Application of an epigenomic-based classifier to identify cancer signal of origin on liquid biopsy in cancer of unknown primary cases.

3073 Background: Cancer of unknown primary (CUP) lacking resolution to a cancer type (Cancer Signal Origin; CSO) leads to suboptimal outcomes. While several approaches currently exist for identifying a cancer type for CUP samples, they rely on clinical approaches that often require tissue biopsies for immunohistochemistry (IHC), and even still they often fail to resolve a CSO (reported to occur in 50-80% clinical cases). This requirement for tissue, and typically lengthy diagnostic journey, create a large unmet need for CSO identification in CUP individuals. Here, we present feasibility data from a high accuracy Liquid Biopsy method for CSO identification in CUP, bypassing the need for a tissue biopsy and quickly returning a CSO to individuals with CUP. Methods: We developed a CSO prediction algorithm on Guardant360 utilizing DNA methylation signatures across thousands of cancer-specific differentially methylated regions for 14 cancer types. We applied the CSO classifier to 1,128 CUP samples in which circulating tumor DNA was detected. Accuracy was assessed by comparing CSO predictions to suspected diagnoses based on clinicopathologic and molecular findings. Results: The CSO prediction algorithm was evaluated on 1,128 CUP samples; lung (285/1128, 25.3%) and bile duct (166/1128, 14.7%) were the most common predicted CSOs, aligning with reported prevalence in the literature. Of the 1,128 samples, 12 had a suspected clinical diagnosis. These 12 spanned 8 tumor types. The top CSO prediction aligned with suspected diagnosis in 91.6% (11/12) of cases. The CSO algorithm also provides confidence scores to quantify the confidence of CSO prediction. Out of the total 12 cases, 7 CSOs had high confidence, 3 had moderate confidence, and 2 had low confidence. 7/7 high, 2/3 moderate, and 2/2 low confidence predictions were correct. The single incorrect moderate sample was diagnosed as CRC but predicted to be lung. Conclusions: These findings show the feasibility of using plasma-based epigenomic profiling to assign CSO with acceptable accuracy. While interventional clinical studies are necessary to demonstrate clinical utility, this has significant potential for guiding treatment decisions and improving outcomes in CUP patients without ready access to tissue.