cfDNA-based signatures of nucleosome organization to enable prostate cancer subtyping.

5130 Background: Neuroendocrine prostate cancer (NEPC) can arise from prostate adenocarcinoma (PRAD) via therapy-associated lineage plasticity and is characterized by reduced AR signaling and activation of neuroendocrine (NE) regulatory pathways. Nucleosome positioning, inferable from cell-free DNA (cfDNA) fragmentation patterns, reflects lineage-specific gene regulation. Recent studies have shown that active, open chromatin is associated with specific nucleosome spacing patterns, leading to an enrichment of longer cfDNA fragments, including di-nucleosome fragments. We hypothesized that fragmentomic signatures reflecting nucleosome positioning could distinguish PRAD from NEPC, providing a simple means to detect lineage plasticity from cfDNA. Methods: We analyzed cfDNA samples from 47 patients (8 with NEPC, 39 with PRAD) with hybrid-capture sequencing. Baits were designed to target adenocarcinoma (AC, 715 sites) and NE-associated regulatory elements (508 sites). For each sample, fragments overlapping each enhancer set were aggregated into 5-bp binned fragment-length distributions (18-500 bp). We quantified di-nucleosome (250-400 bp) and mono-nucleosome (120-180 bp) fragment counts at NE and AC regulatory elements. We calculated the log-ratio of NE versus AC di/mono values: log2 [(di/mono)_NE/(di/mono)_AC]. We also computed a position-free Transcriptional Activation Score (pfTAS) for each enhancer set using a published framework for estimating enhancer activity from cfDNA fragment size distributions, deriving an analogous log-ratio: log2 [pfTAS_NE/pfTAS_AC]. We tested if these ratios could distinguish NEPC versus PRAD, evaluating the area under the receiver operating characteristic curve (AUC) in the full cohort and by leave-one-out cross-validation (LOOCV). Results: The enhancer-specific di/mono log-ratio achieved complete separation between PRAD (median: -0.209; range: -0.449 to -0.090) and NEPC (median: 0.030; range: -0.056 to 0.218), yielding an AUC of 1.00 both in-sample and by LOOCV. The pfTAS log-ratio was also discriminative, with an AUC of 0.99 in-sample and an AUC of 0.89 by LOOCV. Overall, nucleosome fragmentation patterns provide a chromatin-based fragmentomic signal that enables accurate discrimination between PRAD and NEPC. Conclusions: Leveraging hybrid-capture cfDNA, we identified a biologically grounded chromatin-based signature that robustly discriminates NEPC from PRAD. By reflecting lineage-associated nucleosome positioning, this liquid biopsy-based approach provides a cost-effective and non-invasive molecular metric derived from routinely generated cfDNA sequencing data. It may be particularly useful in clinically challenging settings, including cases with mixed morphology or when a biopsy is difficult to obtain.