Abstract
Background Cancer genomes are shaped by DNA damage and repair. DNA Repair deficiencies in cancers may result in characteristic mutational patterns, as exemplified by deficiency of BRCA1/2 and efficacy prediction for PARP-inhibitors. We systematically evaluated the ability to identify and predict mono- or biallelic deficiencies of 736 DDR genes from associations with genome-wide mutational patterns, including structural variants, across 6,065 whole-genome sequenced cancers.
Results We assembled 535 cancer-specific patient cohorts with shared deficiency of a DNA damage response gene. We trained and evaluated a predictive model for each cohort, and shortlisted 24 gene deficiencies that could be predicted with high accuracy from mutational patterns. These included expected gene deficiency models for BRCA1/2, MSH3/6, TP53, and CDK12. CDK12 is associated with tandem-duplications, and we demonstrate that this association can predict gene deficiency with high accuracy (AUROC=0.97) in prostate cancers.
Our novel associations include deficiencies of ATRX, IDH1, HERC2, CDKN2A, PTEN, and SMARCA4. We found that loss of ATRX or IDH1, which are often co-mutated, were associated with the same mutational phenotype in central nervous system cancers; HERC2 loss was associated with an increased number of short deletions in skin cancers and significantly co- occurring with TP53 loss; and PTEN loss was associated with decreased levels of structural rearrangements in cancers of the uterus and the central nervous system.
Conclusion Our systematic, generic approach yielded a catalogue of predictive models, which may provide targets for research and development of treatment, and potentially help guide therapy.
Competing Interest Statement
The authors have declared no competing interest.
