Nucleosome patterns in circulating tumor DNA reveal transcriptional regulation of advanced prostate tumor phenotypes

Advanced prostate cancer comprises distinct tumor phenotypes, but classifying these tumors clinically remains challenging. Circulating tumor DNA (ctDNA) may enable the non-invasive detection of tumor phenotypes to guide treatment. However, uncovering the transcriptional regulation of tumor phenotypes from ctDNA is not fully explored. Here, we sequenced whole genomes of ctDNA from mouse plasma in 26 patient-derived xenograft (PDX) models of androgen receptor active (ARPC) and neuroendocrine (NEPC) prostate cancers. We developed methods to predict the transcriptional activity by analyzing nucleosome patterns reflected in ctDNA at regions of gene promoters, histone modifications, transcription factor binding, and accessible chromatin. The activity of key transcription factors, including AR, ASCL1, HOXB13, HNF4G, and NR3C1, associated with tumor phenotypes from ctDNA. Then, we developed prediction models to classify NEPC from ARPC in 159 patients across three clinical cohorts with 97-100% sensitivity and 85-100% specificity. These results highlight that ctDNA serves as a useful analyte for accelerating diagnostic applications in precision oncology.