Equal-depth sequencing of white blood cells (WBC) and plasma from prostate cancer (PCa) liquid biopsies (LBx) and association with clonal hematopoiesis (CH) confounders in clinically relevant genes.

Abstract

221 Background: CH results from fitness-enhancing mutations in hematopoietic stem cells that accumulate with age. Deep sequencing of LBx, a standard of care for guiding therapy selection in advanced PCa, can be confounded by sensitive detection of variants (vars) arising from both ctDNA and CH-derived DNA. Here we report the prevalence of CH vars in PCa and other cancers and describe an algorithmic method for accurately distinguishing CH from tumor signal, even at low variant allele frequencies (VAF). Methods: Plasma cell-free DNA from 1813 patients with cancer (n = 270 PCa) was sequenced using FoundationOneLiquid CDx (F1LCDx) in a CLIA lab setting. In parallel, DNA from buffy coats was extracted, sheared, and sequenced on F1LCDx to identical depth, to establish ground truth for CH var identity. This cohort was used to train a machine learning variant origin prediction (VOP) model incorporating fragmentomics and other sequencing features to assign probabilities of origin (germline, tumor, or CH) to all short variants across the 324 genes baited on the assay. Only pathogenic vars were considered. Results: 1247/1813 (69%) of LBx detected ≥1 CH var, including 209/270 (77%) of PCa LBx (mean 1.8 vars/sample). CH contributed confounding vars in clinically relevant genes including CHEK2, ATM, BRAF, TP53, and BRCA2 in 90/270 (33%) of PCa LBx (Table). Median VAF of CH vars in these genes was 0.3% in LBx and 0.7% in WBC. CH vars enriched in PCa vs other cancers were: TET2 (23.0% of PCa LBx vs 13.6% of non-PCa), ASXL1 (14.4% vs 7.5%), SF3B1 (7.0 vs 2.3%), and TERT promoter (5.2% vs 1.3%), FDR <0.05 for all. In the pan-cancer cohort, the VOP algorithm identified CH vars with 95.2% (2613/2745) sensitivity, 94.1% (3741/3975) specificity, and 91.8% (2613/2847) positive predictive value (PPV). For CH vars with VAF<1%, VOP had 95.0% (1622/1707) sensitivity, 1021/1169 (87.3%) specificity, and 91.6% (1622/1770) PPV. Conclusions: Equal-depth DNA sequencing of plasma and matched WBC reveals higher CH prevalence in LBx than previously reported: 77% of PCa LBx with a CH var, including 33% with a CH var in clinically relevant genes . Many potentially actionable vars have low VAF in LBx, and filtering CH via shallower depth WBC sequencing than that of LBx risks leaving low VAF CH vars to be mistaken for tumor vars. Equal-depth WBC sequencing or an algorithmic method validated using equal-depth WBC sequencing is required for accurate identification of low VAF CH confounders in LBx. Percentages of LBx with only CH vars/total LBx with vars detected in genes clinically relevant in PCa. Gene PCa Pan-cancer CHEK2 19/25 (76%) 143/168 (85%) ATM 18/29 (62%) 156/227 (69%) BRAF 2/5 (40%) 10/48 (21%) TP53 46/120 (38%) 235/996 (24%) BRCA2 2/12 (17%) 3/37 (8%) PALB2 0/3 (0%) 4/17 (24%) BRCA1 0/2 (0%) 5/28 (18%) CDK12 0/7 (0%) 3/17 (18%)