PO-339 Implications of organ-wise extraction of cancer mutational signatures using 2577 whole genomes

Abstract

IntroductionWith the advent of next generation sequencing of whole exomes and whole genomes, we can now obtain a comprehensive view of the somatic mutations that are present in tumours. Using latent variable discovery approaches such as non-negative matrix factorisation (NMF), recurrent somatic mutational patterns, called mutational signatures, have been identified, some of which have been attributed to mutational processes, such as sunlight exposure or homologous recombination deficiency. As more cancer whole genomes are available, the identification of new signatures and disambiguation of known signatures become possible. We can now begin to ask new questions, such as whether the same mutational process may induce different, organ-specific, mutational signatures.Material and methodsWe used somatic mutations from 2577 tumours from the PCAWG dataset, organised in 21 organs. We then performed 21 independent local organ-specific signature extractions and one global signature extraction with all the samples. Extraction is performed using NMF. The optimal number of signatures for each extraction is determined by the clustering properties of repeated NMF runs. Cluster analysis based on cosine similarity was performed to determine similarity of signatures across organs.Results and discussionsSeveral of the mutational signatures obtained from organ-specific extractions resemble known COSMIC mutational signatures. COSMIC signature 1 can be found in almost all organs, with the exception of the Liver. Other signatures, such as COSMIC signatures 2, 3, 13, 17 and 18, are independently obtained from multiple organs. All the signatures that are found in multiple organs present organ-specificity. Further investigation is required to determine how much of this specificity is biological and how much is due to a dataset bias.ConclusionOur work is the first attempt to determine the variability of mutational signatures across different organs. It reveals that some mutational signatures are more robust than others, and that the same mutational process may induce slightly different mutational signatures in different organs, though part of these differences may be attributed to a dataset bias. The use of organ-specific mutational signatures may be critical for the correct assignment of mutations in a tumour to signatures and, in turn, the correct identification of mutational processes that are at work in a tumour.