一种用于预测p53恢复疗法响应并对适应证进行优先排序的转录组学框架

Introduction: TP53 has been recognized as a promising therapeutic target because of its pivotal role in tumor suppression and the high prevalence of its functional alterations in cancer. However, sensitivity to p53 restoration varies markedly across cancer cell lines and is not determined solely by the presence or absence of intact p53. These findings suggest that additional, pathway-level factors may modulate responsiveness to p53 reactivation. In this study, focusing on p53 restoration therapy, we aimed to (1) elucidate molecular mechanism underlying variable responses, (2) build a transcriptomic prediction model capable of estimating restoration sensitivity, and (3) apply this framework to prioritize clinical indications across tumor types. Methods: In vitro results of p53 restoration were curated from 28 cell lines (20 sensitive [S], 8 resistant [R]). Transcriptomic differences between S and R were evaluated by Gene set enrichment analysis (GSEA) and gene set variation analysis (GSVA) using p53-related gene sets from MSigDB. RNA expression data were normalized by Non-paranormal (NPN) transformation, and singscore values were computed to quantify pathway activity. A logistic regression model was trained to estimate the probability of S or R classification, with the cutoff (specificity ≥0.9) stabilized through bootstrap resampling and fixed at the median. External datasets - CCLE, TCGA, cBioPortal, and GEO (GSE271757, GSE223463, GSE169321) - were analyzed to explore appropriate clinical indications. Results: GSEA revealed a distinct downregulation of DNA elongation and Mismatch repair pathways in the S group, indicating reduced baseline DNA repair activity in S cells. Among gene sets tested, Signature A, achieved the strongest discriminatory power between S and R (PR AUC = 0.754). For indication prioritization, integrative analysis combining Signature A-based predictions from CCLE and TCGA, together with TP53 alteration prevalence from cBioPortal, identified lung, head and neck, ovarian cancers as Tier 1 indications. By investigating the biological nature of Signature A, we observed that DNA-damaging therapies may enhance p53 restoration sensitivity. Consistently, analysis of GEO cohorts with pre- and post-treatment RNA-seq data revealed an increased fraction of S from 52% to 96% in ovarian cancer, and 85% to 92% in pancreatic cancer following DNA-damaging therapy. Conclusion: Signature A, a transcriptomic signature, stratifies p53 restoration responsiveness across cancer types and provides a robust, reproducible framework for indication prioritization and for understanding the mechanistic context of p53 restoration during the patient treatment journey.