Methylation-based tumor fraction monitoring identifies patients with putative molecular progression who may benefit from comprehensive genomic profiling

Background: Longitudinal monitoring of circulating tumor DNA (ctDNA) using methylation-based tumor fraction (TF) provides sensitive detection of disease dynamics in cancer patients over time. Patients with increasing TF may be undergoing molecular progression (MP) preceding clinical progression. They may have developed clinically relevant variants (CRVs, associated with response or resistance to targeted therapies) detectable by comprehensive genomic profiling (CGP) that can guide treatment options. Here, we describe a workflow that applies TF change thresholds to identify patients with putative MP, maximizing detection of CRVs. Methods: To establish the thresholds for significant TF change, we used in-silico dilutions (ISD) from clinical samples to train an input-dependent model targeting 90% analytical specificity. We also determined an absolute TF threshold to maximize detection of CRVs. These thresholds were analytically evaluated by comparing two aliquots from the same blood draw (n=1061, simulating “no TF change” between consecutive samples), as well as longitudinal pan-cancer samples from patients undergoing treatment (n=2103, representing real-world TF changes). The model was also clinically evaluated in the RADIOHEAD cohort 1 (n=116) in which patients with advanced solid tumors received standard-of-care immune checkpoint inhibitors (ICI). Adjusted hazard ratios (aHR) and p-values (p) were determined via Cox proportional hazards with sex, age, and baseline TF as covariates. Results: Significant TF change between consecutive timepoints of >50% increase, along with an absolute TF of >0.1% were yielded from simulation and optimization approach as thresholds for putative MP assessment. When evaluated in replicate plasma samples that simulate no biological change, these thresholds demonstrated empirical specificity of 97%. In serial samples from advanced cancer patients undergoing treatment, 29.5% met the criteria for MP. Critically, 55% of patients with putative MP harbored CRVs not detected at baseline. In the RADIOHEAD cohort, patients with putative MP had significantly worse real-world progression-free survival (rwPFS) compared to those with stable/decreasing TF (HR=4.30, 95% CI: 2.93-6.33, p<0.001). Among patients with putative MP, 65.1% had new CRVs, compared to 26.2% in patients with TF decrease. Conclusions: We developed criteria for putative MP that achieve 97% specificity and are associated with shorter rwPFS in patients receiving ICI. CGP in patients meeting these criteria detects new actionable variants in 55-65% of cases, supporting its utility for guiding therapeutic decision-making upon MP during longitudinal ctDNA monitoring. References 1. Liang S, et al. Cancer Res Commun. 2025;5(8):1384.