A Multi-component Bioinformatics Study on the Construction of a Prognostic Signature of Genes Associated with Diverse Programmed Cell Death in Acute Leukemia and a Multi-perspective Mechanism Exploration

Abstract

Abstract Objective To evaluate the predictive value of diverse PCD related genes on the prognosis of AML patients and explore their roles in the development of AML and immunomodulatory therapy using bioinformatics methods. Methods We downloaded clinical and transcriptome sequencing data of AML patients from TCGA, GEO, and GTEX databases. Then, we obtained 12 PCD patterns related genes, including apoptosis, necroptosis, pyroptosis, ferroptosis, cuproptosis, intrinsic cell death, NETosis, dependent cell death, lysosome-dependent cell death, autophagy-dependent cell death, intracellular alkalinization-induced cell death, and reactive oxygen species-induced cell death. We randomly divided the complete AML samples into training and validation sets. A machine learning algorithm was used to establish a 6-gene signature that quantifies the risk score of AML's cell programmed death (PCD-Risk). We validated the predictive performance of PCD-Risk in multiple databases. We determined the molecular subtypes associated with AML through unsupervised clustering analysis. We constructed a bar plot by combining PCD-Risk with clinical features. Additionally, we analyzed the correlation between PCD-Risk and immune checkpoint genes, tumor microenvironment components, and drug sensitivity. Results We successfully constructed a prognosis model consisting of 6 PCD-related genes using a machine learning algorithm and validated its predictive accuracy in multiple datasets. The PCD-Riskscore exhibited good predictive performance for AML patients, with an AUC value greater than 0.70 in both the training and validation sets and up to 0.85. We identified two AML-related molecular subtypes through unsupervised clustering analysis, which have different essential biological processes. We constructed a high-predictive bar plot by combining PCD-Risk with clinical features. Moreover, we analyzed the correlation between PCD-Risk and drug sensitivity. The results showed that high-risk scores were resistant to AML chemotherapy drugs (5-fluorouracil, dasatinib, cisplatin, docetaxel, imatinib, paclitaxel, mitoxantrone, olaparib, oxaliplatin, rapamycin, vincristine, and zoledronic acid). Therefore, drugs targeting these genes' regulation may be a potential therapeutic target for AML chemotherapy-resistant patients. Finally, through comprehensive analysis of the overall and single-cell transcriptome, we found that PCD-Riskscore is associated with immune checkpoint genes and tumor microenvironment components. Conclusion Our study comprehensively analyzed various PCD pattern-related genes and successfully constructed a new prognosis model that can predict AML patients' prognosis and drug sensitivity.