The high failure rate of novel cancer drugs in human clinical trials necessitates complementary models for accelerating drug development. Spontaneous canine tumors, affecting one in three dogs, present a promising model due to shared environmental factors, intact immune systems, histological similarities, therapeutic responses, and genetic targets with human cancers. Comparative oncology studies have grown, leveraging the canine genome's significant orthology with the human genome and higher sequence similarity to humans than rodents. While genomic analyses of canine tumors have identified oncogenes and tumor suppressor genes, knowledge regarding prognosis and treatment response remains limited. The increasing resources for canine cancer treatment and the application of next-generation sequencing (NGS) provide a unique opportunity to study the prognostic effects of targeted therapies and identify biomarkers predictive of treatment outcomes. This study utilizes real-world clinical-genomic data from the FidoCure® Precision Medicine Platform to identify these biomarkers, applying predictive modeling similar to that used in large-scale human clinico-genomic data analysis. The aim is to demonstrate the value of spontaneous canine cancer models in advancing precision medicine for humans.

Existing literature supports the use of canine cancers in translational and clinical research, highlighting the numerous shared genetic variations and somatic driver mutations between human and canine cancers. Studies have demonstrated the utility of canine models in understanding cancer biology, including therapeutic response, acquired resistance, metastasis, and molecular targets. However, a knowledge gap exists in connecting genetic profiles to prognosis and therapeutic response in canine cancers. The utilization of NGS and large-scale datasets in human studies has demonstrated the potential for predicting treatment outcomes using molecular profiles; this study adapts similar approaches to canine data.

The study utilized data from the FidoCure® Precision Medicine Platform, encompassing 2702 cases. After filtering for data completeness and DNA NGS quality, 1108 dogs with survival outcomes and genomic data were analyzed, with 792 having targeted treatment information. The data included 19 tumor types, mutations in 48 targeted genes (average of 2.4 mutations per dog), and 10 targeted therapies. Statistical analysis involved Cox proportional-hazards models. The first model (Equation 1) assessed the hazard ratios of each tumor type relative to others, controlling for dog weight, sex, reproductive status, age at diagnosis, and time from diagnosis to treatment. A second model (Equation 2) investigated the effect of somatic mutations in the 48 genes on overall survival, including the same control variables as the first model and one-hot encoding for gene mutations. A third model (Equation 3) analyzed the prognostic effects of treatments conditioned on genomic alterations, using a subset of the data with treatment information and controlling for various factors. One-hot encoding was used for treatment and tumor type. In training this model, highly correlated covariates were removed due to multicollinearity, and genes with insufficient sample sizes were excluded. Germline mutations were excluded from primary analyses based on previous publications. Kaplan-Meier curves were used to visualize survival estimates. Data on human studies were also analyzed for comparison. Next-generation sequencing (NGS) was performed on formalin-fixed paraffin-embedded (FFPE) tumor tissue using a targeted panel of genes commonly mutated in human and canine cancers. Sequence read pairs were mapped to the canine reference genome (CanFam3.1 and CanFam4) using BWA-MEM. Variant calling was performed using a pipeline for germline and somatic calling.

Analysis revealed statistically significant hazard ratios associated with nine tumor types (Table 1). Hemangiosarcoma had the highest hazard ratio (2.07), while thyroid carcinoma had the lowest (0.35). Five genes (*TP53*, *PIK3CA*, *NRAS*, *ATM*, and *KIT*) showed statistically significant hazard ratios (Table 2). *TP53* and *PIK3CA* mutations were associated with worse prognosis, while *NRAS*, *ATM*, and *KIT* mutations were associated with better prognosis. Four gene-drug combinations had hazard ratios less than one and p-values less than 0.05 (Table 3). Specifically, BRAF mutant tumors responded favorably to lapatinib, ARID1A mutant tumors to trametinib, and BRCA1 mutant tumors to dasatinib. Analysis including germline and somatic variants showed similar results. Concordance was observed between canine and human data for the relative risk of *TP53*, *ATM*, and *KIT* mutations, while *PIK3CA* and *NRAS* showed differences, possibly due to the high frequency of hemangiosarcoma in the canine dataset. The analysis also found that patients with ALK mutations have a median survival of 424 days compared to 349 days in the group without ALK mutations, for an overall survival hazard ratio of 0.64 (0.42, 0.97) and p-value of 0.04. The study also noted that TP53 is the most common mutated gene in both human and canine tumors, and its mutation is associated with a worse prognosis in several cancer types, including osteosarcoma, in both species. Hemangiosarcoma, the most common tumor type in this study, also exhibited a high frequency of TP53 mutations and a poor prognosis. Finally, there was concordance between the hazard ratios and statistical significance in canine and human tumors for TP53, ATM, and KIT, while PIK3CA and NRAS showed differences.

This study demonstrates the concordance between canine and human cancers in terms of prognostic impact of mutations in several key genes. The findings underscore the value of canine spontaneous tumors as a model for studying human cancers and testing the efficacy of targeted therapies. The high frequency of hemangiosarcoma in the canine data may explain some differences in the results for PIK3CA and NRAS compared to human data. The identification of specific gene-drug combinations with positive prognostic associations provides promising leads for human cancer drug development. The limitations of the study are noted below; further research should focus on tumor type-gene mutation interactions to gain a more comprehensive understanding of the concordance between canine and human cancer models.

This study, using the largest clinico-genomic canine cancer dataset to date, reveals significant concordance between canine and human tumor genomics and treatment response. The identified gene-drug associations offer valuable insights for precision oncology drug development. Future research should focus on larger datasets and additional gene-drug combinations, and explore the nuances of the differences between canine and human cancer models.

The study's limitations include potential biases due to the specific population of dogs included, differences in tumor type distributions between canine and human populations, and the relatively limited sample sizes for certain gene-drug combinations. Further research with larger, more diverse datasets and expanded clinical information is needed to confirm and generalize the findings. The study's reliance on data from a single platform (FidoCure®) could also limit the generalizability of the findings. The sample size limitations also prevented the researchers from reporting tumor type-gene mutation interactions with statistical significance. Further studies are needed to explore such interactions.