Detecting oncogenic selection through biased allele retention in The Cancer Genome Atlas

Abstract

AbstractBackgroundThe loss of genetic diversity in segments over a genome (loss-of-heterozygosity, LOH) is a common occurrence in many types of cancer. By analysing patterns of preferential allelic retention during LOH in approximately 10,000 cancer samples from The Cancer Genome Atlas (TCGA), we sought to systematically identify genetic polymorphisms currently segregating in the human population that are preferentially selected for, or against during cancer development.ResultsExperimental batch effects and cross-sample contamination were found to be substantial confounders in this widely used and well studied dataset. To mitigate these we developed a generally applicable classifier (GenomeArtiFinder) to quantify contamination and other abnormalities. We provide these results as a resource to aid further analysis of TCGA whole exome sequencing data. In total, 1,678 pairs of samples (14.7%) were found to be contaminated or affected by systematic experimental error. After filtering, our analysis of LOH revealed an overall trend for biased retention of cancer-associated risk alleles previously identified by genome wide association studies. Analysis of predicted damaging germline variants identified highly significant oncogenic selection for recessive tumour suppressor alleles. These are enriched for biological pathways involved in genome maintenance and stability.ConclusionsOur results identified predicted damaging germline variants in genes responsible for the repair of DNA strand breaks and homologous repair as the most common targets of allele biased LOH. This suggests a ratchet-like process where heterozygous germline mutations in these genes reduce the efficacy of DNA double-strand break repair, increasing the likelihood of a second hit at the locus removing the wild-type allele and triggering an oncogenic mutator phenotype.