Abstract

Background

Random mutation and deterministic selection both shape the pattern of driver mutations in a cancer’s genome. However, the relative importance of the two evolutionary forces remains unclear.

Method

Here we have used public cancer sequencing data to infer the independent effects of mutation and selection on driver mutation complement. We used non-negative regression based on mutational signatures to identify historic mutational process activity in 10,188 cancer samples. First, we detect associations between a range of mutational processes, including those linked to smoking, ageing, APOBEC and DNA mismatch repair (MMR) and the presence of key driver mutations across cancer types. Second, we quantify differential selection between well-known alternative driver mutations, including differences in selection between distinct mutant residues in the same gene.

Results

We find 56 associations (FDR = 0.05) between key driver mutations and mutational signatures across cancer types. Of particular note, MMR-linked signatures appear to drive FBXW7 R465C mutations in both colorectal cancer and stomach cancer and PIK3CA H1047R mutations in both breast cancer and stomach cancer. Normalising for mutational processes, our results indicate selective differences between key driver mutations. IDH1 R132H appears selected above IDH1 R132C in low grade glioma, and PIK3CA H1047R appears selected above PIK3CA E545K and PIK3CA E542K in breast cancer.

Conclusion

Our results provide initial links between mutational processes and specific driver mutations, and reveal an element of predictability in the driver mutation acquisition of cancers that are exposed to the same mutational forces. Our analysis of selective differences between mutations further informs our understanding of the evolutionary context of cancer development, by quantifying the effects of selective differences between driver mutations.