Cell-free DNA (cfDNA) fragmentomes predict tumor burden in metastatic colorectal cancer (mCRC)

• Published in: Journal of clinical oncology Vol. 40; no. 16_suppl; p. 3541
• Main Authors: Leal, Alessandro, van 't Erve, Iris, Lumbard, Keith, Keefer, Laurel, Carey, Jacob, Wu, Tony, Jakubowski, Debbie, Butler, Denise, Rongione, Michael, Chesnick, Bryan, Punt, Cornelis J. A., Meijer, Gerrit A., Dracopoli, Nicholas C, Maddala, Tara, Scharpf, Robert, Bach, Peter Brian, Velculescu, Victor E., Fijneman, Remond
• Format: Journal Article
• Language: English
• Published: American Society of Clinical Oncology 01.06.2022
• ISSN: 0732-183X; 1527-7755
• DOI: 10.1200/JCO.2022.40.16_suppl.3541

3541Background: Measurement of plasma mutant allele fraction (MAF) in patients with cancer provides prognostic information, but this approach typically relies on prior tumor tissue analyses or knowledge of specific mutations. There is a clinical need to develop rapid and accurate noninvasive plasma-only approaches to estimate disease burden dynamics. Comprehensive genome-wide analyses of cfDNA fragmentomes has become a useful tool for noninvasive cancer characterization. Here we present a novel application to predict MAF from genome-wide fragmentation-related profiles (fMAF) as a prognostic marker in mCRC patients with initially unresectable liver-only metastases enrolled in the phase III CAIRO5 study (NCT02162563). Methods: 693 longitudinal plasma cfDNA samples were obtained from patients with RAS/BRAF-mutant mCRC (training arm, N = 78) and patients with RAS/BRAF wild-type mCRC (validation arm, N = 75) treated with first-line fluoropyrimidine-based chemotherapy were collected. MAFs as measured by digital droplet PCR (ddPCR) were obtained for patients in the training arm. We trained an initial regression model to predict MAFs with genome-wide fragmentation features using subject-level random intercepts. Out of sample predictions for participants in the RAS/BRAF-mutant mCRC arm were obtained through cross-validation. Cox proportional hazards models were used to evaluate the association between fMAF and PFS and overall survival (OS) at pre- and first post-treatment blood draws. Standardized hazard ratios (sHR) and 95% confidence intervals (CI) are reported. Results: There were 68 and 61 evaluable participants with median (range) age of 62 y (41-79 y) and 58 y (27-76 y), and 38% and 36% female in the training and testing arms, respectively. Median time between pre- and post-treatment blood draw was 8 weeks. In the training cohort, median fMAFs dropped from pre- to post-treatment initiation (17.8% to 1.8%). Median PFS was 9 months. For both fMAFs and RAS/BRAF MAFs, the association with PFS was stronger for the pre-treatment than the post-treatment timepoint (Table). Post-treatment fMAFs were not associated with PFS in the training cohort and similar trends were observed for OS. Conclusions: Initial modeling demonstrates the ability of cfDNA fragmentomes to estimate cell-free DNA tumor burden with performance comparable to standard approaches. Ongoing modeling efforts will result in evaluation of a final model in the testing arm. The development of a noninvasive tissue-independent diagnostic approach that does not rely on mutation detection in the circulation has the potential to expand the use of liquid biopsies for advanced disease monitoring. Clinical trial information: NCT02162563.fMAF (Training Arm)MAF (Training Arm)sHR (95% CI)sHR (95% CI)Pre-treatmentPFS1.23 (0.96 to 1.58)1.51 (1.16 to 1.97)OS1.40 (1.06 to 1.85)1.63 (1.22 to 2.18)Post-treatmentPFS0.98 (0.77 to 1.24)1.19 (0.92 to 1.54)OS1.10 (0.89 to 1.36)1.54 (1.16 to 2.06)