HRDex: a tool for deriving homologous recombination deficiency (HRD) scores from whole exome sequencing data

Abstract

BackgroundBreast and ovarian tumors in patients with biallelic BRCA1 and BRCA2 mutations either by germline mutations accompanied by allele-specific loss of heterozygosity (LOH) or truncal somatic mutations respond to PARP inhibition. The repair of double stranded DNA breaks in tumors these tumors leads to homologous recombination deficiency (HRD), which can be measured using a variety of genomic and transcriptomic signatures. However, the optimal biomarker for BRCA deficiency is unknown. MethodsWe developed HRDex to determine HRD and its composite scores from allele specific copy number data analysis of whole exome sequencing (WES) data and examined the discriminatory ability of HRDex and other genomic and transcriptomic measures to identify BRCA deficiency in breast and ovarian tumors from The Cancer Genome Atlas (TCGA). ResultsHRDex scores have high correlation with SNP array based HRD scores in both breast and ovarian cancers. HRDex scores have high discriminatory accuracy to distinguish BRCA deficient breast tumors, similar to SNP array based scores (AUC 0.87 vs 0.90); however, discriminatory ability for ovarian tumors was lower (AUC 0.79 vs 0.90). HRD-LST had the best discriminatory ability of the three composite HRD scores. HRDex had higher discriminatory ability for identification of BRCA deficiency than RNA expression based scores (eCARD, tp53, RPS and PARPi7) in breast and ovarian tumors. Tumor mutational burden (TMB) was associated with BRCA deficiency in breast but not ovarian cancer. Combining HRDex score with mutational signature 3 modestly increased discriminatory ability for BRCA deficient breast and ovarian tumors (breast: AUC 0.90 vs 0.87; ovarian: AUC 0.83 vs 0.79). ConclusionsWES based HRD scores perform similarly to SNP array HRD scores, and better than other genomic or transcriptomic signatures, for identification of tumors with BRCA deficiency due to biallelic BRCA loss.