A Ferroptosis-Related Gene Prognostic Index Associated With Biochemical Recurrence and Radiation Resistance for Patients With Prostate Cancer Undergoing Radical Radiotherapy

Abstract

Background: Ferroptosis is a new type of programmed cell death which has been reported to be involved in the development of various cancers. In this study, we attempted to explore the possible links between ferroptosis and prostate cancer (PCa), and a novel ferroptosis-related gene prognostic index (FGPI) was constructed to predict biochemical recurrence (BCR) and radiation resistance for PCa patients undergoing radical radiotherapy (RRT). Moreover, the tumor immune microenvironment (TME) of PCa was analyzed. Methods: We merged four GEO datasets by removing batch effects. All analyses were conducted with R version 3.6.3 and its suitable packages. Cytoscape 3.8.2 was used to establish a network of transcriptional factor and competing endogenous RNA. Results: We established the FGPI based on ACSL3 and EPAS1. We observed that FGPI was an independent risk factor of BCR for PCa patients (HR: 3.03; 95% CI: 1.68-5.48), consistent with the result of internal validation (HR: 3.44; 95% CI: 1.68-7.05). Furthermore, FGPI showed high ability to identify radiation resistance (AUC: 0.963; 95% CI: 0.882-1.00). LncRNA PART1 was significantly associated with BCR and might modulate the mRNA expression of EPAS1 and ACSL3 through interactions with 60 miRNAs. Gene set enrichment analysis indicated that FGPI was enriched in epithelial-mesenchymal transition, allograft rejection, TGF beta signaling pathway, and ECM receptor interaction. Immune checkpoint and m6A analyses showed that PD-L2, CD96, and METTL14 were differentially expressed between BCR and no BCR groups, among which CD96 was significantly associated with BCR-free survival (HR: 1.79; 95% CI: 1.06-3.03). We observed that cancer-related fibroblasts (CAFs), macrophages, stromal score, immune score, estimate score, and tumor purity were differentially expressed between BCR and no BCR groups and closely related to BCR-free survival (HRs were 2.17, 1.79, 2.20, 1.93, 1.92, and 0.52 for cancer-related fibroblasts, macrophages, stromal score, immune score, estimate score, and tumor purity, respectively). Moreover, cancer-related fibroblasts (coefficient: 0.20), stromal score (coefficient: 0.14), immune score (coefficient: 0.14), estimate score (coefficient: 0.15), and tumor purity (coefficient: -0.15) were significantly related to FGPI, among which higher positive correlation between cancer-related fibroblasts and FGPI was observed. Conclusion: We found that FGPI based on ACSL3 and EPAS1 might be used to predict BCR and radiation resistance for PCa patients. CD96 and PD-L2 might be a possible target for drug action. Besides, we highlighted the importance of immune evasion in the process of BCR.

Keywords: biochemical recurrence; ferroptosis-related gene prognostic index; immune checkpoint; prostate cancer; tumor immune microenvironment.

FIGURE 1

Flowchart of this study. WGCNA = weighted gene coexpression network analysis; GO = Gene Ontology; KEGG = Kyoto Encyclopedia of Genes and Genome; GSEA = gene set enrichment analysis; FGPI = ferroptosis-related gene prognostic index; mRNA = messenger RNA; lncRNA = long noncoding RNA; RP = radical prostatectomy; RRT = radical radiotherapy.

FIGURE 2

Identification of the FGPI score. (A) Modules and phenotype showing tumor-related genes using blue, salmon, brown, and magenta modules; (B) volcano plot presenting the results of DEGs between 248 normal and 476 tumor tissues in the three GEO datasets (Kuner et al., 2013; Penney et al., 2015; Sinnott et al., 2017); (C) Venn diagram showing the results of the intersection of tumor-related genes, DEGs, and ferroptosis-related genes; (D) identifying three genes (ACSL3, EPAS1, and NEDD4L) using the Lasso regression analysis; (E) identifying genes independently associated with BCR in GSE116918 (Jain et al., 2018) using the COX regression; (F) risk factor plot showing the prognostic data and mRNA expression of ACSL3 and EPAS1; (G) determining the prognostic value of the FGPI score for BCR-free survival through the univariate and multivariate COX analyses including the clinical features of PCa patients undergoing RRT in GSE116918 (Jain et al., 2018); (H) ROC curve discriminating BCR from no BCR for PCa patients undergoing RRT in GSE116918 (Jain et al., 2018) using the FGPI score; (I) time-dependent ROC curve discriminating BCR from no BCR for PCa patients undergoing RRT in GSE116918 (Jain et al., 2018) using the FGPI score. FGPI = ferroptosis-related gene prognostic index; ROC = receiver operating characteristic; BCR = biochemical recurrence; GEO = Gene Expression Omnibus.

FIGURE 3

Clinical values and interaction network. (A) Sankey plot showing flow trend of clinical data and outcomes in GSE116918 (Jain et al., 2018); (B) comparison of FGPI score between BCR and no BCR groups for PCa patients in GSE116918 (Jain et al., 2018); (C) PCa patients in GSE116918 (Jain et al., 2018) were divided into high- and low-risk groups based on the median of the FGPI score and comparison of Gleason score between high- and low-risk groups; (D) Kaplan–Meier curve showing the difference of BCR-free survival for PCa patients undergoing RRT in GSE116918 (Jain et al., 2018) according to the median of the FGPI score; (E) Kaplan–Meier curve showing the difference of metastasis-free survival for PCa patients undergoing RRT in GSE116918 (Jain et al., 2018) according to the median of the FGPI score; (F) Kaplan–Meier curve of the internal validation showing the difference of BCR-free survival for 70% of PCa patients undergoing RRT in the GSE116918 (Jain et al., 2018) according to the median of the FGPI score; (G) Kaplan–Meier curve showing the difference of metastasis-free survival for PCa patients undergoing RP in TCGA database according to the median of the FGPI score; (H) ROC curve showing the diagnostic accuracy of the FGPI score for radioresistance; (I) protein–protein interaction network showing genes might interact with ACSL3 and EPAS1 using the GeneMANIA database (Warde-Farley et al., 2010); (J) Venn plot showing potentially interacting miRNAs of ACSL3, EPAS1, and PART1; (K) interaction network of competing endogenous RNAs and transcription factors. ROC = receiver operating characteristic; FGPI = ferroptosis-related gene prognostic index; BCR = biochemical recurrence; RRT = radical radiotherapy; RP = radical prostatectomy.

FIGURE 4

Functional enrichment analysis. (A) GO analysis showing the results of BP; (B) GO analysis showing the results of MF and KEGG analysis; (C) GSEA analysis of high- and low-risk groups (according to the median of the FPGI score) for PCa patients undergoing RRT in GSE116918 (Jain et al., 2018) using the subset of “c2.cp.kegg.v7.4.symbols.gmt;” (D) GSEA analysis of high- and low-risk groups (according to the median of the FPGI score) for PCa patients undergoing RRT in GSE116918 (Jain et al., 2018) using the subset of “h.all.v7.4.symbols.gmt.” GO = Gene Ontology; KEGG = Kyoto Encyclopedia of Genes and Genome; GSEA = gene set enrichment analysis; BP = biological process; MF = molecular function.

FIGURE 5

Drug sensitivity analysis and TME analysis. (A) Plot showing the top 30 potentially sensitive drugs to ACSL3 and EPAS1 using the CTRP; (B) plot showing the top 30 potentially sensitive drugs to ACSL3 and EPAS1 using the GDSC; (C) Venn plot showing the commonly sensitive drugs to ACSL3 and EPAS1 in the CTRP and GDSC; (D) violin plot showing the difference of the mRNA expression of m6A genes between BCR and no BCR groups for PCa patients in GSE116918 (Jain et al., 2018); (E) violin plot showing the difference of the mRNA expression of immune checkpoints between BCR and no BCR groups for PCa patients in GSE116918 (Jain et al., 2018); (F) patients in GSE116918 (Jain et al., 2018) were divided into high- and low-expression groups according to the median score of CD96, and Kaplan–Meier curve presented the difference of BCR-free survival for high- and low-expression groups; (G) forest plot showing the difference between BCR and no BCR groups and prognostic values of the TME parameters for PCa patients in GSE116918 (Jain et al., 2018); (H) heatmap showing correlation among FGPI score and TME parameters. GDSC = genomics of drug sensitivity in cancer; CTRP = the cancer therapeutics response portal; TME = tumor immune microenvironment; FGPI = ferroptosis -related gene prognostic index; CAF = cancer-associated fibroblasts; BCR = biochemical recurrence.