First, we obtained data from the TCGA database (https://portal.gdc.cancer.gov/) and organized the related gastric cancer (GC) gene expression data, clinical data, mutation data and copy number data. The data were processed and analyzed using the limma, dplyr, ggalluvial, and ggplot2 packages in R software (version 4.1.3) for data organization, and the survival, caret, glmnet, survminer, and timeROC packages for model screening. Additionally, Perl software (version 5.30.0) was utilized for data processing and analysis. First, the expression matrix for the gastric cancer samples was generated. Then, the lncRNA and mRNA expression matrices were generated, and the disulfidptosis mRNA expression matrix and disulfidptosis lncRNA expression matrices were compared to identify the DRLs. The correlations between disulfidptosis-related mRNAs and disulfidptosis-related lncRNAs were subsequently calculated.

GC-related data were extracted from the TCGA, DRGs were identified, and a comprehensive analysis was carried out. The lncRNAs associated with the DRGs were identified via R software, and the significantly related lncRNAs were subsequently screened via univariate analysis. Then, univariate and multivariate Cox regression (22) analysis and LASSO regression analysis (23–28) were performed to establish a new prediction model; our filtering conditions were more stringent in these analyses (P < 0.0005). We randomly grouped the samples and performed more than 100 iterations (in practice, a model can be applied only 3–5 times) with 100 completely different prediction models. Then, survival analysis, receiver operating characteristic (ROC) (29) curve analysis, clinical statistics, and principal component analysis (PCA) (30) were used to confirm a valid signature. The expression formula (31) is as follows, where Coefi is the coefficient and Xi is the expression quantity of the lncRNA: Riskscore = ∑ i = 1 n Coef i ∗ X . Finally, the correlations between the DRGs and lncRNAs were analyzed, and a heatmap was generated.

R software was used to conduct subgroup analysis of all samples related to disulfidptosis genes in the TCGA database. First, the samples were randomly divided into a training set and a testing set, with 204 samples in the training set and 203 samples in the testing set. The baseline characteristics are presented in Table 1 . Second, the two subgroups were subjected to survival analysis according to receiver operating characteristic (ROC) curves risk curves and survival status. The lncRNAs of the prediction model were subsequently analyzed via cluster analysis, and a heatmap was drawn. Finally, progression-free survival (PFS) analysis was carried out to further verify the value of the prediction model.

Univariate Cox analysis, multivariate Cox analysis, ROC curve analysis and Concordance index (32) (C-index) analysis were carried out for age, sex, grade, stage and risk score, respectively, and a nomogram (33) was constructed; then, the predictive accuracy of nomogram was verified via a calibration curve.

Survival analysis was performed based on the risk score across various clinical characteristics, including age, G stage, M stage, sex, N stage, overall stage, and T stage, to evaluate the accuracy of the risk score in assessing clinical features of gastric cancer. Subsequently, principal component analysis (PCA) was conducted on four distinct sample sets: all gene samples, disulfidptosis gene samples, disulfidptosis LncRNA samples, and risk LncRNA samples. The results demonstrated that the risk score exhibited superior discriminative ability in distinguishing gastric cancer samples, further validating its prognostic significance.

The samples were grouped according to the risk score of the prediction model, and differential expression analysis was conducted. Significant differentially expressed genes were screened out, and GO (34–39), KEGG (34, 35, 40) and GSEA (34, 35, 37, 41, 42) analysis were conducted on these differentially expressed genes to identify the potential related pathways and functions.

To comprehensively assess the relevance of the risk score to immune features in gastric cancer, the tumor microenvironment (TME) (43), CIBERSORT (34, 35, 44), immCor, immFunction, tumor mutational burden (TMB) (34, 45, 46), TMB stratified analysis, and Tumor Immune were analyzed by predictive model Dysfunction and Exclusion (TIDE) (47) analysis; Fully evaluate the relevance of risk models to gastric cancer immunity.

Drugs that may have significant differences in efficacy in gastric cancer treatment were screened via a predictive model and the OncoPredict (48) package in R software.

Primer design for genes FRMD6-AS2 and GAPDH (Table 2).

In this study, GES1 (BNCC337969) and HGC-27 (BNCC338546) cells were purchased from Beina Biology (China). GES1 cells were cultured in complete DMEM-H supplemented with 1% penicillin−streptomycin (Sigma, USA), and HGC-27 cells were cultured in complete RPMI-1640 supplemented with 1% penicillin−streptomycin (Sigma, USA).

The cell precipitate was collected, and 1 mL of TRIzol (CWBIO, China) was added, and the sample was mixed and incubated on ice for 5 min for cell lysis. Then, 200 μL of chloroform (High Crystal Chemical Industry, China) was added to the EP tube, which was shaken vigorously, left at room temperature for 3 min, and centrifuged at 12000 rpm at 4°C for 15 min, after which the supernatant was collected. Then, an equal volume of isopropyl alcohol (McLean, China) was added, mixed upside down, left for 10 min at room temperature, and centrifuged at 12000 rpm at 4°C for 10 min, after which the supernatant was discarded. Then, 1 mL of 75% ethanol (McLean, China) was added, the mixture was centrifuged at 12000 rpm at 4°C for 3 min, the supernatant was discarded, the lid was opened to allow air-drying, 30 μL of DEPC water was added, and the sample was stored at -80°C. Then, reverse transcription was performed with a 20 μL reaction volume. The amount of each component was 3 μL of RNA, 4 μL 5× Hifair^® II Buffer (11119ES60 yeasen), 2 μL Hifair^® II Enzyme Mix, 0.5 μL + 0.5 μL random primers N6 (50 μm) + or oligo (dT)18 (50 μM), and 20 μL RNase-free H2O. The samples were added to the EP tube according to the above reaction system, mixed with a pipette and put into a PCR (BIO-RAD T100) amplification apparatus. The program was set at 25°C for 5 min according to the following conditions: 42°C for 30 min; 85°C for 5 min; hold at -20°C.

The volume of the fluorescence quantitative PCR reaction system was 20 μL. The amount of each component was as follows: UGreener Flex qPCR 2X mix, 10 μL; forward primer (10 µM), 0.5 μL; reverse primer (10 µM), 0.5 μL; cDNA, 1 μL; and ddH₂O, 8 μL. The samples were added to 96-well plates for fluorescent quantitative PCR according to the above reaction system and mixed successively. The samples were put into the fluorescent quantitative PCR instrument (ABI 7500, USA), and the program was set according to the following conditions: 45 cycles were carried out in the constant temperature stage is set at 95°C for 30 seconds (denaturation), followed by the cycling stage of 95°C for 10 seconds (denaturation) and 60°C for 30 seconds (annealing) for a total of 45 cycles. The melt stage consists of 95°C for 15 seconds (denaturation), 60°C for 1 minute (annealing), 95°C for 15 seconds (denaturation), and 60°C for 15 seconds (annealing).

In the bioinformatics analysis, statistical analysis was performed using R software (version 4.1.3), with a P-value of less than 0.05 considered statistically significant. For the experimental part, all data were replicated three times. Differences between two groups were compared using the Student t-test, and statistical analysis and graphing were conducted using GraphPad Prism 8.0 software. A P-value of less than 0.05 was considered statistically significant.

First, the expression matrix of the lncRNAs was extracted via the data matrix processed with R software (version 4.1.3) and Perl software (version 5.30.0). The correlations between lncRNAs and disulfidptosis genes were mapped via the R package ggalluvial ( Figure 2A ). The lncRNAs whose expression significantly differed between groups were screened via univariate Cox analysis and are displayed in a forest plot ( Figure 2B ). Subsequently, LASSO regression and multivariate Cox regression were used to construct a prognostic signature associated with disulfidptosis in gastric cancer patients ( Figures 2C, D ); the analysis included more than 100 iterations, and the final signature was evaluated via survival analysis, receiver operating characteristic (ROC) curve analysis, clinical statistics and PCA ( Table 3 ). The formula for the signature risk score was as follows: risk score = (0.544735914395105 * AC107021.2 expression) + (0.705013376452246 * AC129507.1 expression) + (0.433534323181848 * ‘FRMD6-AS2’ expression); heatmap showing the correlation between the individual DRLs and the risk score ( Figure 2E ). Next, model lncRNAs and related DRGs were screened out ( Table 4 ).

The samples used to construct the DRL model were subsequently randomly divided into two groups of equal size (a test group and a training group). We used the DRL signature to score each sample for the entire cohort, the test group sample and the training group and divided the patients into high-risk and low-risk groups. Survival analysis, risk score analysis, survival time analysis, enrichment analysis, PFS (49) analysis were performed for each group. Survival analysis revealed that the high-risk patients in each group had a worse prognosis than the low-risk patients (P < 0.05) ( Figures 3A-C ). Risk score analysis revealed that the risk score trend of each group was generally consistent, the risk score of the low-risk group was less than 1, and the risk score of the high-risk group was greater than 1 ( Figures 3D-F ). Similarly, we conducted survival time analysis for each group and found that the survival time distribution of patients in each group was generally consistent: the low-risk group had a lower mortality rate than the high-risk group ( Figures 3G-I ). Through enrichment analysis and heatmapping of signature lncRNAs in each group, AC107021.2, AC129507.1 and FRMD6-AS2 were found to be high-risk lncRNAs. The PFS analysis revealed that the high-risk group had lower PFS rates at a given time point than did the low-risk group ( Figure 3M ). The median survival time (in years) is: 1.152055. Independent survival analysis for each lncrna model reveals that each lncrna can independently influence gastric cancer prognosis ( Figures 4A-I ).

First, univariate Cox analysis, multivariate Cox analysis, clinical characteristic ROC curve analysis, survival ROC curve analysis and concordance index analysis were performed for the risk score and clinical characteristics of age, sex, grade and stage, respectively. Univariate and multivariate Cox analyses revealed that age, grade, stage and the risk score had significant effects on the prognosis of gastric cancer (P < 0.05), whereas sex had no significant effect on the prognosis of gastric cancer (P > 0.05) ( Figures 5A, B ). Clinical characteristic ROC curve analysis revealed that the area under the curve (AUC) (50–52) value of age, grade, stage and the risk score was > 0.5, whereas the AUC value of sex was < 0.5 ( Figure 5C ). ROC curve analysis of survival years revealed that the AUC values at 1, 3 and 5 years were all greater than 0.6 ( Figure 5D ). Concordance index analysis revealed that the risk score was the main factor influencing the prognosis of gastric cancer, whereas sex had the least effect on the prognosis of gastric cancer ( Figure 5E ). The risk score, age, sex, grade, stage, M stage, N stage and T stage were used to establish a nomogram, which revealed that the risk score and age were independent influencing factors for the prognosis of gastric cancer ( Figure 5F ). Finally, calibration curves were generated to verify the accuracy of the nomogram predictions ( Figure 5G ).

The patients were then grouped by age(< 60 years old and ≥ 60 years old). Survival analysis of the two groups revealed that the survival rate of the high-risk group was lower than that of the low-risk group among each of these age groups (P < 0.05) ( Figures 6A, B ). The samples were divided into G1–G2 groups and G3 groups by G stage, and survival analysis was conducted on the two groups. There was no significant difference in the survival rate between the high- and low-risk groups among the G1–G2 group (P > 0.05). However, there was a significant difference in the survival rate between the high- and low-risk groups among the G3 group (P < 0.05) ( Figures 6C, D ). We divided the samples by M stage into M0 and M1 groups; survival analysis of the M0 group revealed significant differences between the high- and low-risk groups (P < 0.05). There was no significant difference in survival between the high- and low-risk groups among the M1 group (P > 0.05) ( Figures 6E, F ). The patients were divided into a female group and a male group, and survival analysis revealed significant differences between the high- and low-risk groups for both females and males. The survival rates of the high-risk groups were lower than those of the low-risk groups (P < 0.05) ( Figures 6G, H ). By N stage, the high-risk group was divided into the N0-1 group and the N2-3 group. Survival analysis revealed that there was a significant difference between the high- and low-risk groups among patient in these subgroups and that the high-risk groups had lower survival rates (P < 0.05) ( Figures 7A, B ).The patients were divided by stage into Stage I-II and Stage III-IV groups. The survival analysis of the high- and low-risk samples in each group revealed significant differences, and the survival rates of the high-risk groups were lower (P < 0.05) ( Figures 7C, D ). In the T stage, the samples were divided into a T1-2 group and a T3-4 group. The survival analysis of the high- and low-risk samples in each group revealed that there was no significant difference in the survival rate between the high- and low-risk groups in the T1-2 subgroup (P > 0.05). The survival rate of the high- and low-risk groups in the T3-4 subgroup was significant, and the survival rate of the high-risk subgroup was lower (P < 0.05) ( Figures 7E, F ). PCA of the all-gene, DRG, DRL and risk lncRNA groups revealed that the risk-related lncRNA model presented the greatest degree of differentiation among samples ( Figures 7G-J ).

First, differential expression analysis was conducted on samples grouped according to the risk score, and significantly differentially expressed genes were identified. GO analysis of the differentially expressed genes revealed that the main enriched biological process (BP) were enriched external encapsulating structure organization, extracellular matrix organization, extracellular structure organization, muscle system process, etc. The main enriched cellular component (CC terms were collagen-containing extracellular matrix, contractile fiber, myofibril, sarcomere, I band, etc. The main enriched molecular function (MF) terms were extracellular matrix structural constituent, glycosaminoglycan binding, heparin binding, integrin binding, sulfur compound binding, etc. ( Figures 8A-D ), ( Supplementary 1 ). KEGG analysis revealed that the main enriched pathways were the cytoskeleton in muscle cells, dilated cardiomyopathy, ECM-receptor interaction, hypertrophic cardiomyopathy, arrhythmogenic right ventricular cardiomyopathy and other pathways ( Figures 9A, B ) ( Supplementary 2 ). GSEA revealed enrichment of pathways and functions including DILATED CARDIOMYOPA, HYPERTROPHIC CARDIOMYOPATHY HCM, PROTEASOME, FOCAL ADHESION, ARRHYTHMOGENIC RIGHT VENTRICULAR CARDIOMYOPATHY ARVC, CALCIUM SIGNALING PATHWAY, ECM RECEPTOR INTERACTION, DNA REPLICATION, CITRATE CYCLE TCA CYCLE, AMINOACYL TRNA BIOSYNTHESIS ( Figures 9C-F ) ( Supplementary Materials 3, 4 ). As shown in Supplementary Material 3 , we found that FLNA is enriched in the FOCAL ADHESION pathway and that TLN1 is enriched in the FOCAL ADHESION pathway. MYH10 was enriched in the TIGHT JUNCTION, VIRAL MYOCARDITIS, and REGULATION OF ACTIN CYTOSKELETON pathways, and as shown in Table 2 , FLNA was positively correlated with AC129507.1; TLN1 was significantly positively correlated with AC129507.1; and MYH10 was significantly positively correlated with AC107021.2. In summary, our differential gene expression analysis identified key biological processes and molecular functions linked to the risk score. GO analysis revealed enrichments in extracellular matrix and muscle system processes, while KEGG and GSEA pinpointed important pathways like cytoskeleton dynamics and cardiomyopathy. Proteins FLNA, TLN1, and MYH10 were strongly associated with specific DRLs and crucial pathways such as focal adhesion and tight junctions. These insights enhance our understanding of gastric cancer prognosis and suggest potential therapeutic targets.

The samples were divided into high- and low-risk groups according to the risk score. Analysis of the TME scores revealed significant differences in the stromal score, immune score and ESTIMATE score between the high- and low-risk groups (P < 0.05) ( Figure 10A ). We estimated the levels of 22 kinds of immune cells (naive B cells, memory B cells, memory B cells, memory T cells, naive CD8+ T cells, resting memory CD4+ T cells, activated memory CD4+ T cells, follicular helper T cells, regulatory T cells (Tregs), gamma delta T cells, resting NK cells, activated NK cells, monocytes, M0 macrophages, M1 macrophages, M2 macrophages, resting dendritic cells, activated dendritic cells, resting mast cells, activated mast cells, eosinophils and neutrophils) in each sample, and a histogram was drawn ( Figure 10B ). The high- and low-risk groups were compared to determine whether there were significant differences in the levels of these 22 kinds of immune cells. The levels of naive B cells, plasma cells, resting memory CD4 T cells, activated memory CD4 T cells, resting NK cells, activated NK cells (determined via calculation), M0 macrophages, M2 macrophages, resting mast cells, activated mast cells and eosinophils were significantly different between the high- and low-risk groups (P < 0.05) ( Figure 10C ). This further confirms that the immune cell infiltration patterns differ among different risk groups. We assessed 29 immune cells/functions (aDCs, APC coinhibition, APC costimulation, B cells, CCR, CD8+ T cells, checkpoint, cytolytic activity, DCs, HLA, iDCs, inflammation-promoting, macrophages, mast cells, MHC class I, neutrophils, NK cells, paraination, pDCs, T cell coinhibition, T cell costimulation, T helper cells, Tfhs, Th1 cells, Th2 cells, TIL, Treg, type I IFN response, and type II IFN response), and we found significant differences in the enrichment of related genes between the high- and low-risk groups for B cells, mast cells, MHC class I cells, neutrophils, pDCs, T helper cells, Th1 cells, Th2 cells, TILs, the type I IFN response, the type II IFN response, and 11 immune-related responses ( Figure 10D ). This indicates that high-risk groups may promote the infiltration of immunosuppressive cells and increase the expression of immune checkpoint genes (ICGs), leading to immune escape and poor prognosis. For the analysis of the relationship between the risk score and TMB of gastric cancer, the top 20 genes (TTN, TP53, MUC16, LRP1B, ARID1A, CSMD3, SYNE1, FAT4, FLG, PCLO, ZFHX4, ACVR2A, HMCN1, DNAH5, OBSCN, RYR2, SPTA1, FAT3, CSMD1, and KMT2D) had higher mutation rates in the low-risk group ( Figures 11A, B ). In the risk score-related samples, there were significant differences in TMB between the high- and low-risk groups ( Figure 11C ). There were significant differences in survival between patients with high and low TMB (H/L TMB) (P < 0.05) ( Figure 11D ). Stratified survival analysis was performed on H/L-TMB samples, and the samples were divided into four subgroups: H-TMB + low risk, H-TMB + high risk, L-TMB + high risk, and L-TMB + low risk. The survival rate of the high-risk group was lower than that of the low-risk group (P < 0.05) ( Figure 11E ). This further confirms the negative correlation between TMB levels and survival prognosis. Analysis of tumor immune dysfunction and exclusion (TIDE) revealed that the high-risk group had a higher TIDE score and was more prone to immune escape (P < 0.05) ( Figure 11F ).This suggests that tumors in the high-risk group may suppress immune responses through multiple mechanisms, leading to poor prognosis. In summary, the analysis revealed significant differences in TME scores and immune cell levels between high- and low-risk groups. Notably, various immune cells such as naive B cells and NK cells, and immune functions, showed significant variation. High-risk groups had higher TIDE scores, indicating a greater tendency for immune escape. Additionally, the low-risk group exhibited higher mutation rates in key genes and better survival outcomes compared to the high-risk group.

We set the P value of screening to pFilter = 0.00000000001 for screening effective drugs. Finally, three drugs with significantly higher sensitivity in the high-risk group were screened: BMS-754807, dabrafenib, and JQ1 ( Figures 12A-C ). As shown in Figure 11 , BMS-754807 and JQ1 had lower scores in the high-risk group, and these two drugs were more effective in the high-risk group. However, dabrafenib scored lower in the low-risk group and was more effective in the low-risk group. These findings indicate that the model can be used at an individualized level to screen drugs likely to be effective against gastric cancer.

The results revealed that FRMD6-AS2 was highly expressed in HGC-27 cells and expressed at low levels in AGS cells compared with normal GES-1 cells ( Figure 13 ), but there were significant differences between the two groups (P < 0.05).