Liquid biopsies can detect actionable mutations and infer broad tumor states from genome-wide cfDNA measurements, but quantitative transcriptome-like phenotyping at single gene resolution still largely requires tissue. Here, we asked whether 6-base whole-genome sequencing that jointly quantifies 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC) could infer gene expression directly from plasma. We applied this framework to plasma from patients enrolled in a phase 2 clinical trial of the PARP inhibitor olaparib plus the PD-L1 inhibitor durvalumab for metastatic castration-resistant prostate cancer. Inferred plasma transcriptomes distinguished adenocarcinoma vs. neuroendocrine phenotypes and identified a noncanonical WNT5A-associated signature linked to poor clinical response. Integrating longitudinal cfDNA methylomic profiles with phylogenetic reconstruction further revealed two resistance trajectories: one featuring high tumor heterogeneity with persistent AR signaling, and another marked by an AR-independent, stem-like program with metabolic reprogramming. These findings demonstrate that ctDNA can inform phenotype-driven tumor biology at gene-level resolution, integrating epigenetic modifications, inferred transcriptional programs, and clonal dynamics as a function of treatment response.

I’ll describe how we used biomodal’s 6-base sequencing on serial liquid biopsies to infer tumor gene activity directly from plasma and capture clinically relevant tumor states. I’ll then show how combining this with deep ctDNA sequencing and longitudinal sampling reveals baseline signatures and evolving resistance patterns that track with treatment response.