Development of bioinformatic tools for canine cancer biomarker discovery and validation

The dog is an increasingly recognised spontaneous model of cancer in comparative oncology, sharing with humans common environmental exposures and, for many histotypes, similar clinical and histopathological tumour features. Unlike human oncology, however, veterinary medicine still lacks a standardised, publicly accessible pan-cancer genomic resource. The candidate work presents the Canine Cancer Genome Atlas (CCGA), the first pan-cancer genomic atlas in the dog, built through the systematic aggregation, harmonisation and reanalysis of publicly available whole-exome sequencing data. Starting from 770 cases identified in the Sequence Read Archive, tumour samples with matched normal tissue were collected across ten histotypes. After stringent quality control, 667 cases were retained in the final cohort. A standardised computational pipeline was developed for the calling of somatic single-nucleotide variants and small indels, copy-number alterations, structural variants, microsatellite instability and mutational signatures, and applied uniformly across the entire cohort. The analysis identified 44,192 somatic SNVs, 1,515 MNVs and 2,272 indels, totalling 47,979 high-confidence coding mutations, with a median tumour mutational burden of 1.47 mutations/Mb and up to a five-fold variation across histotypes. Driver gene identification revealed 39 recurrently altered pan-cancer genes, led by TP53 (23.2%) and the PI3K/AKT/mTOR pathway (34.7% collectively), with 84.4% concordance with known human cancer drivers. A total of 89,247 copy-number events were detected, including amplifications of MYC (29.8%) and KIT/PDGFRA (24.9%) and deletions of PTEN (22.6%), alongside 57,119 coding structural variants across 4,759 genes, with CARD10 among the most recurrent and a chromothripsis rate of 23.7% in osteosarcomas. Mutational signature analysis identified 27 SBS signatures, 19 of which matched human COSMIC signatures, including a universal ageing signature and homologous recombination deficiency in 15.5% of cases. Microsatellite instability was present in 2.4% of tumours (MSI-H), and a mutational burden threshold of 10 mutations/Mb or higher achieved 100% sensitivity and 99.5% specificity in identifying these cases, positioning it as an immediately actionable screening biomarker. The comparative analysis highlighted both strong concordances with human oncology, as in the case of osteosarcoma, and important mechanistic divergences, such as the near-total absence of the UV-radiation mutational signature in canine melanoma despite histopathological concordance with its human counterpart, underscoring the need for species-specific genomic characterisation rather than simple interspecies extrapolation. The CCGA is made available as a queryable public resource, designed to support both translational research and clinical application. Building on these findings, three prospective experimental directions are proposed: anti-PD-1 immunotherapy in MSI-H tumours selected via mutational burden threshold, PI3K-pathway inhibitors in PIK3CA-mutant mammary carcinoma and PTEN-deleted haemangiosarcoma, and PARP inhibitors in osteosarcomas stratified by homologous recombination deficiency. Overall, this work provides both a foundational resource for veterinary cancer genomics and a solid basis for comparative medicine and the clinical translation of genomic findings in canine patients.