Development of an immune‐related prognostic model for pediatric acute lymphoblastic leukemia patients

• Published in: Molecular genetics & genomic medicine 2020-09, Vol.8 (9), p.e1404-n/a
• Main Authors: Quan, Xi, Zhang, Nan, Chen, Ying, Zeng, Hanqing, Deng, Jianchuan
• Format: Article
• Language: English
• Subjects: Acute lymphoblastic leukemia; Antiinfectives and antibacterials; Bioinformatics; Biomarkers; Calcium-binding protein; Cancer therapies; Chemotaxis; Chemotherapy; Gene expression; Genes; Hematology; Immune response (humoral); immune‐related genes; Leukemia; Lymphatic leukemia; Malignancy; Medical prognosis; Model accuracy; Optimization; Original; Patients; pediatric acute lymphoblastic leukemia; Pediatrics; Performance prediction; prognosis; Regression analysis; Risk assessment; Risk groups; S100 protein; Software; Survival; Survival analysis; Time dependence
• ISSN: 2324-9269; 2324-9269
• DOI: 10.1002/mgg3.1404

Background Acute lymphoblastic leukemia (ALL) is the most common hematological malignancy in pediatrics, and immune‐related genes (IRGs) play crucial role in its development. Our study aimed to identify prognostic immune biomarkers of pediatric ALL and construct a risk assessment model. Methods Pediatric ALL patients’ gene expression data were downloaded from Therapeutically Applicable Research to Generate Effective Treatments (TARGET) database. We screened differentially expressed IRGs (DEIRGs) between the relapse and non‐relapse groups. Cox regression analysis was used to identify optimal prognostic genes, then, a risk model was constructed, and its accuracy was verified in different cohorts. Results We screened 130 DEIRGs from 251 pediatric ALL samples. The top three pathways that DEIRGs may influence tumor progression are NABA matrisome‐associated, chemotaxis, and antimicrobial humoral response. A set of 84 prognostic DEIRGs was identified by using univariate Cox analysis. Then, Lasso regression and multivariate Cox regression analysis screened four optimal genes (PRDX2, S100A10, RORB, and SDC1), which were used to construct the prognostic risk model. The risk score was calculated and the survival analysis results showed that high‐risk score was associated with poor overall survival (OS) (p = 3.195 × 10−7). The time‐dependent survival receiver operating characteristic curves showed good prediction accuracy (Area Under Curves for 3‐year, 5‐year OS were 0.892 and 0.89, respectively). And the predictive performance of our risk model was successfully verified in testing cohort and entire cohort. Conclusions Our prognostic risk model can effectively divide pediatric ALL patients into high‐risk and low‐risk groups, which may help predict clinical prognosis and optimize individualized treatment. Differentially expressed immune‐related genes between the relapse and non‐relapse groups of 251 pediatric acute lymphoblastic leukemia patients were analyzed. Four optimal genes (PRDX2, S100A10, RORB and SDC1) were screened, and then an immune‐related prognostic model based on these four genes were constructed.