A Bayesian framework for evaluating somatic copy number variants into comparative genomic analysis of multifocal non-small cell lung cancer

Introduction: Clinical staging of multifocal non-small cell lung cancer (NSCLC) is facilitated by comparative genomic analysis of separate tumor nodules, but the role of genome-wide copy number profiling for this purpose is uncertain. Methods: We analyzed single nucleotide variants (SNVs), gene fusions, small copy number variants (gene amplifications and homozygous deletions) (sCNVs), and large-scale copy number variants (lCNVs) in a retrospective cohort of 34 patients with >1 lung cancer nodule tested on a previously validated tumor-only 515-gene next generation sequencing (NGS) panel (MayoComplete Solid Tumor Panel). For patients with <2 shared SNVs, a Bayesian approach was formulated to calculate the probability of shared versus independent events occurring in different tumor nodules in the same patient. The prevalence of specific lCNVs in NSCLC was derived from an independent dataset of 100 NSCLC peers tested by the same panel, and the prevalence of SNVs, fusions, and sCNVs in NSCLC was derived from publicly available data. Alterations unique to different tumor nodules and suspected germline alterations were excluded from the Bayesian analysis. Results: There were 54 total comparisons for 34 patients. Eleven of the patients (32.4%) had >=2 shared SNVs along with one or more shared lCNVs also identified in each patient. Eight patients (23.5%) had no shared SNVs, gene fusions, sCNVs, or lCNVs. Using the Bayesian approach for the remaining fifteen (44.1%) patients with at least one shared alteration (SNVs, sCNVs, and lCNVs), the probability of clonal relatedness ranged from 73% for a shared whole arm or whole chromosome lCNV to >99% for tumors with two or more shared alterations including multiple or complex lCNVs and SNVs. Whole chromosome or whole arm lCNVs were sometimes shared between tumor nodules with different driver SNVs. Conclusions: Incorporating lCNVs into comparative genomic analysis provides additional evidence of clonal relatedness in a subset of multifocal NSCLCs. However, differences in bioinformatics pipelines for calling lCNVs, concerns over the imprecision of genome-wide copy number profiling, and uncertainty regarding the temporal relationship between lCNVs and carcinogenesis (including somatic events possibly related to “field cancerization”) create challenges for integrating lCNVs into comparative genomic analysis with standard clinical workflows for tumor-only NGS on paraffin embedded tissue sections. Our simplified approach mitigates some of the uncertainty introduced by incorporating lCNVs into clonality determination.