF pathway (Table 6). Similarly, in the GSE84437 dataset, genes are significantly enriched in pathways including NOTCH1 regulation of human endothelial cell calcification, WNT signaling, TGFBETA receptor signaling and IL6 pathway (Table 7). We selected representative pathways in the datasets TCGA-STAD, GSE62254 and GSE84437, respectively, and then presented them in ridge plots (Fig. 6A, Fig. S1A, F) and pathway diagrams (Fig. 6B-E, Fig. S1B-E, G-J).

Table 5.

GSEA analysis of dataset TCGA-STAD

Description	setSize	enrichmentScore	NES	pvalue	p.adjust
cell cycle checkpoints	292	0.752748118	2.527524811	0.00265252	0.022525933
DNA_replication	128	0.775482348	2.397136336	0.002293578	0.021420214
mitotic metaphase and anaphase	236	0.726493972	2.385819067	0.002506266	0.021666303
mitotic spindle checkpoint	111	0.789914454	2.380424835	0.002298851	0.021420214
M phase	414	0.683532759	2.379915265	0.003003003	0.024392994
retinonlastoma gene in cancer	87	0.813785608	2.375351366	0.002304147	0.021420214
mitotic prometephase	202	0.728350286	2.366512959	0.002415459	0.021420214
G2_M_checkpoints	168	0.742063722	2.36410018	0.002347418	0.021420214
S_phase	162	0.746914236	2.352976479	0.00245098	0.021446078
separation of sister chromatids	191	0.724115432	2.341066092	0.002386635	0.021420214
resolution of sister chromatid cohesion	126	0.759453986	2.336240498	0.002298851	0.021420214
transcriptional regulation by TP53	358	0.574872804	1.980576741	0.002832861	0.02349202
gene and protein expression by JAK STAT signaling after interleukin 12 stimulation	38	0.67994323	1.731004985	0.002053388	0.021420214
signaling by NOTCH4	82	0.547204902	1.588865317	0.002232143	0.021420214
dectin 1 mediated noncanonical NF-κB signaling	62	0.549020742	1.522077859	0.00856531	0.049599608

GSEA, Gene Set Enrichment Analysis. TCGA, The cancer genome atlas. STAD, Stomach adenocarcinoma

Table 6.

GSEA analysis of dataset GSE62254

Description	setSize	enrichmentScore	NES	pvalue	p.adjust
allograft rejection	83	0.753226973	2.528609148	0.001455604	0.016590974
immunoregulatory interactions between a lymphoid and a non lymphoid cell	124	0.706724503	2.514126824	0.001364256	0.016590974
leishmania infection	67	0.770707018	2.489571357	0.001501502	0.016590974
chemokine receptors bind chemokines	54	0.795296677	2.465065674	0.001572327	0.016590974
antigen processing and presentation	74	0.744167244	2.451169782	0.001474926	0.016590974
retinoblastoma gene in cancer	86	0.727559161	2.448366679	0.001459854	0.016590974
cell cycle checkpoints	242	0.643340193	2.427356567	0.00122549	0.016590974
graft versus host disease	35	0.84624879	2.417220538	0.001652893	0.016590974
type I diabetes mellitus	39	0.816044777	2.37490739	0.001642036	0.016590974
G2 M checkpoints	125	0.663599932	2.366971735	0.00135318	0.016590974
cell cycle mitot5ic	474	0.596505325	2.362589492	0.001124859	0.016590974
IL23 pathway	37	0.756019262	2.186585817	0.001633987	0.016590974
TNFR2 non canonical NF- κB pathway	95	0.608718782	2.086140565	0.001416431	0.016590974
JAK STAT signaling pathway	152	0.486874702	1.759864522	0.001321004	0.016590974
NFAT TFpathway	45	0.556589808	1.654630126	0.004983389	0.03135169

GSEA, Gene Set Enrichment Analysis

Table 7.

GSEA analysis of dataset GSE84437

Description	setSize	enrichmentScore	NES	pvalue	p.adjust
core matrisome	242	0.633688041	2.642869916	0.001801802	0.070208214
ecm glycoproteins	169	0.61949633	2.485734599	0.001805054	0.070208214
elastic fiber formation	44	0.730966843	2.353722351	0.001883239	0.070208214
collagens	40	0.731405404	2.310564983	0.001869159	0.070208214
collagen chain trimerization	40	0.731405404	2.310564983	0.001869159	0.070208214
extracellular matrix grganization	283	0.534906588	2.261530548	0.00173913	0.070208214
degradation of the extracellular matrix	130	0.564797302	2.196990634	0.001851852	0.070208214
ecm proteoglycans	74	0.623696563	2.195708174	0.001821494	0.070208214
molecules associated with elastic fibers	37	0.71337589	2.195256904	0.001904762	0.070208214
collagen degradation	58	0.647819784	2.172192363	0.001865672	0.070208214
assembly of collagen fibrils and other multimetic structures	56	0.648636809	2.162788252	0.001851852	0.070208214
NOTCH1 regulation of human endothelial cell calcification	17	0.785963334	2.039053985	0.001879699	0.070208214
WNT signaling	112	0.477930134	1.808852074	0.001831502	0.070208214
TGFBETA receptor signaling	53	0.509569891	1.68058577	0.003731343	0.085605114
IL6 pathway	21	0.558246958	1.529250929	0.035916824	0.254074839

GSEA, Gene Set Enrichment Analysis

Fig. 6.

GSEA of the TCGA-STAD dataset. A. GSEA of the TCGA-STAD dataset mainly focused on 4 biological features. B-E. Genes in the TCGA-STAD dataset significantly enriched in pathways including transcriptional regulation by TP53 (B), gene and protein expression by JAK STAT signaling after interleukin 12 stimulation (C), signaling by NOTCH4 (D), and dectin 1 mediated noncanonical NF-κB signaling (E). The criteria for significant enrichment in GSEA were set as P.adj < 0.05 and FDR value (q.value) < 0.25

GSVA

In order to investigate the differences in hallmark gene sets between high-risk (group: High) and low-risk (group: Low) in STAD, we executed a Gene Set Variation Analysis (GSVA) on the expression of ICD prognosis-associated genes in the TCGA-STAD, GSE62254, and GSE84437 datasets. The GSVA results from the TCGA-STAD dataset revealed that 25 hallmark gene sets such as Hypoxia and Cholesterol Homeostasis exhibited differences between different risk group in STAD (Fig. 7A-B, Table 8). The GSVA results from the GSE62254 dataset showed that 28 hallmark gene sets including TNFA signaling via NF-κB and Mitotic spindle displayed differences betweendifferent risk group in STAD (Fig. S2A-C, Table 9). Similarly, the GSVA results from the GSE84437 dataset indicated that 24 hallmark gene sets like Hypoxia and WNT β-catenin signaling showed differences between different risk group in STAD (Fig. S2B-D, Table 10).

Fig. 7.

GSVA of TCGA-STAD Dataset. A. Enriched pathways in the comparison of high and low-risk groups in gastric cancer from GSVA in the TCGA-STAD dataset; B. Heatmap of functional scores from GSVA in the TCGA-STAD dataset, where the horizontal axis represents samples, the vertical axis represents biological functions, and node colors indicate activation or inhibition of corresponding functions (blue for inhibition, red for activation). * denotes P value < 0.05, indicating statistical significance; ** denotes P value < 0.01, indicating high statistical significance; *** denotes P value < 0.001, indicating extreme statistical significance

Table 8.

GSVA analysis of dataset TCGA-STAD

ontology	logFC	AveExpr	t	P.Value	adj.P.Val
HALLMARK_MTORC1_SIGNALING	-0.291002466	-0.062003754	-10.63575554	1.80E-23	9.00E-22
HALLMARK_G2M_CHECKPOINT	-0.354273574	-0.032396228	-9.957017606	4.83E-21	1.21E-19
HALLMARK_UNFOLDED_PROTEIN_RESPONSE	-0.249845877	-0.065367368	-9.371426023	5.04E-19	8.41E-18
HALLMARK_MYC_TARGETS_V1	-0.32057918	-0.043471587	-8.58540764	1.95E-16	2.44E-15
HALLMARK_E2F_TARGETS	-0.338096294	-0.024536438	-8.377072357	8.93E-16	8.93E-15
HALLMARK_MITOTIC_SPINDLE	-0.212273798	-0.054922749	-7.281376104	1.73E-12	1.44E-11
HALLMARK_SPERMATOGENESIS	-0.145186604	-0.048299842	-6.206446657	1.34E-09	9.56E-09
HALLMARK_PROTEIN_SECRETION	-0.193996749	-0.064225845	-6.161893003	1.73E-09	1.08E-08
HALLMARK_MYC_TARGETS_V2	-0.217210746	-0.023584113	-5.019337837	7.77E-07	4.32E-06
HALLMARK_PI3K_AKT_MTOR_SIGNALING	-0.121936921	-0.088146756	-4.941798388	1.13E-06	5.67E-06
HALLMARK_GLYCOLYSIS	-0.122230206	-0.07166342	-4.771836508	2.55E-06	1.16E-05
HALLMARK_MYOGENESIS	0.139566038	-0.041322775	4.240790864	2.76E-05	0.000115026
HALLMARK_DNA_REPAIR	-0.115944457	-0.064905715	-4.057349298	5.95E-05	0.000228978
HALLMARK_CHOLESTEROL_HOMEOSTASIS	-0.118516762	-0.065834705	-4.000071194	7.52E-05	0.000268731
HALLMARK_ANDROGEN_RESPONSE	-0.103077911	-0.071407537	-3.646903055	0.000299949	0.000999829
HALLMARK_UV_RESPONSE_UP	-0.083914605	-0.077126764	-3.421049306	0.00068705	0.002147032
HALLMARK_PEROXISOME	-0.079669494	-0.070107986	-3.134755795	0.001844915	0.005426219
HALLMARK_KRAS_SIGNALING_DN	0.079744105	-0.033700197	3.089507182	0.00214277	0.00595214
HALLMARK_NOTCH_SIGNALING	0.0825477	-0.082012611	2.649957698	0.00836459	0.022012079
HALLMARK_FATTY_ACID_METABOLISM	-0.072926997	-0.055702484	-2.599278491	0.009682289	0.024205724
HALLMARK_PANCREAS_BETA_CELLS	0.080998211	-0.017046845	2.539832594	0.011462339	0.027291284
HALLMARK_BILE_ACID_METABOLISM	0.064579135	-0.066566708	2.416564376	0.016107771	0.036608571
HALLMARK_HYPOXIA	0.065229175	-0.060467738	2.26284711	0.024172503	0.05254892
HALLMARK_ESTROGEN_RESPONSE_EARLY	0.060767302	-0.065609127	2.232487482	0.026127241	0.054431753
HALLMARK_APICAL_JUNCTION	0.06492846	-0.051374784	2.122111554	0.034433113	0.068866226

GSVA, Gene Set Variation Analysis. TCGA, The cancer genome atlas. STAD, Stomach adenocarcinoma

Table 9.

GSVA analysis of dataset GSE62254

ontology	logFC	AveExpr	t	P.Value	adj.P.Val
HALLMARK_PI3K_AKT_MTOR_SIGNALING	-0.18307445	-0.000984165	-7.928686269	3.97E-14	1.99E-12
HALLMARK_MTORC1_SIGNALING	-0.197387861	0.019847775	-7.031697525	1.29E-11	3.23E-10
HALLMARK_ALLOGRAFT_REJECTION	-0.235467258	0.01279032	-6.346579352	7.74E-10	1.29E-08
HALLMARK_INTERFERON_GAMMA_RESPONSE	-0.235033831	0.011326523	-5.796869359	1.65E-08	2.07E-07
HALLMARK_IL6_JAK_STAT3_SIGNALING	-0.21040704	0.003820907	-5.631618798	3.98E-08	3.98E-07
HALLMARK_INTERFERON_ALPHA_RESPONSE	-0.230982429	0.003192057	-5.119528405	5.37E-07	4.47E-06
HALLMARK_SPERMATOGENESIS	-0.124800654	0.005416617	-5.035890764	8.06E-07	5.76E-06
HALLMARK_UNFOLDED_PROTEIN_RESPONSE	-0.130340449	0.006721259	-4.800607609	2.46E-06	1.53E-05
HALLMARK_INFLAMMATORY_RESPONSE	-0.175507135	0.005422508	-4.765079409	2.90E-06	1.53E-05
HALLMARK_E2F_TARGETS	-0.206285913	0.009874688	-4.75372391	3.05E-06	1.53E-05
HALLMARK_G2M_CHECKPOINT	-0.183410245	0.003845996	-4.669325063	4.50E-06	2.04E-05
HALLMARK_COMPLEMENT	-0.144345919	0.0130183	-4.586163181	6.54E-06	2.73E-05
HALLMARK_MYC_TARGETS_V1	-0.155338392	-0.001652378	-4.110283966	5.06E-05	0.000194484
HALLMARK_KRAS_SIGNALING_UP	-0.124211775	0.006996365	-3.975789991	8.72E-05	0.000311491
HALLMARK_IL2_STAT5_SIGNALING	-0.114792467	0.002199093	-3.889216633	0.000122923	0.000409742
HALLMARK_UV_RESPONSE_UP	-0.087396645	0.000466886	-3.613603838	0.000351862	0.001036232
HALLMARK_PROTEIN_SECRETION	-0.107241346	0.006152716	-3.613252992	0.000352319	0.001036232
HALLMARK_APOPTOSIS	-0.101899656	0.002009734	-3.592629981	0.000380206	0.001056128
HALLMARK_DNA_REPAIR	-0.096437951	-0.001931103	-3.553800626	0.000438423	0.001153745
HALLMARK_MYOGENESIS	0.094415725	-0.017581741	3.210749135	0.001461838	0.003654594
HALLMARK_MITOTIC_SPINDLE	-0.07947735	-0.004575013	-3.044016968	0.00253306	0.006031094
HALLMARK_P53_PATHWAY	-0.068328272	0.010124858	-2.753667065	0.006238607	0.014178652
HALLMARK_TNFA_SIGNALING_VIA_NFKB	-0.100703623	-0.002262218	-2.632232824	0.008904385	0.019357358
HALLMARK_KRAS_SIGNALING_DN	0.057823488	0.003702876	2.603408156	0.009671255	0.020148447
HALLMARK_ANDROGEN_RESPONSE	-0.068873539	0.002364036	-2.491213177	0.013250022	0.026500045
HALLMARK_WNT_BETA_CATENIN_SIGNALING	0.075382876	-0.011708188	2.456469701	0.014575285	0.028029393
HALLMARK_FATTY_ACID_METABOLISM	-0.062280641	0.00763542	-2.254690937	0.024845706	0.046010568
HALLMARK_MYC_TARGETS_V2	-0.093685778	0.001668596	-1.989590967	0.047508538	0.084836676

GSVA, Gene Set Variation Analysis

Table 10.

GSVA analysis of dataset GSE84437

ontology	logFC	AveExpr	t	P.Value	adj.P.Val
HALLMARK_HEDGEHOG_SIGNALING	-0.184337339	0.011547906	-6.814368294	3.07E-11	1.54E-09
HALLMARK_NOTCH_SIGNALING	-0.14288899	0.003739774	-5.477528785	7.21E-08	1.31E-06
HALLMARK_TGF_BETA_SIGNALING	-0.143937688	0.011419886	-5.461394295	7.85E-08	1.31E-06
HALLMARK_WNT_BETA_CATENIN_SIGNALING	-0.125829497	-0.001628685	-4.692082071	3.60E-06	4.50E-05
HALLMARK_UV_RESPONSE_DN	-0.100519728	-0.003936085	-3.95269933	8.98E-05	0.000897941
HALLMARK_EPITHELIAL_MESENCHYMAL_TRANSITION	-0.125021785	0.004269709	-3.592959452	0.000363244	0.003027037
HALLMARK_MYOGENESIS	-0.090786029	0.000424425	-3.472398711	0.000565994	0.003952077
HALLMARK_ANGIOGENESIS	-0.110039447	0.014847417	-3.399432316	0.000735769	0.003952077
HALLMARK_HYPOXIA	-0.075762739	0.004291951	-3.393571671	0.000751286	0.003952077
HALLMARK_SPERMATOGENESIS	0.066944273	-0.003972611	3.379279639	0.000790415	0.003952077
HALLMARK_XENOBIOTIC_METABOLISM	-0.067952925	0.007183446	-3.257173836	0.001210917	0.005504168
HALLMARK_HEME_METABOLISM	-0.060295303	0.001167438	-3.11067373	0.001986006	0.008275026
HALLMARK_INTERFERON_ALPHA_RESPONSE	0.108058174	-0.010785889	3.015066514	0.002715452	0.010444046
HALLMARK_ALLOGRAFT_REJECTION	0.086693573	-0.004334194	2.886003606	0.004090485	0.014608876
HALLMARK_G2M_CHECKPOINT	0.088398528	0.006067352	2.76602951	0.005909314	0.019697714
HALLMARK_PANCREAS_BETA_CELLS	-0.074505725	-0.000482083	-2.686246582	0.007494976	0.023421801
HALLMARK_ADIPOGENESIS	-0.057202901	0.004712167	-2.638796438	0.008610486	0.025324957
HALLMARK_E2F_TARGETS	0.091433135	0.009624214	2.51666284	0.012196034	0.033877872
HALLMARK_APICAL_JUNCTION	-0.059263675	0.005146952	-2.474103002	0.013727217	0.036124256
HALLMARK_PROTEIN_SECRETION	-0.06079844	0.011423253	-2.343780455	0.019526365	0.048815913
HALLMARK_ESTROGEN_RESPONSE_LATE	-0.044607564	0.009609537	-2.203879337	0.028041622	0.066765767
HALLMARK_INTERFERON_GAMMA_RESPONSE	0.066501645	-0.004885314	2.071860628	0.03885107	0.088297886
HALLMARK_APICAL_SURFACE	-0.048951024	0.01820085	-1.982392609	0.048046844	0.100057045
HALLMARK_COAGULATION	-0.04919402	0.019186707	-1.973075635	0.049102255	0.100057045

GSVA, Gene Set Variation Analysis

PPI network

We conducted a protein–protein interaction (PPI) analysis on 22 prognosis-related ICD genes, utilizing the STRING database (see Table 1). The confidence parameter in the STRING database was set a minimum required interaction score of 0.400. A PPI network of the 22 ICD prognosis-related genes was created, which was then visualized using Cytoscape software (Fig. 8A). We calculated scores for the ICD prognosis-related genes using 5 algorithms from the Cytohubba plugin (the closeness, the degree, EPC, MCC, MNC) and ranked the ICD prognosis-related genes accordingly based on these scores. Additionally, we utilized the top 10 ICD prognosis-related genes from each of the 5 algorithms to generate PPI networks, namely: the closeness (Fig. 8B); the degree (Fig. 8C); EPC (Fig. 8D); MCC (Fig. 8E); MNC (Fig. 8F). Finally, we determined hub genes for STAD by taking the intersection of genes identified by the 5 algorithms, resulting in 8 hub genes: CD4, HSP90AA1, CD274, HMGB1, IFNB1, IFNGR1, PDCD1, PDIA3.

Fig. 8.

The PPI regulatory network of ICD prognosis-related genes and the PPI regulatory network of the top 10 genes as determined by the CytoHubba algorithm. A. PPI network of ICD prognosis-related genes. B. The PPI network of the top 10 ICD prognosis-related genes derived from the closeness algorithm is depicted, with rectangle colors ranging from yellow to red indicating increasing scores. C. The PPI network of the top 10 ICD prognosis-related genes derived from the degree algorithm is depicted, with square colors ranging from red to yellow indicating scores from high to low. D. The PPI network of the top 10 ICD prognosis-related genes derived from the EPC algorithm is depicted, with square colors ranging from red to yellow indicating scores from high to low. E. The PPI network of the top 10 ICD prognosis-related genes derived from the MCC algorithm is depicted, with square colors ranging from red to yellow indicating scores from high to low. F. The PPI network of the top 10 ICD prognosis-related genes derived from the MNC algorithm is depicted, with square colors ranging from red to yellow indicating scores from high to low

Prognostic clinical correlation analysis

To further validate the LASSO regression prognostic model we created, we conducted statistical analysis on the clinical information of STAD patients obtained from the TCGA-STAD dataset in this study (Table 11). Then, within the TCGA-STAD dataset, we employed univariate and multivariate Cox regression analyses to evaluate the relationship between the expression of hub genes (CD4, HSP90AA1, CD274, HMGB1, IFNB1, IFNGR1, PDCD1, PDIA3) and different clinical variables, as well as the impact of these genes' expression on prognosis. Initially, we conducted univariate Cox regression analysis on the expression levels of hub genes and different clinical variables, factors with P < 0.1 were chosen for integration into the multivariate Cox regression analysis to construct a multivariate Cox regression model (Table 12), which was visualized in a forest plot (Fig. 9A). Ultimately, seven genes were included in the Cox regression analysis, while IFNB1 was not incorporated due to data missingness. After multivariate Cox regression analysis, only HMGB1 retained its independent prognostic significance after adjusting for clinical factors, while the other genes did not demonstrate independence. The results indicated a significant correlation between clinical T stage, clinical N stage, clinical M stage, pathology stage, age and prognosis (P < 0.05). Subsequently, we conducted a nomogram analysis to evaluate the predictive capacity of the model, we illustrated the results in a nomogram (Fig. 9B). In addition, We executed a prognostic calibration analysis for 1-year (Fig. 9C), 3-years (Fig. 9D), and 5-years (Fig. 9E) on the univariate and multivariate Cox regression model nomogram, from which we generated the calibration curves (Fig. 9 C-E). According to the figure, the blue line corresponding to the 3-year is closest to the gray ideal case line, suggesting that the predictive efficacy of the model is superior in the 3-year, compared to its performance in the 1st and 5th years.We subsequently employed DCA analysis to evaluate the clinical utility of the constructed LASSO-Cox regression prognostic model over 1 year (Fig. 9F), 3 years (Fig. 9G), and 5 years (Fig. 9H), and presented the results (Fig. 9F-H). As can be seen from the figures, the blue line representing the model is stably surpasses the red line for 'all positive' and the grey line for 'all negative'. The range of x-values is largest at 3 years, and smaller at 1 and 5 years, suggesting that the predictive performance of the model initially improves with time, then decreases. The values of the prediction index for 1-year, 3-year, and 5-year periods are shown in Figure S3 of the supplementary materials.

Table 11.

Patient Characteristics of STAD patients in the TCGA datasets

Characteristic	levels	Overall
n		375
T stage, n (%)	T1	19 (5.2%)
	T2	80 (21.8%)
	T3	168 (45.8%)
	T4	100 (27.2%)
N stage, n (%)	N0	111 (31.1%)
	N1	97 (27.2%)
	N2	75 (21%)
	N3	74 (20.7%)
M stage, n (%)	M0	330 (93%)
	M1	25 (7%)
Pathologic stage, n (%)	Stage I	53 (15.1%)
	Stage II	111 (31.5%)
	Stage III	150 (42.6%)
	Stage IV	38 (10.8%)
Gender, n (%)	Female	134 (35.7%)
	Male	241 (64.3%)
Age, n (%)	<  = 65	164 (44.2%)
	> 65	207 (55.8%)
OS event, n (%)	Alive	228 (60.8%)
	Dead	147 (39.2%)
DSS event, n (%)	Alive	263 (74.3%)
	Dead	91 (25.7%)
PFI event, n (%)	Alive	251 (66.9%)
	Dead	124 (33.1%)

STAD, Stomach adenocarcinoma. TCGA, The cancer genome atlas

Table 12.

Cox regression to identify hub genes and clinical features associated with OS

Characteristics	Total(N)	Univariate analysis		Multivariate analysis
		Hazard ratio (95% CI)	P value	Hazard ratio (95% CI)	P value
T stage	362
T1	18	Reference
T2	78	6.725 (0.913–49.524)	0.061	5.139 (0.659–40.105)	0.118
T3	167	9.548 (1.326–68.748)	0.025	7.211 (0.836–62.208)	0.072
T4	99	9.634 (1.323–70.151)	0.025	6.358 (0.714–56.643)	0.097
N stage	352
N0	107	Reference
N1	97	1.629 (1.001–2.649)	0.049	1.243 (0.619–2.495)	0.541
N2	74	1.655 (0.979–2.797)	0.060	1.505 (0.640–3.542)	0.349
N3	74	2.709 (1.669–4.396)	< 0.001	2.036 (0.865–4.791)	0.103
M stage	352
M0	327	Reference
M1	25	2.254 (1.295–3.924)	0.004	1.188 (0.490–2.882)	0.703
Pathologic stage	347
Stage I	50	Reference
Stage II	110	1.551 (0.782–3.078)	0.209	1.058 (0.373–3.000)	0.916
Stage III	149	2.381 (1.256–4.515)	0.008	0.976 (0.249–3.829)	0.972
Stage IV	38	3.991 (1.944–8.192)	< 0.001	1.996 (0.502–7.944)	0.327
Age	367
<  = 65	163	Reference
> 65	204	1.620 (1.154–2.276)	0.005	1.977 (1.353–2.890)	< 0.001
Gender	370
Female	133	Reference
Male	237	1.267 (0.891–1.804)	0.188
CD4	370
Low	185	Reference
High	185	1.203 (0.865–1.672)	0.272
HSP90AA1	370
Low	184	Reference
High	186	1.028 (0.741–1.427)	0.869
CD274	370
Low	185	Reference
High	185	0.958 (0.690–1.330)	0.799
HMGB1	370
Low	184	Reference
High	186	0.724 (0.521–1.007)	0.055	0.620 (0.429–0.896)	0.011
IFNGR1	370
Low	186	Reference
High	184	0.935 (0.674–1.297)	0.687
PDCD1	370
Low	184	Reference
High	186	0.783 (0.564–1.088)	0.145
PDIA3	370
Low	183	Reference
High	187	1.043 (0.752–1.447)	0.801

OS, overall survival

Fig. 9.

Prognostic Correlation Analysis. A-B. Forest plot (A) and nomogram plot (B) of univariate and multivariate Cox regression analysis of hub genes. C-E. Calibration curve plots for 1-year (C), 3-year (D) and 5-year (E) from nomogram analysis of univariate and multivariate Cox regression models. F–H. Decision curve analysis (DCA) plots for 1-year (F), 3-year (G), and 5-year (H) of the LASSO-Cox regression prognostic model. The x-axis in the DCA plot represents the probability threshold or threshold probability, while the y-axis represents net benefit

Additionally, we individually plotted prognosis survival subgroup KM curves for the 8 hub genes (CD4, HSP90AA1, CD274, HMGB1, IFNB1, IFNGR1, PDCD1, PDIA3) and subtype clinical variables, including T stage (T1, T2, T3, T4), N stage (N0, N1, N2, N3), M stage (M0, M1), Pathologic stage (stage I, stage II, stage III, stage IV), gender (female, male) and age (≤ 65, > 65). We considered P < 0.05 as statistically significant, leading to the identification of 8 hub gene-subtype clinical variable combinations meeting the criteria, namely: HMGB1-N stage (N1) (P = 0.036, Fig. 10A), HMGB1-Pathologic stage (stage III) (P = 0.044, Fig. 10B), HMGB1-T stage (T4) (P = 0.043, Fig. 10C), PDCD1-N stage (N2) (P = 0.024, Fig. 10D), PDCD1-N stage (N3) (P = 0.017, Fig. 10E), PDCD1-gender (female) (P = 0.006, Fig. 10F), HSP90AA1-N stage (N1) (P = 0.028, Fig. 10G), IFNGR1-Pathologic stage (stage II) (P = 0.041, Fig. 10H).

Fig. 10.

Subgroup KM Prognostic Analysis. A-H. Prognostic analysis Kaplan–Meier curves for HMGB1-N stage (N1) (A), HMGB1-Pathologic stage (stage III) (B), HMGB1-T stage (T4) (C), PDCD1-N stage (N2) (D), PDCD1-N stage (N3) (E), PDCD1-gender (female) (F), HSP90AA1-N stage (N1) (G), IFNGR1-Pathologic stage (stage II) (H)

ROC curve and chromosomal localization analysis

To further explore the expression differences of these 8 hub genes (CD4, HSP90AA1, CD274, HMGB1, IFNB1, IFNGR1, PDCD1, PDIA3), we subsequently plotted ROC curves for the 8 hub genes in the TCGA-STAD dataset and presented the results (Fig. 11A-H). From the ROC curves, it was observed that among these 8 hub genes: the expression of HSP90AA1 (AUC = 0.965, Fig. 11B) showed a significant correlation with the occurrence of STAD; the expressions of HMGB1 (AUC = 0.825, Fig. 11D), IFNGR1 (AUC = 0.786, Fig. 11F) and PDIA3 (AUC = 0.839, Fig. 11H) exhibited relatively high correlations with the occurrence of STAD; while the expressions of CD4 (AUC = 0.644, Fig. 11A), CD274 (AUC = 0.693, Fig. 11C), IFNB1 (AUC = 0.653, Fig. 11E) and PDCD1 (AUC = 0.607, Fig. 11G) showed moderately high correlations with the occurrence of STAD. Subsequently, we generated a chromosomal localization plot (Fig. 11I) for the 8 hub genes. The results indicated that except for the genes CD274 and IFNB1 located on chromosome 9, the other genes were situated on different chromosomes, with PDCD1 on chromosome 2, IFNGR1 on chromosome 6, CD4 on chromosome 12, HMGB1 on chromosome 13, HSP90AA1 on chromosome 14, and PDIA3 on chromosome 15.

Fig. 11.

The receiver operating characteristic (ROC) Curve and Chromosomal Localization Plot. A-H. Results of ROC curves for hub genes CD4 (A), HSP90AA1 (B), CD274 (C), HMGB1 (D), IFNB1 (E), IFNGR1 (F), PDCD1 (G), PDIA3 (H) in the TCGA-STAD dataset. I. Chromosomal localization results for hub genes

Immunohistochemistry analysis

We conducted immunohistochemistry analysis in STAD tumor tissues and normal gastric tissues, using the HPA database to examine the expression of hub genes (HSP90AA1, HMGB1, IFNGR1, PDIA3) that showed good correlation in the ROC curve results. The staining was performed using DAB (3,3'-Diaminobenzidine) and hematoxylin counterstain. The immunohistochemistry analysis revealed that compared to normal gastric tissues (Fig. 12B, D, F, H), the expression levels of genes HSP90AA1, HMGB1, IFNGR1, PDIA3 were higher in STAD tumor tissues (Fig. 12A, C, E, G).

Fig. 12.

Immunohistochemistry Analysis. A. HSP90AA1 (STAD tissue), B. HSP90AA1 (Normal stomach tissue), C. HMGB1 (STAD tissue), D. HMGB1 (Normal stomach tissue), E. IFNGR1 (STAD tissue), F. IFNGR1 (Normal stomach tissue), G. PDIA3 (STAD tissue), H. PDIA3 (Normal stomach tissue)

Discussion

Gastric cancer is knowned as a highly prevalent malignant tumor of the digestive system globally [31], its pathogenesis and prognosis have been a hot topic of research concerning [32]. The prediction of GC prognosis is primarily based on clinical features, such as TNM stage, inflammatory infiltration, tumor budding, together with certain tumor biomarkers [33–35]. However, the predictive performance of these factors is limited. Therefore, it is important to find efficient biomarkers, which can assist in judging patient prognosis and making treatment decisions. With the rapid development of bioinformatics technologies, many multigene models, like the prognostic model of ICD-related genes, show significant predictive performance in cancer. ICD, as a special form of cell death, initiates anti-tumor immune responses by releasing or exposing DAMPs from the dying tumor cells [36–38]. It holds significant importance for cancer treatment [39–41] and prognosis [42–44]. This study focused on ICD-related genes, aiming to provide new theoretical basis for precise treatment and prognosis assessment of gastric cancer by analyzing the expression changes of these genes in gastric cancer and their relationship with prognosis.

In this study, cross-validation was performed using data from TCGA and two independent GEO cohorts, which enhanced the robustness of the model. An 8-gene signature was refined through a "two-step screening" approach, balancing both predictive performance and clinical operability. A combined predictive model was constructed by integrating core genes with clinical staging variables, and its clinical application value was validated through nomograms and DCA.The differential expression of HSP90AA1, HMGB1, IFNGR1, and PDIA3 in gastric cancer tissues versus normal tissues was verified using immunohistochemistry results from the HPA database. We successfully constructed a prognostic model closely related to gastric cancer prognosis containing 22 ICD-related genes. We also identified differential expression of the 22 ICD prognosis-related genes among different risk groups of STAD in the TCGA-STAD, GSE62254 and GSE84437 datasets. Furthermore, we explored the relationship between biological processes, molecular functions, cellular components, biological pathways and STAD related to these genes. The findings suggest that the primary enrichment of these differentially expressed genes was observed in biological processes such as lymphocyte differentiation and cytokine secretion, as well as molecular functions like cytokine binding and co-receptor activity. The occurrence and progression of gastric cancer are intimately linked to these biological processes and pathway, further confirming the significant role of ICD-related genes in gastric cancer. Particularly, genes upregulated in the high-risk group might serve as potential targets for gastric cancer treatment. For instance, The non-histone chromosomal protein, High Mobility Group Box 1 (HMGB1), exhibits increased expression in the high-risk group of STAD. The P value of HMGB1 in multivariate Cox regression analyses was 0.055. Although slightly higher than 0.05, the high expression of HMGB1 had a hazard ratio (HR) of 0.724. This suggests that this gene is associated with a better prognosis and could be considered a prognostic indicator with potential clinical significance (Fig. 9A). HMGB1 has a dual function in cancer: excessive HMGB1 production caused by chronic inflammatory response can contribute to tumorigenesis [45], while HMGB1 also plays a protective role in the suppression of tumor growth and enhances the effects of anti-tumor therapy [46], this provides the possibility for combined anti-tumor therapy. GSEA and GSVA revealed additional pathways involving ICD-related genes in the biological processes of gastric cancer, such as TP53 transcriptional regulation [47] and the JAK-STAT signaling pathway [48]. The activation or inhibition of these pathways correlate with the immune microenvironment, cell proliferation and apoptosis, invasion, and migration processes in GC [49, 50].

We identified 8 hub genes closely related to gastric cancer prognosis, including CD4, HSP90AA1, CD274, HMGB1, IFNB1, IFNGR1, PDCD1 and PDIA3 through PPI network analysis. These genes might be closely associated with the biological processes of gastric cancer, providing new insights for the future precise treatment of gastric cancer. For instance, HSP90AA1 could regulate oncogene products or signaling transduction factors, promoting tumor progression, invasion, and drug resistance [51]. PDCD1 (PD-1) plays a vital role in tumor immune escape mechanisms and is a target for immunotherapy [52]. The drugs related to PDCD1 that are currently used in clinical practice mainly include PD-1 inhibitors, such as pembrolizumab, nivolumab, and cemiplimab. These inhibitors can block the binding of PD-1 to its ligands, thereby activating the immune system to attack tumors.The application of immune checkpoint inhibitors has led to a better prognosis for patients with gastric cancer (GC). Inhibiting the expression of CD274(PD-L1) on tumor cells can enhance immune surveillance and reduce the function of PD-L1-derived immune checkpoint pathways [53]. Cytotoxic T cells are important effector factors in anti-tumor immunity. CD4 + T cells not only express key molecules associated with cytolytic granules but also possess direct cytotoxicity. This serves as the foundation for both pathogenic and protective immunity, including in the context of cancer [54]. HMGB1 can bind to a variety of receptors, including RAGE, and subsequently activate certain key cellular signaling pathways (such as NF-κB, p38, and MAPK), leading to cancer progression and metastasis [55]. HMGB1 can also regulate the progression of gastric cancer, for instance, vitexin inhibits the development of gastric cancer (GC) cells by suppressing the activation of the HMGB1-mediated PI3K/AKT/mTOR/HIF-1α pathway [56]. IFN-β was initially identified as an immunoregulatory cytokine due to its antiviral activity. Further characterization of its biological effects has revealed a broad range of potential anti-tumor actions, such as inhibiting cell proliferation, inducing cell death, and causing cell cycle arrest [57]. Studies have shown that the expression of IFNGR1 can enhance anti-tumor immunity, and targeting the stability of IFNGR1 can overcome intrinsic resistance to immunotherapy in colorectal cancer patients [58]. PDIA3 plays an oncogenic role in gastric cancer (GC). Downregulation of PDIA3 can significantly inhibit the proliferation, invasion, and migration of GC cells TMK1 and AGS, leading to cell cycle arrest at the G2/M phase [59].The above findings indicate that these 8 hub genes are closely associated with the occurrence and development of tumors and have the potential to serve as therapeutic targets. It should be noted that this study is a discovery-oriented research and requires further validation.

The clinical value of hub genes for gastric cancer prognosis was verified in particular, univariate and multivariate Cox regression analyses were conducted to explore the role of hub genes (including CD4, HSP90AA1, CD274, HMGB1, IFNB1, IFNGR1, PDCD1, PDIA3) and clinical variables in prognosis (Fig. 9A and Table 12). Among the 22 ICDRGs and 8 hub genes, except for HMGB1, the other genes exhibit limited clinical independent prognostic value and are more inclined to serve as potential indicative factors rather than independent prognostic indicators. Preliminary univariate analysis selected factors with P < 0.1 for inclusion in the multivariate model, and the results showed that T3 and T4 in T staging, III and IV stages in pathological staging, N3 staging, M staging, and age were significantly associated with patient prognosis (P < 0.05). Forest plots of univariate and multivariate Cox regression analyses (Fig. 9A) presented hazard ratios for each variable, further supporting the predictive ability of the model. Future research can increase the sample size and further validation to clarify its impact. Meanwhile, the Kaplan–Meier survival analysis explored the impact of specific hub genes and combinations of clinical variables on prognosis. Significant combinations included HMGB1—N stage (N1), pathologic stage (stage III), T stage (T4), PDCD1—N stage (N2, N3), gender (female), HSP90AA1—N stage (N1) and IFNGR1—pathologic stage (stage II) (Fig. 10 A-H). The results of this analysis are exploratory findings and require further validation in independent cohorts and experiments. Subsequently, by plotting the nomogram (Fig. 9B) and calibration curves (Fig. 9 C-E), the model's predictive ability at 1 year, 3 years and 5 years was validated, with the best prediction effect at the 3rd year (Fig. 9D). The clinical utility of the model was further evaluated using Decision Curve Analysis (DCA). The results indicated that the prognosis model for 3 years had the greatest utility (Fig. 9G). In summary, although the model demonstrated good predictive ability at the 3-year mark, further optimization is needed at different time points. Future research could improve the applicability and stability of the model in different populations by increasing sample size, multi-center validation and in-depth mechanism studies, and explore its practicality and translational potential in clinical settings.

Furthermore, ROC curves for the 8 hub genes were plotted in the TCGA-STAD dataset, revealing the expression of HSP90AA1 (AUC = 0.965) showed significant correlation with the occurrence of STAD. Immunohistochemistry analysis was conducted using the HPA database to examine the expression levels of hub genes (HSP90AA1, HMGB1, IFNGR1, PDIA3) which had shown good correlation in the ROC curve analysis, in both STAD tumor tissues and normal gastric tissues, revealing higher expression levels of genes HSP90AA1, HMGB1, IFNGR1, PDIA3 in STAD tumor tissues compared to normal gastric tissues.

Although this study has made certain discoveries regarding the relationship between gastric cancer and immunogenic cell death-related genes, it still has certain limitations. Firstly, this research is primarily based on bioinformatics analysis and lacks experimental validation. In the future, we aim to further validate these findings through experimental means. Secondly, given that the number of tumor samples in TCGA significantly exceeds that of normal control samples, there exists a limitation of imbalanced sample sizes, which may lead to statistical bias. We have already made efforts to minimize bias through standardization and multi-cohort validation. In the future, we will continue to expand our data sources and adopt more appropriate statistical methods. Finally, the lack of direct clinical validation analysis is another limitation that could be addressed in future research to confirm the prognostic roles of identified key genes. We will continue to explore the potential applications of these key genes in the treatment of gastric cancer, aiming to provide new strategies for individualized treatment of gastric cancer.

Conclusions

In a word, this study, by constructing a prognostic model based on ICDRGs, further study the mechanisms of ICDRGs in gastric cancer and their relationship with prognosis, revealing the significant role of these genes in the prognosis assessment of gastric cancer. The research results provide new theoretical basis for precise treatment and prognosis assessment of gastric cancer. These findings lay the foundation for future research, including laboratory studies and larger-scale clinical validation studies, to confirm the roles of these genes in STAD. These studies might contribute to future prognosis assessment for STAD patients and the development of personalized treatment strategies.

Abbreviations

ICD

Immunogenic cell death

GC

Gastric cancer

ICDRGs

Immunogenic cell death-related genes

TCGA

The Cancer Genome Atlas

GEO

The Gene Expression Omnibus

HPA

The Human Protein Atlas

DAMPs

Damage-associated molecular patterns

STAD

Stomach Adenocarcinoma

LASSO

Least absolute shrinkage and selection operator

ICD prognosis-related genes

Immunogenic cell death prognosis-related genes

DEGs

Differentially expressed genes

GO

The Gene Ontology

BP

Biological process

MF

Molecular function

CC

Cellular component

KEGG

The Kyoto Encyclopedia of Genes and Genomes

FDR

False discovery rate

GSEA

Gene Set Enrichment Analysis

GSVA

Gene Set Variation Analysis

PPI

Protein–protein interaction

EPC

Edge percolated component

MCC

Maximal clique centrality

MNC

Maximum neighborhood component

DCA

Decision curve analysis

KM

Kaplan–Meier

ROC

Receiver Operating Characteristic curve

AUC

The Area Under Curve

HMGB1

High Mobility Group Box 1

Author contributions

YKL, XYW and YF conceived and designed the study, CFM, ZYL and DDG collected the data. YKL prepared Figs. 1, 4, 6, 9, 10, 11, S1, S3 and ZJG prepared Figs. 2, 3, 5, 7, 8, 12 and S2. YKL prepared Figs. 1,4,6 and ZJG prepared Figs. 2,3,5. XYW, ZYL and SQZ prepared the first draft of the manuscript and corrected it.YKL, YF and ZJG reviewed and edited it.All authors read and approved the final manuscript.

Funding

This work was supported partially by grants from: (1) Key project fund of Jiangsu Provincial Health Commission (ZD2022052, K2023016); (2) Suqian science and technology support project fund (KY202203); (3) Social Development Fund of Zhenjiang (SH2024002, SH2024075, JC0004031).

Availability of data and materials

Data is provided within the manuscript information files.The datasets analyzed for this study can be found in the TCGA database (https://portal.gdc.cancer.gov/) and GEO databases (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE62254; https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE84437).

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.