A novel ferroptosis- and endoplasmic reticulum stress-related gene signature for predicting prognosis, immune features and drug sensitivity in gastric cancer

Abstract

Objective

While ferroptosis and endoplasmic reticulum stress (ERS) are implicated in gastric cancer (GC), their integrated prognostic value remains unclear. Our research aimed to construct a novel prediction model based on ferroptosis- and ERS-related genes (F&ERSRGs) to assess prognosis and identify potential therapeutic strategies for GC.

Methods

We analyzed the transcriptome and clinical data of GC cohort (n = 378) from The Cancer Genome Atlas (TCGA) database as a training set and data of three independent cohorts from the Gene Expression Omnibus (GEO) database as validation sets. Differential expression analysis for the ferroptosis- and ERS-related genes (F&ERSRGs) between tumor and normal tissues was performed. Among the differentially expressed F&ERSRGs, prognostic F&ERSRGs screened by univariate Cox regression were included in Least Absolute Shrinkage and Selection Operator (LASSO) analysis to develop a risk model. Kaplan-Meier survival analysis and receiver operating characteristic (ROC) analysis were performed to assess the performance of the model. A nomogram was developed by integrating the risk score and clinicopathological characteristics. Functional enrichment analysis and evaluation of tumor microenvironment (TME) were also conducted. In addition, drug sensitivity was predicted using data from the Genomics of Drug Sensitivity in Cancer (GDSC) database. Finally, the expression of seven signature genes was validated by quantitative real-time PCR (qRT-PCR) in clinical samples.

Results

A seven-gene (NOX4, CHAC1, MYB, SNCA, ALB, CAV1, and FABP4) predictive risk model was finally constructed. Patients were categorized as high- or low-risk using median risk score as the threshold. The area under the ROC curve (AUC) values for the 1-, 3-, and 5-year overall survival (OS) in the training cohort were 0.619, 0.680, and 0.709, respectively. Survival analysis showed a better OS in low-risk patients in the training and validation cohorts. The AUC values of the nomogram for predicting 1-, 3-, and 5-year OS were 0.709, 0.711, and 0.723, respectively. TME analyses revealed a higher M2 macrophage infiltration and an immunosuppressive TME in the high-risk group. Furthermore, High-risk patients tended to be more sensitive to sepantronium bromide.

Conclusion

A novel F&ERSRGs based signature was built for prognosis and treatment prediction of GC patients.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-026-04562-8.

Introduction

Gastric cancer (GC) remains a particularly high burden in global public health. In 2022, GC accounted for approximately 968,000 and 660,000 deaths globally, ranking it the 5th in terms of both incidence and cancer-related mortality in the world [1]. Despite significant advancements in chemotherapy, targeted therapies and immune checkpoint inhibitors, the prognosis of advanced gastric cancer remains suboptimal [2]. Conventional prognostic strategies in GC, notably tumor staging and histopathological evaluation, fail to reliably predict patient outcomes due to intertumoral heterogeneity present even within identical stage groupings and histological subtypes [3, 4]. Developing more accurate methods to predict GC prognosis and guide clinical management has positive clinical significance.

Ferroptosis, a term introduced in 2012 by Dr. Brent R. Stockwell’s laboratory [5], represents a distinct iron-dependent modality of regulated cell death with distinct properties and recognizing functions involved in cancers, neurodegeneration, nervous system diseases, tissue injury, inflammation, blood diseases, and infection [6, 7]. It is primarily triggered by dysregulated iron metabolism and is implicated in chemoresistance [8]. The endoplasmic reticulum (ER), an essential organelle in eukaryotic cells, facilitates lipid metabolism, calcium (Ca²⁺) homeostasis, and protein processing, folding, and transport [9]. Endoplasmic reticulum stress (ERS) is triggered by hypoxic conditions, genetic mutations, nutrient deficiencies, and oxidative stress. These stimuli cause accumulation of misfolded/unfolded proteins within the ER lumen, activating the unfolded protein response (UPR) to mitigate cellular stress [10]. However, persistent high-intensity ERS exceeding the adaptive capacity of the UPR triggers cell death and contributes to diverse pathologies, including cancer, atherosclerosis, diabetic retinopathy, and ischemic nephropathy [11–13]. The persistent activation of endoplasmic reticulum (ER) stress sensors equips malignant cells with enhanced capabilities for tumorigenesis, metastasis, and chemoresistance [14]. Besides, researches have revealed that ERS signaling transforms the tumor microenvironment, fostering tumor growth [15, 16]. Furthermore, accumulating evidence indicates that ferroptosis and endoplasmic reticulum stress (ERS) share interconnected regulatory pathways, with their crosstalk modulating disease pathogenesis in multiple disorders including cancer [17–19]. Comprehensive research on ferroptosis and ERS that provides novel targets and insights is still needed.

Recent advances in transcriptome analysis have enabled the identification of survival-associated genes to guide personalized therapeutic strategies for cancer patients [20–23]. Building upon this foundation, we systematically integrated ferroptosis- and ERS-related genes (F&ERSRGs) and established a robust prognostic signature through bioinformatics approaches. The immunological profile and biological functions of gastric cancer patients classified as high-risk and low-risk were subsequently investigated. Collectively, our findings demonstrate that this prognostic model based on ferroptosis and ERS serves as a reliable predictor of clinical outcomes in GC. Furthermore, it provides mechanistic insights into GC pathogenesis and potential therapeutic targets.

Materials and methods

Data collection

Transcriptomic data and corresponding clinical records of gastric cancer (GC) patients were retrieved from the XENA database (https://xenabrowser.net/datapages/). Detailed clinical characteristics of patients (who had complete gene expression data available) are provided in Supplementary Data Sheet 1. Additionally, three independent validation cohorts (GSE15459, GSE62254, GSE84437) with normalized microarray expression profiles and matched clinical data were acquired from the GEO repository (http://www.ncbi.nlm.nih.gov/geo/). Ferroptosis-related genes (FRGs, n = 564) were sourced from FerrDb (http://www.zhounan.org/ferrdb/current/), while endoplasmic reticulum stress-related genes (ERSRGs, n = 3363) were obtained from GeneCards (https://www.genecards.org/, Relevance score > 5). The intersection of ferroptosis- and endoplasmic reticulum stress-related genes (F&ERSRGs) which were detected in all four cohorts (GSE15459, GSE62254, GSE84437, and TCGA-STAD) were used for further analysis.

Identification of differentially expressed prognostic F&ERSRGs

After collection and preprocessing the data of GC, differential expression analysis for the F&ERSRGs between tumor and normal tissues was performed using the “DESeq2” package in R (version 4.2.3) [24]. The criteria for identifying differentially expressed genes (DEGs) was defined as an absolute log fold change (|logFC|) > 1 and an adjusted p-value (adj. P) < 0.05. The Univariate Cox regression analysis was further performed on the differentially expressed F&ERSRGs to identify F&ERSRGs with prognostic value (P < 0.05).

Construction and evaluation of F&ERSRGs based signature

Candidate F&ERSRGs were subjected to least absolute shrinkage and selection operator (LASSO) Cox regression analysis using the R package glmnet (version 4.1-8) to construct a prognostic gene signature. The analysis was performed with the glmnet function over a predefined sequence of λ values. The optimal penalty parameter λ was determined by 20-fold cross-validation based on the criterion that minimizes the cross-validated error (lambda.min). Genes with nonzero coefficients at the optimal λ were retained for signature building. The concordance index (C-index) was calculated to evaluate the predictive performance of the final multivariable model. Each patient’s risk score was computed using the formula:

Where n represents the number of signature genes, Coefficient mRNA(i) denotes the LASSO regression coefficient of gene *i*, and Expression mRNA(i) indicates the expression level of gene *i*.

For the TCGA-STAD and GEO cohorts, the following sequential criteria were applied: (1) Patients with available transcriptomic data were initially retained. (2) Only one sample (the primary tumor sample) per patient was included. (3) Patients were required to have complete overall survival (OS) information (both survival time and vital status). Patients with any missing values in these fields were excluded from the prognostic analysis.

Patients in the TCGA discovery cohort were classified into high/low-risk groups according to the median risk score. The model underwent rigorous external validation in three independent cohorts (GSE15459, GSE62254, GSE84437), where identical risk score calculation and median-based stratification were applied. The prognostic utility of the signature was confirmed through Kaplan-Meier analysis and ROC curve assessment, implemented via the R packages survival (version 3.5-3) and timeROC (version 0.4).

Independent prognostic analysis and nomogram construction

To evaluate the ferroptosis and ERS (F&ERS) signature as an independent prognostic factor in gastric cancer, we performed multivariate Cox regression analysis. Subsequently, a nomogram integrating patient age, gender, tumor stage, and risk score was developed using the “rms” R package to predict 1-, 3-, and 5-year overall survival (OS).

Functional enrichment analysis

DEGs between risk groups (|log₂FC| > 1, adjusted P < 0.05) were identified via “DESeq2” R package. Functional enrichment including Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways was assessed by hypergeometric testing in “clusterProfiler” R package [25], visualized with “circlize” R package [26]. KEGG pathway enrichment was further validated by Gene Set Enrichment Analysis (GSEA) with the Kolmogorov–Smirnov (KS) test.

Risk model’s association with TME

Tumor immune microenvironment (TIME) features were profiled using Immuno-Oncology Biological Research (IOBR, v0.99.9). Enrichment scores for TME-associated signatures were computed per sample. Differential expression of immune checkpoint genes between risk groups was assessed, followed by Pearson correlation analysis with risk scores. Immune cell infiltration quantitation was calculated with CIBERSORT (relative abundance) and ESTIMATE (immune/matrix scores) algorithms within IOBR.

Drug sensitivity analysis

Drug sensitivity profiles were predicted using the Genomics of Drug Sensitivity in Cancer (GDSC) database [27]. Based on transcriptional data, half maximal inhibitory concentrations (IC₅₀) were derived for individual cell lines. Differences in drug response between risk groups were quantified by comparing area under the dose–response curve (AUC) values using Wilcoxon rank-sum tests.

Validation of the expression of signature genes via human protein atlas

The immunohistochemistry (IHC) results of seven signature genes in normal and gastric cancer tissues were obtained from the Human Protein Atlas (HPA) (https://www.proteinatlas.org/).

Expression of signature genes by real time quantitative reverse-transcriptase polymerase chain reaction (qRT-PCR)

A gastric cancer cDNA microarray comprising 15 paired cancer and adjacent non-tumor gastric tissues was obtained from OUTDO (Shanghai OUTDO Biotech Co., LTD Shanghai, China). We performed qRT-PCR analyses using an Applied Biosystems™ QuantStudio™ 1 Plus Real-Time PCR System (Applied Biosystems, Waltham, MA, USA). Relative gene expression levels were normalized to the levels of β-actin using the △CT method. The primer sequences used for amplification are listed in Supplementary Data Sheet 2.

Statistical analysis

Statistical analyses were performed in R v4.2.3. Unless otherwise indicated, all statistical significance was set at P < 0.05.

Results

Identification of prognosis−related F&ERSRGs with differential expression in GC patients

The study design is schematized in Fig. 1. Differential expression analysis between gastric cancer and adjacent normal tissues in TCGA-STAD cohort was conducted. Among the F&ERSRGs, we identified 15 upregulated and 27 downregulated genes (Fig. 2A). Univariate Cox regression revealed 18 F&ERSRGs significantly associated with OS (Fig. 2B). Integrative analysis defined 11 consensus genes with dual differential expression and prognostic significance (Fig. 2C).

Fig. 1.

Overall workflow of the study

Fig. 2.

Identification of prognosis-related F&ERSRGs with differential expression. A Volcano plot illustrating the differentially expressed F&ERSRGs between gastric cancer and normal tissues in TCGA-STAD dataset; B The forest plot demonstrating 18 prognosis-related F&ERSRGs obtained by univariate Cox regression analysis in the TCGA-STAD dataset; C Veen diagram demonstrating 11 overlap genes based on univariate Cox regression and differential expression analysis

Construction prognostic F&ERSRGs based signature in GC

We constructed a predictive prognostic model comprising 7 F&ERSRGs (NOX4, CHAC1, MYB, SNCA, ALB, CAV1, and FABP4) by LASSO regression analysis (Figs. 3A–C). The coefficients of the genes in the ferroptosis- and ERS-related genes (F&ERSRGs) signature derived from the LASSO Cox regression are presented (Fig. 3A). A linear prediction model was developed based on the weighted regression coefficients of 7 prognostic F&ERSRGs, calculated as risk score = (0.265 × NOX4 exp) + (-0.105 × CHAC1 exp) + (-0.0830 × MYB exp) + (0.0404 × SNCA exp) + (0.0310 × ALB exp) + (0.00973 × CAV1 exp) + (0.00767 × FABP4 exp). Of these, CHAC1 and MYB showed significant inverse correlations with risk scores, while NOX4, SNCA, ALB, CAV1 and FABP4 showed significant positive correlation with risk scores.

Fig. 3.

Construction of the prognostic risk model. A LASSO Cox regression analysis for signature construction. B and C Lasso-Cox regression analysis revealed 7 F&ERSGs associated with prognosis

Validation of the F&ERSRGs signature

After establishing the predictive prognostic model based on 7 prognostic F&ERSRGs in GC, we analyzed the distribution of risk score, the survival status and expression patterns of these genes among GC patients in the TCGA cohort (Fig. 4A). The risk curves and scatter plots implied that mortality was positively related to the risk score in the TCGA cohort. The heatmap illustrates the expression profiles of the seven prognostic F&ERSRGs. Kaplan-Meier analysis revealed significantly better overall survival in low-risk patients (P < 0.001; Fig. 4B). Time-dependent ROC analysis indicated AUC values of 0.619 (1-year), 0.680 (3-year), and 0.709 (5-year) (Fig. 4C). To further assess the generalizability and robustness of the model, three independent GEO cohorts (GSE15459, GSE62254, and GSE84437) were used as the external validation cohorts. The Kaplan-Meier curves consistently showed significant stratification between high- and low-risk groups across all validation sets, with the low-risk group exhibiting a pronounced survival advantage (P < 0.001, P < 0.001, and P = 0.004, respectively; Fig. 4D–F). The ROC curve further confirmed the model’s predictive accuracy. The AUC values at 1-, 3-, and 5- years for the GSE15459 cohort were 0.662, 0.674, and 0.718, respectively (Fig. 4G). The AUC values for the GSE62254 cohort were 0.685, 0.633, and 0.621, respectively (Fig. 4H). The AUC values for the GSE84437 cohort were 0.600, 0.609, and 0.606, respectively (Fig. 4I). These results collectively demonstrate the robust prognostic performance and translational applicability of the F&ERSRGs signature in GC.

Fig. 4.

Validation of the F&ERSG signature. A The distribution of risk score, survival status and heatmap of TCGA-STAD dataset. B Kaplan-Meier survival curves of OS between low-risk (blue) and high-risk (red) groups in the TCGA-STAD dataset. C Time-dependent ROC curves of 1-, 3-, and 5-years of GC patients TCGA-STAD dataset. D–F Kaplan-Meier survival curves of OS between low-risk and high-risk groups in the GSE15459 (D), GSE62254 (E), and GSE84437 cohorts (F), respectively. G–I Time-dependent ROC curves of 1-, 3-, and 5-years of GC patients in the GSE15459 (G), GSE62254 (H), and GSE84437 (I) cohorts, respectively

Creation of nomograms based on F&ERS signatures combined with clinical characteristics

To further evaluate reliability and clinical value of the F&ERSRGs signature as a prognostic predictor, we conducted multivariate Cox regression analysis incorporating clinical characteristics (Supplementary Data Sheet 3). The results showed that tumor stage (P < 0.001), age (P < 0.001) and risk score (P < 0.001) were independent prognostic factors (Fig. 5A). Based on the above analysis, in order to be able to predict patients’ prognosis quantitatively and to inform clinical decision-making, we subsequently integrated risk score with clinical indicators to construct a nomogram predicting 1-, 3-, and 5-year overall survival probabilities (Fig. 5B). Kaplan-Meier analysis revealed significantly better overall survival in low-risk patients stratified by nomogram total score (P < 0.001; Supplementary Figure S1). Further, time-dependent ROC analysis comparing the nomogram, risk score, and clinicopathological features (Fig. 5C) demonstrated: risk score outperformed individual clinical features, and the nomogram model achieved superior accuracy versus risk score (1-year: 0.709 vs. 0.619; 3-year: 0.711 vs. 0.680; 5-year: 0.723 vs. 0.709). Further, DeLong test was conducted to compare the AUC of the risk score and the nomogram total score with gender and age (Supplementary Data Sheet 5). The results indicated that at 1 year, the AUC of the nomogram total score was significantly higher than that of age (p = 0.002), and gender (p < 0.001). At the longer-term follow-up points of 3 and 5 years, both the risk score and the nomogram score showed significantly better predictive efficacy than gender alone (all p < 0.05). These results indicate that the F&ERSRGs signature-based nomogram provides enhanced clinical utility for prognostic stratification in gastric cancer.

Fig. 5.

Development and assessment of the nomogram model in TCGA-STAD. A Multivariate Cox analysis demonstrated that risk score was an independent prognostic factor associated with OS in TCGA-STAD dataset. B Nomogram was drawn to predict the survival of the model. C ROC curves comparing the predictive accuracy of the nomogram, risk score, and clinical variables. *** P < 0.001

Identification of DEGs between high-risk and low-risk groups and function enrichment analysis

Differential expression analysis was performed between high- and low-risk groups in TCGA-STAD dataset. We identified 2,909 DEGs with a significance threshold of adjusted P < 0.05 and |log₂FC| > 1. Among these, 2,669 genes were upregulated, and 240 genes were downregulated, as visualized in the volcano plot (Fig. 6A). The complete lists of upregulated and downregulated genes are provided in Supplementary Data Sheet 4. Gene Ontology (GO) enrichment analysis categorized results into three functional domains: biological processes (significantly enriched terms included external encapsulating structure organization, extracellular matrix organization, and extracellular structure organization), cellular components (significantly enriched terms included collagen-containing extracellular matrix and synaptic membrane), and molecular functions (notably enriched terms included extracellular matrix structural constituent, glycosaminoglycan binding, extracellular matrix structural constituent conferring tensile strength, and heparin binding) (Fig. 6B). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis revealed significant enrichment in neuroactive ligand-receptor interaction, cytoskeleton regulation in muscle cells, calcium signaling pathway, hormone signaling, and cAMP signaling pathway (Fig. 6C). Gene Set Enrichment Analysis (GSEA) further identified these pathways as significantly enriched in high-risk samples, with ridge plot visualization demonstrating prominent enrichment of calcium signaling, cAMP signaling, cytoskeleton in muscle cells, hormone signaling, and neuroactive ligand-receptor interaction pathways (Fig. 6D). The enrichment patterns of key pathways in high-risk groups are detailed in GSEA plots (Fig. 6E).

Fig. 6.

Functional enrichment analysis of DEGs between risk groups. A Volcano plot displaying 2,909 DEGs identified between the high- and low-risk groups in TCGA-STAD dataset. B GO enrichment analysis of DEGs between high- and low-risk groups. C KEGG enrichment analysis of DEGs between two groups. D Ridgeplot of KEGG by GSEA. E GSEA plot highlighting significantly enriched pathways

Immune signatures between high-risk and low-risk groups

To further characterize the immune microenvironment disparities between the high- and low-risk groups, we compared the enrichment scores of tumor microenvironment (TME)-related cell signatures. The results indicated significantly elevated signature score in the high-risk group for T cell-related signatures [T cells Bindea et al. [28], T cell accumulation, T cell exhaustion, T cell regulatory [29] and tumor-associated macrophage-related signatures [Macrophages Rooney et al., Macrophages Bindea et al. [28], TAM_Peng_et_al [29] (Supplementary Figure S2). Further assessment of TME characteristics revealed an immunosuppressive, exclusive, and exhausted TME phenotype in high-risk patients (Supplementary Figure S3A–C). Moreover, high-risk patients showed elevated scores in epithelial–mesenchymal transition (EMT) (Supplementary Figure S3D). In contrast, the low-risk group showed higher scores in cell cycle, DNA damage response (DDR), mismatch repair (MMR), and homologous recombination pathways (Supplementary Figure S3E), implying potential heightened sensitivity to immunotherapy. Collectively, these findings indicate a profoundly immunosuppressive TME in the high-risk group. Subsequent evaluation of immune cell infiltration within the TCGA cohort was then assessed. The results of ESTIMATE algorithm demonstrated elevated stromal score, Immune score and ESTIMATE score in the high-risk group (Fig. 7A). Using the CIBERSORT algorithm, we quantified the proportions of 22 immune cell types per sample and compared these between low- and high-risk groups (Fig. 7B). Significantly reduced infiltration of follicular helper T cells (P < 0.01), resting NK cells (P < 0.01), and M0 macrophages (P < 0.05) was observed in high-risk patients. Conversely, high-risk samples exhibited markedly increased fractions of monocytes (P < 0.001), M2 macrophages (P < 0.0001), resting dendritic cells (P < 0.0001), and resting mast cells (P < 0.001). We further analyzed the correlation between risk scores and expression levels of key immune checkpoint genes, including PDCD1 (PD-1), CD274 (PD-L1), CTLA4, PDCD1LG2 (PD-L2), HAVCR2 (TIM-3), LAG3, and TIGIT [30]. Notably, PDCD1LG2, HAVCR2 and TIGIT were upregulated in the high-risk group (Supplementary Figure S4A), and the expression of PDCD1LG2 correlated positively with risk score (Supplementary Figure S4B).

Fig. 7.

Immune signatures between high-risk and low-risk groups. A Difference of immune infiltration between high- and low-risk in the TCGA-STAD dataset analyzed by ESTIMATE algorithm. B The proportion of immune cell components analyzed by CIBERSORT in the TCGA-STAD dataset. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, ns, not significant

Relationship between risk scores and response to chemotherapy

To identify effective therapeutics for patients in high-risk group, we further analyzed chemosensitivity differences between risk groups using the GDSC database. High-risk patients exhibited significantly reduced sensitivity to AZD4547, LGK974, Ulixertinib, and Wee1 inhibitor, but heightened sensitivity to sepantronium bromide (Fig. 8).

Fig. 8.

F&ERSRGs signature predicts chemosensitivity. A sepantronium bromide; B AZD4547; C LGK974; D Ulixertinib; E Wee1 inhibitor

Protein expression patterns of the prognostic signature in clinical samples

To investigate the protein-level basis of our signature genes, we examined their expression in stomach normal tissues and gastric cancer tissues using immunohistochemistry (IHC) data from the HPA database (Fig. 9). CHAC1 was expressed at a high level in stomach normal tissues and had diverse expression levels in gastric cancer tissues, ranging from not detected and low to medium expression. MYB was expressed at a high to medium level in both stomach normal tissues and gastric cancer tissues. SNCA was not detected in both stomach normal and GC tissues. ALB was expressed at a low and medium to high levels in stomach normal tissues and had diverse expression levels in gastric cancer tissues, ranging from not detected and low to medium expression. FABP4 was expressed at a medium level or not detected in stomach normal tissues but was not detected or expressed at low level in gastric cancer tissues. Data for NOX4 were lacking and therefore not presented.

Fig. 9.

Immunohistochemistry images of seven genes in normal and gastric cancer samples

Transcriptional expression of the signature genes in paired clinical samples

To explore the transcriptional expression of the signature genes in clinical samples, we performed qRT-PCR in paired gastric cancer and adjacent normal tissues (Fig. 10). The results showed that the expression of NOX4 and MYB was elevated in GC tissue, whereas the expression of CHAC1, SNCA, and FABP4 was decreased. No significant difference was observed in the expression of ALB and CAV1 between normal and GC tissues.

Fig. 10.

Expression of seven F&ERS-related signature genes in normal and GC tissue samples. *P < 0.05, **P < 0.01, ***P < 0.001, ns, not significant

Discussion

GC, a highly prevalent and lethal digestive tract malignancy, continues to pose a significant clinical challenge. Although immunotherapy and molecular targeted therapies have improved outcomes in recent years, the prognosis for patients with advanced GC remains poor. The TNM staging system, as defined by the American Joint Committee on Cancer (AJCC), plays a critical role in guiding prognostic evaluation and treatment strategies [31], yet current stratification methods are often inadequate for predicting outcomes due to tumor heterogeneity [3, 4]. Therefore, there is an urgent need to identify novel and effective molecular biomarkers to enhance prognostic precision and guide personalized treatment. To address this unmet need, we developed a risk model based on ferroptosis and ERS.In recent years, ferroptosis and ERS have garnered wide attention due to their roles in regulating tumor growth, invasion, metastasis and chemoresistance through several mechanisms in cancers [32–37]. Accumulating evidence further suggests that ferroptosis and ERS are interconnected through shared regulatory pathways, with their crosstalk contributing to disease pathogenesis—including cancer progression [17–19]. For instance, it is indicated that ferroptotic agents induce endoplasmic reticulum (ER) stress and augment the expression of the pro-apoptotic molecule PUMA via the ER stress-mediated PERK-eIF2α-ATF4-CHOP signaling pathway [38]. Guo et al. demonstrated the ferroptosis induced by anlotinib through ER stress indicating that targeting ferroptosis may be a promising therapeutic approach for anaplastic thyroid cancer [39]. In gastric cancer, it is reported PB2 exerts its antitumor effects by inducing endoplasmic reticulum stress, promoting ferroptosis, and triggering oxidative stress-related cell death [40]. Despite these advances, to the best of our knowledge, no studies to date have investigated a prognostic model based on ferroptosis and ERS in gastric cancer.

In this study, we integrated expression profiles of F&ERSRGs from the TCGA gastric cancer (GC) dataset and identified 7 genes to construct a novel prognostic model using LASSO regression analysis. The resulting F&ERSRGs signature was established as an independent prognostic factor for GC with a significant divergence in survival outcomes observed between high- and low-risk groups. Notably, the hazard ratio (HR) of the risk score in multivariate Cox regression analysis exceeded that of tumor stage, indicating that the risk score may hold greater prognostic value than conventional staging. A nomogram incorporating age, gender, tumor stage and risk score exhibited strong predictive accuracy for 1-, 3-, and 5-year overall survival in GC patients. This integrative tool shows potential in supporting clinical decision-making and facilitating personalized treatment strategies.

We further reviewed the biological functions of seven signature genes. NADPH oxidase 4 (NOX4) promotes ferroptosis by oxidative stress-induced lipid peroxidation via the impairment of mitochondrial metabolism [41]. It is reported that NOX4 inhibition attenuates ERS, evidenced by reduced levels of ERS-related proteins and their transcripts, directly supports a functional role for NOX4 in promoting the ERS response [42]. It has been observed that NOX4 plays a pivotal role in sustaining the immune-suppressive phenotype of cancer-associated fibroblasts (CAFs) within tumors. Inhibition of NOX4 enhances immunotherapeutic efficacy by alleviating the exclusion of CD8 + T cells mediated by CAFs [43]. As a γ-glutamyl cyclotransferase, CHAC1 facilitates the degradation of glutathione, thereby influencing calcium signaling pathways and mitochondrial functionality [44]. Furthermore, CHAC1 plays a pivotal role in modulating intracellular reactive oxygen species (ROS) levels, which ultimately affects the sensitivity of gastric cancer cells to chemotherapy [45]. Specifically, ATF4/CHAC1 axis regulating ER stress-dependent ferroptosis [19]. MYB serves as a transcriptional conductor, governing two essential elements of T cell exhaustion: the attenuation of effector functions and the sustained maintenance of self-renewal potential [46]. SNCA plays a pivotal role in the profound immunosuppression observed in cases of malignant tumor ascites, and inhibiting SNCA has shown significant efficacy in markedly enhancing the immune status of the host [47]. In breast cancer, SNCA inhibits epithelial-mesenchymal transition (EMT) and the spread of metastasis with its expression levels potentially serving as valuable indicators for predicting chemosensitivity and assessing the immune microenvironment associated with this malignancy [48]. ALB exerts a significant influence on clear cell renal cell carcinoma through the activation of the endoplasmic reticulum stress (ERS) and the modulation of the immune microenvironment [49]. It is reported CAV1 plays a key role in regulating the interactions between tumors and their host environment, through facilitating tumor progression, metastasis, chemoresistance, and cellular survival [50]. Furthermore, it is indicated that CAV1 serves as a crucial molecular conduit linking ketogenic metabolism to ferroptosis in pancreatic cancer [51]. Additionally, CAV1 alleviates endoplasmic reticulum stress and suppresses pyroptosis via regulation of the FXR/NR1H4 and its downstream effectors ABCG5/ABCG8 [52]. The inhibition of FABP4 offers a safeguard against acute kidney injury (AKI) induced by rhabdomyolysis by alleviating endoplasmic reticulum (ER) stress and mitigating mitochondrial dysfunction [53]. The role of FABP4 in lipid metabolism is pivotal in modulating breast cancer stem cells activity and influencing the progression of triple negative breast cancer [54].

The enrichment analysis revealed that samples in the high-risk group exhibited significant enrichment in the calcium signaling pathway. Ca²⁺ signaling is closely implicated in ferroptosis and endoplasmic reticulum stress. Ca²⁺ primarily influence ferroptosis through the regulation of reactive oxygen species (ROS) and glutathione (GSH) concentrations [55]. Besides, the imbalance of iron homeostasis in ferroptosis is often accompanied by irregularities in calcium signaling. The excessive intracellular calcium can serve as an indicator of the dysregulation of the cellular redox environment, which is marked by iron accumulation during the initial stages of ferroptosis. Such conditions may precipitate disturbances in calcium homeostasis and signaling pathways [56]. The disruption of cellular homeostasis triggers adaptive responses to ERS, prominently including the unfolded protein response (UPR). Central to these stress responses, Ca²⁺ signaling is vital, given that the ER serves as the principal storage organelle for Ca²⁺ and acts as a crucial source for cellular signaling. Within the ER, a multitude of proteins are dedicated to the processes of Ca²⁺ import, export, storage, inter-organelle movement of Ca²⁺, and the replenishment of Ca²⁺ stores in the ER [57].

The interplay among immune cell infiltration, stromal remodeling, and immune checkpoint expression significantly shapes both patient survival and responses to immunotherapy [58–60]. The identification of robust, immune-related biomarkers that accurately reflect tumor biology and inform treatment strategies is essential for advancing clinical decision-making and improving patient outcomes [61, 62]. As ferroptosis and ERS play significant roles in remodeling the immune environment in cancer, we further compared immune landscapes between the high- and low-risk groups. An immunosuppressive signature was identified in the high-risk group, prompting us to speculate that these patients may respond poorly to immunotherapy, given that immune checkpoint inhibitors (ICI) are less effective in such contexts [63]. Further supporting this notion, the expression of immune checkpoint molecules HAVCR2, PDCD1LG2, and TIGIT was significantly upregulated in the high-risk group, and PDCD1LG2 expression positively correlated with the risk score. Future investigations incorporating tumor mutation burden (TMB) and human leukocyte antigen (HLA) status are warranted to fully delineate the relationship between our risk score and immunotherapy outcomes.

Analyzing immune cell infiltration in different risk groups of gastric cancer (GC) patients provides valuable insights into the patient-specific immune landscape. Our analysis revealed that the high-risk group exhibited reduced infiltration of M0 macrophages but elevated infiltration of M2 macrophages, along with enrichment of tumor-associated macrophage (TAM)-related gene signatures. These findings suggest a predisposition toward M2 macrophage polarization in high-risk patients. The results indicates that M2 macrophage polarization is more likely to appear in high-risk group. Polarized M2 macrophages have been demonstrated to facilitate the formation of a premetastatic niche and to enhance the metastasis of gastric cancer by secreting TGFB1 [64]. Ye et al. found that exosomal let-7 g-5p, derived from gastric cancer cells, mediated by SERPINE1 can induce the macrophage M2 polarization, which promotes the gastric cancer progression [65]. Therefore, we hypothesize that the polarization of M2 macrophages may contribute to unfavorable outcomes observed in high-risk patients.

Chemoresistance remains a leading cause of mortality in cancer patients. To assess the association between our F&ERSRGs signature and therapeutic response, we analyzed drug sensitivity data from the GDSC database. Our analysis revealed that patients in the high-risk group showed reduced sensitivity to several agents, including AZD4547, LGK974, ulixertinib, and Wee1 inhibitors. Conversely, sepantronium bromide, a selective survivin suppressant, was predicted to exhibit greater efficacy in the high-risk group. It is reported that sepantronium bromide holds the potential to treat gemcitabine-resistant bladder cancer and head neck squamous cell carcinoma [66, 67]. Therefore, further investigation into its efficacy in high-risk gastric cancer patients is warranted. These results suggest that the F&ERSRGs signature could serve as a predictive biomarker for chemotherapy response.

Key strengths of our work include the use of well-curated public datasets (TCGA and GEO) with substantial sample sizes to construct and validate a reproducible prognostic signature. Moreover, the model’s dependence on a limited number of genes enhances its translational potential by lowering economic and technical barriers to implementation. Clinically, the combined nomogram not only improves prognostic stratification but also helps guide tailored therapeutic strategies, highlighting its broad utility in gastric cancer treatment.

The robust discriminatory power of the F&ERSRGs signature across multiple retrospective cohorts supports its potential for clinical translation. To realize this potential, a key next step is to establish a fixed and universally applicable risk score cutoff for prospective use. While cohort-specific medians were employed herein for unbiased validation, a definitive threshold should be derived and locked based on a large, well-characterized reference cohort or optimized in a prospective pilot study. This predefined cutoff will ensure stable risk classification in future clinical settings, enabling consistent patient stratification and facilitating the design of interventional trials tailored to different risk groups.

Several limitations of this study should be acknowledged. First, further external validation in prospective, multi-center clinical cohorts is necessary to confirm the generalizability and clinical applicability of the prognostic model. Second, the reliance on public genomic databases inherently restricts the availability of comprehensive clinical annotations—such as detailed treatment history and molecular subtypes—which may influence prognostic accuracy. Finally, although we have quantitatively verified the expression of signature genes via qPCR in clinical samples, the biological mechanisms through which these F&ERSRGs modulate the tumor microenvironment and affect immunotherapy response remain unclear and warrant further investigation through functional assays in vitro and in vivo.

In conclusion, this study developed and validated a novel ferroptosis and endoplasmic reticulum stress-related gene signature for prognostic prediction in gastric cancer. The model demonstrated robust independence and predictive accuracy through rigorous external validation. These results not only provide mechanistic insights into the roles of F&ERS in GC but also offer a potential clinical tool to aid in individualizing treatment strategies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Author Contributions: Hongyu Chen contributed to the main data analysis and wrote the first draft of the main manuscript. Zhilei Chen prepared the main data. Ke Cao performed the quantitative real-time PCR and related analysis. Zhenjun Wang contributed to the conception and design of the study, and was responsible for supervision. All authors reviewed the manuscript.

Funding

None.

Data availability

The datasets analyzed in this study are publicly available in the following repositories: The RNA-seq data for the TCGA-STAD cohort is available via the UCSC Xena platform (https://xenabrowser.net/). The gene expression profiles for the validation cohorts are available in the Gene Expression Omnibus (GEO) under the accession numbers GSE15459, GSE62254, and GSE84437 (https://www.ncbi.nlm.nih.gov/geo/).

Declarations

Ethics approval and consent to participate

The study protocol for the use of human tissue samples was reviewed and approved by the Ethics Committee of Shanghai Outdo Biotech Company (China). Informed consent was obtained from all participants, and all procedures were conducted in compliance with the ethical standards of the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.