Abstract
BACKGROUND: Evaluation for activating mutations in KRAS, NRAS, and BRAF in colorectal cancer (CRC) and in KRAS in pancreatic ductal adenocarcinoma (PDAC) is essential for clinical care. Plasma cell-free DNA (cfDNA) next-generation sequencing (NGS) allows convenient assessment of a tumor's molecular profile, however low tumor DNA shedding limits sensitivity. We investigated mutant allele frequency (MAF) of other oncogenic dominant genes to identify a threshold for accurate detection of KRAS, NRAS, and BRAF (RAS/RAF) mutations in cfDNA. METHODS: Molecular and clinical data were obtained from the Duke Molecular Registry of Tumors and the SCRUM-Japan GOZILA study. Patients with CRC or PDAC and a KRAS, NRAS, or BRAF activating single nucleotide variant (SNV) present on tissue NGS and with available cfDNA assays were included. Recursive partitioning and Wilcoxon-rank statistics methods identified potential cut-points for discriminative MAF values. RESULTS: One hundred and thirty-five CRC and 30 PDAC cases with 198 total cfDNA assays met criteria. Greatest non-RAS/RAF dominant gene MAF of 0.34% provided maximum discrimination for predicting RAS/RAF SNV detection. Sensitivity for RAS/RAF SNVs increased with dominant gene MAF, with MAF ≥1% predicting sensitivity >98%, MAF between 0.34 and 1% predicting sensitivity of 84.0%, and MAF £0.34% predicting sensitivity of 50%. For 43 cfDNA assays that did not detect RAS/RAF SNVs, 18 assays detected 34 other oncogenic variants, of which 80.6% were not also detected on tissue. CONCLUSIONS: Non-RAS/RAF dominant oncogenic mutation MAF ≥1% on cfDNA NGS predicts high sensitivity to detect RAS/RAF oncogenic SNVs in CRC and PDAC. MAF £0.34% indicates an assay may not reliably detect RAS/RAF SNVs, despite detection on tissue testing. Most variants from assays that did not detect RAS/RAF had MAF <1% and were not detected on tissue, suggesting potential confounding. These data suggest a practical approach to determining cfDNA assay adequacy, with implications for guiding clinical decisions in CRC and PDAC.