The mRNA expression profiles of GPX4 between the tumor tissues and the corresponding normal tissues across different cancer types were explored using R on Xiantao platform (https://www.xiantao.love/), where the RNAseq data in transcripts per million reads (TPM) format from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression database (GTEx) was converted by log2 for analysis and comparison. The Human Protein Atlas (https://www.proteinatlas.org/) was applied to explore the expression levels of GPX4 in tissue through immunohistochemical method. The DNA methylation and protein level of GPX4 in different cancer types was analyzed by CPTAC platform in UALCAN (http://ualcan.path.uab.edu/index.html). We also used the GEPIA2 (http://gepia2.cancer-pku.cn/#index) and the TISIDB website (http://cis.hku.hk/TISIDB/index.php) to examine the connection between the expression level of GPX4 and pathological characteristics of the tumor.

The Cox regression and Kaplan-Meier (KM) survival curves were used to assess the prognostic value of GPX4, including overall survival (OS), disease-specific survival (DSS), and progression-free interval (PFI). Based on data from the TIDE database (http://tide.dfci.harvard.edu), we evaluated the association between methylation levels of GPX4 and OS of patients with different cancers. In TIDE, first, they rank all patients by the molecular level of the gene in the test. Second, at each threshold position, they cut the gene value into 1 (higher than the current threshold) and 0 (otherwise) and run the Cox-PH regression on binary gene values to compute the z-score associated with the target gene covariate. Finally, they select the threshold where the z-score had the same sign (+/-) as the z-score from continuous regression without a cutoff and achieved the largest absolute value (13).

Using the cBioPortal platform (https://www.cbioportal.org/), genetic alterations of GPX4 across TCGA cancers were identified, encompassing mutations, structural variants, amplifications, and deep deletions. Furthermore, survival analyses were conducted to compare cancers with and without genetic alterations in GPX4.

At the single-cell level, CancerSEA (http://biocc.hrbmu.edu.cn/CancerSEA/home.jsp) was used to examine the correlation between GPX4 expression and tumor biological functions. To visualize the expression distribution of GPX4 in cancer, the t-distributed stochastic neighbor embedding (t-SNE) diagram was generated.

The association between GPX4 expression and various immune infiltrating cells, such as cancer-associated fibroblasts (CAF), neutrophils, macrophages, B cells, dendritic cells, CD8+ T cells, T-regulatory cells, NK cells, and monocytes, was assessed using TIMER2.0 (http://timer.cistrome.org/). Concurrently, the correlation between GPX4 expression and the expression of immune checkpoint genes was explored using TIDE (http://tide.dfci.harvard.edu).

Additionally, the correlation between GPX4 expression levels and scores related to stroma, immune cells, and the ESTIMATE algorithm was determined using the Sangerbox 3.0 platform (http://www.sangerbox.com/home.html), with the R software package ESTIMATE (version 1.0.13, https://bioinformatics.mdanderson.org/public-software/estimate/) being utilized for the calculations.

The gene network analysis tool BioGRID (https://thebiogrid.org/) was utilized to examine the networks linked to GPX4. With the top 100 genes that showed significant correlation with GPX4 identified from the GEPIA2 database, GO and KEGG analyses were performed to elucidate the biological and molecular functions of GPX4. The analysis was carried out using the R package.

The human colorectal cancer cell line SW480, and the gastric cancer cell line AGS were obtained from the American Type Culture Collection (ATCC; Manassas, VA, USA). SW480 cells were cultured in RPMI 1640 medium (10-040-CV, Corning, Manassas, VA, USA), whereas AGS were cultured in F-12K medium (10-025-CVR, Corning, Manassas, VA, USA), supplemented with 10% fetal bovine serum (A5669701, FBS; Gibco, Paisley, UK), and maintained at 37 °C in a humidified incubator with 5% CO2. Transfections were carried out using Lipofectamine 3000 reagent (L3000-015, Invitrogen, Carlsbad, CA, USA) following the manufacturer’s instructions. After transfection for 48 hours, the cells were used for following assays.

Cells were disrupted using chilled RIPA buffer from Beyotime (P0013, Shanghai, China), and samples were prepared in sodium dodecyl sulfate (SDS) buffer, separated by 12% SDS-PAGE, and then transferred onto a polyvinylidene fluoride (PVDF) membrane. Then, membranes were first blocked with 5% non-fat milk solution for 1 hour at ambient temperature before being incubated with the corresponding primary antibodies (GPX4: 67763-1-Ig, proteintech, China; β-actin: 66009-1-Ig, proteintech, China) at 4 °C for an overnight period. Finally, after 1-hour incubation with HRP-conjugated secondary antibodies (SA00001-1, proteintech, China) at room temperature, protein detection was achieved using Western ECL substrates from BIO-RAD (170-5061, Hercules, CA, USA).

For the MTS test, a total of 1000 cells were plated into each well of the 96-well cell culture plates 48 hours post-transfection, with each group having four replicates. To measure cell growth, 20 μL of AQueous Cell Proliferation Assay solution (G3581, Promega, Madison, WI, USA) was added to each well and then incubated at a temperature of 37 °C for a duration of 2 hours. Optical density (OD) readings at 490 nm were taken using a microplate spectrophotometer on day 0 (six hours post-cell seeding) as well as on days 1-3.

The EdU assay was conducted utilizing the Cell-Light™ EdU Apollo^®643 In Vitro Imaging Kit from RiboBio (C10310-1, Guangzhou, China), adhering to the provided guidelines. The treated cells were seeded into 24-well plate 24 hours prior to the assay. EdU was diluted in the culture medium at a ratio of 1:1000 and then incubated for 2 hours. After fixation with a 4% paraformaldehyde solution and permeabilization with 0.5% Triton-100, the cells were incubated with 100μL of 1× Apollo staining reaction solution for 30 minutes in the dark with gentle agitation. After thorough washing, the cell nuclei were stained with Hoechst33342 solution for 30 minutes. Finally, the cellular fluorescence was examined under a fluorescence microscope (Olympus IX51, Japan).

As for in vitro experiment result, the graphical results were presented as the mean ± standard deviation (SD) of three independent replicates, and Student’s t-test was performed for the statistical analysis using GraphPad Prism 8.3 software. A p-value < 0.05 was considered to indicate statistical significance. * indicated p < 0.05, ** indicated p < 0.01, and *** indicated p < 0.001.

According to TCGA data alone, GPX4 expression was upregulated in colon adenocarcinoma (COAD), esophageal carcinoma (ESCA), head and neck squamous cell carcinoma (HNSC), kidney renal clear cell carcinoma (KIRC), kidney renal papillary cell carcinoma (KIRP), liver hepatocellular carcinoma (LIHC), lung adenocarcinoma (LUAD), prostate adenocarcinoma (PRAD), rectum adenocarcinoma (READ), sarcoma (SARC), stomach adenocarcinoma (STAD), thyroid carcinoma (THCA), and uterine corpus endometrial carcinoma (UCEC), but downregulated in breast invasive carcinoma (BRCA) ( Figure 1A ). Next, we estimated GPX4 expression in paired cancer tissues and adjacent normal tissues using TCGA datasets. GPX4 expression was significantly higher in COAD, ESCA, KIRC, KIRP, LIHC, STAD, and THCA compared to paired adjacent normal tissues, but significantly lower in BRCA ( Figure 1B ). Since normal tissue is limited in TCGA dataset for some types of cancer, we combined data from TCGA and GTEx. In 27 out of 33 cancer types, GPX4 expression differed significantly. The overexpression of GPX4 was observed in several cancers, including adrenocortical carcinoma (ACC), COAD, diffuse large B-cell lymphoma (DLBC), ESCA, glioblastoma multiforme (GBM), HNSC, kidney chromophobe (KICH), KIRC, KIRP, brain lower grade glioma (LGG), LIHC, LUAD, lung squamous cell carcinoma (LUSC), ovarian serous cystadenocarcinoma (OV), pancreatic adenocarcinoma (PAAD), PRAD, READ, SARC, skin cutaneous melanoma (SKCM), STAD, THCA, thymoma (THYM), UCEC, and uterine carcinosarcoma (UCS). However, GPX4 was downregulated in BRCA, LAML, and testicular germ cell tumors (TGCT) compared to normal tissues ( Figure 1C ). Subsequently, we further analyzed protein levels of GPX4 using a large-scale proteome dataset from the National Cancer Institute’s CPTAC dataset. A significant increase in GPX4 expression in lung cancer and a decrease in breast cancer was observed, which was consistent with the transcriptional data ( Supplementary Figure S1A ). Consistent with the sequencing results, the results of immunohistochemistry also support that the expression level of GPX4 in colon cancer and liver cancer was significantly higher than that in the corresponding normal tissues ( Supplementary Figures S1B, C ).

In view of the pathological and clinical feature of GPX4 across different cancer types, we investigated whether GPX4 could be used as a potential biomarker for the diagnosis of human cancers. As shown in Figure 1D and Supplementary Table S1 , GPX4 expression levels were highly reliable in separating cancer from normal tissue, especially in THCA, KIRP, UCEC, READ, LIHC, ESCA, and COAD, where the area under the curve (AUC) exceeded 0.8. Considering that there were a limited number of normal tissues in the TCGA database, we further included the GTEx data into the analysis and found that the discriminating value of GPX4 in some tissues was even improved. AUCs exceeded 0.9 in PAAD, LAML, THYM, TGCT, DLBC, THCA, and LIHC ( Supplementary Table S2 ).

We also observed significant correlations between GPX4 expression and disease stage in ESCA, HNSC, THCA, and TGCT ( Supplementary Figure S2A , Supplementary Table S3 ). Furthermore, analysis using the TISIDB database revealed that expression of GPX4 was associated with tumor grade in HNSC, KIRC, LGG, LIHC, STAD, and UCEC ( Supplementary Figure S2B ).

Since GPX4 is highly expressed in various cancer tissues, we hypothesize that it plays an oncogenic role in these tumors. To test this hypothesis, we conducted functional experiments in colorectal and gastric cancer cell lines. Initially, we transfected AGS gastric cancer cells and SW480 colorectal cancer cells and with GPX4 overexpression plasmid, and Western blot (WB) experiments confirmed the efficiency of overexpression ( Figure 2A ). Further functional experiments revealed that GPX4 overexpression significantly promoted the proliferation of AGS and SW480 cells ( Figure 2B ). Concurrently, we observed that GPX4 overexpression increased the proportion of cells in the proliferative phase ( Figure 2C ). The above results suggest that GPX4 plays a pro-oncogenic role in colorectal and gastric cancers.

The GPX4 gene may serve as an accurate predictor of survival outcomes across a wide range of cancers. According to Cox regression, GPX4 expression levels were significantly associated with OS in six cancer types: BRCA, COAD, LAML, THCA, UCEC, and UVM ( Figure 3A ). The results from the KM survival curves also demonstrated that elevated GPX4 levels were correlated with poorer OS in COAD, LAML, and UVM, but with better OS in BRCA, THCA, and UCEC ( Figure 3B ). In terms of DSS analysis, the forest plot derived from Cox regression outcomes indicated that GPX4 expression had a positive correlation with the hazard ratios for DSS in STAD, but a negative correlation in UCEC ( Figure 3C ). Similarly, KM analysis revealed that increased GPX4 expression predicted worse DSS in STAD patients, but better DSS in UCEC patients ( Figure 3D ). Both Cox regression and KM analyses indicated that high GPX4 expression was linked to poor PFI in ACC, STAD, and UVM ( Figures 3E, F ).

Using the cBioPortal web server, we investigated genetic alterations of GPX4 across a range of cancers. As depicted in Figure 4A , SARC had the highest rate of GPX4 mutations, approximately 8%, with amplification being the predominant form of genetic alterations. ACC, Glioma, miscellaneous neuroepithelial tumors, and cholangio carcinoma (CHOL) also showed amplification-dominated variants. In OV and cervical squamous cell carcinoma and endocervical (CESC), both female reproductive system cancers, deep deletion emerged as the most prevalent type of mutation.

Subsequently, we investigated the mutation types and mutation sites within the GPX4 genomic sequence. A total of 27 mutations were identified, with missense mutations being the most common (18 cases). Truncating mutation, splice mutation, and SV/fusion mutation were each observed in 3 cases. Notably, the truncating mutation G174A/D could potentially be a driver mutation in cancers ( Figure 4B ). A genetic alteration in GPX4 was associated with improved overall survival in patients with GBM (p=0.0257) ( Figure 4C ). In contrast, HNSC patients with GPX4 genetic alteration displayed a poor DSS (p=0.0481), although no significant was observe in OS (p=0.3870; Figure 4C ). In other cancer types, GPX4 genetic alterations did not significant effect OS or DFS ( Supplementary Figure S3 ).

It has been established that alterations in DNA methylation patterns influence gene expression (14). Consequently, we investigated the methylation levels of GPX4 across TCGA pan-cancers utilizing the UALCAN database. Compared to normal samples, GPX4 promoter methylation levels were found to be decreased in LIHC, TGCT, THCA, CHOL, READ, LUAD, KIRP, BLCA, ESCA, GBM, and PAAD. In contrast, PRAD, BRCA, and PCPG exhibited significantly elevated GPX4 promoter methylation levels compared to their normal counterparts ( Figure 5A , Supplementary Figure S4A ). Moreover, no significant alterations in GPX4 methylation levels were observed in other types of cancer ( Supplementary Figure S4B ). These findings imply that promoter methylation of GPX4 might be accountable for the dysregulated expression of GPX4 in these cancers.

Subsequently, we utilized the TIDE database to assess the relationship between methylation levels and survival outcomes. As shown in Figure 5B , an increased methylation level of GPX4 was associated with improved OS in glioma, PAAD, and SKCM, but with poorer OS in HNSC and KIRC.

The CancerSEA database was used to investigate the expression pattern of GPX4 at the single-cell level and its functional implications in cancer biology. As summarized in Supplementary Figure S5A , GPX4 expression was correlated with various functional states across cancers. Specifically, in NSCLC, GPX4 expression showed negative correlations with DNA damage, DNA repair, and the cell cycle, but positive correlations with metastasis, angiogenesis, inflammation, epithelial–mesenchymal transition (EMT), and quiescence ( Supplementary Figure S5B ). In UM, GPX4 expression was negatively associated with quiescence, DNA damage, apoptosis, DNA repair, metastasis, hypoxia, invasion, and differentiation ( Supplementary Figure S5C ). The single-cell expression distributions of GPX4 in NSCLC and UM are further displayed in Supplementary Figures S5D and S5E , respectively. Based on these findings, we propose that GPX4 plays a critical role in cancer development and progression.

Through comprehensive analysis, we investigated the potential relationship between GPX4 expression and tumor-infiltrating immune cells. Using seven computational algorithms including TIMER, EPIC, MCPCOUNTER, CIBERSORT, CIBERSORT-ABS, QUANTISEQ, and XCELL, we evaluated immune cell infiltration levels across all TCGA tumor types. A strong association was observed between GPX4 expression and macrophage infiltration in COAD, DLBL, ESCA, HNSC-HPV^-, SARC, and UCEC. Notably, ESCA, HNSC-HPV^-, and SARC exhibited infiltration of M2 type macrophages, suggesting a positive link between GPX4 expression and the presence of M2 macrophages ( Figure 6A ). Additionally, GPX4 expression was significantly correlated with the infiltration of cancer-associated fibroblasts (CAFs) in ESCA, HNSC, STAD, and TGCT, neutrophils in CESCs and HNSCs, B cells in HNSC-HPV⁺ and TGCT, CD8+ T cells in HNSC-HPV+, and regulatory T cells (Treg) in HNSC ( Figures 6B, C , Supplementary Figures S6A–C ). However, no significant correlation was observed between GPX4 expression and the infiltration of dendritic cells (DCs), natural killer (NK) cells, CD4+ T cells, monocytes, or other immune cell types ( Supplementary Figures S6D–I ).

Furthermore, we also computed immune and stromal scores using the ESTIMATE algorithm ( Supplementary Table S4 ). GPX4 expression showed the most significant correlation with immune scores in GBMLGG, SARC, and pan-kidney cancer (KIPAN) ( Figure 6D ). Regarding stromal scores, significant correlations between GPX4 expression and stromal scores were observed in GBMLGG, KIPAN, and KIRC ( Figure 6E ). As shown in Figure 6F , GPX4 expression was positively correlated with ESTIMATE scores in GBMLGG and SARC, but negatively correlated in KIPAN.

We next investigated the correlation between GPX4 expression and immune modulator pathways as well as genes associated with immune checkpoints using data from the TCGA database. The heatmap displayed that GPX4 expression was highly correlated with the levels of numerous chemokines and chemokine receptors, such as CXCL16, CCL28, CX3CL1, CCL5, CCR10, CXCR5, and CXCR3, across the majority of cancer types ( Supplementary Figure S7 ). Furthermore, the expression of GPX4 was closely associated with MHC molecules, immune activation genes, and immunosuppressive genes such as HLA-A, HLA-B, CD160, TNFRSF18, TNFRSF4 and CD276 in most cancer types. Subsequent analysis also demonstrated a clear relationship between GPX4 expression and key immune checkpoints such as VEGFB, TGFB1, CD276, TNFRSF18, and TNFRSF4 across most cancer types ( Supplementary Figure S7 ).

TMB and MSI are comprehensive genetic markers employed to forecast the effectiveness of immune checkpoint inhibitors (ICIs). Thus, we assessed the correlation between GPX4 expression and both MSI and TMB. GPX4 expression levels were positively correlated with TMB in UCEC, LUAD, and STES, but negatively correlated with TMB in LAML and HNSC ( Supplementary Figure S9A ). In UCEC, PRAD, THCA, LIHC, STES, and KIPAN, a positive correlation was observed between GPX4 expression and MSI, whereas a negative correlation was found in GBMLGG ( Supplementary Figure S9B ).

The therapeutic potential of GPX4 in cancer was investigated using publicly available datasets. Across 21 immunotherapy cohorts, GPX4 exhibited predictive capability for response to immune checkpoint-targeted therapy in 10 cohorts, with AUC values exceeding 0.5. Notably, in the Gide 2019 cohort, the predictive AUC for GPX4 was 0.83, surpassing the MSI score, TIDE, CD274, CD8, T. Clonality, and B. Clonality ( Supplementary Figure S10 ). Moreover, GPX4 has been demonstrated prognostic value in patients undergoing immune checkpoint blockade therapy. Higher GPX4 expression was associated with improved clinical outcomes in melanoma trials involving immune checkpoint inhibitors ( Supplementary Figure S11 ).

Finally, we performed functional enrichment analysis of GPX4-associated genes in cancer. GPX4-interacting proteins were explored using BioGRID ( Figure 7A ). To further explore the molecular mechanisms involving GPX4, we extracted the top 100 genes associated with GPX4 from GEPIA2 ( Supplementary Table S5 ). GPX4 consistently exhibited strong correlations with the top 10 most-associated genes across multiple cancer types ( Figure 7B ).

Subsequently, we carried out GO and KEGG functional enrichment analyses for GPX4. The biological processes (BP) primarily encompassed assembly of the mitochondrial respiratory chain complex, the electron transport chain, and the NADH dehydrogenase complex. The cellular components (CC) were predominantly located in complexes containing mitochondrial proteins, the mitochondrial respirasome, the respirasome itself, and the respiratory chain complex. Additionally, the enriched molecular functions (MF) pertained to electron transfer activity and the activity of oxidoreduction-driven active transmembrane transporters, aligning with its function in reducing reactive oxygen species (ROS). KEGG pathway analysis highlighted associations with Huntington’s disease, amyotrophic lateral sclerosis, thermogenesis, prion diseases, and oxidative phosphorylation ( Figure 7C ). Notably, many of these pathways and functions are linked to lipid peroxidation, underscoring the pivotal role of GPX4 in ferroptosis.