Mutation Load Measured Using A 315 Gene Panel Predicts Genome-Wide Mutation Load

High tumor mutation load has been associated with better response to immunotherapy. Measuring mutation load via whole-genome DNA sequencing can be cost prohibitive and requires extensive analysis and data management. Targeted genomic, on the other hand, are an appealing alternative. However it is unclear how mutation load assessed using a sample of hundreds of genes relates to genome-wide mutation burden. Here we use mutation data from The Cancer Genome Atlas (TCGA) to investigate if the exonic mutation load in a small set of genes can be used to predict the genome-wide exonic mutation load. Using a gene panel composed of 315 genes, we observed a strong (R = 0.72) positive correlation between the total mutation burden and the gene panel mutation burden. To determine whether these genes allow high and low mutation burden samples to be accurately identified, we derived various classifiers on a training set of TCGA samples and evaluated their performance on held-out test data. High mutation load was defined as greater than 181 non-synonymous mutations. This threshold best distinguished microsatellite instable (MSI) high training set samples from MSI low and microsatellite stable (MSS) samples (95% true positive rate and 15% false positive rate). Receiver operating characteristic (ROC) curve analysis revealed that the 315-gene panel had excellent power to discriminate high and low mutation samples, with the area under the ROC curve evaluated on held-out test data ranging from 0.85 to 0.97 across indications. Our analysis suggests that the optimal threshold for identifying high mutation load samples may vary by indication. Our models were trained and tested using data generated by the TCGA consortium on pre-treatment patient biopsies. Direct application of our classifiers to clinical data would assume similar mutation calling sensitivity between platforms and similar underlying patient populations. To test these assumptions, we compared the mutational loads between TCGA and clinical samples within the same indication. We further performed a sensitivity analysis using a simulation-based approach which showed how deviations from these underlying assumptions would be expected to affect classification performance. These results demonstrate the feasibility of using mutation burden in a cancer related gene panel as a biomarker for genome-wide mutation load. This approach may be useful in identifying patients more likely to respond to cancer immunotherapies, but may require development be tailored to the specific sequencing platform and patient-population of interest. Citation Format: Artur Veloso, Z. Alexander Cao, Derek Y. Chiang, Hans Bitter, Kenzie D. MacIsaac. Mutation load measured using a 315 gene panel predicts genome-wide mutation load. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 854.