Cell-free DNA nucleosome and genomic signatures as predictors of outcomes with radium-223 in metastatic castration-resistant prostate cancer.

5039 Background: Radium-223 (Ra-223) is a bone-seeking alpha emitter that induces double-stranded DNA breaks. Previous studies have shown Ra-223 can prolong survival of men with mCRPC with symptomatic bone metastases and no visceral metastases. However, clinical outcomes are heterogeneous, and predictive biomarkers for lack of benefit or progression with visceral disease are lacking. We evaluated whether genomic and epigenetic features derived from cell-free DNA (cfDNA) could identify tumor characteristics associated with favorable or unfavorable outcomes with Ra-223 therapy. Methods: Pre-treatment plasma cfDNA was analyzed from 137 mCRPC patients who received Ra-223. Ultra-low pass and deep whole genome sequencing were performed. Tumor fraction and copy number alterations were estimated using ichorCNA. Transcription factor (TF) activity was derived by nucleosome profiling of cfDNA fragments around transcription starts sites (TSS) and TF binding sites using the tools Griffin and TritonNP. Genomic and epigenetic features were associated with clinical outcomes including completion of six Ra-223 cycles, overall survival, and development of new nodal, visceral, or liver metastases. A multimodal binary classifier integrating copy number variation and TF activity was constructed using LASSO feature selection and XGBoost with repeated, nested 5-fold cross-validation. External validation was performed using independent cohorts of mCRPC patients treated with docetaxel (n=46) or other agents (n=142) for the prediction of new metastatic involvement. Results: Higher pre-treatment tumor fraction and TF activity at specific loci including HIF3A, ZNF770, and PRDM6 were associated with adverse outcomes. A binary classifier predicting completion of Ra-223 demonstrated strong discrimination in cross-validation, with similar performance for prediction of liver metastasis progression. External validation in independent cohorts demonstrated moderate discrimination, supporting generalizability across treatment contexts. Conclusions: Our findings demonstrate that analysis of cfDNA-derived genomic and epigenomic features can identify biologically relevant tumor characteristics associated with benefit from Ra-223 in mCRPC. These findings support the development of minimally invasive biomarkers to refine patient selection and risk stratification. Our ongoing analyses look to integrate mutations, amplifications, and deletions in key disease-related genes to further explore tumor genotype-phenotype associations with Ra-223 outcomes.