Abstract 2059: AI - optimized genomic homologous recombination deficiency test (HRDScar) to predict platinum and PARP inhibitor responses in high-grade serous ovarian cancer

Abstract

Abstract High-grade serous ovarian cancer (HGSC) is the most common and most lethal subtype of ovarian cancer. More than half of the HGSCs are defective in Homologous Recombination DNA repair (HRD), and sensitive to Poly-ADP Ribose Polymerase (PARP) inhibitors. Recently, a genomic HRD test based on three types of genomic scarring events (Large scale transitions; LST, Telomeric Allelic Imbalance TAI, Loss of Heterozygosity LOH) was shown to predict which patients benefit the most from PARP inhibitors. The HRD test, however, was originally designed and optimized in breast cancer, and therefore the details of genomic scarring events occurring in ovarian cancers are unknown. To characterize the genomic scarring events and define an optimal cut-off for HRDScar biomarker in ovarian cancer, we are using large publicly available dataset from the TCGA. We selected 103 HRD samples based on somatic and germline mutations, gene deletions or hypermethylation in the BRCA1/2 and RAD51 paralog genes, and 34 HR proficient (HRP) samples without any mutation, deletion or hypermethylation of the HR genes. To identify the detailed features of LOH, LST and TAI scarring events, we employed state-of-the-art machine-learning algorithm and statistics to optimally separate the HRD-samples from HRP in the TCGA SNP array dataset. Our new optimized genomic footprints and cut-offs showed improved accuracy to separate HRD from HRP compared to the previous algorithm (accuracy of 0.89 vs 0.79). The optimized HRDScar showed reliable performance in NGS-derived data and correlated with mutational signature 3 (p=2.2e-16, r2=0.5). Interestingly, the HRDScar levels also positively correlated with an HRD score derived from an ex-vivo RAD51-based functional assay for HRD performed in the prospective HERCULES samples (n=72). Using two independent validation cohorts (PCAWG, HERCULES), our optimized HRDScar more accurately predicted progression-free survival (PFS) and overall survival (OS) when compared to previous algorithms. Importantly, improved prediction of PFS was detected especially in patients without BRCA1/2 alterations (p= 1.9e-04, HR=0.70). We are in the process of analyzing HRDScar from a clinical trial involving PARP inhibitor Niraparib (TOPACIO/Keynote-162 (NCT02657889)) enabling direct association of the HRDScar to clinical outcomes. In conclusion, HRDScar shows promise as a fully optimized algorithm that can be used for improved selection of patients for PARP inhibitor therapies in HGSC. Citation Format: Fernando Perez-Villatoro, Jaana Oikkonen, Manuela Tumiati, Julia Casado, Sakari Hietanen, Johanna Hynninen, Elizabeth P. Garcia, Panagiotis Konstantinopoulos, Sampsa Hautaniemi, Liisa Kauppi, Anniina Färkkilä. AI - optimized genomic homologous recombination deficiency test (HRDScar) to predict platinum and PARP inhibitor responses in high-grade serous ovarian cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 2059.