PMC: A more precise classifier of POLE mutations to identify candidates for immune therapy.

Abstract

3548 Background: Specific somatic mutations in DNA polymerase epsilon ( POLE) can cause a hypermutant phenotype with tumor mutation burden (TMB) in excess of 100 mutations per megabase. It has been reported that POLE mutant tumors are enriched in response to immune therapy and this association is being tested in multiple active clinical trials. However, most POLE mutations are passenger mutations and have no pathogenic role. Current methods to classify POLE mutations are limited in both accuracy and completeness, which could lead to inappropriate use of immune agents in tumor such as MSS CRC, where response rate is 5% or less. Here we present a new classifier, POLE Mutation Classifier or PMC, based on the unique trinucleotide mutation signature caused by selective loss of the proofreading function (LOP) of POLE. Methods: cBioPortal was queried to identify all tumors with POLE mutation. TMB was calculated for each, additionally, trinucleotide mutation signatures were obtained for all POLE mutant tumors in TCGA. Using OncoKB to identify a gold standard of 12 functional POLE mutations (n = 98 tumors) a POLE mutational signature was created. A combination of mutational signature, amino acid location, and TMB was used to classify each POLE variant. Results: Among all 48035 unique tumors the overall frequency of POLE mutations was 2.5% (n = 1184), however only 9.2% (n = 110) were determined to cause the selective LOP. The incidence of LOP POLE mutation was highest in uterine carcinoma and CRC, these tumors also had the highest ratio of LOP to passenger mutations. In a pan-cancer analysis the overall survival of LOP POLE patients was significantly better than those with passenger mutations (not-yet-reached vs. 51 mo, HR = 4.4, p < 0.0001). A similar analysis performed using the polyphen-2 classifier to identify functional POLE mutations did not show a difference in overall survival (HR = 1.0, p-value = 0.57). To further validate the improved specificity of the PMC classifier TMB was used as a surrogate marker, using the PMC classifier 98% of tumors with LOP showed hypermutation (TMB > 20mut/Mb), vs. 53% called functional by polyphen-2. A retrospective analysis of MD Anderson CRC patients identified 25 patients with LOP POLE mutation, who had improved OS relative to 267 CRC patients with passenger POLE mutation (not-yet-reached vs. 70 mo, HR:4.2, p = 0.028). Four metastatic CRC patients with LOP POLE mutation were treated with immune therapy (nivolumab, or ipilimumab/nivolumab) in 2nd or 3rd line, all four achieved objective response and remain on therapy (mean time on treatment 15 mo). Conclusions: The PMC classifier specifically identifies mutations in POLE that cause loss of the proofreading function, outperforming both manually curated databases and machine learning-based methods. Clinical trials that use POLE mutation as a selection criteria for immune therapy should be restricted to just those POLE mutations that cause LOP.