DOI: 10.1200/go.2023.9.supplement_1.89 ISSN: 2687-8941

Assessing consistency of calculating homologous recombination deficiency (HRD) score using different segmentation method.

Xiaopei Chao, Lei Li, Zhentian Kai, Shuilong Zhang, Lisha Huang, Hongsheng He
  • Cancer Research
  • Oncology

89

Background: A number of similar tests were developed since FDA approval of using niraparib in HRD positive (HRD score >=42) patients. However, various factors can substantially influence HRD score estimation. We hereby evaluated the concordancy of bioinformatics pipelines, especially different copy number variation (CNV) algorithms, in inferring HRD scores. Methods: A total of 195 ovarian cancer patients enrolled from April 2021 to Dec 2022 were used for retrospective analysis in this study. Both tumor tissue and blood samples were sequenced with Topgen HRD panel, which covered ~35,000 single nucleotide polymorphisms (SNPs). Sequence data were trimmed and aligned to hg19 reference genome. Two commonly used CNV pipeline (i.e. Sequenza, and PureCN which have tumor-normal mode and tumor-only mode) were adopted to generate allele specific copy number (ASCN) profile. HRD scores were then inferred from ASCN profile using R package scarHRD. Pearson correlation was used for evaluating correlation between HRD scores generated with different algorithms. Paired T-test and Wilcoxon signed-rank test was used in comparing differences in HRD scores and segment counts from different algorithms. Percentage agreement was evaluated based on cutoff of HRD score >=42. Results: HRD scores produced from Sequenza and PureCN (tumor-normal mode) showed significant correlation (R = 0.83, P value < 0.001). Sequenza tended to over-segment than PureCN (median segment counts: 139 vs 82, P value < 0.001) and have higher HRD scores (mean HRD scores: 55.0 vs 23.5, P value < 0.001). Using Sequenza, HRD-positivity was observed in 70.0% of the total population, whereas PureCN classified only 39.0% as HRD-positive. HRD scores inferred from two modes of PureCN showed improved correlation (R = 0.89, P value < 0.001). Tumor-only mode generated higher mean HRD score than tumor-normal mode (42.7 vs 36.0, p-value < 0.001). Prevalence of HRD-positivity was 47.7% using tumor-only mode. Using tumor-normal mode as the reference test, the percentages of agreement were shown in the Table. Conclusions: Comparison of different CNV algorithms demonstrated significant variances in HRD scores estimation. Despite good correlation between different pipelines was observed on HRD scores, the estimated prevalence of HRD-positive patients varied significantly from 39% to 70%. Our results urge the necessity of harmonisation on bioinformatics pipeline during HRD score calculation. [Table: see text]

More from our Archive