Liquid biopsy to reveal colorectal cancer molecular subtype information from ctDNA epigenetics.

238 Background: Consensus molecular subtypes (CMS1-4) have shown prognostic value in colorectal cancer (CRC) and may predict therapy response, including enhanced benefit from irinotecan in CMS4 and potential immunotherapy response in CMS1. However, molecular subtyping in metastatic CRC is challenging due to limited tissue availability. We investigated whether ctDNA methylation profiling could identify CRC molecular subtypes from liquid biopsy. Methods: We performed a retrospective analysis of 724 blood samples from advanced cancer patients that were submitted to BillionToOne’s CLIA/CAP accredited clinical laboratory for liquid biopsy. Comprehensive genomic profiling and DNA methylation analysis of >2000 differentially methylated regions were performed using the Northstar Select and Response assays, respectively. Unsupervised clustering analyses were then applied to ctDNA epigenetic features to identify molecular patterns. Results: As a methodological control, we first confirmed the ability to differentiate between ctDNA of distinct histologies, i.e. breast, hepatocellular, pancreatic, gastric, and colorectal tumors, by unsupervised machine learning methods (PCA, tSNE and UMAP), exclusively on high dimensional methylation signals. Breast, hepatocellular, pancreatic and colorectal tumors each form distinct tSNE groups. This confirms that the high dimensional analysis of Northstar Response assay is able to classify cancers with different biological origins. Moreover, the separation of colorectal and hepatocellular samples is achieved on cfDNA with < 1% tumor fraction, displaying the Northstar Response assay ability to resolve tumor type within an organ system at low tumor fraction levels. We then identified several distinct ctDNA methylation clusters on the colorectal cancer samples alone. The 237 colorectal samples analyzed formed 9 clusters using Louvain clustering, and 4 distinct groups with UMAP. Notably, one cluster (5%) showed high enrichment of microsatellite instability high (MSI-H) and BRAF-positivity, which are characteristic features of the CMS1 (MSI/immune) subtype. One other cluster (20%) was enriched for high MYC copy number amplification, which is a characteristic of the CMS2 (canonical) subtype. Conclusions: ctDNA methylation profiling can distinguish tissue of origin with high sensitivity and likely identifies CRC molecular subtypes. DNA methylation has previously been shown to be the dominant biomarker for molecular subtyping of colorectal tumor tissues, supporting the feasibility of ctDNA methylation for CRC subtyping by liquid biopsy. Future investigation should investigate whether non-MSI CMS1 tumors identified by ctDNA methylation also respond to immune checkpoint blockade, potentially expanding immunotherapy eligibility beyond MSI status.