HRD status prediction in patients with advanced breast, prostate, ovarian and pancreatic cancers in a liquid biopsy assay.

Abstract

3074 Background: Homologous recombination and repair (HRR) deficiency (HRD) is characterized by genomic instability associated with mutations in BRCA1/2 or other HRR genes. HRD can also be detected by copy number variant (CNV) features, indels, and SNVs (HRD signature). Patients with canonical BRCA-associated cancers harboring an HRD signature with or without HRR mutations derive clinical benefit from PARPi therapy. Here, we present a method of predicting HRD status using Guardant Infinity in patients with these advanced cancers. Methods: We developed an ensemble logistic regression model to predict HRD status, inferred from genome-wide somatic SNV and CNV signatures indicative of deficiency in HRR genes, including ploidy-adjusted large-scale state transitions (LST), whole-genome tumor loss of heterozygosity (LOH) and telomeric allelic imbalance (TAI). The model was trained on clinical samples processed on Guardant Infinity, a next-generation platform evaluating both genomics and epigenomics, to assess the sensitivity and accuracy of detecting biallelic loss-of-function in BRCA1/2. The aggregated model was tested on an independent pan-tumor clinical cohort and pre-treatment samples from a subset of patients enrolled in TRITON2, a phase 2 single arm study evaluating rucaparib in metastatic castration resistant prostate cancer (mCRPC) patients with HRR mutations. HRD status association with radiographic progression free survival (rPFS) was evaluated with Cox-proportional hazards model. Results: Our model demonstrated high sensitivity in patients with BRCA1/2 biallelic loss and high specificity in HRR-wildtype patients, with an AUC of 0.95 in a pan-tumor cohort with tumor fraction (TF) >10%. In an independent cohort of breast prostate ovarian and pancreatic samples, HRD detection ranged from 79-100% in samples with BRCA1/2 biallelic loss and >10% TF. In breast (n = 703) and prostate (n = 655) cancers, HRD was detected in 14.9% and 13.5% of samples with > 10% TF (3.6% and 3.7% in all TF), with 5.5% and 6.2% attributed to samples not harboring deleterious mutations in HRR genes, respectively, potentially reflecting non-genomic drivers of HRD. In a pilot cohort (n = 15) from TRITON2, HRD was detected in 100% of patients enrolled with either BRCA1/2, or PALB2 mutations (n = 10), where rucaparib demonstrated meaningful activity as measured by independent radiology review objective response rate. HRD was not detected in patients with mutations in CDK12, FANCA, or NBN (n= 5). HRD detected status was associated with prolonged rPFS (HR = 0.07, p = 0.03). Conclusions: Guardant Infinity can predict HRD status in patients with advanced canonical BRCA-associated cancers, with preliminary results indicating potential for predicting PARPi benefit in mCRPC. Further studies are warranted to determine PARPi response for breast, prostate, ovarian, and pancreatic cancers with detected HRD status.