Identifying prognostic genes related to oxidative stress and Ferroptosis

Following the intersection of 566 ORGs and 484 FRGs, 76 OFRGs were identified in total (Fig. 1A), and enrichment analyses were performed on these genes. GO analysis showed that they were primarily enriched in the “response to oxidative stress” of biological process (BP), the “NADPH oxidation complex” and “oxidoreductase complex” of cellular component (CC), and the “superoxide generating NAD (P) H oxidation activity” and “antioxidant activity” of molecular function (MF). According to KEGG analysis, these genes were associated with the “Chemical cancer genes reactive oxygen species” and “TNF signaling pathway” (Fig. 1B). Gene mutation analysis of OFRGs revealed that TP53 exhibited the highest mutation frequency, and missense mutation was the most prevalent type of mutation. A waterfall plot was created using the top 20 genes with the highest frequency of mutations (Fig. 1C).

Fig. 1

Genetic, expression, and mechanism analysis of OFRGs in gastric cancer. (A) Determine 76 OFRGs through Veen diagram. (B) GO and KEGG analyses of OFRGs. (C) The mutation state of OFRGs in somatic cells. (D) The frequency of CNVs gain and loss in OFRGs. (E) Locations of CNV alterations in OFRGs on chromosomes. (F) Determine OFRG associated with prognosis through univariate Cox analysis. (G) The expression of OFRGs in normal gastric samples and gastric cancer samples in the TCGA database. ns p > 0.5; *p < 0.05; **p < 0.01; ***p < 0.001. OFRGs, oxidative stress and ferroptosis-related genes; GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; CNV, copy number variation; TCGA, The Cancer Genome Atlas

Additionally, we examined the CNV levels of OFRGs and observed that GAIN variation occurred more frequently than LOSS variation. We also created a copy number circle diagram representing these genes’ chromosomal locations (Fig. 1D, E). Through univariate COX analysis, 15 prognostic-related OFRGs were determined (Fig. 1F), with their K-M survival curves displayed in Figure S1. Subsequent differential analysis of tumor and adjacent samples from the TCGA cohort revealed significant differences in the expression of the majority of OFRGs (Fig. 1G).

Consensus clustering analysis

The researchers used the STRING website to generate a PPI network of OFRGs and imported Cytoscape software to screen out the five most critical OFRGs (Fig. 2A). Consensus clustering analysis was performed, dividing all GC samples into two OFRG clusters (K = 2) (Fig. 2B, Table S2). PCA and t-SNE demonstrated significant dispersion between the two clusters, which is helpful for distinguishing patients (Fig. 2C). Furthermore, a statistically significant distinction in survival outcomes was observed between the two patient groups (p < 0.05), with patients in cluster A showing a more prolonged OS(Fig. 2D). We used the ssGSEA approach to measure the expression levels of immune cells in two clusters. We identified significant differences in 20 different types of immune cells, with cluster B displaying higher immune-suppressive cell infiltration (Fig. 2E). Subsequently, we employed GSVA analysis to generate a heat map of the KEGG pathway between the two clusters. The enrichment pathways of cluster A, namely “BASE_EXCISION_REPAIR”, “PEROXISOME”, and “CITRATECYCLE_TCA_CYCLE”, primarily focus on cellular function and metabolism. The pathways enriched in cluster B, such as “TGF_BETASIGNALING_PATHWAY” and “FOCAL_ADHENSION,” are linked to matrix and cancer activation (Fig. 2F). Subsequently, we conducted differential analysis on clusters A and B (| logFC |>1.0 and FDR < 0.05) and performed enrichment analysis on DEGs. The GO analysis implied that these DEGs were primarily enriched in pathways, including “Leukocyte migration”, “myeloid Leukocyte migration”, and “response to lipopolysaccharide” (Fig. 2G). According to KEGG findings, DEGs were associated with pathways that promote inflammation and carcinogenesis, such as the “IL-17 signaling pathway” and the “PI3K-Akt signaling pathway” (Fig. 2H).

Fig. 2

Consensus clustering analysis. (A) PPI network of OFRGs constructed utilized STRING website and Cytoscape software. (B) Consensus clustering analysis based on OFRGs. (C) PCA and t-SNE analysis of OFRG clusters. (D) K-M analysis of OFRG clusters. (E) Evaluate the level of immune cell infiltration through the ssGSEA algorithm. (F) GSVA analysis of OFRG clusters. (G, H) GO and KEGG analysis of DEGs between OFRG clusters. ns p > 0.5; *p < 0.05; **p < 0.01; ***p < 0.001. PPI, Protein-Protein Interaction; OFRGs, oxidative stress and ferroptosis-related genes; PCA, principal component analysis; t-SNE, t-distributed stochastic neighbor embedding; K-M, Kaplan-Meier; ssGSEA, single sample gene set enrichment analysis; GSVA, gene set variation analysis; DEGs, differentially expressed genes

Construction and validation of a prognostic risk signature

Upon identifying prognostic-related genes by univariate COX analysis on the aforementioned DEGs (Fig. 3A), we categorized all GC samples into gene clusters A or B (K = 2) using clustering analysis (Fig. 3B, Table S1). Expression levels of JUN, IL6, and PTGS2 were lower in gene cluster B, whereas SRC expression was elevated. TP53 did not significantly change between the two gene clusters (Fig. 3C). In addition, the K-M curve indicated that patients in gene cluster B had a superior prognosis (Fig. 3D). Then, we performed LASSO analysis to mitigate the risk of overfitting and ultimately constructed a prognostic signature for OFRGs, including SLC7A2, CDH19, and CCN1, through multivariate analysis (Fig. 3E, F). The risk score for each sample was calculated as follows: Risk score= (0.128282677879963) × SLC7A2 expression + (0.13087784597397) × CDH19 expression + (0.120683245893589) × CCN1 expression. GC patients were classified as high-risk or low-risk groups based on the median risk score. Table S4 displays the results of the risk score. Notably, the high-risk group exhibited elevated expression levels of SLC7A2, CDH19, and CCN1 (Fig. 3G). Furthermore, K-M analysis revealed that the groups with higher expression levels of these three genes had worse prognostic outcomes and a shorter OS period (Fig. 3H). Afterward, we generated a Sankey plot of OFRG clustering, gene clustering, risk signature, and survival outcomes (Fig. 4A). Figure 4B and C illustrated that the risk scores of cluster B and gene cluster A groups were significantly higher, supporting the consistency and reliability of the previous analysis. K-M analysis indicated that the OS of high-risk scoring populations in the TCGA and GSE84437 cohorts was significantly shorter (Figs. 4D and 5A), and the scatter plot of risk scores and patient survival statistics also demonstrated that patients with higher risk scores had a higher risk of death (Figs. 4E and 5B). The ROC curve displayed area under curve (AUC) values of 0.631, 0.637, and 0.652 for 1, 3, and 5 years, respectively. Compared to other clinical features, the risk signature has the highest 5-year AUC value (Fig. 4F), and consistent results were shown in the GSE84437 cohort (5 C). Through PCA method dimensionality reduction analysis, we found that the high and low-risk groups exhibited two distinct development trend directions with good dispersion, suggesting that the risk score can effectively differentiate GC patients (Fig. 4G).

Fig. 3

Development of a prognostic risk signature for OFRGs. (A) Genes associated with prognosis in DEGs of clusters A and B. (B) Divide gene clusters based on consensus clustering analysis. (C, D) Expression and prognostic analysis of gene clusters. (E) Lasso Cox regression analysis and cross-validation. (F) Multivariate Cox analysis for determining the optimum signature genes. (G) Expression of signature genes in high and low-risk subgroups. (H) K-M analysis of signature genes in the risk signature. *p < 0.05; **p < 0.01; ***p < 0.001. OFRGs, oxidative stress and ferroptosis-related genes; DEGs, differentially expressed genes; LASSO, Least Absolute Shrinkage and Selection Operator; K-M, Kaplan-Meier

Fig. 4

Evaluation of the risk signature. (A) Sankey plot between OFRG clusters, gene clusters, risk signature, and GC prognosis. (B) The relationship between OFRG clusters and risk scores. (C) The relationship between gene clusters and risk scores. (D) K-M analysis of the risk signature in the TCGA-STAD queue. (E) Distributions of risk scores and survival statuses. (F) ROC curve for the risk signature and other clinical characteristics. (G) PCA between risk subgroups based on the OFRGs signature. (H) Cox regression analyses of the signature and other clinical parameters in the TCGA cohort. OFRGs, oxidative stress and ferroptosis-related genes; GC, gastric cancer; K-M, Kaplan-Meier; ROC, receiver operating characteristics curve; PCA, principal component analysis. OFRGs, oxidative stress and ferroptosis-related genes; GC, gastric cancer; K-M, Kaplan-Meier; TCGA, The Cancer Genome Atlas

Fig. 5

Validation of the risk signature. (A) K-M analysis in the GSE84437 queue. (B) Distributions of risk scores and survival statuses in the GSE84437 cohort. (C) ROC curve for the risk signature and clinical characteristics in the GSE84437 cohort. (D) Univariate and multivariate Cox regression analyses in the GSE84437 cohort. (E) GSVA analysis for the risk signature. (F) A clinical nomogram constructed based on age, stage, and risk scores. (G) Calibration plot for the nomogram. (H) ROC curve for the nomogram and other clinical characteristics. K-M, Kaplan-Meier; ROC, receiver operating characteristics curve; GSVA, gene set variation analysis

In addition, to further evaluate the significance of risk score in predicting the prognosis of GC patients, we carried out COX regression analysis on risk score and several clinical characteristics, including age, gender, grade, and stage. Univariate Cox analysis displayed that Age (HR = 1.022, 95% CI = 1.006–1.039, p < 0.01), Stage (HR = 1.596, 95% CI = 1.294–1.970, p < 0.001), and Risk score (HR = 1.685, 95% CI = 1.339–2.120, p < 0.001) can significantly affect the OS of GC. In multivariate Cox analysis, Age (HR = 1.032, 95% CI = 1.015–1.050, p < 0.001), Stage (HR = 1.676, 95% CI = 1.346–2.087, p < 0.001), and Risk score (HR = 1.717, 95% CI = 1.353–2.180, p < 0.001) were independent prognostic factors. Cox regression analysis of univariate (HR = 3.261, 95% CI = 1.521–6.997, p = 0.002) and multivariate (HR = 3.812, 95% CI = 1.761–8.253, p < 0.001) in the GSE84437 cohort also confirmed the best predictive effect of risk score (Fig. 5D). Later, we conducted mechanistic analysis on the high and low-risk groups using GSVA, suggesting that the low-risk group’s enriched pathways were primarily related to cellular metabolism and function. In contrast, the high-risk group was linked to signaling pathways associated with cancer occurrence and development, such as “TGF_BETA,” “HEDGELOG,” “MTOR,” and “MAPK,” as well as the matrix secretion and activation pathways like “FOCAL_ADHESION” and “GAP_JUNCTION” (Fig. 5E).

Establishment and evaluation of a nomogram

In order to predict GC individuals’ survival time and survival rate, researchers constructed a clinical nomogram combining clinical pathological features and risk scores (Fig. 5F, Table S5). For instance, the nomogram predicts a patient’s 1-year, 3-year, and 5-year survival rates to be 62.1%, 23.8%, and 13.6%, respectively, with a total score of 159. The C-index value of calibration curve and AUC of the ROC curve for years 1, 3, and 5 of the nomograms were all greater than 0.65, confirming the accuracy of the prediction capacity of the nomogram. Moreover, we examined the correlations between the risk score and the clinical pathological features of GC, discovering that higher grades, later T stages, and death populations all had considerably higher risk scores (Figure S2 A-H). Additionally, low-score patients had significantly longer OS in the age ≤ 65, male, G3 level, T3-T4 stage, N1-N3 stage, and M0 subgroups (Figure S2 I-O).

Assessment of immune cell infiltration and immunological function

As seen in Fig. 6A, the stromal and immune scores of the high-risk group were significantly higher, indicating that the GC population with high-risk scores had a higher proportion of stromal cell and immune cell infiltration in the TME. After that, we explored the correlation between risk scores and immune function. Our analysis revealed that the low-risk group was primarily associated with functions like MHC class I and Th2 cells, whereas the high-risk group had higher levels of immune cell infiltration, including dendritic cells (DC), neutrophils, regulatory T cells (Treg), tumor-infiltrating lymphocytes (TIL), and so on (Fig. 6B). Likewise, we investigated the relationship between TIICs and risk scores using multiple methods. The findings demonstrated that, with the exception of activated CD4 T cells and type 17 T helper cells, which were negatively correlated with risk scores, the majority of tumor-infiltrating immune cells (TIICs), such as macrophage, myeloid-derived suppressor cells (MDSCs), and Treg, were positively correlated with risk scores. These cells demonstrated a higher proportion of infiltration in the high-risk group (Fig. 6C-F, Table S6).

Fig. 6

The correlation between the risk signature and immune cell infiltration. (A) The stromal, immune, and estimate scores of GC patients. (B) Immune function analysis for risk subgroups. (C-D) Analyzing the infiltration levels of TIICs in high and low-risk groups using several algorithms. (E) Heatmap between the risk signature and immune infiltrating cells. (F) Spearman analysis between the risk signature and several TIICs, including activated CD4 T cell and regulatory T cell. (G) TIDE, dysfunction, and exclusion in high and low-risk groups. (H) Expression of ICIs in risk groups. *P < 0.05; **P < 0.01; ***P < 0.001. GC, gastric cancer; TIICs, tumor-infiltrating immune cells; TIDE, tumor immune dysfunction and exclusion; ICIs, immune checkpoint inhibitors

Prediction of immunotherapy efficacy

The use of immunotherapy in cancer treatment has extremely high clinical value. However, its efficacy is limited to a subset of patients, as tumor cells can evade immune detection and develop resistance to immunotherapy [37]. The TIDE score can reflect the possibility of immune escape during immunotherapy and evaluate the potential clinical efficacy of immunotherapy in different risk groups. As shown in Fig. 6G, the TIDE, Dysfunction, and Exclusion scores of the high-risk scoring group were substantially higher than those of the low-risk scoring group. This suggests that patients in the high-risk scoring group are more likely to have immunological dysfunction and to develop resistance to immune therapy. We also studied the association between risk scores and several common immunological checkpoints (Fig. 7B). Additionally, we discovered a negative correlation between risk score and RNAss (Fig. 6H).

Fig. 7

Prediction of immunotherapy efficacy. (A) Analysis of risk scores and microsatellite state. (B) The correlation between risk scores and RNAss. (C) Predict the efficacy of anti-PD-1 and anti-CTLA-4 antibodies in the risk subgroup. (D) Waterfall diagram of somatic mutations in high and low risk scoring groups. (E) TMB scores in high and low-risk groups. (F) K-M curve of OS in high and low-TMB groups. (G) K-M curve survival curves among the four groups that combined TMB with risk signature. *P < 0.05; **P < 0.01; ***P < 0.001. RNAss, RNA stemness scores; TMB, tumor mutation burden; OS, overall survival. K-M, Kaplan-Meier; OS, overall survival

Microsatellites are repetitive sequences of small fragments of nucleic acids present in the genome with high mutagenicity. Functional defects in mismatch repair (MMR) proteins cause microsatellite instability (MSI), which raises the risk of tumor formation by inducing a high mutation phenotype in the genome. Microsatellites are classified into three groups based on their status: MSS, MSI-L, and MSI-H. The proportion of MSI-H varies significantly among tumor types, with a higher incidence observed in solid tumors such as GC, colorectal cancer (CRC), and endometrial cancer (EC) [38]. The detection of MSI is crucial for the diagnosis, treatment, and prognosis of various solid tumors, including CRC and EC. MSI-H is an independent prognostic marker for stage II colorectal cancer. Compared to MSS patients, those with MSI-H have a better prognosis for GC and small intestine adenocarcinoma [39, 40]. Likewise, MSI-H patients are more responsive to immunotherapy, benefiting from ICIs regardless of cancer type [41, 42]. Consequently, we analyzed the connection between risk scores and microsatellite status. Our findings revealed that the MSI-H group had the lowest risk score, and the low-risk group had a higher proportion of MSI-H (27% vs. 10%), indicating that the GC population with low-risk scores was more likely to benefit from immunotherapy (Fig. 7A, Table S7). Additionally, this study uncovered a relationship between risk scores and the effectiveness of ICIs. As Fig. 7C illustrated, patients in the low-risk group responded better to several ICI groups, including ctla4_neg_pd1_pos, ctla4_pos_pd1_neg, and ctla4_pos_pd1_pos.

We next investigated the somatic mutation data from the TCGA-STAD cohort and visualized the results using a waterfall plot. Researchers observed that the low-risk scoring group had a higher mutation frequency (93.68% vs. 88.04%), with missense mutation being the most common mutation and TTN being the gene with the greatest mutation frequency (Fig. 7D). In recent years, TMB has received great attention in studying ICI-related biomarkers. TMB can indirectly reflect the ability and degree of tumors to produce new antigens and has been proven to predict the efficacy of immunotherapy for various malignancies [43]. Treatment with ICIs is more likely to be beneficial for patients with high TMB (TMB-H). Therefore, we calculated the TMB value for each patient with GC and discovered that the low-risk group had a higher TMB score (Fig. 7E), and the TMB-H group had a longer OS time (Fig. 7F). Moreover, in the combined analysis of TMB and risk scores, the population with high TMB and low-risk scores had the longest OS and the best prognosis (Fig. 7G). Lastly, we also forecasted drug sensitivity based on the risk scores, finding that the low-risk score group exhibited lower IC50 values and higher sensitivity for a few commonly utilized chemotherapeutic agents, such as 5-Fluorouracil, Oxaliplatin, Irinotecan, and Cisplatin FigureS3A).

Determination CDH19 as a potential biomarker for GC

The researchers evaluated the diagnostic efficacy of genes in the signature model using ROC curves. The AUC for SLC7A2, CDH19, and CCN1 were 0.524, 0.793, and 0.487, respectively, with CDH19 showing the best diagnostic efficacy (Fig. 8A). Therefore, we decided to conduct further research on CDH19. Firstly, a PPI network of genes closely associated with CDH19 was constructed through the STRING website, and then imported into Cytoscape software for processing (Fig. 8B). The correlation heatmap showed that CDH19 had the highest correlation with CDH10, CDH11, and CDH18 (correlation coefficient > 0.4), with all p-values less than 0.001(Fig. 8C). To detect the expression of CDH19 in GC and normal gastric tissues, the researchers included STAD samples from the GTEx database to reduce sample bias. After integration, a total of 620 samples were obtained, comprising 210 normal gastric samples and 410 GC samples. Visualization using the “limma” package revealed that CDH19 expression was higher in GC, suggesting that CDH19 may be play a role in the occurrence and development of GC. Subsequently, the researchers validated through the GSE54129 cohort that the expression level of CDH19 in GC was also higher than that in normal gastric tissue (Fig. 8D). Therefore, to verify CDH19 expression levels, we performed the qRT-PCR experiment and found that CDH19 was considerably up-regulated in GC cell lines (AGS, HGC-27, and MKN-7) (Fig. 8E). Additionally, we divided all individuals into high and low groups based on the expression of CDH19 and performed K-M analysis to better understand the impact of CDH19 on the prognosis of GC patients. The results revealed that OS, PFS, and DFS were shorter in the population with high CDH19 expression (Fig. 3H, S3B). As a result, we speculate that CDH19 may serve as a pro-oncogene and that patients with higher expression levels may have shorter lifetimes and worse prognoses. According to GSEA analysis, the high CDH19 expression population was associated with “PPAR_SIGNALING_PATHWAY”, “MAPK_SIGNALING_PATHWAY”, “HEDGEHOG_SIGNALING_PATHWAY”, “FOCAL_ADHESION” and “GAP_JUNCTION“(Figure S3C), which were linked to the occurrence of cancer and matrix activation. Furthermore, ssGSEA analysis showed that the low CDH19 expression group had more activated CD4T cells infiltration, whereas the high CDH19 group had a higher proportion of immunosuppressive cells, such as macrophages, MDSCs, and Tregs (Fig. 8F). Similarly, for immunotherapy-related indicators and biomarkers, the CDH19 high expression group exhibited higher TIDE, Dysfusion, and Exclusion scores (Fig. 8G) but lower MSI-H ratios and TMB scores (Fig. 8H-I), illustrating that CDH19 can be used to direct clinical immunotherapy and forecast immune efficacy.

Fig. 8

Identifying CDH19 as a potential gastric cancer biomarker. (A) ROC curves for diagnosing gastric cancer, including SLC7A2, CDH19 and CCN1. (B) PPI network of proteins closely related to CDH19. (C) Heatmap of CDH19 and related genes. (D) The expression levels of CDH19 in GC tissue and normal gastric tissue, including GTEx database, TCGA-STAD cohort, and GSE54129 cohort. (E) Determination of CDH19 expression levels in gastric cancer cell lines and normal gastric epithelial cells by qRT -PCR, and the experiment was repeated three times. (F) Immune cell infiltration analysis of high and low CDH19 expression groups. (G) The tumor immune dysfunction and exclusion score of different CDH19 expression groups. (H) Evaluation of CDH19 expression and microsatellite status. (I) The relationship between CDH19 expression, TMB score, and prognosis. ns P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001. ROC, receiver operating characteristics curve; PPI, Protein-Protein Interaction; qRT-PCR, quantitative real-time polymerase chain reaction; GC gastric cancer; TCGA, The Cancer Genome Atlas; TMB, tumor mutation burden

CDH19 affected the biological behavior of GC cells

We constructed RNAi negative control (sh-NC) and knockdown CDH19 (sh-CDH19) lentiviral vectors to evaluate the role of CDH19 in promoting GC cell proliferation, invasion, and metastasis. After transfecting them into the AGS and HGC-27 cell lines, we validated the knockdown effect using qRT-PCR assay. Figure 9A and B demonstrated that sh-CDH19-1 and sh-CDH19-2 in AGS, as well as sh-CDH19-2 and sh-CDH19-3 in HGC-27, exhibited the most pronounced knockdown effects. Thus, we selected the cells above for subsequent phenotypic experiments. Cell viability and proliferation were assessed using the MTT assay and colony formation assay. The findings indicated that the AGS knockdown group (Fig. 9C, E, H) and the HGC-27 knockdown group (Fig. 9D, F, H) had considerably lower cell viability and proliferation capacity than the control group (sh-NC). Following this, a 48-hour wound-healing experiment revealed that the migration ability of GC cells in the sh-CDH19 group was significantly reduced (Fig. 9G). All experimental data in Table S8. Thus, we preliminarily infer that knocking down CDH19 expression inhibited the growth and migration of GC.

Fig. 9

CDH19 promotes the proliferation and migration of GC cells. (A, B) The expression of CDH19 in gastric cancer cell control (sh-NC) and knockdown (sh-CDH19) groups was examined utilized the qRT-PCR assay and select the two cell lines with the most obvious knockdown for subsequent experiments, repeating the experiment three times. (C-F) The MTT assay was employed to evaluate the proliferation and viability of AGS and HGC-27 GC cells. (G) The wound-healing experiment assessed the GC cell’ migration capacity and calculate its migration rate. (H) Colony formation experiments of AGS and HGC-27 cells lines, including sh-NC and sh-CDH19 groups. (I) Iron staining of AGS and HGC-27 cells lines. ns P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001. **** P < 0.0001. GC, gastric cancer; qRT-PCR, quantitative real-time polymerase chain reaction

CDH19 affects ferroptosis and oxidative stress in GC cells

As shown in Fig. 9I, researchers used FerroOrange reagent for iron staining and found that the intensity and concentration of iron ion staining were higher in sh-CDH19 cells. The expression level of ACSL4 is lower in GES-1 cells, while the expression level of GPX4 is higher in AGS and HGC-27 cells (Fig. 10A), ACSL4, a lipid metabolism enzyme, is known to promote ferroptosis when its expression or activity is elevated [44], indicating that ferroptosis may be involved in the occurrence of GC. Subsequently, Western blot results exhibited an increase in ACSL4 transcription levels in the sh-CDH19 group, and a decrease in GPX4 transcription levels (Fig. 10B). Therefore, CDH19 may be involved in regulating ACSL4 and GPX4 related pathways, thereby affecting the occurrence of ferroptosis.In addition, to explore whether CDH19 influences the oxidative stress process, we used DCFH-DA and DHE reagents to measure ROS levels in GC cells. The study discovered that ROS was significantly increased in the sh-CDH19 groups of AGS and HGC-27 cell line (Fig. 10C, D), implying that suppressing CDH19 expression may enhance oxidative stress and increase ROS accumulation.

Fig. 10

CDH19 participates in ferroptosis and oxidative stress. (A) Expression levels of ACSL4 and GPX4 in GES-1, AGS and HGC-27 cell lines. (B) Exploring the effect of CDH19 on ferroptosis-related proteins GPX4 and ACSL4 in GC cells through Western-blot experiments. (C-D) Exploring the effect of CDH19 on reactive oxygen species levels in GC cells through immunofluorescence staining. Above experiments were repeated three times. GC, gastric cancer; GPX4, glutathione peroxidase 4