Background: Prostate cancer is the second leading cause of cancer death for men in the U.S., and metastatic castration-resistant prostate cancer (mCRPC) is the most advanced form. In previous work, we have reported prognostic significance of PTEN-PI3K-AKT pathway and DNA damage response aberrations detected in cell-free DNA (cfDNA) of mCRPC. Here we utilized nucleosome profiling pattern at androgen receptor (AR) binding sites to further investigate AR signaling pathway dysregulation of mCRPC. Methods: Using the next-generation sequencing PredicineScore liquid biopsy assay (low-pass whole genome sequencing, LP-WGS), we profiled cfDNA fragment coverage centered around AR binding sites (ARBS). For 118 mCRPC patient samples, we quantified sample-level ARBS nucleosome profiling abnormality score (ARBS score) by comparing its centric fragment coverage with normal plasma background using standard Z-score. ARBS score was correlated with clinical outcomes including PSA response rate (PSA RR), progression-free survival and overall survival (OS). Results: By comparing ARBS nucleosome profiling and normal plasma background, we found that mCRPC plasma samples had significant ARBS nucleosome depletion, reflective of transcription factor binding and persistent AR signaling. Notably, ARBS nucleosome depletion was prostate cancer specific, and did not occur in other examined cancer types. ARBS score was highly correlated with tumor fraction (R=0.82), indicating that high tumor fraction mCRPC sample had higher AR binding. Importantly, we found that high ARBS score (>2) in pre-treatment samples (n=86) was associated with significantly shorter OS and lower PSA RR compared with low ARBS (≤2) group for patients receiving AR pathway inhibitor (ARPI) or taxane chemotherapy (13.4 vs. 19.4 months, p=7.3×10^-3, 45.7% vs. 77.5%, p=5.2×10^-3). By integrating genome-wide DNA methylation profiling, we further found that ARBS was hypo-methylated in mCRPC patients, and identified a strong negative correlation between DNA methylation beta value and ARBS score (R=-0.83). Conclusions: ARBS nucleosome profiling abnormality inferred from LP-WGS can serve as a key prognostic tool for mCRPC patients. Patients with high ARBS score had worse clinical outcomes, highlighting the potential clinical utility of cfDNA fragmentomics analysis for mCRPC patient stratification. Integration of mutation, copy number variation and fragmentomics profiling of ARBS may further enhance our understanding of molecular aberrations and treatment outcomes in mCRPC.