4588 Background: APEX (Associating Plasma Epigenomics with eXpression) is a machine-learning framework that infers genome-wide gene expression in cancer from plasma cell-free chromatin Immunoprecipitation sequencing (cfChIP) by integrating signal from multiple histone marks and fragmentomic features, enabling enhanced transcriptional readouts from liquid biopsy without tissue sampling. We investigated whether APEX can noninvasively quantify NECTIN4 expression to predict outcomes with EV, a NECTIN4-targeted antibody-drug conjugate, in metastatic bladder cancer. Methods: Baseline plasma (1 mL) was collected from patients with metastatic bladder cancer within 90 before to 8 days after start of EV monotherapy and profiled by cfChIP-seq, followed by APEX-based tumor gene-expression inference. APEX-inferred NECTIN4 expression was dichotomized into high and low groups by the cohort median and tested for association with objective response (CR/PR vs SD/PD). Progression-free survival (PFS) and overall survival (OS) were analyzed using log-rank test and multivariable Cox regression model, accounting for the presence of bone or liver metastases and cfDNA tumor fraction. APEX performance was then compared against NECTIN4 locus signal from individual histone marks, plasma tumor fraction, and NECTIN4 copy-number status/amplification. Results: In the EV-treated cohort ( n =24), baseline APEX-inferred NECTIN4 expression was significantly higher in responders versus non-responders (p = 0.002) and outperformed NECTIN4 estimates derived from single histone-mark features, tumor fraction, and NECTIN4 copy number. Patients with high plasma-inferred NECTIN4 had an objective response rate of 58%, whereas no responses were observed among those with low inferred NECTIN4 , supporting strong negative predictive value. High baseline APEX-inferred NECTIN4 was also significantly associated with improved progression-free and overall survival (HR = 0.22, 95%CI: 0.08 – 0.65, p = 0.005 and HR = 0.27, 95%CI: 0.10 – 0.75, p = 0.008, respectively), with stronger associations than plasma tumor fraction, NECTIN4 copy number/amplification, or individual histone-mark coverage. In multivariable Cox models, APEX-inferred NECTIN4 remained independently associated with survival. Moreover, responders were enriched for urothelial luminal genes known to associate with NECTIN4 expression and favorable outcomes, while non-responders showed increased activation of epithelial-mesenchymal transition-related genes, known to associated with worse outcomes. Conclusions: A machine learning framework for plasma-based inference of tumor gene expression identifies plasma-based NECTIN4 as a clinically actionable, expression-based biomarker that predicts EV response and survival in metastatic bladder cancer.
Nawfal et al. (Wed,) studied this question.