Abstract 5884: Cancer subtype classification by cfDNA fragmentomics analysis

Introduction: Nucleosome occupancy patterns at transcription factor binding sites (TFBS) exhibit differences between circulating tumor DNA (ctDNA) in cancer patients and normal plasma controls, as well as among different tumor subtypes. Therefore, cell-free DNA (cfDNA) fragmentomics can be employed for cancer subtyping analysis. The tumor fraction in cfDNA can affect the performance of nucleosome profiling, and a minimal tumor fraction is required to obtain high sensitivity of cancer subtyping. This study investigated the detection sensitivity - specifically, the minimal tumor fraction required - for different TFBS analyses. Methods: We utilized the Griffin framework to classify tumor subtypes based on nucleosome profiling of cancer-specific TFBS from low-pass whole genome sequencing data of cfDNA. Androgen receptor (AR) and ASCL1 binding sites were used to distinguish between AR-dependent prostate cancer (ARPC) and neuroendocrine prostate cancer (NEPC). Estrogen receptor (ER+) ATAC-seq peaks were employed for subtyping of ER-positive breast cancer. For each tumor type, we selected three tumor samples with highest nucleosome profiling abnormality scores for simulation studies. Each cancer sample was down-sampled at various percentages and mixed with six down-sampled normal samples, with three replicates per normal sample using different random seeds, to obtain simulated cancer samples with a serious of predetermined tumor fractions. Results: The specificity of nucleosome profiling at AR, ASCL1 and ER binding sites was high (> 95%) for all 87 normal samples. The limit of detection (LoD) for tumor fraction of AR binding sites ranged between 10% and 14%, with the average of 12% and standard deviation of 0.0098. Similarly, the LoD for ER binding sites was between 10% and 13%. The LoD of ASCL1 binding sites was higher, ranging from 20% to 25%. When genome-wide DNA methylation data were incorporated, both the specificity and sensitivity of nucleosome profiling increased. Interestingly, we observed methylation loss at AR and ER binding sites in prostate and breast cancer patients respectively. Conclusion: Prostate and Breast Cancer cfDNA samples sequenced by low-pass whole-genome sequencing can be assigned to subtypes according to nucleosome profiling of AR, ASCL1 and ER binding sites. The minimum tumor fraction required for AR and ER is at least 10%, while for ASCL1 it is as high as 20%. Integrating DNA methylation data into nucleosome profiling analysis can enhance sensitivity and lower the limit of detection, which we are actively investigating. Citation Format: Lisha Zhu, Chao Dai, Shuang Gan, Shidong Jia, Pan Du. Cancer subtype classification by cfDNA fragmentomics analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 5884.