Copy number loss and somatic loss of heterozygosity calling in liquid- and tissue-based genomic profiling

Introduction: Copy number loss (CNL) alterations are common oncogenic driver events important for therapy selection and clinical trial matching in several cancer types. We developed a CNL detection approach that pairs a purpose-built panel with a ploidy-aware, multi-signal model to sensitively detect loss of heterozygosity (LoH) and homozygous deletion (homdel) events from ~1MB to chromosome scale in both plasma- and FFPE-derived samples. Method: We employed hybrid-capture assays (Guardant360 Liquid and Guardant360 Tissue, Guardant Health, Palo Alto, CA) that target more than 700 cancer-associated genes and incorporate a dense tiling of common single-nucleotide polymorphisms (SNPs) to optimize segmentation resolution. Read depth and SNP minor allele fraction (MAF) are integrated within a ploidy-aware likelihood framework that jointly estimates tumor fraction (TF), ploidy, and allele-specific copy number (CN). Highly polymorphic loci (e.g. HLA) are handled with a unique target design that mitigates polymorphism and artifacts while preserving true deletions. A single caller operates across specimen types, with material-specific adjustments for ctDNA and FFPE tissue. Validation of the caller performance established deletion limit of detection (LoD), sensitivity, specificity and limit of blank (LoB) using a combination of clinical and contrived cell-free DNA (cfDNA) and genomic DNA (gDNA) samples. Result: Joint tumor fraction (TF)/CNL inference reduces miscalls and improves allele-specific state assignment in polyploid genomes at low TF. The CNL reportable range extends to >500 genes including coverage of the HLA locus, homologous recombination (HRR) and DNA mismatch repair (MMR) pathways. LoD was established at 20% TF in cfDNA and 30% in tissue-derived gDNA, and LoB was demonstrated with a per-sample false positive rate of <5%. Both analytes demonstrated ≥90% sensitivity across the reportable range, and clinical precision established at ≥90% PPA for samples above LoD. HLA deletion calling, while known to be technically challenging, demonstrated similar overall performance to other genes. Among 117k plasma and 43k tissue samples from patients with advanced cancer, this method yielded reportable deletion 10.8% (plasma) and 18.6% (tissue) cases, substantially improving diagnostic yield. Conclusion: We demonstrate that ploidy- and TF-aware deletion caller approach sensitively and accurately detects CNLs irrespective of gene identity, including in traditionally difficult-to-map regions like the HLA locus, in both cfDNA and gDNA inputs. This ability is a critical component of modern comprehensive genomic profiling, which must include robust detection of clinically relevant copy number alterations to better inform cancer treatment decisions.