Evaluation of drug sensitivity, immunological characteristics, and prognosis in melanoma patients using an endoplasmic reticulum stress-associated signature based on bioinformatics and pan-cancer analysis
Alphonse Houssou Hounye,Bingqian Hu,Zheng Wang,Jiaoju Wang,Cong Cao,Jianglin Zhang,Muzhou Hou,Min Qi
DOI: https://doi.org/10.1007/s00109-023-02365-w
Abstract:We aimed to develop endoplasmic reticulum (ER) stress-related risk signature to predict the prognosis of melanoma and elucidate the immune characteristics and benefit of immunotherapy in ER-related risk score-defined subgroups of melanoma based on a machine learning algorithm. Based on The Cancer Genome Atlas (TCGA) melanoma dataset (n = 471) and GTEx database (n = 813), 365 differentially expressed ER-associated genes were selected using the univariate Cox model and LASSO penalty Cox model. Ten genes impacting OS were identified to construct an ER-related signature by using the multivariate Cox regression method and validated with the Gene Expression Omnibus (GEO) dataset. Thereafter, the immune features, CNV, methylation, drug sensitivity, and the clinical benefit of anticancer immune checkpoint inhibitor (ICI) therapy in risk score subgroups, were analyzed. We further validated the gene signature using pan-cancer analysis by comparing it to other tumor types. The ER-related risk score was constructed based on the ARNTL, AGO1, TXN, SORL1, CHD7, EGFR, KIT, HLA-DRB1 KCNA2, and EDNRB genes. The high ER stress-related risk score group patients had a poorer overall survival (OS) than the low-risk score group patients, consistent with the results in the GEO cohort. The combined results suggested that a high ER stress-related risk score was associated with cell adhesion, gamma phagocytosis, cation transport, cell surface cell adhesion, KRAS signalling, CD4 T cells, M1 macrophages, naive B cells, natural killer (NK) cells, and eosinophils and less benefitted from ICI therapy. Based on the expression patterns of ER stress-related genes, we created an appropriate predictive model, which can also help distinguish the immune characteristics, CNV, methylation, and the clinical benefit of ICI therapy. KEY MESSAGES: Melanoma is the cutaneous tumor with a high degree of malignancy, the highest fatality rate, and extremely poor prognosis. Model usefulness should be considered when using models that contained more features. We constructed the Endoplasmic Reticulum stress-associated signature using TCGA and GEO database based on machine learning algorithm. ER stress-associated signature has excellent ability for predicting prognosis for melanoma.