Identify non-mutational p53 loss of function in human cancers

Abstract

An accurate assessment of TP53 status is critical for cancer genomic medicine. Tumors with non-mutational p53 inactivations that failed to identify from DNA sequencing are largely misclassified as p53-normal, which leads to inaccurate prognosis and downstream association analyses. Here we built the support vector machine (SVM) models to systematically evaluate p53’s functional status in multiple TCGA cancer cohorts. Cross-validation using independent samples demonstrated the excellent performance of the SVM models with mean AUC = 0.9845, precision = 0.9844 and recall = 0.9825. Our model revealed that most (87–100%) wild-type TP53 (TP53^WT) tumors are loss-of-function. Further analyses revealed that these genetically normal but functionally impaired tumors (TP53^WT-pLoF) exhibit similar genomic characteristics as those p53 mutants with significantly increased tumor mutation burden, copy number variation burden, aneuploidy score, and hypoxia score. Clinically, compared to p53-normal tumors, patients with TP53^WT-pLoF tumors have significantly shortened overall survival and disease-free survival and exhibited increased sensitivity to platinum- based chemotherapy and radiation therapy. The increased sensitivity to radiation therapy is further validated in glioblastoma patient-derived xenografts models.