Development of a methylation-based, tissue-agnostic test for the detection of molecular residual disease by circulating tumor DNA.

266 Background: Clinical validation studies support tumor-informed molecular residual disease (MRD) as a prognostic biomarker for disease recurrence across multiple solid tumor types. However, these tests are not always feasible due to the occasional lack of tumor tissue. Here, we discuss the design of a test for tissue-agnostic MRD detection and its application to a cohort of patients with colorectal cancer (CRC). Methods: A targeted panel composed of differentially methylated regions was developed. A machine-learning model was trained on differential methylation patterns in order to classify plasma samples as MRD-positive or MRD-negative. Performance of the independently trained classifier was assessed in a cohort of 247 patients enrolled in the Bespoke CRC trial (NCT04264702). These patients had MRD results available using a tumor-informed circulating tumor DNA (ctDNA) assay (Signatera), of whom 163 were persistently MRD-negative without clinical progression and 84 had MRD-positive results. Tissue-agnostic MRD results were compared to the tumor-informed results by calculating the percent positive agreement (PPA). Additionally, the differentially methylated allele fraction (DMAF) from the tissue-agnostic test was compared with variant allele frequencies (VAFs) from the tumor-informed test. Results: In the Bespoke CRC clinical cohort (72% non-Hispanic White, 54% male, mean age 61.4±12.3 years), 71 (28%) patients had stage II CRC, and 147 (60%) had stage III CRC. Overall, PPA was 86% (95% CI: 70-100%) and specificity was 97% (95% CI: 93-100%). When categorizing based on tumor-informed MRD VAF levels, PPA was 97% for VAF >0.2%, 100% for VAF 0.1-0.2%, 89% for VAF 0.04-0.1%, and 68% for VAF <0.04%. The DMAFs strongly correlated with the VAFs from the tumor-informed test and the correlation was independent of disease stage, histology, age, and sex. Conclusions: This is the first study of its kind demonstrating high concordance between a tissue-agnostic MRD test and a clinically validated tumor-informed ctDNA assay. These findings demonstrate that in cases where tissue is not available or of inadequate quality, a methylation-based tissue-agnostic assay may serve as a potential alternative for MRD detection.