Identification of an optimal circulating tumor DNA (ctDNA) shedding threshold to detect actionable driver mutations in colorectal and pancreatic adenocarcinoma.

Abstract

3571 Background: In colorectal cancer (CRC), mutations in KRAS, NRAS, and BRAF predict resistance to anti-EGFR therapies. In pancreatic ductal adenocarcinoma (PDAC), ~90% of patients harbor KRAS mutations, while KRAS wild-type tumors often have clinically actionable fusion alterations. Thus, in both cancers, the accurate ascertainment of RAS and BRAF driver status is essential. Sequencing of cell-free DNA (cfDNA) from plasma allows convenient assessment of a tumor’s molecular profile, but sensitivity can be limited by low ctDNA shedding. We sought to establish a ctDNA shedding threshold at which actionable driver mutations can be reliably detected. Methods: Molecular reports and matched clinical data were obtained from the Duke Molecular Registry of Tumors and the SCRUM-Japan GOZILA and GI-SCREEN. CRC or PDAC patients with a pathogenic KRAS, NRAS, or BRAF activating point mutation (“driver”) present on tissue next-generation sequencing (NGS) assays and who also had cfDNA assay available were included. Tissue NGS included Foundation One CDx and Oncomine Comprehensive Assay. Guardant 360 (G360) was the sole plasma cfDNA assay. 131 CRC and 24 PDAC cases with 189 total G360 assays met criteria and were included. Samples were analyzed according to detection of the driver mutation and the maximum mutant allele frequency (MAF) of non-driver mutations on G360. An optimal cut-point for max MAF was explored among the CRC and PDAC patients using a maximally selected Wilcoxon rank statistic method. Results: 76.8% of driver mutations were in KRAS, 22.6% in BRAF, and 1.9% in NRAS with an overlap of 1 BRAF and 1 NRAS mutation with a KRAS mutation. Overall sensitivity of G360 for drivers was 83.0% for CRC and 54.2% for PDAC. No variants were detected on G360 in 9.1% of CRC and 37.5% of PDAC. Sensitivity for driver mutations increased with higher maximum non-driver MAF, with MAF > 1% predicting sensitivity > 98% (Table). Optimal cut-point analysis identified MAF of > 0.34% (p < 0.0001), above which the driver was identified in 97% of patients and below which only 27%. Conclusions: In our study, non-driver MAF > 1% on cfDNA NGS predicts high sensitivity for RAS and BRAF mutations and thus is adequate to guide clinical decisions such as anti-EGFR therapy in CRC, evaluation for fusions in PDAC, and validity in clinical trials. MAF ≤0.34% is a clear threshold to consider an assay inadequate and thus seek alternative testing. Sensitivity rises for MAF between 0.35 and 1% but will require greater patient numbers to establish clinically relevant sensitivity thresholds. These results will be updated with additional data for final presentation. [Table: see text]