Global and local redistribution of somatic mutations enable the prediction of functional XPD mutations in bladder cancer

Xeroderma pigmentosum group D (XPD) is a DNA helicase with critical functions in transcription initiation and nucleotide excision repair. Missense mutations in XPD are putative drivers in around 10% of bladder cancers (BLCA), but the associated mutational process remains poorly understood. Here, we examine the somatic mutational landscape of XPD wild-type (n=343) and mutant (n=39) BLCA whole genomes. The genome-wide distribution of somatic mutations is significantly altered in XPD mutants, affecting both APOBEC and non-APOBEC associated mutational processes. Specifically, XPD mutants are enriched in T[C>T]N mutations (SBS2) with altered correlation with replication timing. At a locoregional genomic level, mutant XPD BLCA had striking T>G mutation hotspots at CTCF-cohesin binding sites (CBS) with evidence linking XPD to genomic uracil repair. Leveraging differential distribution of somatic mutations, we developed a machine-learning model for predicting pathogenic XPD mutations, which we validated in an independent TCGA cohort with 100% accuracy. Our model enabled the discovery of missed XPD mutation calls and uncovered pathogenic non-hotspot XPD mutations in bladder cancer. Our study reveals how XPD mutations redistribute somatic mutations in cancer genomes and provides a genome sequencing approach to differentiate driver and passenger XPD mutations.