The development and validation of a novel programmed cell death-related signature based on machine learning analysis of 15 programmed cell death patterns for predicting prognosis and therapeutic response in high-grade serous ovarian carcinoma.

BACKGROUND: High-grade serous ovarian carcinoma (HGSOC) is the most prevalent and aggressive histological type of ovarian cancer. Predicting the prognosis of HGSOC remains challenging. Dysfunction of programmed cell death (PCD) is confirmed to be involved in the development and progression of ovarian cancer. The study aims to develop a PCD-related prognostic model to predict prognosis and treatment response in HGSOC patients.

METHODS: Through mining The Cancer Genome Atlas of Ovarian Serous Carcinoma (TCGA-OV) dataset, we characterized the molecular patterns of HGSOC based on prognostic PCD-related genes via consensus cluster analysis and established a prognostic signature using least absolute shrinkage and selection operator (LASSO) Cox and multivariate step Akaike information criterion (AIC) analyses. Kaplan-Meier (K-M) and receiver operating characteristic (ROC) curves were used to evaluate its performance. The robustness of the signature was validated by GSE32062, GSE9891 [Gene Expression Omnibus (GEO)] and the International Cancer Genome Consortium (ICGC) database-Australia (ICGC-AU) datasets. The relationship between different molecular patterns and risk groups was visualized via a Sankey diagram. Differences in clinical features, pathways, TIME, and chemotherapy sensitivity were analyzed between the different risk groups.

RESULTS: A total of 26 PCD-related genes with prognostic value were identified. Through the unsupervised clustering approach, three distinct molecular patterns (C1, C2, and C3) were discerned. K-M and TIME analysis indicated C2 was an immune-active subtype with favorable prognosis. Subsequently, a seven-gene prognostic signature was constructed. K-M analysis and time-dependent ROC curves demonstrated the excellent prognostic value of the signature [area under the curve (AUC)1-year =0.67, AUC3-year =0.65, AUC5-year =0.65] and analyses of three independent validation cohorts yielded similar results. Moreover, based on the prognostic signature and clinical parameters, we constructed and validated nomogram models for patients with HGSOC, with reliable prognostic values in training and validation datasets. We further found that the majority of patients in the high-risk group had C1 and C3 patterns, while the bulk of patients with C2 were in the low-risk group, suggesting that patients in low-risk group may benefit from immune therapy. Subsequent analysis, including functional, pathway enrichment, and TIME analysis revealed that the immune activity-related pathways were significantly active in the low-risk group, which exhibited an abundance of CD8+ T cells and a high expression of immune checkpoint genes. Therefore, this group may have better sensitivity to immunotherapy. Chemotherapy drug sensitivity analysis indicated that insulin-like growth factor-1 receptor (IGF-1R) inhibitors, protein kinase B (Akt) inhibitors, and phosphatidylinositol 3-kinase (PI3K) inhibitors may be more beneficial for patients in the high-risk group. Finally, RAB38 was revealed to be significantly downregulated in HGSOC and was associated with favorable prognosis. Its mechanism may be related to the regulation of the TIME.

CONCLUSIONS: We established a novel PCD-related prognostic signature that can be used to effectively predict HGSOC prognosis and treatment response.