1 Introduction

Gastric cancer (GC) is one of the most prevalent malignancies worldwide, ranking fifth in incidence and fourth in mortality rate [1]. As a complex multifactorial disease, GC is influenced by environmental and genetic factors, with Helicobacter pylori (Hp) infection being an important factor [2, 3]. Despite increasing treatment options for GC, such as surgery, radiotherapy, and chemotherapy, patient prognosis remains poor [4]. Currently, the commonly used biomarkers are less effective in determining appropriate treatment for GC, and the discovery of new prognostic biomarkers is necessary to identify the different risk levels of patients with GC and new targets for immunotherapy.

Programmed cell death (PCD) is actively executed by intracellular death programs [5]. Common PCD types include apoptosis, pyroptosis, necroptosis, ferroptosis, and other forms of cell death [6]. Recently, the discovery of a completely new form of cell death, disulfidptosis, has garnered considerable attention. A previous study found that SLC7A11-mediated cystine uptake promotes cell death under glucose starvation conditions [7]. Excessive depletion of intracellular NADPH during this process leads to an abnormal accumulation of disulfides, such as cystine, in cells, and this induces disulfide stress and is often highly toxic to cells [8, 9]. This event ultimately leads to disulfidptosis by triggering the collapse of the actin cytoskeleton. Disulfidptosis may have a substantial effect on tumor progression. However, the precise mechanism underlying this phenomenon remains poorly understood, and further research is needed to provide new directions for the clinical treatment of tumors.

In this study, we investigated the expression of and mutations in disulfidptosis-related genes (DRGs) using GC-related data obtained from public databases [Gene Expression Omnibus (GEO) and The Cancer Genome Atlas (TCGA)]. We divided GC samples into clusters according to their gene expression and the available clinical data. Subsequently, a prognostic model based on six-gene biomarkers was constructed to better predict patient prognosis, and the potential associations between the risk score (RS) and disease progression, immune characteristics, antitumor drug sensitivity, and tumor microenvironment (TME) were determined. Finally, we selected GPC3, a less studied model gene, for further study and found that it has potential as a therapeutic target for GC. In addition, we conducted essential experiments (qRT-PCR, western blot, apoptosis, transwell, migration, and invasion) to demonstrate the potential of GPC3 as a biomarker in gastric cancer cells.

2 Materials and methods

2.1 Source and preprocessing of data

The research methodology of this study is shown in Fig. 1. RNA sequencing (RNA-seq) data, tumor mutation data, gene copy numbers, and clinical information of patients with GC were obtained from the TCGA-STAD database; furthermore, RNA-Seq data and clinical information were obtained from the GSE84437 database. In total, 439 samples (403 samples of patients with GC and 36 normal samples) were collected from the TCGA database, and 483 samples from patients with GC were obtained from the GSE84437 dataset. The data from the two databases were organized into matrix files, and then merged and batch-corrected using “limma” and "SVA" in the R package [10].

Fig. 1
figure 1

Simple flowchart for this article

Full size image

2.2 Unsupervised clustering

Twenty-three DRGs were found to form different DRG clusters [7,8,9, 11]. Based on these genes, an unsupervised clustering analysis of samples of patients with GC was performed using the "ConsensusClusterPlus" R package [12]. The expression of DRGs in different clusters was visualized using the "pheatmap” R package.

2.3 Correlation enrichment analysis and difference analysis of DRG clusters

The “clusterProfiler” R package was used to performed enrichment analyses, including Gene Ontology (GO) term and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses, and the significant results were visualized. The "gene set variation analysis (GSVA)" R package was used to identify differences between DRG clusters in terms of biological function [13]. The two DRG clusters were analyzed using the “limma” R package to identify DEGs. Secondary unsupervised clustering was also performed on these genes using the following filtering criteria: p < 0.05, |fold change| ≥ 1.5 [14]. A heat map was generated to display the general distribution of DEGs among clusters using the "pheatmap” R package.

2.4 Construction and validation of disulfidptosis-related prognostic models

According to the DEGs in different DRG clusters, a univariate Cox regression analysis was performed, and prognostically associated DEGs were obtained. Subsequently, all samples from the combined TCGA and GEO datasets were randomly assigned on a 1:1 scale and designated as the training and test groups, respectively. Lasso–Cox analysis was implemented to identify the genes used to build the prognostic model. According to the model genes, calculations were performed using the following equation to obtain the risk scores of the samples, subsequently categorizing them into two groups according to the scores: high risk (HR) and low risk (LR).

$$Risk Score=\sum \left[{Exp}_{\left(DEGs\right)}\times {coef}_{\left(DEGs\right)}\right]$$

Survival analysis was performed separately using the “survival” R package on both training and test groups to examine whether there was a survival discrepancy between the HR and LR groups. Receiver operating characteristic (ROC) curves were used to assess the precision of the models. Using the R package, a nomogram was established to assess the survival of patients with GC. A calibration curve was constructed to confirm the accuracy of the plot.

2.5 Evaluation of immune cell infiltration and TME

A single-sample gene set enrichment analysis (ssGSEA) was used to score immune cells among different DRG clusters. The "CIBERSORT" tool was employed to determine the level of infiltration of multiple immune cells in GC samples and to visualize the association among model genes, RSs, and immune cells [15]. The tumor mutational burden (TME) scores of the GC samples were computed using the “estimate” R package and visualized via a violin plot to compare the differences in the three scores between the HR and LR groups [16]. We used single-sample Gene Set Enrichment Analysis (ssGSEA) to assess immune cell infiltration across different disulfidptosis-related gene (DRG) clusters. For intra-sample immune cell abundance comparisons, we employed the CIBERSORT tool, which uses linear support vector regression (SVR) for estimation. Notably, CIBERSORT reflects relative abundance within individual samples, and for inter-sample comparisons, CIBERSORT-ABS would be a more appropriate tool. However, due to limitations in available resources, we were unable to apply CIBERSORT-ABS in this study.

2.6 Immunotherapy analysis and drug sensitivity prediction

Tumor immune dysfunction and exclusion (TIDE), TMB, and microsatellite instability (MSI) analyses were performed to investigate the immunotherapy aspects. The strength of tumor immune escape was analyzed according to the TIDE score generated using the Harvard online tool [17]. Data from the URL http://tide.dfci.harvard.edu was downloaded, and violin plots were used to visualize the results between the groups. Mutation data from GC samples were analyzed using the “Maftools” R package and represented in the form of waterfall plots [18]. Box plots and bar charts were plotted using the “ggpubr” R package to visualize the TMB and MSI results between the risk score groups. The IC50 values of several classical antineoplastic drugs were determined using the R package "pRRophetic" (version:0.5) [19]. Sensitivity to drug treatment was compared between the HR and LR groups.

2.7 Single-gene analysis of GPC3

The harmonized and standardized pan-cancer dataset was downloaded from the UCSC (https://xenabrowser.net/) database and the expression data and clinical data of ENSG00000147257 (GPC3) was further extracted from individual samples. The “limma” R package was used for differential expression analysis, the “survival” R package was used to analyze the prognostic relationship between expression and tumor, the “GSVA” R package was used for immune cell infiltration analysis, and visualization was performed with the “ggplot2” R package. Furthermore, gene single-cell analysis was performed using the TISCH (http://tisch.comp-genomics.org/home/) database by selecting “Dataset” from the web page, selecting “STAD” (Stomach Adenocarcinoma) for Cancer type, selecting “Human” for Species, selecting “No treatment” for Treatment, selecting “Primary” for Primary/Metastatic, and finally selecting the GSE167297 dataset for analysis.

2.8 Immunohistochemistry

Immunohistochemical staining results of DEGs in GC tissue and normal gastric tissue was obtained from the HPA database. The website is: http://www.proteinatlas.org.

2.9 Cell culture and transfection

In this study, the human gastric mucosal cell line GES-1, as well as the human gastric cancer cell lines NCI-N87, BGC823, HGC-27, and MKN45, were employed. These cell lines were obtained from the Cancer Institute at the Chinese Academy of Medical Sciences (Beijing, China). Cells were maintained in RPMI 1640 medium, supplemented with 10% fetal bovine serum (FBS) and 1% penicillin–streptomycin (Gibco, Grand Island, NY, USA), and incubated at 37 °C in a humidified atmosphere with 5% CO2. To induce GPC3 overexpression, BGC823 cells were transfected with a GPC3 overexpression plasmid using Lipofectamine Transfection Reagent. The transfection efficiency was subsequently assessed through real-time quantitative PCR.

2.10 Quantitative real-time fluorescence polymerase chain reaction

RNA was isolated using TRIzol reagent (Invitrogen, Thermo Fisher Scientific, Waltham, MA, USA). Subsequently, complementary DNA (cDNA) was synthesized from the extracted RNA using a reverse transcription kit (#R333; Vazyme, Nanjing, China). GAPDH was used as an internal control, and the relative gene expression levels were determined using the 2−ΔΔCt method. Detailed primer sequences utilized in the experiments are provided in Supplementary Table S1.

2.11 Cell proliferation assay

Cell viability was assessed using the CCK-8 assay kit (Seven, China). A 96-well plate was seeded with a suspension containing 5 × 103 cells per well. At 24 h intervals, 10 μL of CCK-8 reagent was added to each well, followed by a 2 h incubation period. Thereafter, the optical density (OD) at 450 nm was determined using a multifunctional enzyme-linked immunosorbent assay reader in order to quantify cell viability.

2.12 Cell apoptosis

Cell apoptosis was detected using an Annexin V-mCherry and SYTOX Green Cell Apoptosis Detection Kit (Beyotime, Shanghai, China) in accordance with the manufacturer’s protocol. Cells were incubated with Annexin V-mCherry and SYTOX Green for 20 min in the dark. Thereafter, cells were washed with PBS and resuspended in 400 µL of binding buffer for 30 min. Apoptosis rates were subsequently measured using a FACSCanto II flow cytometer (BD Biosciences, San Jose, CA).

2.13 Cell migration and invasion assays

Transwell chambers with 8.0 μm pores (Procell Life Science & Technology, China) were used to assess both cell migration and invasion. A 500 μL medium containing 10% FBS was added to the lower chamber, whereas 1 × 104 cells were seeded onto the upper chamber in a serum-free medium. For the invasion assays, the transwell membrane was coated with 1 mg/ml Matrigel (Procell Life Science & Technology, China). After incubating the cells for 24 h at 37 °C, non-migrated or non-invaded cells were gently removed using cotton swabs. The cells that successfully migrated or invaded the lower membrane surface were stained with crystal violet, counted, and then fixed in 4% paraformaldehyde for preservation.

2.14 Western blot

Cell lysates were prepared on ice for 30 min using radioimmunoprecipitation assay (RIPA) buffer. Thereafter, protein concentrations were determined using a bicinchoninic acid (BCA) assay kit (Beyotime, China). For sample preparation, the protein extracts were mixed with 5 × loading buffer (Beyotime, China) at a 1:4 ratio and heated at 100 °C for 10 min. Proteins were separated by 10% SDS-PAGE and transferred onto PVDF membranes (Millipore, USA). The membranes were blocked with 5% skim milk at room temperature for 2 h. After blocking, the membranes were incubated overnight at 4 °C with the following primary antibodies: GAPDH (1:5000; Proteintech, USA), cleaved caspase-3 (1:1000; Abcam, USA), Bcl-2 (1:1000; Abcam, USA), E-cadherin (1:1000; PTM Biolabs, China), and Vimentin (1:1000; PTM Biolabs, China). The membranes were washed thrice for 10 min each at room temperature and incubated with a secondary antibody at room temperature. Protein bands were visualized using enhanced chemiluminescence (Thermo Scientific, USA).

2.15 Statistical analyses

All statistical analyses were performed using R software (version 4.2.1). Statistical significance was set at p < 0.05. The Wilcoxon rank-sum test was used to analyze the differences in DRGs, TME, and risk scores.

3 Results

3.1 Expression and mutation of DRGs in GC

Eighteen DEGs were identified by analyzing the expression of all 23 DRGs in the GC and normal tissue samples (Fig. 2a). All DRGs were highly expressed in the tumor tissue, suggesting that they may be involved in the progression of GC. We obtained immunohistochemical staining results for these DEGs from the Human Protein Atlas (HPA), and the results were consistent with those of previous analyses (Fig. 3). Approximately 138 of the 431 samples (32.02%) contained DRG mutations, with the highest incidence observed in MYH10 (Fig. 2b). More than 50% of the DRGs exhibited an increase in copy number, with IQGAP1 exhibiting the most pronounced increase (Fig. 2c). The copy number variations (CNVs) of the DRGs on the chromosome are shown in Fig. 2d.

Fig. 2
figure 2

Variations in the expression of disulfidptosis-related genes (DRGs) in gastric cancer (GC). a Differential expression of DRGs between normal and tumor tissues. b Mutation frequency of DRGs in patients with GC. c Copy number variation (CNV) status of DRGs in patients with GC. d Status of the CNVs of DRGs on different chromosomes. *p < 0.05, **p < 0.01, ***p < 0.001

Full size image
Fig. 3
figure 3

Protein expression of DRGs in gastric cancer and normal gastric tissue

Full size image

3.2 Clustering of patients with GC based on DRGs

By collating data from the TCGA and GSE84437 datasets, we obtained survival information of 871 patients with GC. Using univariate COX and correlation analyses, we investigated the association between these DRGs and disease prognosis. Fourteen DRGs were found to be negative factors for the prognosis of GC and nine were protective factors (all p < 0.05) (Fig. 4a). These results suggest that crosstalk among different DRGs may be associated with disease prognosis and tumor heterogeneity in patients with GC.

Fig. 4
figure 4

DRG clustering and subsequent analysis of patients with GC. a Interaction among DRGs in GC. The degree of significance of DRGs is indicated by the size of the circles. b Two clusters (k = 2) were established. c Principal component analysis was used to determine consistency within clusters. d Overall survival (OS) analysis between the clusters. e The status of the two DRG clusters in terms of clinical information and DRG expression. f Status of immune cell infiltration. g Gene set variation analysis (GSVA) of the two DRG clusters in terms of biological pathways. *p < 0.05, **p < 0.01, ***p < 0.001

Full size image

To further investigate the relationship between patient prognosis and disulfidptosis, a consensus cluster analysis of the 871 samples was performed, and two clusters were identified (Fig. 4b). The reliability of differentiating these two clusters based on the expression levels of 23 DRGs was verified using principal component analysis (PCA) (Fig. 4c). DRG cluster B exhibited a better overall survival rate than DRG cluster A (Fig. 4d). The heat map showed a greater abundance of highly expressed DRGs in cluster A than in cluster B (Fig. 4e). These results suggest that higher DRG expression may correlate with a poor prognosis in GC.

3.3 Immunological properties and biological behavior of the two DRG clusters

To investigate the effect of DRGs on the TME, we used the ssGSEA to observe differences in the levels of immune cell infiltration in different clusters. Eighteen immune cell types differed in infiltration levels between the DRG clusters (Fig. 4f). To investigate the differences in the biological behavior of different DRG clusters, the KEGG-related GSVA algorithm was used (Fig. 4g). Multiple pathways related to the regulation of the actin cytoskeleton and cardiovascular disease were enriched in DRG cluster A, whereas pathways associated with the metabolism of glutathione, pyrimidine, alanine, aspartate, and glutamate were markedly enriched in DRG cluster B.

3.4 Clustering of patients with GC based on DEGs

To determine the relevant biological behavior of the two disulfidptosis-associated subtypes, 1186 DEGs were identified using the "limma" R package. Enrichment analysis of these DEGs revealed that they were primarily expressed in tumors and cardiovascular-related signaling pathways (Fig. 5a, b). The GC samples were divided into two gene clusters using an unsupervised clustering analysis (Fig. 5c). The survival analysis results indicated that the survival rate of Cluster A was better than that of Cluster B (Fig. 5d). In total, 18 DRGs differed in terms of their expression levels between the gene clusters (Fig. 5e). Patients in gene clusters A and B corresponded to those in DRG clusters B and A (Fig. 5f). The results suggest a good correspondence between these gene clusters and those associated with disulfidptosis.

Fig. 5
figure 5

Gene clusters were identified based on differentially expressed genes (DEGs). a, b Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses of DEGs. c Two gene clusters (k = 2) were obtained using the consensus clustering algorithm. d Differences in the survival status between the gene clusters. e Differential expression of DRGs in the two gene clusters f Clinical information and expression of DEGs in different DRG clusters and gene clusters. *p < 0.05, **p < 0.01, ***p < 0.001

Full size image

3.5 Construction of the prognostic model related to disulfidptosis

First, using the univariate Cox analysis, we obtained 907 prognosis-associated genes from 1186 DEGs among the clusters. Thereafter, we used Lasso–Cox regression analysis to conduct further screening of the 907 genes with prognostic value (Fig. 6a, b), and 6 key genes, namely, HEYL, CD109, GRP, DEFB1, GPC3, and GPA33, were finally identified. A prognosis-related RS model was constructed based on these 6 key genes. RS = (0.1542*HEYL) + (0.2498*CD109) + (0.1306*GRP) + (0.1247*DEFB1) + (0.1231*GPC3) + (− 0.0648*GPA33). The GC samples were classified into HR and LR groups based on the RSs. Among the 23 genes associated with desulfurization, NCKAP1, ACTB, CAPZB, DSTN, FLNA, MYH10, MYL6, MYH9, and TLN1 were expressed at higher levels in the HR group, whereas SLC7A11, GYS1, LRPPRC, CD2AP, FLNB, INF2, and PDLIM1 were expressed at lower levels (all p < 0.05) (Fig. 6c). The survival analysis indicated a better prognosis in the LR group than in the HR group (Fig. 6d). This result was confirmed in both training and test groups (Fig. 6e, f). The ability of the risk score model to evaluate the survival status of patients with GC was validated using ROC curves (Fig. 6g–i).

Fig. 6
figure 6

Prognostic model building and correlation analysis. a, b Lasso regression analysis results and cross-validation errors. c The expression levels of disulfide-related genes in both risk groups. d–f Survival curves for the total sample, training, and test groups. g–i Receiver operating characteristic (ROC) analysis of the total sample, training, and test groups. *p < 0.05, **p < 0.01, ***p < 0.001

Full size image

3.6 Prognostic value and clinical significance of RSs

Analysis of the survival status of the patients from both risk groups revealed that the majority of patients in the HR group died within a shorter period than those in the LR group. Furthermore, we found that the other five model genes, except GPA33, were highly expressed in the HR group (Fig. 7a). This result was validated by trends in survival status and model gene expression in the training and test groups (Fig. 7b, c). DRG cluster B had lower RSs than DRG cluster A (p < 0.001) (Fig. 7d). Among the gene clusters, cluster A had lower RSs than cluster B (p < 0.001) (Fig. 7e). The association among DRG clusters, gene clusters, RSs, and prognosis was demonstrated using a Sankey diagram. The LR group mainly consisted of patients with DRG cluster B and gene cluster A and exhibited a better prognosis. In contrast, most patients with a poorer prognosis were in the HR group, consisting mainly of DRG cluster A and gene cluster B (Fig. 7f). To improve the clinical usefulness of the risk model, a nomogram was established to assess the survival of patients with GC by combining their clinical information (Fig. 7g). The predictive power of the nomogram was assessed using calibration curves and obtained good results (Fig. 7h).

Fig. 7
figure 7

Predictive and clinical value of RSs. a–c Risk profiles, survival status, and expression of model genes in the total sample, training, and test groups. d–e Differences in the RSs between DRG clusters and gene clusters. f Sankey diagram displaying the relationship among DRG clusters, gene clusters, RSs, and survival status. g Nomogram predicting the survival of patients with GC. h Calibration curves of nomograms. ***p < 0.001

Full size image

3.7 Evaluation of immune cell infiltration and TME

It is well known that the TME can significantly influence tumor progression and affect a patient’s response to immunotherapy [20, 21]. We analyzed the degree of infiltration of different immune cells and explored their correlation with the model genes and RSs. To address potential variability among different estimation methods, we conducted a comparative analysis using ESTIMATE and TIMER in addition to ssGSEA and CIBERSORT. Despite some inconsistencies in specific immune cell types, the overall trends in immune infiltration supported our primary findings. These results highlight the importance of method selection in immune cell analysis. The analysis indicated a link between most immune cells and model genes. Ten immune cells significantly correlated with the RS. Among these cells, follicular helper T cells, CD8 T cells, activated memory CD4 T cells, plasma cells, and resting natural killer (NK) cells were negatively associated with the RS, whereas resting memory CD4 T cells, monocytes, resting mast cells, M2 macrophages, and naïve B cells were positively associated with the RS (all p < 0.01) (Fig. 8a). Using ESTIMATE analysis, we found that the immune and ESTIMATE scores were higher in patients in the HR group than in those in the LR group (p < 0.001) and that the difference in stromal scores between the HR and LR groups was not significant (p > 0.05) (Fig. 8b).

Fig. 8
figure 8

Role of risk score in the tumor microenvironment (TME) and immunotherapy. a The relationship among model genes, risk scores, and immune cells. b Violin plot shows the level of TME-related scores in the high-risk (HR) and low-risk (LR) groups. c–e Tumor mutational burden (TMB) and analysis of differences between the groups. f Correlation between TMB and risk scores. g Differences in risk scores among MSS, microsatellite instability (MSI)-L, and MSI-H. h Differences in MSI characteristics between the HR and LR groups

Full size image

3.8 Prediction of immunotherapy

TMB and MSI are valid biomarkers for immunotherapy [22, 23]. By analyzing the somatic mutation data of patients with GC, we found that the percentage of mutations was 87.1% in HR-GC patients (Fig. 8c) and 95.1% in LR-GC patients (Fig. 8d). In addition, the difference between the groups was significant (p < 0.001) (Fig. 8e). TMB was inversely associated with RSs (r = − 0.42, p < 0.001) according to the correlation analysis (Fig. 8f). Specifically, an increase in TMB was associated with decreased RSs. The patients in the microsatellite instability-high (MSI-H) group had lower RSs than those in the microsatellite stability (MSS) and microsatellite instability-low (MSI-L) groups (Fig. 8g). We analyzed the distribution of MSI in the HR and LR groups and found that the LR group had a higher proportion of MSI-H patients (Fig. 8h). Subsequently, the TIDE score was used to determine the likelihood of tumor immune escape in the HR and LR group patients. The analysis revealed that the HR group had higher TIDE scores than the LR group (p < 0.001) (Fig. 9a). Using the stem cell correlation analysis, we found a negative correlation between the stem cell index and RS (r = − 0.59, p < 0.001) (Fig. 9b).

Fig. 9
figure 9

Tumor immune dysfunction and exclusion (TIDE) and drug sensitivity. a Violin plot showing the difference between the risk groups, “***” means p < 0.001. b Correlation between stem cell index and risk score. c–f Sensitivity analysis of four drugs

Full size image

3.9 Analysis of drug sensitivity

To enhance the usefulness of RSs for predicting drug treatment in patients with GC, we analyzed the sensitivity of patients in different risk groups to various drugs. All 87 effective drugs were retrieved according to the genes generated by our model. Four of these drugs were selected for demonstration (Fig. 9c–f), whereas the remaining results are shown in the supplementary material (Supplementary Figures S1–S4). The results revealed that patients in the LR group were more sensitive to ABT888 (veliparib) and BIBW2992 (afatinib), whereas patients in the HR group were more sensitive to AMG706 (motesanib) and dasatinib (p < 0.001).

3.10 Analysis and validation of GPC3

To delve deeper into the role of prognostic models, we analyzed and validated GPC3. We found that GPC3 exhibited low tumor tissue expression in 33 cancers (n = 17) (Fig. 10a). A similar result was observed in terms of TCGA-STAD (Fig. 10b). The results of the qRT-PCR experiments were consistent with the abovementioned findings (Fig. 10c). The ROC curve of GPC3 was analyzed, as shown in Fig. 10d. We performed a prognostic analysis of GPC3 in 33 tumors and indicated that its high expression in these tumors did not have a consistent correlation with prognosis, with a tendency for high expression to be associated with a poorer prognosis in gastric cancer (Fig. 10e). Subsequently, we evaluated GPC3 expression in gastric cancer immune cells and found that high GPC3 expression was associated with high scores in most immune cells and was positively correlated with these immune cells (Fig. 10f, g). We analyzed GPC3 using single-cell sequencing through the GSE167297 dataset and visualized the cellular components (Fig. 11a, b) and clustering of cellular components within tissues (Fig. 11c). We discovered that GPC3 was mainly clustered in the fibroblasts of tumor tissues (Fig. 11d–f).

Fig. 10
figure 10

Pan-cancer analysis of GPC3. a Differential analysis of GPC3 expression between normal and tumor tissues in 33 tumors. b Expression of GPC3 in gastric cancer and paracancerous tissues. c qRT-PCR validation of GPC3 expression in gastric cancer cells and gastric mucosal cells. d AUC curve of GPC3 predicting gastric cancer. e Hazard ratio of GPC3 in different cancers. f Correlation analysis of GPC3 immune cell infiltration. g Differential expression analysis of GPC3 immune cell infiltration. *p < 0.05, **p < 0.01, ***p < 0.001

Full size image
Fig. 11
figure 11

Single-cell sequencing analysis. a, b Content of various cell types. c, d Two-dimensional distribution of cells in tissues. e Two-dimensional distribution of GPC3 in tissues. f Differences in cellular expression between normal and tumor tissues

Full size image

3.11 GPC3 overexpression promoted the malignant behavior of gastric cancer cells

To further investigate the role of GPC3 in gastric cancer, we selected the BGC823 cell line, in which the difference between GPC3 in normal and GC cells is most pronounced. We observed significant overexpression of GPC3 (Fig. 12a), where the proliferative capacity of GC cells (BGC823) was significantly increased after the overexpression of GPC3 (Fig. 12b). The WB results indicated that GPC3 overexpression decreased the expression of the apoptosis-promoting protein, c-caspase3, and increased the expression of the apoptosis-suppressing protein, bcl-2 (Fig. 12c). Flow cytometry quadrant plots indicated that GPC3overexpression inhibited late apoptosis in BGC823 cells (Fig. 12d, e). Overall, GPC3 overexpression resulted in reduced apoptosis in BGC823 cells (Fig. 12f). Transwell assay results revealed that overexpression of GPC3 facilitated the migration and invasion of BGC823 cells (Fig. 12g), displaying a significant difference (Fig. 12h). In addition, WB results indicated that GPC3 overexpression increased vimentin protein expression and decreased E-cadherin expression. These results suggest that GPC3 overexpression was positively correlated with the malignant behavior of gastric cancer cells.

Fig. 12
figure 12

Experimental validation of GPC3. a Overexpression of GPC3. b CCK8 assay to evaluate the proliferative activity of BGC823 cells after overexpression of GPC3. c Western blot detection of apoptosis, migration, and invasion-related proteins after overexpression of GPC3. d–f Detection of apoptosis in BGC823 cells after overexpression of GPC3 by flow cytometry. g, h Transwell assay to evaluate the migration and invasion ability of BGC823 cells before and after overexpression of GPC3. *p < 0.05, **p < 0.01, ***p < 0.001

Full size image

4 Discussion

GC is a common malignant tumor of the digestive system and owing to the high risk of recurrence and metastasis, patients often have a less than satisfactory prognosis [24]. Surgery is the main treatment for GC [25]. However, because of a lack of obvious and concrete signs and symptoms in the initial stages, GC is usually diagnosed at a locally advanced or metastatic stage, which makes surgical treatment less effective [26,27,28]. Chemotherapy is also used as a primary treatment for GC [29]; however, chemotherapy resistance remains challenging in certain cases [30]. Immunotherapy, a treatment option that has emerged in recent years, has increased the treatments available for patients with GC.

The discovery of disulfidptosis as a novel form of cell death offers a new direction for treating tumors. Disulfidptosis is classified as a form of PCD, which is of importance in tumor development [31]. For example, promoting tumor cell apoptosis is an important mechanism in tumor treatment [3, 30]. Therefore, it is possible that disulfides could be used to treat tumors. Several studies have indicated that disulfide disulfiram has positive therapeutic effects against breast, cervical, and colorectal cancers [32,33,34]. To address the limitations of traditional histological and anatomical classification methods in guiding the clinical management of GC, we investigated the potential relationship between disulfidptosis and GC, and identified disulfidptosis-associated subtypes to aid in the individualized treatment of patients with GC.

First, we analyzed the expression levels and genetic variations of 23 DRGs. The expression levels of 18 DRGs differed between normal and tumor tissues, and all 18 genes were highly expressed in tumor tissues. The most common mutation was in MYH10 (8%). Subsequently, we performed an unsupervised cluster analysis on the transcriptome data collected from 871 GC samples and found that clustering worked best when samples were divided into two subtypes: DRG cluster A and DRG cluster B. There were significant differences between DRG clusters A and B in terms of both survival time and the level of immune cell infiltration. We identified two gene clusters by further analyzing 1186 DEGs between the DRG clusters, which exhibited different prognostic statuses.

To develop a more comprehensive prognostic model for assessing the risk of GC, we selected six key genes from the 1186 differentially expressed genes (DEGs) identified between DRG clusters A and B. While these genes—HEYL, CD109, GRP, DEFB1, GPC3, and GPA33—were not part of the original DRG list, they were chosen through Lasso–Cox regression for their strong associations with patient survival. This approach was based on the understanding that DRG clusters capture broader biological differences in GC, reflecting interactions between disulfidptosis and other critical pathways influencing patient outcomes. The absence of direct DRGs in the final model underscores the complexity of cancer biology, where key prognostic factors may arise from indirect or downstream effects of multiple interacting pathways. This broader analysis provided a more robust and clinically relevant prognostic model. To understand the clinical value of the prognostic model, we performed further studies on the model genes. Liu et al. showed that HEYL overexpression could accelerate the rate of GC progression by activating CDH11 [8, 9]. As a downstream target gene of NTOCH, HEYL-mediated inhibition of Mybl2 expression is an important mechanism that limits the proliferation of breast cancer cells [35]. Dai et al. found that CD109 can be used as a biomarker to predict the prognosis of GC [36]. Numerous studies have found that CD109 is significantly associated with the progression and metastasis of cervical cancer, pancreatic ductal carcinoma, and lung cancer [37,38,39]. GRP promotes gastric acid production as a neuropeptide [40]. GRP also plays a major role in the process of tumor angiogenesis and is highly expressed in tumor tissues, such as gastric, lung, and prostate cancer tissues [41]. Additionally, GRP is a growth factor in small cell lung, breast, and colon cancers [42]. DEFB1 is an essential component of the antimicrobial peptide series and is correlated with the resistance of epithelial surfaces to microbial colonization [43]. DEFB1 is a tumor suppressor that promotes apoptosis in tumor cells [44]. Donald et al. demonstrated that DEFB1 expression decreased in kidney and prostate cancer tissues [45]. GPC3 is a carcinoembryonic antigen overexpressed in hepatocellular carcinoma [46]. GPC3, a key component of Wnt signaling, has been implicated in managing hepatobiliary function, controlling cytodifferentiation, and regulating hepatocellular carcinoma development [47]. GPA33 is a colon cancer antigen expressed in more than 95% of human colon cancers [48]. In addition, Lopes et al. suggested that GPA33 is a strong indicator of good prognosis in patients with GC [49]. Together, these studies demonstrate that our model genes play critical roles in the development of several cancers.

TMB and MSI provide a means of assessing a patient’s response to immunotherapy. Patients with a high level of TMB reportedly exhibit enhanced responses to immunotherapy [50]. Some studies have also shown that patients with high MSI may benefit more from immunotherapy. For example, one study on colorectal cancer found that patients with high MSI scores were more sensitive to immune checkpoint inhibitors (ICIs) and reaped more benefits from immunotherapy [51]. In our study, LR-GC patients had higher TMB scores and a higher MSI overall, indicating that the LR group would benefit more from immunotherapy than the HR group. We also found that patients in the LR group had relatively lower TIDE scores and that patients in the LR group had a higher effective immune response and were more suitable for immunotherapy. Studying the differences in the sensitivity of patients in different risk groups to anticancer drugs offers an alternative way of tackling chemotherapy resistance. The results of the drug sensitivity analysis indicated that LR group patients were more responsive to veriparib and afatinib, whereas the HR group patients were more responsive to motesanib and dasatinib. Veliparib is a PARP inhibitor, and a previous study has shown that veliparib in combination with SAHA promotes prostate cancer cell death [52]. Afatinib, an EGFR tyrosine kinase inhibitor, improves objective remission rates and progression-free survival in patients with Squamous Cell Carcinoma of the Head and Neck (SCCHN) [53]. Motesanib is a vascular endothelial growth factor receptor (VEGFR) inhibitor that induces the regression of tumors, such as breast and colon cancers [54]. A study on pancreatic cancer reported that a combination of dasatinib and paclitaxel had an inhibitory effect on tumor cells [55]. These results suggest that RSs may guide more individualized immunotherapy regimens and more appropriate chemotherapy strategies in clinical settings.

To study the model genes in depth, we selected GPC3, which was stably differentially expressed in normal and tumor cells, for single-gene analysis. GPC3 is a member of the acetylheparin sulfate (HS) proteoglycan family. Previous studies have indicated that GPC3 can be used as a biomarker for the effective diagnosis of hepatocellular carcinoma [56]. In addition, overexpression of GPC3 promotes lung squamous cell proliferation and inhibits lung squamous cell apoptosis [57]. However, the specific role of GPC3 in GC remains controversial. Our experimental results indicated that excess GPC3 promotes the proliferation, migration, and invasion of GC cells (BGC823). This is consistent with previous studies; Jiang’s experiments also demonstrated that the downregulation of GPC3 in GC may inhibit gastric cancer cell metastasis and affect the tumor immune microenvironment [58]. We first evaluated the expression of GPC3 in 33 cancers and observed low expression in most cancers when compared to normal tissues, and certainly in GC. However, in GC, high GPC3 expression was positively correlatead with poor prognosis, which is consistent with our experiments. Although GPC3 is more highly expressed in normal gastric tissues, its positive correlation with the risk score (RS) in our model reflects its context-dependent role, particularly within cancer-associated fibroblasts (CAFs) in the tumor microenvironment. Although GPC3 may have a regulatory function in normal tissues, its expression in CAFs could promote tumor aggressiveness, contributing to poorer prognosis in gastric cancer (GC) patients. This dual role of GPC3 explains its association with higher RS in our model despite higher expression in normal tissues. Thus, we further analyzed GPC3 in terms of single-cell sequencing and found that it was highly abundant in the fibroblasts of tumor tissues. Fibroblasts expand after activation and produce growth factors, and cancer cells can acquire favorable growth, migration, and survival properties from these released growth-promoting factors [59]. Li et al. found that GPC3 is predominantly upregulated in cancer-associated fibroblasts (CAF) in GC tissues, which is consistent with our study [60]. GPC3 is a member of the acetylheparin sulfate (HS) proteoglycan family and has been identified as a potential biomarker in several cancers. Although not directly implicated as a primary driver of disulfidptosis, GPC3's prominent role in cancer-associated fibroblasts (CAFs) within the tumor microenvironment suggests it may influence disulfidptosis-related pathways indirectly. Our analysis showed that GPC3 is highly expressed in fibroblasts of tumor tissues, where it may modulate the cellular environment to promote tumor survival, aggressiveness, and resistance. This interaction between GPC3 and the broader cellular processes within disulfidptosis-related subtypes supports its focus in our study as a potential contributor to the disulfidptosis-related cellular environment that affects tumor progression and prognosis. Therefore, we conclude that GPC3 may promote tumor development and influences prognosis through CAF.

Our study linked DRGs to GC and elucidated differences in immune infiltration status and prognosis between disulfidptosis subtypes in patients with GC. Key genes associated with prognosis were also screened, and an RS model associated with disulfidptosis was established, identifying GPC3 as a new biomarker for predicting patient survival status and sensitivity to immunotherapy. While our study primarily focused on mRNA expression levels to identify disulfidptosis-related genes in gastric cancer, we observed certain discrepancies when comparing these data with protein expression patterns from proteinatlas.org. For instance, IQGAP1 and RPN1 exhibited high protein expression in normal gastric tissues, which was not fully consistent with their mRNA expression patterns. These differences may arise from post-transcriptional regulation, differential protein stability, or tissue-specific factors that affect protein levels independently of mRNA expression. These findings highlight the necessity of integrating multi-omics approaches and verifying findings at the protein-level in future studies. However, this study has several limitations. Our data were obtained from public databases (GEO and TCGA), and although the expression and basic behavior of the model gene (GPC3) was verified, their exact mechanisms of action remain unclear. Therefore, additional basic experiments are required to elucidate the specific mechanisms and functions of the model genes.

5 Conclusions

In this study, we established disulfidptosis-related subtypes and built a six-gene RS model based on inter-subtype DEGs, allowing for better assessment of the prognosis of patients with GC and guiding individualized clinical immunotherapy and chemotherapy. Although our findings suggest that GPC3 may be a promising therapeutic target owing to its significant association with prognosis and its role within the tumor microenvironment, we acknowledge that further experimental validation, including in vitro and in vivo studies, is necessary to confirm its therapeutic potential in GC.