Noninvasive genomic profiling of somatic mutations in oral cavity cancers

Abstract

ObjectivesSomatic mutations may predict prognosis, therapeutic response, or cancer progression. We evaluated targeted sequencing of oral rinse samples (ORS) for non-invasive mutational profiling of oral squamous cell carcinomas (OSCC). Materials and methodsA custom hybrid capture panel targeting 42 frequently mutated genes in OSCC was used to identify DNA sequence variants in matched ORS and fresh-frozen tumors from 120 newly-diagnosed patients. Receiver operating characteristic (ROC) curves determined the optimal variant allele fraction (VAF) cutoff for variant discrimination in ORS. Behavioral, clinical, and analytical factors were evaluated for impacts on assay performance. ResultsHalf of tumors involved oral tongue (50 %), and a majority were T1-T2 tumor stage (55 %). Median depth of sequencing coverage was 260X for OSCC and 1,563X for ORS. Frequencies of single nucleotide variants (SNVs) at highly mutated genes (including TP53, FAT1, HRAS, NOTCH1, CDKN2A, CASP8, NFE2L2, and PIK3CA) in OSCC were highly correlated with TCGA data (R = 0.96, p = 2.5E−22). An ROC curve with area-under-the-curve (AUC) of 0.80 showed that, at an optimal VAF cutoff of 0.10 %, ORS provided 76 % sensitivity, 96 % specificity, but precision of only 2.6E−4. At this VAF cutoff, 206 of 270 SNVs in OSCC were detected in matched ORS. Sensitivity varied by patient, T stage and target gene. Neither downsampled ORS as matched control nor a naïve Bayesian classifier adjusting for sequencing bias appreciably improved assay performance. ConclusionTargeted sequencing of ORS provides moderate assay performance for noninvasive detection of SNVs in OSCC. Our findings strongly rationalize further clinical and laboratory optimization of this assay, including strategies to improve precision.