High-resolution copy-number mutational signatures for ovarian cancer patient stratification

Abstract

During tumour evolution, diverse mutational processes can affect copy number state across the genome, by deleting or duplicating sections of genomic material. The activity of mutational processes can be identified through the patterns - mutational signatures - they leave in genomic data.

Mutational signatures of copy number have previously been extracted from micro-array data. These signatures have been used to predict overall survival and drug response at the point of diagnosis. However, these signatures are of limited use in ovarian cancer, which is dominated by smaller scale changes that are not accurately captured by low-resolution micro-array data. These changes can be resolved with higher resolution approaches, such as deep whole genome sequencing (WGS).

Here, existing low-resolution signatures are quantified in downsampled high-resolution WGS data, and are related to these more complex genomic features of ovarian cancer. This has not previously been attempted in an ovarian-specific dataset of this size. A negative correlation between two of the existing signatures was evident in the cohort. Hierarchical clustering separated the cohort into two patient subgroups, those with homologous recombination deficiency (HRD) and chromoplexy (whereby chains of translocations and deletions occur across multiple chromosomes), and those with variation caused by chromosome segregation errors. However, a survival analysis suggested no significant difference in survival time or time to relapse between the two groups.

Three copy-number signatures were then extracted de novo from the complete high-resolution data, which, when quantified in the same cohort, also produced two patient subgroups. Again, these were separated by the presence or absence of HRD and chromoplexy. These groups showed clear separation in the survival analysis, suggesting the high-resolution signatures are more effective prognostic indicators than low-resolution signatures quantified in downsampled data.

The results suggest that these high-resolution mutational signatures may have the potential to provide actionable clinical information, at the point of diagnosis, for the prediction of prognosis and to inform treatment plans.