Poster #36: Katrin Donetski

Computational Detection of Candidate Pathogenic Mosaic Variants in High-Risk Breast Cancer Patients

Katrin Donetski¹, Mingyi Xia², Bradley Wubbenhorst², Mwanga Akamandisa³, Susan Domchek, MD²˒⁴, Katherine Nathanson, MD²˒⁴ ¹ Cornell University, Department of Biomedical Engineering, Ithaca, NY ² University of Pennsylvania, Division of Translational Medicine and Human Genetics, Philadelphia, PA ³ Yale School of Medicine, New Haven, CT ⁴ Basser Center for BRCA at Penn Medicine, Philadelphia, PA

Breast cancer (BC) affects approximately 13% of women, and inherited pathogenic variants (PVs) in susceptibility genes such as BRCA1 and BRCA2 contribute to disease risk. Conventional clinical germline testing is primarily designed to detect constitutional PVs present in nearly all cells, at a variant allele frequency of 50% when heterozygous. It may therefore be less sensitive to low-level mosaic variants, present in only a subset of cells (<30% VAF). Prior clinical germline testing may not have reliably detected these variants, particularly if they occur in regions with inadequate sequencing coverage. We performed high-depth targeted sequencing, with a mean target coverage exceeding 350X, on blood or saliva derived DNA samples from 160 individuals at high risk for BC. No PV in a known BC susceptibility gene had been identified in the participants’ prior clinical genetic testing records. Candidate variants were identified using Mutect2 and DeepVariant and annotated with clinical and research databases, including ClinVar, OncoKB and gnomAD. A computational pipeline was developed to remove likely sequencing artifacts and prioritize candidate mosaic PVs based on ClinVar pathogenicity classifications, OncoKB oncogenicity annotations, gnomAD population allele frequencies, and additional custom filtering criteria. We reviewed candidate mosaic PVs in the Integrative Genomics Viewer (IGV) to verify they have at least 5 supporting reads, which identified a mixture of likely germline variants and candidate low- or high-level mosaic variants. Three candidates in BARD1, CHEK2, and TP53 showed read patterns consistent with mosaicism. Specific mosaic variant callers, including MosaicForecast and DeepMosaic, will be used to further validate candidates in silico, followed by experimental validation with Sanger sequencing. Clonal hematopoiesis will also be evaluated as a possible origin of low-VAF variants. Together, these approaches may reveal previously undetected contributors to breast cancer susceptibility and advance personalized genomic risk assessment.