Assessing Variability Across HRD Assays: Findings From the Friends’ HRD Harmonization Project

Association for Molecular Pathology (AMP) 2022 PRESENTATION
Authors Hillary Stires, Zhiwei Zhang, Lisa McShane, Jonathan Bieler, Li Chen, Vincent Funari, Mohit Gupta, Alexander J. Lazar, Brittany McKelvey, Sarabjot Pabla, Jerod Parsons, Daniel Saul, Omar Serang, Ethan Sokol, Elizabeth Starks, Shuang Yang, Jennifer Yen, Mark Stewart, Jeff Allen

Introduction – Homologous recombination deficiency (HRD) assays determine eligibility for treatment with PARP inhibitors and other DNA repair targeting drugs. The assays measure several factors to define homologous recombination (HR) status including causes (i.e., inactivation in HR pathway genes) and consequences (i.e., genomic instability) of HRD. Methodological variability across HRD assays has not been investigated thoroughly, and an empirical assessment of assay variability may support broader adoption of HRD and strengthen clinical interpretation of test results.

Methods – Friends of Cancer Research (Friends) initiated a unique partnership with HRD assay developers and other key stakeholders to characterize differences in assay factors and assess levels of agreement and variability across HRD assays. First, we surveyed HRD developers (n=20) about factors their assays measure to determine HR status. Subsequently, a subset of assay developers (n=11) measured in silico and reported HR status and the contributing factor(s) for 348 TCGA ovarian cancer samples. We performed pairwise comparisons of assay’s HR status calls to determine the level of agreement and considered specific factors measured by each assay to identify potential sources of variation. Additionally, we analyzed HR status agreement for BRCA1/2 mutated versus wild type BRCA1/2 samples.

Results The 20 surveyed HRD assays are heterogeneous in the factors they measure. While all assays consider BRCA1/2 mutations, assays also variously consider genomic loss of heterozygosity (gLOH; 75% of assays), additional HRR genes (55%), telomeric allelic imbalance (TAI; 45%), and large-scale state transitions (LST; 45%). For assays involved in the TCGA analysis, the range of percent positivity (% patients with HRD) was 9-67% with a median of 49%. Rates of HRD were higher in assays that included gLOH, TAI, and/or LST. Median positive percent agreement (PPA) was 74% and median negative percent agreement was 81%. The presence of BRCA1/2 mutations was associated with an increase in PPA. The median Spearman correlation for pairwise comparisons of ranked continuous HRD scores was 0.66 and 0.80 for %gLOH.

Conclusions – Preliminary findings demonstrate variation in the factors measured and the HR status calls made across HRD assays. Some of the variation in HR status calls could be due to the nature of the TCGA dataset, and future studies will aim to understand assay agreement from freshly extracted formalin-fixed paraffin-embedded human archival ovarian tumor samples. Understanding the agreement among assays will help to inform assay interpretation and improve consistency between HR status calls and alignment of HRD scores across HRD assays to help patients and providers make appropriate treatment decisions.