Introduction

New approaches to the development of cancer therapeutics are urgently needed to improve the current 89% failure rate of novel drugs in clinical trials1,2,3, and to improve patient outcomes. Spontaneous cancer in companion animals represents a unique opportunity for investigation of novel therapeutics for human and veterinary use4,5,6,7,8,9,10,11,12,13,14,15. Dogs develop spontaneous tumors that are highly similar to human cancers in terms of histological features and clinical presentation, but canine tumors typically progress more rapidly7,10,11,16. Canine cancers have also been found to have similar genetic and molecular targets to human malignancies8,10,17,18,19,20,21,22,23,24,25,26,27,28,29,30, and thus affected dogs present an opportunity to test novel therapeutics in a treatment-naive setting that is not currently feasible in human medicine11. Dogs represent a large animal model with an intact immune system, enabling comparative studies of therapeutic efficacy, immunotherapy, tumor evolution, and tumor microenvironment7,11,31. Studies of new therapeutic agents have begun to include dogs with cancer to help characterize pharmacokinetic and pharmacodynamic properties, efficacy, and tolerability11,31. In addition, the National Cancer Institute’s Center for Cancer Research has founded the Comparative Oncology Program and the Canine Oncology Trials Consortium to support comparative studies in dogs and facilitate integration of these findings with human oncology efforts31.

Canine tumors provide a powerful platform for translational investigation11,12,32. Over the past decade, genomic characterization of canine cancers has highlighted the marked biological and molecular similarities between several canine and human cancers, including lymphoma17,18,33, osteosarcoma8,23,25, hemangiosarcoma20,26,27,28,30,34, glioma29, melanoma22, mammary tumors35,36,37, and urothelial carcinoma19,21. Some of the somatic mutations identified in these canine cancers occur at the orthologous position to known mutational hotspots found in human cancers, including PIK3CA H104720,26, BRAF V588 (V600 for human)19, and FBXW7 R470 (R465 for human)18. These somatic mutations do not always occur in the same cancer type across species, more canine studies are needed to better characterize this association.

Although studies have shown genomic concordance between canine and human cancers, the number of canine tumors that have undergone genomic sequencing lags behind human tumors by an order of magnitude (fewer than 2000 canine tumors have been sequenced38, compared to more than 20,000 human tumors39). Consequently, the landscape of actionable tumor mutations in canine cancers is not fully understood6,25. We sought to address this issue in order to assess the feasibility of matching dogs with spontaneous cancers to targeted therapy, thereby providing treatment opportunities to canine patients while developing a platform that could accelerate a more global understanding of the clinical as well as translational potential from dogs to humans.

To do this, we developed a next-generation sequencing (NGS) panel targeting coding exons of 59 genes frequently mutated in human cancers. Using this panel, we performed the largest sequencing study of canine cancers to date, including 671 tumors of 23 histologic types from dogs representing more than 96 breeds. Importantly, our study revealed 18 canine mutational hotspots, 8 of which were orthologous to hotspots reported in human cancers and clinically actionable. These results demonstrate significant overlap in somatic hotspot mutations between human and canine cancers, further highlighting spontaneous canine cancers as an excellent model for the investigation of targeted therapies.

Results

Cohort demographics

The cohort consisted of 671 dogs with tumors representing 23 cancer types. Hemangiosarcoma are the most common tumor type (n = 166), followed by soft tissue sarcoma (n = 96), melanoma (n = 46), osteosarcoma (n = 46), lymphoma (n = 35) and anal sac carcinoma (n = 31) (Fig. 1a). In total, 337 sarcomas, 203 carcinomas, and 131 other cancer types were included (Fig. 1a; Table S1). For each case of mammary carcinoma, lymphoma and mast cell tumor, we extracted the tumor subtype and grade data available from the original histopathological reports and provided the data in the Table S1.

Figure 1
figure 1

Demographics of enrolled 671 dogs. (a) Distribution of tumor supertypes (inner cycles) and types (outer circles). The numbers inside the parentheses indicate the dog numbers. The locations of the 41 carcinomas are provided in Table S1. (b) Distribution of breed. Breeds with ≥ 10 dogs are specified. (c) Distribution of sex and reproductive status. (d) Distribution of age. (e) Distribution of weight.

Full size image

The cohort consisted of both purebred (n = 457 dogs, 96 breeds), and mixed breed (n = 213, ≥ 45 breeds) ancestry dogs (Table S1) as reported by owner. A total of 9 breeds are represented by ≥ 10 dogs (Fig. 1b; Table S1). The largest breed groups are Golden Retrievers (61 pure and 17 mixed), Labrador Retrievers (56 pure and 29 mixed), German Shepherd Dogs (29 pure and 2 mixed), and American Pit Bull Terriers (20 pure and 15 mixed) (Fig. 1b). A total of 355 cases are male (321 neutered and 34 intact) and 315 are female (304 spayed, 11 intact, and 1 unknown) (Fig. 1c; Table S1). Dogs ranged in age from 1 to 16 years (mean 9.9 ± 2.8) (Fig. 1d, Table S1). Interestingly, dogs with osteosarcoma are significantly younger than dogs with other tumor types, with an average age of 8.4 years compared to 10.1 years respectively (Fig. S1). Weights ranged from 1 to 91 kg (mean 25.5 ± 13.5) (Fig. 1e; Table S1).

Germline-somatic mutation discrimination

The 671 tumors were subjected to targeted sequencing of the FidoCure® NGS panel, which contains 59 oncogenes and tumor suppressors frequently mutated in human cancer. A total of 42,566 mutations (1274 unique mutations) were called by comparing the sequences to the CanFam3.1 genome. These mutations, however, consisted of both germline and somatic mutations. As no normal samples from these dogs were sequenced, we developed a pipeline for germline-somatic mutation discrimination based on known canine germline mutations, variant allele frequency (VAF) distribution, and known human somatic mutations (Fig. 2). Briefly, we first filtered out germline variants published in various databases and literature (see “Methods”)38,40,41,42,43. As a result, 41,430 total (442 unique) mutations were classified as germline variants and were excluded (Fig. 2). We then divided the remaining mutations into two groups, based on their rate of recurrence. For mutations found in ≥ 5 dogs, we examined the VAFs. Mutations with a VAF distribution clustered around 50% or near 100% were classified as heterozygous or homozygous germline mutations respectively. As a result, 29 total mutations (2 unique) mutations were excluded (Fig. 2). For mutations found in < 5 dogs, we identified those for which the human counterparts are known somatic mutations in human cancers, and considered them somatic mutations (Table S2). As a result, 592 total (529 unique) mutations were classified as somatic (Fig. 2), with the remaining 306 total (287 unique) mutations with the germline/somatic status unknown (unclassifiable in Fig. 2).

Figure 2
figure 2

Germline-somatic mutation discrimination pipeline. The pipeline first filtered out known germline mutations reported in literatures and databases, or identified in normal samples. The remaining mutations were then divided into two groups. For mutations identified ≥ 5 dogs, variant allele fraction (VAF) distribution of each mutation was examined to determine if the mutation is germline (clustered at 0.5 for heterozygous germline and near 100% for homozygous germline) or somatic mutation (random distribution). For mutations identified in < 5 dogs, a mutation would be classified as somatic if its human orthologous mutation was found in COSMIC or cBioPortal (both databases host human somatic mutations). Total and unique mutation numbers, along with VAF distributions, were shown for each major step.

Full size image

Our pipeline classified 801 total (543 unique) mutations as somatic (Fig. 2). The VAF distribution resembled those of known somatic mutations (p > = 0.32), including TP53 and PIK3CA mutations, but differed from those of known and identified germline mutations in these tumor samples (p < 1 × 10−6) (Fig. 3a). Importantly, we repeated the same sequencing and GATK mutation calling to samples from 20 healthy dogs, collected using cheek swabs. The VAF distribution of these normal samples resembled germline mutations (p > 0.99), but not somatic mutations (p < 1 × 10−15), found in the tumor samples (Fig. 3a). Importantly, among 411 unique variants called in normal samples and 546 unique somatic mutations identified by our pipeline (Fig. 2) in tumors samples, only 6 (1%) mutations were shared (Table S2). Lastly, the identified somatic mutations had a base substitute type pattern matching that of somatic mutations, but not germline mutations (Fig. 3b), e.g., G > A/C > T mutations being dominant. These observations indicate that our pipeline (Fig. 2) is effective.

Figure 3
figure 3

Identified germline-somatic mutation comparison. (a) VAF distributions of germline mutations and somatic mutations. TP53 and PIK3CA mutations are all somatic (via manual examination). The VAF distribution of somatic mutations identified by Fig. 2, along with the VAF distribution of mutations detected in 20 normal samples sequenced, are also shown. (b) The distribution of the 6 base substitution types of somatic, germline and unclassified mutations identified from Fig. 2.

Full size image

Somatic mutational landscape

Many of the somatic mutations discovered by our pipeline are consistent with published studies18,20,21,22,23,25,26,29,38. For example, TP53 is the most frequently mutated gene across the cohort, mutated in 151 out of 671 animals (22.5%) (Fig. 4a). PIK3CA is the third most mutated gene across the tumor types, with the mutation especially common in hemangiosarcoma, mutated in 13% of the samples (Fig. 4a; Table S2). ERBB2 and BRAF are the most frequently mutated genes in pulmonary adenocarcinoma and urothelial carcinoma, mutated in 50% and 40% of the samples respectively (Fig. 4a).

Figure 4
figure 4

Somatic mutational landscape in canine tumors. (a) Oncoprint of the 22 most frequently mutated genes across the cohorts in each tumor sample, grouped by tumor supertype (sarcoma, carcinoma, and other indicated at the top), tumor types, and mutations. Tumor mutation burden (TMB), breed, age, weight and tumor location are also indicated. LSA lymphoma, MCT mast cell tumor. (b) Distribution of TMB in each tumor type. Each dot represents a sample, while the black horizontal lines indicate the median TMB in the respective cancer types.

Full size image

Our pipeline identified known mutational hotspots. These included PIK3CA H1047R (n = 16), ERBB2 V659E (n = 13), BRAF V588E (n = 13), NRAS Q61R (n = 6), NRAS Q61K (n = 6), TP53 R209H and R226H (n = 6 each) (Table S2). These mutations are known or likely cancer drivers, and most of them are activating or gain-of-function changes.

Our study revealed findings not previously reported. For example, chromatin remodeler ARID1A is the second most frequently mutated gene of the cohort, mutated across tumor types at a rate of 8% (Fig. 4a). Moreover, about 78% (40/51) of mutations in ARID1A are in-frame deletions (Fig. 4a; Table S2), the significance of which remains to be determined. One deletion, ARID1A 1038_1040del variant was called in one normal sample (Table S2), likely because it locates in a GCC repetitive region and has a higher chance of being deleted. The increased frequency of ARID1A 1038_1040del in tumor samples may arise from the increased instability of tumor genomes. Other chromatin modeler genes KMT2D, KMT2C, SETD2, and CREBBP are the third, fourth, seventh, eighth and fifteenth most mutated genes across the cohort (Fig. 4a). Those mutations appear to be randomly distributed among the tumor supertypes and types (Fig. 4a). SETD2 is mutated in 18 tumors, 7 of which harbored SETD2 truncation mutations, consistent with previous findings of frequent truncation mutation in SETD21,2.

Our study provided a snapshot of somatic mutations for > 9 canine carcinomas, including previously uncharacterized tumor types such as anal sac carcinoma and neuroendocrine carcinoma, as well as less characterized tumor types including hepatocellular carcinoma and thyroid carcinoma (Fig. 4). These four types of carcinomas appear to have different mutational landscapes than other carcinomas, including depletion of TP53 mutation (Fig. 4a). The mutations also appear more random, lacking a prominent mutated gene like ERBB2 in pulmonary cancer and BRAF in bladder cancer (Fig. 4a). Lastly, they also have lower TMB values overall for genes included in the targeted panel (Fig. 4b).

Other previously less-characterized tumor types include soft tissue sarcoma, histiocytic sarcoma, and mast cell tumor. Our study also provided a more comprehensive mutation landscape for these tumors. The same as in other sarcomas, TP53 is the most frequently mutated gene in soft tissue sarcoma and histiocytic sarcoma, mutated at 22% (21/96) and 35% (10/29) respectively (Fig. 4a). However, soft tissue sarcoma also harbors more mutations in chromatin modeler genes (e.g., ARID1A and KMT2D) and the neurotrophic tyrosine kinase receptor gene NTR1, as well as higher panel-specific TMB, compared to histiocytic sarcoma (Fig. 4). TP53 is also the most frequently mutated gene in mast cell tumors, mutated in 21% (4/19) (Fig. 4a). However, in mast cell tumors, mutations of other genes are relatively rare (Fig. 4a), and the panel-specific TMB was lower overall (Fig. 4b).

Location-specific mutation in hemangiosarcoma

The hemangiosarcoma samples in our study consisted of 129 tumors from the spleen and 37 tumors from non-splenic locations (Fig. 5a; Table S3). TP53 is the most frequently mutated gene in both splenic and non-splenic hemangiosarcoma, mutated at 29% (37/129) and 46% (17/37) respectively. PIK3CA mutations, ≥ 50% of which are H1047R/L, are also common, mutated at 12% (15/129) and 16% (6/37) in splenic and non-splenic hemangiosarcoma, respectively (Fig. 5a). However, while NRAS mutations, 89% of which are Q61R/K/H, are frequent in splenic hemangiosarcoma (mutated in 7% [9/129]), NRAS mutations are not detected in non-splenic hemangiosarcoma (Fig. 5a). Furthermore, in splenic hemangiosarcoma, NRAS mutations are mutually exclusive with TP53 mutations (p < 0.01) and PIK3CA mutations (not significant, likely due to small sample size) (Fig. 5b), consistent with a previous study3.

Figure 5
figure 5

Location-specific mutation and mutually exclusive mutations in canine hemangiosarcoma. (a) Oncoprint of gene and specific mutations indicated in hemangiosarcomas grouped by splenic and non-splenic locations and then ordered by mutation type. (b) Heatmap indicating co-occurrence (green) or mutual exclusion (brown) between mutations shown in splenic and non-splenic hemangiosarcoma.

Full size image

Somatic mutation enrichment and depletion

Consistent with previous findings4, our study indicates that canine cancer mutations were tumor type dependent, but largely breed independent (Table S4). Specifically, TP53 mutations are significantly enriched in sarcomas (p = 1.00 × 10−7), including hemangiosarcoma (p = 5.82 × 10−6) and osteosarcoma (p = 2.92 × 10−2), but depleted in carcinoma (p = 2.58 × 10−7), including anal sac carcinoma (p = 5.73 × 10−4) (Fig. 6a). PIK3CA mutation is also enriched in sarcoma including hemangiosarcoma (p = 4.70 × 10−5) (Fig. 6a). BRAF and ERBB2 mutations, however, specifically BRAF V558E and ERRB2 V659E, are depleted in sarcoma (Fig. 6a,b). These mutations are enriched in carcinoma, with significant enrichment of BRAF V558E in urothelial carcinoma (p = 3.92 × 10−14) and BRAF V695E in pulmonary adenocarcinoma (p = 3.52 × 10−8) (Fig. 6a,b). Interestingly, KIT mutation is enriched in gastrointestinal stromal tumors (p = 7.28 × 10−5) (Fig. 6a). We did not observe significant enrichment and depletion of somatic mutations in specific age or weight groups (Fig. 4a; Table S3).

Figure 6
figure 6

Canine somatic mutation enrichment and depletion across tumor supertype and tumor types. (a) Heatmaps indicating the enrichment (red) or depletion (blue) scores, based on Fisher’s exact test, of genes mutated in > 5 dogs in tumor super type or tumor type with > 20 samples. (b) Heatmaps for individual gene mutation, presented as described above.

Full size image

Comparison of dog–human mutational hotspots

We identified 18 canine mutational hotspots (Table S5), which are more likely to harbor cancer drivers44,45 and anti-cancer targets. Many of the mutational hotspots are in oncogenes including PIK3CA, KRAS, NRAS, BRAF, KIT, ERBB2, and EGFR, as well as in the tumor suppressor TP53 (Fig. 7; Table S5).

Figure 7
figure 7

Comparison of canine and human mutational hotspots. Lollipop plots depict the mutational distribution in TP53 (a), PIK3CA (b), ERBB2 (c), EGFR (d), KIT (e), BRAF (f), KRAS (g), and NRAS (h) in 24,592 human (upper) and 671 canine (lower) tumors. The x-axis indicates amino acid position in the human protein. The y-axis indicates the number of samples with the mutations. The most prominent mutational hotspots are labeled, with the precise human and canine protein position indicated.

Full size image

We compared these canine mutational hotspots to those reported in 24,592 human tumors44,46. The two species share many mutational hotspots, e.g., PIK3CA H1047, KRAS G12, NRAS Q61, and BRAF V600 in human or V588 in canine (Fig. 7). However, species-specific mutational hotspots are also identified, including PIK3CA E542/E545 and ERBB2 S310 in humans, as well as ERBB2 V659 in dogs (Fig. 7).

KRAS and NRAS have the same amino acid sequences between the dog and the human. Importantly, the same as in human tumors, the majority of KRAS mutations identified in canine tumors are located at mutational hotspot G12, including G12A (detected in 2 melanomas), G12V (in 2 mammary carcinomas), G12D (in 1 mammary carcinoma and 1 squamous cell carcinoma), and G12C (in 1 melanoma) (Table S5). For NRAS, all 15 mutations detected in canine tumors are at G13 and G61, both of which are also mutational hotspots in human tumors (Fig. 7). We identified NRAS Q61R/K/L mutations in 14 canine tumors, 57% (8/14) of which were hemangiosarcoma, 28% (4/14) malignant melanoma, 7% (1/14) plasma cell tumors and 7% (1/14) soft tissue sarcoma (Table S5).

Discussion

Cancer genomes harbor actionable gene mutations and clinical sequencing provides immense opportunities for precision medicine in cancer treatment47,48. Indeed, clinical sequencing is routinely used in many hospitals in guiding treatment of lung cancers and other cancers in humans. In dogs, clinical sequencing lags significantly behind and is thus underdeveloped for use in cancer treatment. To address this deficiency, we developed the FidoCure® Personalized Genomic Panel, a targeted sequencing panel containing 59 well-known oncogenes and tumor suppressors frequently mutated in human cancer, with common mutational hotspots. Moreover, as clinical sequencing often sequences only the tumor samples due to cost reasons and/or the lack of matching normal samples, we developed an effective germline-somatic mutation discrimination pipeline that maximizes the use of tumor-only sequencing data. We applied the panel and our pipeline to 671 spontaneous canine tumors across 23 tumor types and 96 breeds common to pet dogs in the US. This study, to our knowledge, represents the largest sequencing study of canine tumors to date, and includes tumor types for which mutations have not previously been characterized (e.g., anal sac carcinoma and neuroendocrine carcinoma), or have been less characterized (e.g., soft tissue sarcoma, hepatocellular carcinoma, thyroid carcinoma and mast cell tumors). Our study adds to the growing body of canine comparative oncology studies showing genomic similarities between human and canine cancers and specifically evaluates hotspot mutations that can be targeted with a precision medicine approach. Our study provides a much-needed resource in canine cancer research, accelerating canine precision medicine and enhancing the canine model in human cancer research.

Our analysis largely captures the landscape of hotspot mutations in canine tumors, which are similar to the mutational landscapes reported by previous whole exome or genome sequencing studies (e.g., TP53, NRAS and PIK3CA mutations in hemangiosarcoma, TP53 and SETD2 mutations in osteosarcoma, ERBB2 mutation in pulmonary carcinoma, BRAF mutation in urothelial carcinoma and FBXW7 mutation in lymphoma)18,20,22,23,24,25,26,27,29,35,36,37,38,49,50. Consistent with previous research18,20,22,23,24,25,26,27,29,35,36,37,38,49,50, our results indicated that TP53 is the most recurrently mutated gene across tumor types. TP53 mutations are significantly more common in sarcomas than in carcinomas. This may be due to differences in the cell of origin and the mechanisms of tumorigenesis in these cancer types. Carcinomas originate from polarized epithelial cells or their progenitors, and alterations of cell polarity genes and loss of cell polarity are likely the major drivers of accelerated cell proliferation in carcinoma development51,52,53. Sarcomas originate from mesenchymal cells, for which loss of function of TP53 leads to defective cell cycle checkpoints and accelerated proliferation. Our studies also identified other frequently mutated genes reported in canine cancer, including chromatin modeling genes (ARID1A, KMT2D and others).

Our study finds TP53 mutated in ~ 46% osteosarcomas, lower than those reported by several publications23,54. One reason for this discrepancy is that we examined only somatic base substitutions and small indels, not including somatic copy number alterations, unlike the other studies23,25. Moreover, we only sequenced the exons of TP53, not the entire gene, and hence were unable to identify intronic translocations and other aberrations found by whole genome sequencing (WGS)25. Indeed, Gardner et al.25 reports that WGS found TP53 mutated in 71% cases, whereas whole exome sequencing (WES) found TP53 muted in only 38% cases. Second, the sample size of our study (46 cases) and other publications (26–66)25,25,54 is not large enough to represent the population, resulting in variations among the studies. Similarly, our study identifies BRAF V588E (equivalent to V600E in humans) in 35% urothelial carcinomas, lower than several publications9,19,55,56. Again, the small sample size (20–66 cases) is a reason for the discrepancy. Another reason is the approach. The cited publications used Sanger sequencing9,19, restriction fragment length polymorphism genotyping19, or droplet digital PCR55 to specifically target the mutation. These methods may be more sensitive and/or have a higher false positive rate, compared to deep sequencing strategies like ours. Indeed, Cronise et al.56 reports that WES identified the mutation in 36% cases, while targeted Sanger sequencing identified the mutation in 70% cases.

Mutational landscape varies by tumor type but is largely breed-independent, consistent with a previous pan-cancer and pan-breed study that investigated whole exome data from 591 canine tumors38. The largest difference is between carcinomas and sarcomas, with significant differences found in the mutational frequency of TP53, PIK3CA, NRAS, ERBB2 and BRAF. Carcinomas are more variable than sarcomas in their mutational spectrum and mutational burden. Pulmonary adenocarcinoma and squamous cell carcinoma are highly enriched in BRAF and TP53 mutations respectively, and have the highest mutation burden among the tumor types investigated. Anal sac carcinoma and thyroid carcinoma, however, are depleted in TP53 mutation and mutations in other prominent genes, and have lower mutation burden. The few mutations detected in canine thyroid carcinomas occur in genes including BRAF and KRAS, both of which are frequently mutated in human thyroid carcinoma58.

We acknowledge that the lack of mutations does not mean that the genes are not altered. For example, ERBB2 mutation is found in only one case of anal sac carcinoma despite the overexpression of this receptor identified in 80% of cases in a previous study57. Differences in gene expression and mutational profile were also seen in urothelial carcinoma, which is characterized by overexpression of EGFR and ERBB2 in approximately 70% and 60% of the tumors, respectively, and clinical response to EGFR/ERBB2 inhibitors. However, mutations are not identified in either gene in our study or research by others59,60,61. Thus, further studies are needed to comprehensively identify alterations in these tumors.

Our study finds that NRAS mutations are mutually exclusive with TP53 mutations in splenic hemangiosarcoma, reaffirming the existence of different molecular subtypes of the same histology type27. Interestingly, NRAS mutations were not identified in non-splenic hemangiosarcoma, but mutations (particularly Q61R) in this gene are commonly seen in the splenic form27. This is the first time a statistically significant difference in genomic profiles of different anatomic locations of hemangiosarcoma has been reported, and may help guide therapeutic strategies.

Our work reveals that numerous mutational hotspots are shared between dogs and humans, including PIK3CA H1047, BRAF V600/V588, KRAS G12 and others. These findings further position dogs as a powerful translational model for human and veterinary oncology, as both existing and novel targeted therapies for these mutations (e.g., PIQRAY for PIK3CA62, PLX4032 for BRAF V600E mutations63) can be assessed in canine cancer patients. Among 20 canine hotspots identified, 13 overlap with those of human cancer. Mutations in five oncogenes are identified as hotspots, representing a unique opportunity to apply targeted therapy translated from human experience.

Our analysis also revealed species-specific mutational hotspots, including PIK3CA E545/2K mutations found only in human cancers. ERBB2 V659EE and TP53 R209H/C, KIT L575P, KRAS Q61H/K/R are identified as canine-only hotspots. Further studies are needed to better understand the mechanisms underlying these differences, which will assist anti-cancer drug development and precision medicine in both species.

The FidoCure® Personalized Genomic Panel and our somatic-germline mutation discrimination pipeline effectively capture the landscape of actionable hotspot mutations in canine tumors. We anticipate that this resource will accelerate canine cancer genomic research, significantly increasing the use of the canine model in precision medicine and anti-cancer drug development for both humans and dogs.

Materials and methods

Ethical statement

This study was performed in accordance with a protocol approved by the Institutional Ethics Committee of the One Health Company. Prior to enrollment, pet-owners were required to sign an informed consent. No additional procedures were performed on client owned dogs thus this trial does not fall under any regulations overseeing experimental animal trials.

Enrollment and sample collection

Client-owned dogs with histologically confirmed cancer diagnoses were enrolled in FidoCure® by 200 veterinarians in clinical practice. A total of 671 individual biopsies taken from May of 2019 until September of 2020 were analyzed through the FidoCure® Precision Medicine Platform, the proprietary name of The One Health Company’s precision medicine unit. Upon enrollment, tissue re-cuts obtained from formalin-fixed paraffin embedded (FFPE) tumor tissue used for histopathologic diagnosis were requested from the appropriate veterinary diagnostic laboratory. These tissues were evaluated by practicing board-certified veterinary pathologists and only tissue confirmed to be neoplastic progressed to genomic sequencing.

Library preparation and next generation sequencing

Genomic DNA (gDNA) was extracted from FFPE tissues using the Mag-Bind® FFPE DNA/RNA kit (Omega Bio-tek). The quality of the extracted gDNA was confirmed using the Agilent Genomic DNA ScreenTape Assay (Agilent) and the amount of gDNA was quantified using the Qubit dsDNA HS assay kit (Thermo Fisher). DNA samples with a major peak of 2000 bp and more than 30% of fragments being > 500 bp were chosen for sequencing. The DNA library was constructed using the SureSelect Low Input library prep kit (Agilent) according to the manufacturer's protocol.

The FidoCure® Precision Medicine Platform targets the coding exons of the genes ABL1, ALK, APC, ARID1A, ATM, BCL2, BCL6, BRAF, BRCA1, BRCA2, BTK, CDK2, CDK4, CDK6, CDKN2A, CREBBP, EGFR, ERBB2, FBXW7, FGFR1, FGFR2, FGFR3, FLT1/VEGFR1, FLT3, FLT4/VEGFR3, HDAC1, HIF1, HNF1, HRAS, JAK1, JAK2, JAK3, KDR/VEGFR2, KIT, KMT2C, KMT2D, KRAS, MEK/MAP2K1, MET, mTOR, NF1, NOTCH1, NRAS, TP53, PARP1, PDGFRa, PDGFRβ, PIK3CA, PTEN, PTPRD, PTPRT, RAF1, RB1, RET, ROS1, SETD2, SMAD4, SMARCA4, and TERT. These genes are commonly mutated in human cancers and targeted by commercially available oncology panels.

Hybrid capture-based enrichment of the targeted genes was performed using the SureSelect custom DNA Target Enrichment Probes and SureSelect XT Hyb and Wash kit following manufacturer’s instructions. The final library was quantified using qPCR and pooled for sequencing on the Illumina® platform (Illumina, California, USA) with a read length configuration of 150 paired-end (PE) for up to 6M PE reads (3M in each direction), yielding target read depth averaging at 500 × and with a distribution shown in Fig. S2. Sequencing was performed in a CLIA-certified CAP-accredited laboratory.

Variant calling and evaluation

The sequence read pairs were mapped to the canine reference genome (CanFam3.1)43 using BWA64 (version 0.7.17). Concordantly and uniquely mapped pairs with at least one read with ≥ 1 bp overlapping a coding sequence (CDS) region of the canFam3 1.99 GTF annotation were used to find mutations. Germline base substitutions and small indels were first called by applying GATK65 3.8.1 HaplotypeCaller to the bam files of individual tumor or normal samples with parameters of dontUseSoftClippedBases and -stand_call_conf 20.0. Variants were then filtered with GATK VariantFiltration with parameters of FS > 30.0 and QD < 2.0. Furthermore, variants with total read coverage < 10 were excluded.

Tumor mutation burden (TMB) calculation

TMB value was calculated by \(TMB = \frac{total\;somatic\;mutations}{{total\;callable\;bases\;in\;million}}\) for each tumor, where the “total somatic mutations” are the sum of somatic mutations indicated in Table S2 for the tumor. Callable bases were identified by applying the GATK CallableLoci function to the realigned and duplicate-removed bam file of the tumor, with a minimum base quality set to 10. Samples with a very small number of callable bases (< 11,143) were excluded from TMB calculation.

Germline-somatic mutation discrimination

Mutations identified above were subjected to our germline-somatic mutation discrimination pipeline outlined in Fig. 2. First, these mutations were compared to > 9M known germline mutations collected from databases and literature38,40,41,42,43 to identify and filter out germline mutations. Then, the remaining were divided into groups. For mutations found in ≥ 5 dogs, those with a VAF distribution clustered around 50% or near 100% were classified as heterozygous or homozygous germline mutations respectively. For mutations found in < 5 dogs, those whose human counterparts have been reported to be somatic mutations in human cancers were considered somatic mutations.

Jensen–Shannon (JS) divergence was calculated between two distributions using R package philentropy (version 0.7.0). For permutation testing, the two distributions of interest were combined, random samples of the same size as the original groups were taken, and the JS divergence was calculated between the random samples. This was repeated 100,000 times. The proportion of JS divergences from random samples greater than the original observed JS divergence was considered the p-value.

Somatic mutations identified above were then annotated with Annovar66 (version 2017Jul16).

Definition of somatic mutational hotspots

Mutational hotspots in each species were annotated using the method developed by Chang et al.44, by identifying positions mutated more frequently than the background mutation rate with a cutoff of recurrence in ≥ 4 samples. Mutations at different nucleotide positions in the same codon of a gene and different nonsynonymous and synonymous base substitutions in the same codon were considered together.

Statistics and reproducibility

Statistical analyses were performed using R (version 4.1.0)67. Fisher’s exact tests were used to compare mutation-positive and mutation-negative groups with categorical features to identify enrichment or depletion of variants in different categories. Multiple testing correction was applied using the Benjamini–Hochberg method to obtain the adjusted p values. For all tests, a two-sided adjusted p-value of < 0.05 was considered statistically significant. Enrichment scores were determined by − log10(adjusted p), with positive values indicating enrichment and negative values indicating depletion.