Integrating multiple omics and machine learning to reveal the prognostic value of endoplasmic reticulum stress gene FKBP10 in gastric cancer

Introduction

Stomach adenocarcinoma (STAD) constitutes a major contributor to the global burden of cancer mortality, constituting a formidable clinical challenge due to the severe paucity of validated prognostic biomarkers and efficacious treatment strategies. This disease, ranking as the fifth most common and fifth deadliest cancer globally, imposes a substantial burden on worldwide health (1). The low early-detection rate of gastric cancer (GC) results in frequent late-stage diagnoses, ultimately contributing to diminished survival rates (2). Consequently, identifying prognostic biomarkers and building robust predictive models is urgently needed to improve outcomes. This is particularly compelling given the established correlation between immune cell infiltration and cancer prognosis, especially in highly metastatic malignancies like GC (3). Clinical evidence demonstrates that immune checkpoint inhibitors (ICIs) have significantly improved treatment outcomes for patients with advanced GC (4), though heterogeneous responses to immunotherapy remain a significant challenge (5). Overall, the discovery of novel gene targets provides a practical foundation for developing targeted therapies and paves the way for advancing precision medicine in GC.

Sustained endoplasmic reticulum stress (ERS), driven by pathological conditions including Helicobacter pylori infection, epigenetic alterations, and tumor microenvironment acidification, disrupts proteostasis in gastric carcinogenesis (6). As a key cellular quality control mechanism, ERS is driven by the accumulation of unfolded proteins. It activates adaptive programs to maintain proteostasis and cellular adaptability (7). The team led by Wen Liu confirmed that the endoplasmic reticulum (ER) responds to stimuli that disrupt its homeostasis by activating a signaling network called the unfolded protein response (UPR), restoring cellular balance and determining cell fate through three key sensors (8). Diseases such as lung cancer, ovarian cancer and GC have all been linked to dysregulated ERS in their pathogenesis (9,10). In addition, studies have shown that 3-bromopyruvate (3-BP) is a hexokinase inhibitor that induces ERS in gastrointestinal tumors and disrupts ER function through the accumulation of free radicals or reactive oxygen species and protein misfolding, leading to apoptosis of human large cell carcinoma cells (11). New insights into the interplay between ERS and tumor immunity emerge from these findings, which collectively underscore its critical regulatory function.

Despite these advances, existing studies on ERS in cancer have predominantly focused on malignancies such as colorectal cancer and cholangiocarcinoma, leaving its role in GC relatively underexplored. However, the majority of existing prognostic models for GC are constructed using conventional Cox regression analyses, often relying on restricted gene panels or single-omics data. This methodological approach may fail to adequately represent the intricate biological complexity of tumors. Notably, few studies have systematically integrated multi-omics data with machine learning algorithms to construct ERS-related prognostic signatures and elucidate their immunomodulatory roles in the gastric tumor microenvironment. This represents a critical gap, as integrative approaches offer the potential to uncover more robust and biologically informative biomarkers by leveraging complementary dimensions of genomic, transcriptomic, and clinical data. Furthermore, while associations between ERS and immune infiltration have been suggested, a systematic dissection of ERS-related molecular subtypes and their immune landscape at single-cell resolution has yet to be performed in GC.

To address these gaps, the present study employs a multi-omics integrative framework combined with machine learning to systematically characterize ERS-associated gene signatures in GC. This study systematically identified an ERS-associated gene signature through comparative analysis of GC patients and controls. A robust 14-gene prognostic model was established using a combination of Cox regression analyses and machine learning, demonstrating significant predictive value. The reliability of the model was assessed using multiple methods, including receiver operating characteristic (ROC) analysis, decision curve analysis (DCA), and calibration curves. Following consensus clustering that defined two GC subtypes, their functional pathways, immune microenvironment, and checkpoint profiles were elucidated. The expression dynamics of relevant signature genes were then interrogated across the tumor cellular atlas at single-cell resolution. We further investigated how FKBP10 reshapes cell communication and found that upregulation of FKBP10 expression can enhance interactions between other cells and epithelial cells. Importantly, we further interrogated the expression dynamics of signature genes at single-cell resolution, providing novel insights into their cellular origins and intercellular communication networks within the tumor microenvironment. Functional experiments confirmed that silencing FKBP10 in GC cell lines (HGC27 and MKN1) significantly suppressed proliferation and induced cell cycle arrest, underscoring its oncogenic role. Through this integrated analytical strategy, our study not only advances the understanding of ERS biology in GC but also establishes a foundation for more individualized prognostic assessment and immunotherapeutic stratification. This integrative approach provides a more comprehensive perspective on ERS-related gene signatures in GC, suggesting potential avenues for future translational research. We present this article in accordance with the TRIPOD reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2817/rc).

Methods

Data collection and preprocessing

The overall design and workflow of this study are illustrated in Figure 1, which provides an overview of the methodology.

Figure 1 The data analysis process. CCK-8, Cell Counting Kit-8; DEG, differentially expressed gene; ERS, endoplasmic reticulum stress; GTEx, Genotype-Tissue Expression; IHC, immunohistochemical; TCGA, The Cancer Genome Atlas; WB, Western Blot.

The data were sourced from The Cancer Genome Atlas (TCGA) (https://tcga-data.nci.nih.gov/tcga/) and Genotype-Tissue Expression (GTEx) databases (https://www.gtexportal.org/), and the RNA sequencing dataset used in this analysis contained 410 samples (of which 24 samples with missing survival information were not included in the subsequent model construction). Batch effect correction was applied to the TCGA and GTEx cohorts using the ComBat function from the sva R package. We screened 1,111 ERS-related genes from GeneCards and Gene Set Enrichment Analysis (GSEA) databases for subsequent analysis. The GSE66229 and GSE14208 datasets in the GEO database are used as external validation queues.

Gene annotation files (gff3) for GENCODE releases v22 and v33 were acquired. Based on these annotations, ENSG_IDs were linked to GeneSymbols, with median expression calculated for multi-mapped genes. Subsequently, rows/columns with >50% missing values were filtered out, and the expression matrix was log₂ (x + 1) transformed for normalization.

Screening of ERS-associated differentially expressed genes (ERSDEG)

ERSDEG were identified by intersecting the results of a DESeq2 analysis (|log₂ fold change| >1, P_adj<0.05) with a curated database of ERS-related genes from GeneCards and GSEA.

Protein-protein interaction (PPI) network analysis

A PPI network was constructed employing the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database (interaction score >0.4) and subsequently visualized and analyzed using Cytoscape (version 3.9.1).

Cox regression analysis

Univariate and multivariate Cox proportional hazards regression analyses were utilized to identify key genes associated with STAD prognosis, with results summarized in forest plots.

Machine learning

Construct a prognostic model using least absolute shrinkage and selection operator (LASSO) regression. By introducing a penalty term to reduce the coefficients of features with low correlation, this technique achieves variable selection, thereby improving the model’s generalization ability. The coefficient of 14 genes used to construct the model can be found in Table S1.

Construction and validation of a risk assessment model

LASSO regression was utilized to derive a risk assessment model by calculating coefficient weights for each ERSDEG. Each patient’s risk score was calculated as the sum of the products of gene expression and their corresponding coefficients: risk score = Σ (gene expression × coefficient). We stratified patients into high-risk and low-risk groups using the median risk score as the cutoff. For external validation, the model was evaluated using the GSE66229 and GSE14208 datasets from the GEO database. To further assess the model’s predictive performance, we conducted DCA and time-dependent ROC analysis. The prognostic stratification based on overall survival was validated using Kaplan-Meier analysis with the log-rank test. Subsequently, a clinically applicable column chart was developed that combines risk score with clinical pathological variables (including age, gender, clinical stage, pathological stage, risk score), and calibrated by plotting the relationship between observed values and predicted probabilities.

Consensus clustering

We applied unsupervised consensus clustering (ConsensusClusterPlus R package) to define molecular subtypes of STAD based on ERSDEG expression. The clustering was run for 1,000 iterations, setting an 80% resampling proportion each time. The stability of clusters and the optimal k value were then evaluated using the cumulative distribution function (CDF) and delta area plot. To visualize patient distribution and validate the cluster segregation, principal component analysis (PCA) was performed on the identified subtypes, projecting them into two-dimensional space to reveal the inherent grouping structure.

Immune infiltration analysis

The immune cell distribution across STAD subtypes and its correlation with the risk score were evaluated using four computational algorithms: ssGSEA, CIBERSORT, ESTIMATE, and Xcell. Using these tools, we profiled immune cell infiltration in the tumor microenvironment (TME) and examined its correlation with our prognostic model. Furthermore, the ESTIMATE algorithm was employed to calculate stromal, immune, and combined (ESTIMATE) scores, with higher scores denoting increased stromal/immune cell presence. This served to profile the tumor immune microenvironment and evaluate its prognostic relevance.

Enrichment analysis

To delineate the biological characteristics of STAD subtypes, Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were conducted using the “ClusterProfiler” R package, with results integrated and visualized in Metascape. Subsequently, Gene Set Variation Analysis (GSVA) was employed to detect subtle, continuous variations in pathway activity. In parallel, we conducted GSEA with the MSigDB C5-GO gene set to assess the enrichment of predefined pathways across phenotypes, yielding a broad view of the transcriptomic landscape.

Single-cell RNA sequencing (scRNA-seq) analysis

To profile ERS expression at single-cell resolution, we analyzed tumor scRNA-seq data from the TISCH2 database (dataset: GSE183904). We processed the scRNA-seq data with Seurat (version 4.3.0) and annotated cell types by referencing established marker genes. Dimensionality reduction was carried out with uniform manifold approximation and projection (UMAP), and differential FKBP10 expression among different cell types was analyzed employing the “FindMarkers” function. This approach allowed us to explore the immune-related prognostic features of STAD within the tumor microenvironment.

Cell-cell communication analysis

To explore how the ERS-related pathway modulates intercellular communication in the GC microenvironment, we utilized CellChat (v2.2.0) to reconstruct and analyze the cellular interaction network. Cells from the GEO dataset were stratified into high- and low-ERS expression groups based on the median gene expression score. After generating Seurat objects for each group, we identified overexpressed ligands, receptors, and their corresponding pairs, and computed communication probabilities. Pathway-level analyses and signal network visualizations were performed using CellChat’s integrated tools. Subsequently, we compared the quantity and intensity of ligand-receptor interactions between the two groups to assess how different cell populations influence adjacent tumor cells.

Research subjects and materials

This study included a retrospective analysis of 90 GC patients admitted to The Affiliated Hospital of Xuzhou Medical University from May 2018 to November 2022. Inclusion criteria: diagnosed with primary GC through pathological examination; successfully performed radical gastrectomy for GC and met the R0 resection criteria; age ≥18 years old; the postoperative follow-up data is complete. Exclusion criteria: malignant tumors from other organ sources combined; cases of perioperative death or loss to follow-up. Collect disease-related information of selected patients: the medical record system collects patient data, including gender, age, tumor clinical stage, lymph node metastasis and vascular infiltration, and histopathological type; telephone follow-up to track the postoperative survival status and cause of death of patients; obtain wax blocks of the patient’s pathological tissue from the pathology department of The Affiliated Hospital of Xuzhou Medical University. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The Ethics Committee of The Affiliated Hospital of Xuzhou Medical University reviewed and granted approval for this study protocol (ethics approval No. XYFY2023-KL277-01). The obtainment of all human tissue samples was contingent upon prior informed consent from the patients.

Immunohistochemical (IHC) analysis

IHC analysis was performed on 90 pairs of STAD and adjacent normal paraffin embedded tissue sections from the Pathology Department of The Affiliated Hospital of Xuzhou Medical University. After dewaxing and rehydration, sections were subjected to antigen repair to block endogenous peroxidase activity, and serum was blocked with 3% bovine serum albumin (BSA) at room temperature for 30 minutes. Then, the slices were incubated overnight with 1:200 diluted anti-FKBP10 primary antibody (ProteinTech, 12172-1-AP) at 4 ℃, followed by standard IHC procedures for secondary antibody incubation, diaminobenzidine (DAB) staining, and hematoxylin counterstaining. The staining is scored based on intensity and percentage of positive cells, and the final score is calculated as the product of these two parameters. Observe stained sections under an optical microscope and perform semi quantitative analysis of staining intensity using ImageJ software (National Institutes of Health, USA).

Cell culture and transfection

In the present study, two human GC cell lines, HGC27 and MKN1, were utilized. The cells were routinely cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum (FBS, Gibco) and 100 U/mL penicillin-0.1 mg/mL streptomycin (Gibco), and maintained at 37 ℃ in a humidified incubator with 5% CO₂.

Two small interfering RNAs (siRNAs) targeting FKBP10 (si-FKBP10) and a negative control siRNA (si-NC) were synthesized by Genepharma (Shanghai, China). The sequences were as follows: si-FKBP10#1, 5'-CCACACCTACAATACCTATAT-3'; si-FKBP10#2, 5'-CCACTACAATGGCTCCTTGAT-3'; and si-NC, 5'-TTCTCCGAACGTGTCACGT-3'. For transfection, cells were seeded in six-well plates and grown to approximately 80% confluence, after which siRNA delivery was performed using Lipofectamine 2000 in accordance with the manufacturer’s protocol.

Cell Counting Kit-8 (CCK-8) assays

Exponentially growing HGC27 and MKN1 cells were seeded into 96-well plates at a density of 1.5×10⁴ cells per well. To assess proliferation over time while avoiding repeated measurements that could compromise cell viability, separate plates were prepared for each time point (0, 1, 2, 3, and 4 days) and incubated in parallel. Following cell attachment (approximately 24 hours after seeding), the medium was replaced with fresh complete medium supplemented with 10% CCK-8 reagent (C917226, Macklin, Shanghai, China). After a 2-hour incubation at 37 ℃ (12), absorbance was measured at 450 nm using a microplate reader.

Western blotting analysis

Total protein was extracted from harvested cells, and the concentration was determined using a bicinchoninic acid assay (BCA) protein assay kit (Beyotime, Inc., Shanghai, China). Equal amounts of protein were resolved by SDS-PAGE on precast gels and then electrotransferred onto PVDF membranes (Sigma Aldrich, Shanghai, China). After blocking with 5% skim milk to block non-specific binding, the membranes were incubated overnight at 4 ℃ with primary antibodies targeting FKBP10 (Wuhan Sanying Biotechnology, Cat No. Ag2814, 1:1,000 dilution, Wuhan, China) and GAPDH (1:1,000 dilution). The membranes were subsequently incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies for 2 h at room temperature. Immunoreactive bands were visualized using an enhanced chemiluminescence (ECL) detection kit (Beyotime, Inc.).

Flow cytometry analysis

Cell cycle progression and apoptosis in HGC27 and MKN1 cells were evaluated by flow cytometry. Cells were plated in 24-well plates at a density of 1×10⁴ cells per well and cultured for 24 hours. For apoptosis assessment, cells were collected using 0.25% trypsin-EDTA, washed with phosphate-buffered saline (PBS, pH 7.4), and stained with 5 µL Annexin V-FITC and 5 µL propidium iodide (PI) for 15 minutes at room temperature in the dark. Flow cytometric analysis was performed on a BD FACSCalibur, with data analyzed using FlowJo software (v10.6.0). For cell cycle analysis, cells were fixed in 75% ice-cold ethanol overnight at 4 ℃. After washing with PBS, cells were incubated with 50 µg/mL PI containing 10 mg/L RNase A for 30 minutes at 37 ℃ in the dark. DNA content was then measured using the same flow cytometer and analyzed with FlowJo software.

Statistical analysis

The statistical analyses were performed using R software (version 4.4.2) (https://www.r-project.org/) and Statistical Product and Service Solutions (SPSS; version 31.0). Adjusted P values were calculated for all tests to assess statistical significance.

Results

Identification of prognostic ERS-related genes

Differential expression analysis of tumor and normal samples from the TCGA and GTEx databases revealed significant alterations in the gene expression profile of STAD. Analysis identified a total of 6,500 differentially expressed genes (DEG), comprising 2,924 downregulated genes and 3,576 upregulated genes (Figure 2A). Furthermore, we identified 305 genes overlapping between ERS-related genes and the DEG (Figure 2B). Univariate Cox regression was applied to evaluate the prognostic value of these genes, based on patient survival status, follow-up time, and gene expression data. This analysis revealed 33 prognostically significant genes. Their hazard ratios and confidence intervals are depicted in the accompanying forest plot (Figure 2C). A subsequent KEGG enrichment analysis of these genes was performed, revealing significant enrichments in pathways including “protein processing in ER”, “cAMP signaling pathway”, and “AGE-RAGE signaling pathway” (Figure 2D). Additionally, protein protein interaction (PPI) analysis reveals the close relationship between these prognostic related genes (Figure 2E).

Figure 2 Identification of prognosis-related endoplasmic reticulum stress genes. (A) Volcano plot showing differential gene expression between tumor and normal tissues; black dots represent genes with no significant difference. (B) Venn diagram analysis was performed to identify overlapping genes between differentially expressed genes and endoplasmic reticulum stress-related genes. (C) Forest plot displaying the P values and risk ratios of the 33 genes with prognostic significance. (D) Bar plot visualizing the KEGG enrichment spectrum derived from the 305-gene set. (E) Exploring prognostic related genes through PPI analysis. CI, confidence interval; DEG, differentially expressed gene; ERS, endoplasmic reticulum stress; HR, hazard ratio; KEGG, Kyoto Encyclopedia of Genes and Genomes; PPI, protein-protein interaction.

Molecular subtyping based on 33 ERS-related genes

Using the expression data from the 33 prognostic genes, consensus clustering was performed on 410 TCGA-STAD patients. The analysis yielded an optimal k of 2 (Figure 3A,3B), stratifying patients into two distinct groups. Kaplan-Meier analysis confirmed a significant survival advantage for Cluster 1 over Cluster 2 (P<0.05; Figure 3C). Furthermore, immune landscape analysis revealed substantial inter-subtype differences. Specifically, Cluster 1 exhibited significantly higher stromal, immune, and ESTIMATE scores (Figure 3D-3F), alongside enriched proportions of activated B and CD8⁺ T cells (Figure 3G), consistent with its more favorable prognosis.

Figure 3 Based on 33 prognostic ERS-related genes, distinct patient subgroups were identified, revealing significant heterogeneity in the tumor immune microenvironment between the clusters. (A) Consensus clustering at k=2: heatmap showing the two stable patient subgroups (Cluster 1 and Cluster 2) under the optimal partitioning of the data. (B) CDF curves: illustrating consensus stability for k values ranging from 2 to 9, used to determine the optimal k. (C) Survival disparity between subtypes: Kaplan-Meier analysis indicated a significant overall survival advantage for Cluster 1 over Cluster 2 (P<0.005). (D-F) Tumor microenvironment scoring: comparative analysis showed significantly higher stromal score, immune score, and ESTIMATE score in Cluster 1, reflecting distinct microenvironment composition. (G) Immune cell infiltration profiling: ssGSEA quantified the abundance profiles of 28 immune cells, highlighting subtype-specific immune landscapes. *, P<0.05; **, P<0.01; ns, non-significant. CDF, cumulative distribution function; ERS, endoplasmic reticulum stress; ssGSEA, single sample gene set enrichment analysis.

DEG and their functional annotations

Differential expression analysis between the two clusters identified 490 DEG, with 386 downregulated and 104 upregulated in Cluster 1 versus Cluster 2 (Figure 4A). Subsequent GSEA using the C5-GO gene set revealed that the DEGs were predominantly enriched in pathways related to DNA replication and positive regulation of interferon production (Figure 4B). Further pathway analyses indicated that these DEGs were frequently downregulated in key biological contexts, including calcium signaling and neuroactive ligand-receptor interaction (KEGG, Figure 4C), as well as in processes like muscle contraction (GO biological process, Figure 4D). The bubble plot illustrates the relationship between P values and scores of relevant pathways (Figure 4E,4F).

Figure 4 Analysis of differentially expressed genes and functional enrichment. (A) Volcano plot depicting the differentially expressed genes between the two subtypes. (B) GSEA plots revealed distinctive functional signatures across the C1 and C2 subtypes. (C) Circle plot of significant signaling pathways identified by KEGG analysis. (D) KEGG enrichment analysis: molecular functions depicted by bubble chart. (E) Circle plot of over-represented biological processes from GO analysis. (F) GO enrichment analysis: molecular functions depicted by bubble chart. GO, Gene Ontology; GSEA, Gene Set Enrichment Analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Cross-cluster profiling of functional features and immune contexts

Comparative analysis of the 33 prognostic genes demonstrated distinct expression profiles between the two groups (Figure 5A). The results of gene GSVA based on the GO dataset indicate that C1 is mainly involved in core life activities within the cell nucleus, such as chromosome separation and DNA replication. On the contrary, C2 is mainly related to the structure of the extracellular matrix, intercellular connections, and support (Figure 5B,5C). The Xcell algorithm exhibited a marked elevation in the level of infiltration of CD4⁺ naive T cells, CD8⁺ T cells, and ImmuneScore in C1 (Figure 5D).

Figure 5 Characterization of immune infiltration and biological functions across ERS-based clusters. (A) Expression of 33 prognostic ERS-related genes: boxplots comparing the expression levels of the 33 differentially expressed endoplasmic reticulum stress-related genes between the two clusters. (B) GSVA of GO gene sets: bar graph presenting the GSVA enrichment scores for GO terms, highlighting functional differences between clusters. (C) GSVA of KEGG pathways: bar chart displaying the GSVA results for KEGG pathways, comparing pathway activities across clusters. (D) Immune cell infiltration profiles: boxplots illustrating the differential infiltration levels of immune cell types between clusters C1 and C2, as calculated by the Xcell algorithm. ns, non-significant; *, P<0.05; **, P<0.01; ***, P<0.001. ERS, endoplasmic reticulum stress; GO, Gene Ontology; GSVA, Gene Set Variation Analysis; KEGG, Kyoto Encyclopedia of Genes and Genomes.

Construction of risk model

To identify the optimal lambda (λ), we utilized LASSO regression coupled with 10-fold cross-validation, which selected 14 prognostic genes (Figure 6A,6B). Using these genes, a risk score model was constructed. Based on the median risk score, patients in the TCGA cohort were stratified into high-risk (n=193) and low-risk (n=193) groups. Scatter plots confirmed a positive correlation between increasing risk scores and patient mortality (Figure 6C,6D). The high-risk group exhibited significantly worse survival compared to the low-risk group in the Kaplan-Meier analysis (Figure 6E). The model demonstrated moderate predictive accuracy, with an area under the curve (AUC) of 0.702 in time-dependent ROC analysis (Figure 6F).

Figure 6 Establishment and assessment of a prognostic signature. (A) LASSO coefficient profiles: traces showing the coefficient paths of genes across lambda values, with 14 feature genes selected at the optimal point. (B) Selection of optimal lambda (λ): plot of cross-validation error, where the optimal λ (minimum criteria) is indicated by vertical dashed lines. (C) Population distribution based on risk scores associated with the 14-gene model. (D) The scatter plot illustrates the difference in patient survival status between the high-risk group and the low-risk group. (E) Comparison of overall survival between high- and low-risk cohorts. (F) ROC curve analysis: receiver operating characteristic curves evaluating the model’s predictive accuracy (AUC >0.7). AUC, area under the curve; LASSO, least absolute shrinkage and selection operator; ROC, receiver operating characteristic.

Independence of the established risk model

To evaluate the independence and clinical relevance of the prognostic model, we performed comprehensive subgroup and regression analyses comparing risk scores across key clinical parameters. No significant differences in risk scores were observed by gender (Figure 7A) or node (N) stage (Figure 7B). In contrast, risk scores varied significantly across age groups (Figure 7C), metastasis (M) stage (Figure 7D), tumor (T) stage (Figure 7E), and overall stage groups (Figure 7F). Notably, the prognostic stratification remained robust within subgroups stratified by age (Figure 7G,7H) and gender (Figure 7I,7J), with high-risk patients consistently exhibiting worse survival. Given the observed associations with T stage and overall stage, we assessed the predictive capacity of the risk score for these clinical classifications via ROC analysis (Figure 7K,7L). The results suggest limited staging predictive utility, possibly owing to sample size, warranting future model refinement. Yet, they establish the model’s independent prognostic value and its relationship with principal clinical covariates in GC.

Figure 7 Clinicopathological features stratified by prognostic risk score. (A-F) Risk score distribution across clinical groups: boxplots comparing risk scores by age, gender, and TNM stage. Significant differences were observed across age groups (C), M stage (D), T stage (E), and overall stage (F), but not by sex (A) or N stage (B). *, P<0.05; **, P<0.01; ***, P<0.001; ns, non-significant. (G,H) Age-stratified survival: Kaplan-Meier curves within age subgroups confirm consistent prognostic separation by risk score. (I,J) Sex-stratified survival: survival analysis within male and female subgroups maintains significant outcome differences between risk groups. (K,L) ROC curve analysis for stage prediction. AUC, area under the curve; ROC, receiver operating characteristic; TNM, tumor-node-metastasis.

Construction, calibration, and external validation of an integrated nomogram (Figure 8)

Figure 8 Constructing and validating a STAD prognostic nomogram. (A-C) Model calibration: calibration curves at 1-, 2-, and 3-year intervals demonstrate strong agreement between predicted and observed survival probabilities. (D) Integrated prognostic tool: the presented nomogram combines the molecular risk score with key clinicopathological features for individualized outcome prediction. (E) Clinical utility assessment: in comparison to other strategies, the nomogram offered a greater net clinical benefit for decision-making, according to DCA. (F,G) External validation: ROC curves from two independent GEO cohorts (GSE66229 and GSE14208) confirm the model’s robust predictive performance and generalizability. AUC, area under the curve; DCA, decision curve analysis; GEO, Gene Expression Omnibus; M, metastasis; N, node; OS, overall survival; ROC, receiver operating characteristic; STAD, stomach adenocarcinoma; T, tumor.

We built a nomogram to better predict prognosis in gastric adenocarcinoma (STAD), thereby enhancing model performance (Figure 8D). This model enables individualized prediction of survival probability at 1-, 2-, and 3-year intervals. Calibration plots demonstrate the degree of agreement between observed and ideal values (Figure 8A-8C). Collectively, the calibration curves confirmed the model’s clinical validity, while DCA indicated that combining multiple predictors (as in the nomogram) provides greater net benefit than any single variable (Figure 8E). These results support the nomogram’s utility as a practical prognostic tool in clinical settings. Additionally, to verify the model’s robustness, validation was performed using the GSE66229 and GSE14208 datasets (Figure 8F,8G). All AUC values were no less than 0.6.

Characterization of the tumor microenvironment in risk groups

To elucidate the biological basis of the prognostic risk stratification, GSEA was performed. The high-risk group was significantly enriched in pathways related to epithelial cell differentiation and tumor microenvironment, whereas the low-risk group showed prominence in protein transport and folate metabolism (Figure 9A). Further analysis of immune checkpoint molecules revealed significantly higher expression of HMGB1 in Group 1, and elevated levels of TGFB1 and TLR4 in Group 2 (Figure 9B). Additionally, risk scores derived from the model varied significantly between DEG-based clustering groups (Figure 9C). We also revealed the differences in gene mutation profiles between high- and low-risk groups through the display of mutation waterfall plots (Figure 9D,9E).

Figure 9 Enrichment analysis, mutation analysis, and immune infiltration analysis. (A) GSEA visualization of enrichment differences between high- and low-risk groups. (B) HMGB1, TGFB1, and TLR4 levels are different between Group 1 and 2, with statistical significance represented by **, P<0.01; ***, P<0.001. (C) Risk score reveals differences between Group 1 and 2. ****, P<0.0001. (D,E) TMB displays gene mutations in high- and low-risk groups. GSEA, Gene Set Enrichment Analysis; TMB, tumor mutation burden.

Profiling gene expression in tumors using scRNA-seq

The GC single-cell dataset GSE183904 was retrieved from the GEO database for comprehensive analysis. Cellular landscape revealed through UMAP included principal lineages such as immune and stromal compartments (Figure 10A), with clear separation between tumor and normal epithelial cells (Figure 10B). A compositional pie chart revealed T cells as the most abundant population (Figure 10C). Violin plot analysis showed differential expression of FKBP10 across cell types, with pronounced upregulation specifically in fibroblasts from tumor samples (Figure 10D). This spatial expression pattern was further corroborated by its localization within tumor-associated cell clusters (Figure 10E). The marked upregulation of FKBP10 in tumor-associated fibroblasts suggests a pivotal role for this stromal subset in GC pathogenesis.

Figure 10 Analyzing STAD at single-cell resolution using RNA sequencing. (A) UMAP map of major cell lineage clustering. (B) UMAP plot depicting the distribution of tumor and normal cells. (C) Cellular composition: pie chart representation. (D) Violin plot depicting FKBP10 expression across distinct cell types in tumor versus normal samples. ****, P<0.0001; ns, non-significant. (E) FKBP10 expression projected in UMAP space. STAD, stomach adenocarcinoma; UMAP, uniform manifold approximation and projection.

FKBP10-driven remodeling of the cellular communication network in GC

To explore the role of FKBP10 in modulating cell communication within the GC microenvironment, we performed CellChat analysis using tissue samples stratified by FKBP10 expression levels. Compared to the low-expression group, samples with high FKBP10 expression exhibited more extensive intercellular interactions (Figure 11A), particularly between fibroblasts and immune cells (Figure 11B). Furthermore, we evaluated the contribution of ligand-receptor pairs from various cell populations to epithelial cells under both conditions. The results revealed that both the number and signaling intensity of ligand-receptor pairs were markedly elevated in the FKBP10-high group relative to the FKBP10-low group (Figure 11C,11D).

Figure 11 Elevated FKBP10 expression in tumor tissues markedly reshaped intercellular communication networks. (A) Higher FKBP10 levels were associated with increased overall interaction density and complexity compared to low-expression conditions. (B) Differential interaction analysis revealed enhanced (red) and diminished (blue) cell-cell communication in the FKBP10-high group relative to the FKBP10-low group. Additionally, ligand-receptor pair analysis was performed for both FKBP10-high (C) and FKBP10-low (D) groups, highlighting distinct signaling patterns associated with FKBP10 expression.

The expression level of FKBP10 is positively correlated with the malignant progression parameters of GC

Patients were stratified into low- and high-FKBP10 expression groups based on the median expression level in tumor tissues. Clinicopathological analysis revealed that the high-expression group had a significantly greater proportion of advanced-stage (III–IV) disease and vascular invasion compared to the low-expression group (Table 1). Furthermore, elevated FKBP10 expression was strongly associated with worse clinical outcomes. Kaplan-Meier survival analysis demonstrated significant stratification, with the high expression group exhibiting markedly inferior overall survival throughout follow-up (log-rank P=0.002; Figure 12). These findings highlight the potential of FKBP10 as a prognostic biomarker for identifying GC patients with unfavorable survival.

Figure 12 Kaplan-Meier analysis demonstrated a significant overall survival difference between patients with high versus low FKBP10 expression.

Reduced proliferation of FKBP10-silenced cells

To validate the findings from bioinformatics analyses, IHC staining was performed on five paired tumor and adjacent normal tissue samples with clear immunoreactivity. FKBP10 expression was markedly elevated in cancer tissues compared with matched non-tumor counterparts (P<0.001; Figure 13A). To further explore its functional role in GC progression, HGC27 and MKN1 cells were transfected with si-FKBP10, which efficiently knocked down both messenger RNA (mRNA) and protein levels of FKBP10 (Figure 13B). FKBP10 knockdown significantly inhibited proliferation in both cell lines by day 4, as determined by CCK-8 assays (Figure 13C,13D). Additionally, flow cytometric analysis demonstrated that FKBP10 depletion led to a reduced proportion of cells in S-phase, indicating cell cycle arrest (Figure 13E).

Figure 13 FKBP10 expression is elevated in gastric cancer tissues and cell lines, and its depletion significantly reduces tumor cell viability. (A) Analysis of five paired tissue samples by immunohistochemistry revealed significant upregulation of FKBP10 in gastric tumor tissues relative to adjacent normal controls (antibody source: ProteinTech; dilution ratio: 1:200). ***, P<0.001. (B) Western blot analysis confirms increased FKBP10 protein levels in HGC27 and MKN1 cells. (C) Flow cytometry was used to analyze cell cycle distribution in HGC27 and MKN1 cells after FKBP10 silencing. (D,E) CCK-8 assays demonstrate that FKBP10 downregulation significantly reduces the viability of HGC27 and MKN1 cells. ****, P<0.0001. CCK-8, Cell Counting Kit-8; PT, peritumoral tissue; siFKBP10, small interfering RNA targeting FKBP10; siNC, small interfering RNA targeting negative control; T, tumor.

Discussion

GC is a leading cause of cancer-related mortality worldwide (13). Delaying treatment initiation adversely affects the prognosis of GC patients, irrespective of primary or metastatic status (14). While immunotherapy has become a standard treatment for STAD globally, patient responses remain heterogeneous, with many deriving limited or no clinical benefit (15). GC management continues to face persistent challenges, including high invasiveness, suboptimal clinical outcomes, and poor prognosis. Meanwhile, ERS is recognized as a key activator of the UPR—a compensatory mechanism that alleviates ER burden and restores proteostasis by attenuating protein synthesis and augmenting folding capacity (16). The role of this process in GC, particularly given the tumor’s invasive nature, remains unclear.

This study presents a comprehensive assessment of ERS-related genes in STAD, integrating multi-omics and machine learning approaches. Based on advanced computational algorithms, we identified ERS genes associated with STAD prognosis and selected 33 with differential expression for molecular subtyping. Consensus clustering analysis (k=2) of the TCGA STAD cohort stratified patients into two subgroups (C1 and C2) with significant prognostic differences. Analyses using CIBERSORT, ssGSEA, and ESTIMATE algorithms revealed markedly distinct immune infiltration landscapes between the subgroups. High immune cell infiltration in the C1 subgroup was associated with a more favorable prognosis. Comprehensive immunological and functional enrichment analyses indicated that these observed disparities in both immune context and pathway activity suggest the potential value of immunotherapy for STAD. Furthermore, evaluation of immune checkpoint genes revealed significant inter-subtype differences between C1 and C2, informing potential strategies for immune checkpoint modulation in GC. Collectively, by integrating ERS with GC prognosis and classification, this work offers novel perspectives on the complexity of tumor heterogeneity in STAD.

Guided by these initial observations, we implemented a machine learning framework comprising LASSO regression and univariate/multivariate Cox regression for further analysis. The process yielded a set of 14 key genes (e.g., FKBP10), enabling the development of a prognostic risk model. The model’s reliable accuracy was validated in an independent external dataset. Our findings reveal that increased risk scores are directly associated with poorer patient prognosis. Marked differences between the risk stratifications were revealed by employing functional enrichment, immune cell profiling, and mutation landscape visualization. Additionally, we constructed a prognostic nomogram for GC patients and validated its predictive accuracy using calibration curves.

FKBP10/precursor protein A/laminin A axis is associated with nuclear atypia and promotion of muscle invasion (17). This correlation between increasing risk scores and mortality underscores the model’s effectiveness. FKBP10, a validated prognostic marker, is also directly implicated in cancer progression (18). Chen et al.’s study showed that FKBP10 promotes M2 polarization of macrophages through the MEK/ERK/CXCL8 axis and promotes tumor progression in clear cell renal cell carcinoma (19).

Analysis revealed distinct pathway enrichments: KRAS-related in C1 and MYOGENESIS in C2. Supporting these findings, Wu et al. noted that ARID1A mutations could improve ICI efficacy, and KRAS inhibitor success in other cancers suggests a targeted approach for Krukenberg tumors (20). The study by Xing et al. showed that early postoperative persistent circulating tumor DNA (ctDNA) positivity (KRAS and TP53 mutations, etc.) combined with Lauren diffuse classification and serosal infiltration can effectively identify high-risk patients with GC peritoneal metastasis (21). In addition, Shen et al. found that TGF-β signaling activation drives lymph node metastasis in GC (22). Li et al. found that ACTG1 promotes proliferation and metastasis of COL21A1 mediated by epithelial-mesenchymal transition (EMT) in GC (23). In parallel, Lv et al. found that cell proliferation-related gene sets—including myogenesis, mitotic spindle, and G2M checkpoint—were enriched in SPIN1-high GCs (24). These convergent findings strengthen the evidence for the central contribution of epithelial cells to GAC pathogenesis and prognosis, suggesting that targeting the related pathways holds therapeutic potential.

Single-cell sequencing has advanced significantly in recent years (25). Single-cell tools allow for in-depth profiling of the tumor immune microenvironment, enhancing our understanding of its role in modulating immune responses (26). Employing single-cell sequencing, we delineated the FKBP10 expression profile in GC, revealing its putative role in shaping the immune microenvironment and suggesting its relevance for immunotherapeutic approaches.

To further investigate how FKBP10 reshapes cell communication, we conducted CellChat analysis on samples stratified by FKBP10 expression levels. High expression of FKBP10 enhanced intercellular interactions, particularly between fibroblasts and immune cells. Ligand receptor pair analysis showed that high expression of FKBP10 significantly enhanced the interaction between other cells and epithelial cells (12).

IHC validation confirmed elevated FKBP10 expression in GC tissues, supporting its potential as an immune-related diagnostic and prognostic biomarker. Functional assays demonstrated that FKBP10 silencing suppressed tumor cell proliferation and induced S-phase arrest, highlighting its role in GC progression. Collectively, these data highlight ERS as a promising target for deeper exploration in the context of STAD.

This study contributes to a broader understanding of ERS-related gene expression patterns and immune interactions in STAD. First, it elucidates the central role of ERS in the origin and progression of GC. Second, it reveals novel cellular interactions—among immune cells, tumor cells, and cancer stem cells—within the gastric adenocarcinoma tumor microenvironment, thereby offering fresh insights into the biology of this disease (27). Secondly, the application of advanced machine learning enhanced both the precision of key gene/pathway identification and the reliability of prognostic evaluation. The integrative approach revealed distinct molecular and immune features associated with ERS-related gene expression in STAD.

Although this study integrates bioinformatics analyses with preliminary in vitro experimental validation, several inherent limitations should be acknowledged. The conclusions are drawn from secondary analysis of public repositories, and the experimental validation, while supportive, is currently limited to in vitro cellular assays. The model’s performance in an external cohort is tempered by the selection bias possible in retrospective public data, a limitation that raises concerns about the generalizability of the findings. Although we integrated scRNA-seq data and preliminary in vitro experiments to demonstrate the proliferative effect of FKBP10 in GC cells, the specific molecular regulatory networks involving FKBP10 and its crosstalk with ERS pathways remain to be fully elucidated. Furthermore, while the prognostic model was validated in multiple external datasets, the inherent selection bias and incomplete clinical information in public databases may limit its broader applicability. In particular, imbalanced sample sizes in certain subgroup analyses may affect statistical power and the stability of the conclusions. It is noteworthy that our key observation on FKBP10 converges with existing published evidence. FKBP10 upregulation in GC, as shown by Chen et al. using public datasets and in vitro models, was found to promote lymph node metastasis through mechanisms involving enhanced adhesion and interstitial transformation (28). Limitations of this work include the moderate significance of select findings, which may reflect constraints in sample size and data diversity; this underscores the importance of subsequent validation in larger independent populations. Despite known associations with GC for some genes, key aspects such as their exact molecular mechanisms, regulatory networks, and functional crosstalk are not yet fully understood. The application of machine learning introduces potential algorithmic bias, necessitating comprehensive sensitivity analyses to mitigate the risk of false discoveries and strengthen conclusion validity. These considerations highlight the necessity of prospective, multi-center validation to assess model robustness across diverse populations, as well as functional studies to advance our understanding of ERS in GC biology. Driven by the goal of improving patient prognosis, future work must focus on mechanistic decipherment, methodological refinement, and the promotion of clinical translation to enable more effective therapeutic strategies.

Despite the promising prognostic value of the 14-gene risk model and the potential biological role of FKBP10, several challenges must be addressed to facilitate their clinical translation. First, the model was developed and validated using retrospective public datasets. Although preliminary in vitro experiments support the oncogenic role of FKBP10, prospective, multi-center studies with standardized protocols are urgently needed to confirm the model’s generalizability and clinical utility. Second, the transition from a bioinformatics-derived signature to a clinically applicable assay requires the development of standardized, cost-effective platforms (e.g., RT-PCR or NanoString) to ensure reproducibility across different laboratories and clinical settings. Third, while our findings implicate FKBP10 in GC cell proliferation and cell cycle progression, its precise molecular mechanisms, upstream regulators, and downstream effectors remain to be fully characterized. Deeper mechanistic exploration—including target specificity, potential off-target effects, and validation in in vivo models—is essential to assess its therapeutic viability.

Looking forward, integrating this risk model with established clinicopathological factors could refine patient stratification for immunotherapy or adjuvant treatment decisions. Ultimately, the clinical utility of these findings must be confirmed through rigorously designed prospective trials. Future efforts should focus on translating these bioinformatics-driven insights into practical tools for precision oncology, achieved through continuous refinement of the predictive model, comprehensive functional exploration of key genes such as FKBP10, and sustained collaboration between computational biologists and clinical researchers.

Conclusions

Using integrated computational and experimental approaches, this work investigated the pivotal role of ERS in GC progression, prognosis, and molecular subtyping. The FKBP10-based prognostic model was validated externally, and significant multi-dimensional differences were observed between the resulting risk groups. Concurrently, analysis identified two distinct molecular subtypes, designated C1 and C2. By mapping the single-cell expression profiles of key genes in the tumor microenvironment, this study reveals context-specific patterns that offer new leads for immunotherapy research. In addition, CellChat analysis showed that FKBP10 drives immune remodeling by enhancing intercellular communication and altering key signaling pathways in the tumor microenvironment. Furthermore, in vitro experiments confirmed that FKBP10 is upregulated in GC tissues, and its silencing significantly suppresses tumor cell proliferation and induces cell cycle arrest, supporting its functional role in disease progression. Our findings identify FKBP10 as a potential prognostic marker, establishing a basis for future research into ERS-related genes within precision oncology. Guided by the objective of improving clinical outcomes, subsequent research must address the underlying mechanisms of ERS, refine the associated methodologies, and evaluate their translational potential to pioneer more effective treatment avenues.

Acknowledgments

The author thanks the staff of Xuzhou Medical University for their contributions in managing various clinical, operational, and laboratory aspects of research. Artificial intelligence tool was used to translate and polish the English content of this article.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2817/rc

Data Sharing Statement: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2817/dss

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2817/prf

Funding: This study was supported by The Affiliated Hospital of Xuzhou Medical University “Pairing Assistance” Scientific Research Project (FXJDBF2024211).

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-2025-1-2817/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study protocol was approved by Medical Ethics Committee of The Affiliated Hospital of Xuzhou Medical University (ethics approval No. XYFY2023-KL277-01). Informed consent was obtained from all participants involved in this study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Liu XX, Wei Y, Guo Y, et al. An integrated AI-enabled system using One Class Twin Cross Learning for early gastric cancer detection. Front Oncol 2025;15:1623394. [Crossref] [PubMed]
2. Li X, Zhang W, Chen Y, et al. Serum tRF-18-HR05X6D2 may serve as a promising potential diagnostic biomarker for gastric cancer. Clin Exp Med 2025;26:15. [Crossref] [PubMed]
3. Jiang Z, Zhuang X, Guo J, et al. Integrative single-cell atlas unveils heterogeneity and prognostic value of cancer-associated fibroblasts in gastric cancer. Front Oncol 2025;15:1559489. [Crossref] [PubMed]
4. Udagawa S, Ooki A, Osumi H, et al. The clinical utility of an mGPS and SII combined score in patients with advanced gastric cancer treated with nivolumab monotherapy. ESMO Gastrointest Oncol 2025;8:100177. [Crossref] [PubMed]
5. Shao Y, Yu X, Yan J, et al. Molecular subtyping and functional characterization of gastric cancer using arginine metabolism-related genes. Front Cell Dev Biol 2025;13:1732490. [Crossref] [PubMed]
6. Gui Z, Qian L, Gao L, et al. Telomere-protecting protein 1 promotes gastric cancer cell metastasis via enhancing endoplasmic reticulum stress. J Biol Chem 2026;302:110998. [Crossref] [PubMed]
7. Kong J, Zhu Z, Hu Y, et al. Momordicine I induces ER stress and inhibits OSCC by targeting ribosomal proteins. J Zhejiang Univ Sci B 2025;27:164-80. [Crossref] [PubMed]
8. Liu W, Gupta A, Kerin M, et al. H19 Is a PERK-Regulated Long Non-Coding RNA That Fine-Tunes UPR Signalling and Inhibits Endoplasmic Reticulum Stress-Induced Cell Death. Int J Mol Sci 2026;27:1658. [Crossref] [PubMed]
9. Luo Q, Gao X, Meng P, et al. Endoplasmic reticulum stress in non-small cell lung cancer: a review of therapeutic agents, mechanistic insights, and implications for therapy. Front Cell Dev Biol 2025;13:1693023. [Crossref] [PubMed]
10. Lu X, Zhu L, Zhang X, et al. A Prognostic Nomogram Based on an Immunogenic Cell Death and Endoplasmic Reticulum Stress-Related Gene Signature for Ovarian Cancer. Int J Womens Health 2025;17:4891-903. [Crossref] [PubMed]
11. Liang G, Ma Y, Deng P, et al. Hexokinases in gastrointestinal cancers: From molecular insights to therapeutic opportunities. Semin Oncol 2025;52:152351. [Crossref] [PubMed]
12. Huang Y, Cao D, Zhang M, et al. Exploring the impact of PDGFD in osteosarcoma metastasis through single-cell sequencing analysis. Cell Oncol (Dordr) 2024;47:1715-33. [Crossref] [PubMed]
13. Zhang M, Xie J, Yao S, et al. The expression signature, prognostic significance and immune cell infiltration of the OAS gene family in gastric cancer. Sci Rep 2025;15:39682. [Crossref] [PubMed]
14. Zhang L, Xu X, Wang Y, et al. Research hotspots and frontiers in the tumor microenvironment of gastric cancer: a bibliometric review from 2005 to 2024. Transl Cancer Res 2025;14:8329-46. [Crossref] [PubMed]
15. Li W, Xu M, Cheng M, et al. Current Advances and Future Directions for Sensitizing Gastric Cancer to Immune Checkpoint Inhibitors. Cancer Med 2025;14:e71065. [Crossref] [PubMed]
16. Zhao Q, Zhang HJ, Han MM, et al. Leveraging ANXA1 to enhance recombinant protein yields in CHO cells: A UPR-Mediated bioprocessing approach. Synth Syst Biotechnol 2026;12:197-208. [Crossref] [PubMed]
17. Zhao X, Wang J, Tian S, et al. FKBP10 Promotes the Muscle Invasion of Bladder Cancer via Lamin A Dysregulation. Int J Biol Sci 2025;21:758-71. [Crossref] [PubMed]
18. Li Y, Zhang C, Chen J, et al. FKBP10 expression and TP53 mutation predict prognosis and chemotherapy response in triple-negative breast cancer. Discov Oncol 2025;16:1409. [Crossref] [PubMed]
19. Chen JW, Li JY, Feng HQ, et al. FKBP10 promotes M2 polarization of macrophage via MEK/ERK/CXCL8 axis and facilitates tumor progression in clear cell renal cell carcinoma. Int J Biol Sci 2026;22:1807-33. [Crossref] [PubMed]
20. Wu J, Jiang S, Shen Q, et al. Postoperative metastatic Krukenberg tumors with ARID1A and KRAS mutations in a patient with gastric cancer treated with oxaliplatin and tegafur: A case report. Oncol Lett 2025;29:262. [Crossref] [PubMed]
21. Xing F, Shi J, Mu Y, et al. Blood ctDNA-specific markers predict the risk of peritoneal metastasis for advanced gastric cancer. Discov Oncol 2025;17:60. [Crossref] [PubMed]
22. Shen K, Liu D, Su J, et al. TRIM26 deficiency drives gastric cancer lymph node metastasis via TGF-β signaling activation and modulates gemcitabine response. Front Cell Dev Biol 2026;14:1746425. [Crossref] [PubMed]
23. Li S, Lan B, Pan L, et al. ACTG1 facilitates proliferation and epithelial-mesenchymal transition-mediated metastasis via COL21A1 in gastric cancer. iScience 2026;29:114666. [Crossref] [PubMed]
24. Lv BB, Cai L, Xiao Y, et al. Gene expression profiling of SPIN1 in gastric cancer: insights into tumorigenesis and potential therapeutic targets. Front Genet 2025;16:1510849. [Crossref] [PubMed]
25. Zhao S, Liu S, Shao W, et al. Research Progress Regarding the Use of Single-Cell Sequencing Technology in Analyzing Tumor Endothelial Cell Pathophysiology. Int J Mol Sci 2025;26:11128. [Crossref] [PubMed]
26. Qu H, Sun J, Wang G, et al. SLC12A7 serves as a prognostic and immunotherapeutic biomarker identified by multi-omics analysis. Front Immunol 2025;16:1712579. [Crossref] [PubMed]
27. Li J, Ren J, Zhang W, et al. Correlation of G protein-coupled receptor and tumor microenvironment with gastric cancer outcomes and therapies. Chin Med J (Engl) 2026;139:65-82. [Crossref] [PubMed]
28. Chen R, Jiang L. A novel m6A/m5C/m1A/m7G-related classification and risk signature predicts prognosis and reveals immunotherapy inclination in gastric cancer. Transl Cancer Res 2024;13:3285-98. [Crossref] [PubMed]