The oncogenome of the domestic cat

Abstract

Cancer is a common cause of morbidity and mortality in domestic cats. As the mutational landscape of domestic cat tumors remains uncharacterized, we performed targeted sequencing of 493 feline tumor-normal tissue pairs from 13 tumor types, focusing on the feline orthologs of ~1,000 human cancer genes. TP53 was the most frequently mutated gene, and the most recurrent copy number alterations were loss of PTEN or FAS, or gain of MYC. By identifying 31 driver genes, mutational signatures, viral sequences, and tumor-predisposing germline variants, our study provides insight into the domestic cat oncogenome. We demonstrate key similarities with the human oncogenome, confirming the cat as a valuable model for comparative studies, and identify potentially actionable mutations, aligning with a ‘One Medicine’ approach.

Introduction

Neoplastic disease is the leading cause of morbidity and mortality in companion animals, in particular pet cats and dogs (1), and their tumors share marked clinicopathologic similarities with human tumors (2). Notably, studying companion animal tumors offers several key advantages over using rodent models: Pets are exposed to the same environmental conditions as their owners, develop similar non-neoplastic related co-morbidities, such as diabetes (3, 4) and cardiovascular disease (5), and most relevantly, tumors arise spontaneously in a naturally heterogeneous population (2, 6, 7). Thus, cross-species comparisons have a key role in advancing precision medicine to improve the survival of both humans and their animal companions. This is in line with a ‘One Medicine’ approach (6, 8, 9), which promotes the two-way flow of data and knowledge between medical and veterinary disciplines to benefit both human and animal health (10).

Over the past decade, molecular characterization of canine tumors has grown exponentially, and extensive investment has been made in molecular biomarker discovery efforts (11). In addition, there is preliminary evidence that molecular analysis of tumors using targeted next-generation sequencing (NGS) panels may aid in the clinical management of canine cancer patients by providing information on prognosis and/or potential therapeutic options (12–14), with a vision to follow the trajectory of how genomics-based management has transformed human oncology. In contrast, less than a handful of NGS investigations have been published on feline neoplastic disease, all using small sample numbers and single tumor types (15–17). Thus, understanding the feline oncogenome could facilitate the development of diagnostic and prognostic biomarkers, targeted therapies, and nominate agents for repurposing in feline patients. Ultimately, this has the potential to improve the clinical management of feline cancer patients, and in a comparative oncology setting, this could also benefit human patients.

In this work, we performed targeted sequencing of 493 feline tumor-normal tissue pairs from 13 tumor types, assessing the mutational status of the feline orthologs of ~1,000 human cancer genes and comparing our findings to human counterparts. This study provides insights into the oncogenome of the domestic cat. In addition, we identified strong similarities to the human oncogenome and revealed several potentially actionable mutations.

Results

Demographics of the feline pan-cancer mutational landscape study

We selected tumor types that align with broad human histopathological classifications, encompassing both benign and highly aggressive malignant tumors, as well as common and rare tumor entities. The inclusion of uncommon or rare tumors was intentional: if cross-species comparisons reveal strong genetic similarities to their well-characterized human counterparts, such findings could identify potential therapeutic avenues for affected cats.

The study group consisted of primary tumors from 13 histologically defined tumor types; basal cell carcinoma (BCC, n=40 cases), cholangiocarcinoma (CCA; intrahepatic, n=30), colorectal adenocarcinoma (CRC, n=34), cutaneous mast cell tumor (cMCT, n=41), cutaneous squamous cell carcinoma (cSCC, n=62), glioma (GLIO, n=7), lung adenocarcinoma (LUCA, n=57), lymphoma (LYM, n=51), mammary carcinoma (MAM, n=47), meningioma (MEN, n=28), osteosarcoma (OSA, n=25), oral squamous cell carcinoma (oSCC, n=42), and pancreatic adenocarcinoma (PANC, n=29; Fig. 1A, Table S1(18)). Some cohorts could be subdivided by phenotype or location, specifically LYM (T- or B-cell), MAM (ER+ or ER-), and OSA (appendicular or axial).

Fig. 1. Overview of somatic mutational landscape of 13 feline tumor types.

(A) Distribution of the tumor-normal samples based on tumor type (pie chart), and frequency of the age, sex, and breed of cats in the study cohort (bar plots). CNS, central nervous system; F, female; MC, Maine Coon; M, male; NK cell, natural killer cell; Siam, Siamese; U, unknown. (B) Distribution of the tumor mutational burden (TMB) in each tumor type. Each dot represents a sample, and the black horizontal line indicates the median TMB of the respective tumor type (only samples with >0 mutations that passed variant filtering are shown). LYM_Bcell, B-cell lymphoma; LYM_T-cell, T-cell lymphoma. (C) Genes identified as being under positive selection pressure in each tumor type (driver genes). Within each tumor type, the driver genes are listed in order of decreasing significance (q-global < 0.1) and those shown in color are present in more than one tumor type. The driver genes shown for LYM are for the B-cell subtype, because no driver genes were identified for the T-cell subtype. (D) Oncoplot of the top 5 most frequently mutated genes from each tumor type. Mutations include SNVs, multinucleotide variants (MNVs) and small insertions/deletions (indels; <100bp).

The study population included more females than males (56% vs. 42%, respectively Fig. 1A), primarily due to the MAM cohort being entirely female. The tumor distribution was primarily towards older cats (with a median age of 11 years, range: 0.7 – 21 years; Fig. 1A, Table S1), consistent with cancer registries recording 8-11 years as the mean age of tumour presentation (depending on the country) (19–22). The most common breed was Domestic Shorthair (DSH; a general term for a non-pedigreed cat with a short coat; 73.1%), followed by Domestic Longhair (DLH; 10.4%), Siamese (2.2%), and Maine Coon (1.8%), with the “Others” category composed of 24 different breeds (n=1 to 10 cats each; Fig. 1A). All sample details, including signalment data, are provided in Table S2 (18).

The somatic mutational landscape across cancer types

To provide an investigation of cancer-associated genes that may be altered in feline tumors, we compiled a list of 1,039 human cancer-associated genes (see materials and methods) and identified their feline orthologs; ultimately, 978 feline genes were included in our targeted panel. Tissue samples were from the original diagnostic formalin-fixed, paraffin-embedded (FFPE) blocks, with the diagnosis independently verified by a board-certified veterinary pathologist, who then selected the tumor and normal areas for sampling. Genomic DNA extracted from the 493 tumor-normal tissue pairs was hybridized with the targeted panel, and the captured DNA was subsequently analyzed by NGS. To gain an overview of the feline somatic mutational landscape, we first identified the somatic mutations in each tumor, including somatic single nucleotide variants (SNVs), multi-nucleotide variants, and insertions/deletions, relative to the FelCat9 reference assembly (Table S3 (18) lists the protein-altering variants in each sample). We have also provided the variant locations relative to the Fca126 reference assembly (Table S3 (18)

The tumor mutational burden varied across the tumor types, with a median of ~0.5 to ~20 mutations/megabase (Fig. 1B). Mutational signature analysis identified COSMIC Signature SBS7 in 34/62 (52%) of the cSCC samples (Table S4 (18)). The proposed etiology of SBS7 is ultraviolet (UV) light exposure (23), with this signature predominantly found in UV-associated human skin cancers.

Across the study, we identified 31 driver genes (Fig. 1C, Table S5 (18)). Five genes, TP53, CTNNB1, PTEN, TRAF3, and FBXW7, were drivers in multiple tumor types, whereas most were only drivers of a specific tumor type, such as KIT for cMCT and PIK3CA for MAM. To gain insights into the comprehensiveness of our targeted sequencing approach for driver gene identification, we performed whole-exome sequencing (WES) of freshly collected feline MAM (n=18; Fig. S1, Table S3 (18)), finding FBXW7 and PIK3CA as drivers, in keeping with results from targeted sequencing of our MAM cohort (Table S5 (18)). Across the entire pan-cancer cohort, TP53 was the most recurrently mutated gene (Fig. 1D), and 14 mutation hotspots in 10 genes were identified (Table S6 (18)). The most frequently observed were PIK3CA p.H1047R, found in 19/493 (4%) tumors, most commonly occurring in MAM (30% of the cohort). To rule out the possibility that recurrent mutations were artifacts, somatic hotspot mutations in the MAM and cMCT cohorts were orthogonally validated with capillary sequencing (Fig. S2).

As somatic copy number alterations (CNA) drive many human tumor types, it was important to characterize the CNA landscape of the feline tumors. On average, MAM showed the most CNAs (mean genome fraction altered: 16.94%, range: 0.00-66.72%, and CRC showed the least (mean: 0.59%, range: 0.00 to 9.88%; Fig. 2A). Recurrent whole chromosome gains and losses were seen, with eleven tumor types showing gain of chromosomes F1 or F2, and nine showing loss of chromosome X (Table S7 (18)). Considering both broad (≥10 Mb up to a whole chromosome) and focal (<10 Mb) CNA across the feline tumor types, copy number (CN) gains were consistently seen on chromosomes A3, B4, F1, and F2. The gene most altered by CN gain was MYC (20% all tumors), particularly noticeable in T-cell LYM (57% of cases). CN loss at the start of chromosome D2 was seen in all tumor types except CRC. The genes most affected by CN loss were PTEN, FAS (~20% all tumors), and CDKN2A (15% all tumors). Genes within each of the broad and focal gains/losses are shown in Fig. S3. When considering both the CNAs and somatic mutations within each sample, we observed an inverse relationship, particularly in the highly altered tumors (Fig. 2B); such tumors were replete with either CNAs or mutations, but rarely both, which is consistent with the ‘cancer genome hyperbola’ of human tumors (24). The samples with the highest recurrent mutations were predominantly found in the cSCC cohort, whereas the samples with the highest CNAs came from various tumor types (Table S8 (18)).

Fig. 2. Overview of somatic CNA landscape of 13 feline tumor types.

(A) Somatic CNAs across the 13 tumor types. CN gains/amplifications [log₂ (fold change) ≥ + 0.32] are shown in red and losses/deletions [log₂(fold change) ≤ -0.4] are in blue. The color shown is the proportion of tumors (0-40% of the cohort) showing an overall CN gain or loss at 1 Mb intervals across the feline genome (chromosomes indicated on the x-axis). Tumor types were hierarchically clustered using average linkage and Euclidean distance on their net gain/loss profiles to define the dendrogram structure. (B) Hexagon-binned density plot showing the approximate inverse relationship between focal CNA burden (x-axis) and recurrent mutation burden (y-axis). CNAs were counted as focal events ≤ 10 Mb with log₂ (fold change) ≥ +0.32 (gain) or ≤ −0.40 (loss), summed per sample. Recurrent mutations were counted per sample and restricted to genes mutated in ≥ 2 samples across the cohort. Truncating variants (nonsense, frameshift indel, splice-site, start-loss, nonstop) were always included. Missense and in-frame indels were included for non-hypermutators; for hypermutator tumors (considered as those with an SBS7 signature) they were counted only when occurring in hotspot events, i.e. the same amino-acid position mutated in ≥ 2 tumors with a ±1-residue tolerance. Hexagon colour encodes the number of samples in each bin; only samples with both mutation and CNV data were included.

Next, we examined the feline somatic mutational landscape on a per-tumor-type basis, to define the genetic landscape of each tumor type, which may shed light on their biological behavior. Complete descriptions of the somatic mutations and CNAs identified in each tumor type are included in the Supplementary Text, Figs. S4-S17 and Table S9 (18).

Presence of papillomavirus in cSCC and BCC

Some feline cancers have been associated with viral infection, with the oncogenic potential of feline leukemia virus (FeLV), feline immunodeficiency virus (FIV) and papillomavirus (PV) having been well-established in cats (25, 26). Thus, to look for a possible viral etiology of the tumors in this study, we searched the off-target sequencing reads for the presence of viral DNA sequences (filtering out any reads for FIV or FeLV; see materials and methods). This approach identified strong evidence of two different genera of PV, specifically Dyothetapapillomavirus (DyoPV) and Taupapillomavirus (TauPV), in the cSCC and BCC samples, with only one oSCC sample showing the presence of TauPV sequences (Fig. S18).

Germline predisposition variants

Sequencing of matched normal tissue for each tumor enabled the identification of putative pathogenic germline variants that may have predisposed these cats to cancer. We considered loss-of-function variants and known human cancer predisposition genes (see materials and methods). With this approach we identified variants in 14 genes (Table S10A (18)). Of the 493 cats in this study, 20 had at least one putative cancer-predisposition germline variant, with only one cat having more than one variant. The median age at diagnosis for these 20 cats was 11 ± 3.7 years (SD), which was not significantly different from cats without one of these variants (P=0.5671; two-sided Wilcoxon rank-sum test). CHEK2 was the only gene to have recurrent variants in >2 samples. One of the CHEK2 variants (c.546C>A; p.Y182*), found in 4 cats, mapped to an orthologous site in humans, which is annotated in the ClinVar database as a pathogenic allele in humans (Table S10B (18)). Similarly, additional orthologous variants in CHEK2 and BRIP1 (n=1 each) have been reported in the ClinVar database as pathogenic (Table S10B (18)).

Cross-species comparison with mutations found in human cancers

We first compared the mutational frequency of feline driver genes within tumors of the same type from human pan-cancer datasets (Table S11 (18)). Only genes analyzed in both datasets were considered (Fig. 3A). In some cases, the genes showed similar tumor type mutational frequencies between the species, such as PIK3CA in MAM, and TP53 in OSA, oSCC and MAM. Conversely, in some cases, the mutational frequencies were very different, such as APC in CRC, CTNNB1 in PANC and FBXW7 in MAM. A cross-species comparison of the mutational landscape of each tumor type, including somatic mutations and CNAs, is included in the Supplementary Text.

Fig. 3. Cross-species comparison of the somatic oncogenomic mutational landscape between humans and felines.

(A) Mutation frequencies of the feline driver genes in the feline cohort (y-axis) versus the human cohort (‘China Pan-cancer’, TCGA’ or ‘MSK-IMPACT’; x-axis) per tumor type. The red dotted line indicates the null hypothesis (no difference in mutational frequency between the species). (B) Comparison of the feline and human hotspots. Lollipop plots depict the mutational distribution in TP53, FBXW7, CTNNB1, PIK3CA, and KIT in human tumors (upper panel; obtained from the COSMIC database) and cat tumors from this study (lower panel). Between panels the human protein with relevant domains (based on the canonical transcript as defined by Ensembl v104) is indicated. Domain information obtained from the SMART or Pfam databases. C2_PIK, C2 phosphatidylinositol 3-kinase-type domain; Ig, immunoglobulin subtype; PIK_AB, phosphatidylinositol 3-kinase, adaptor-binding domain; PI3K-PI4K_cat, phosphatidylinositol 3-/4-kinase, catalytic domain; PIK_acc; phosphoinositide 3-kinase, accessory (PIK) domain; PIK_RBD, phosphatidylinositol 3-kinase Ras-binding (PI3K RBD) domain; PTK_cat, tyrosine-protein kinase, catalytic domain; TA, transactivation domain; Tetra, tetramerization domain.

We next compared the frequency and position of mutations in five key driver genes identified in this study with those in human cancers by ‘humanizing’ cat mutations (see materials and methods). Shared recurrent mutations were seen in TP53, predominately in the DNA binding domain; FBXW7, in the WD40 repeats; and CTNNB1, in the N-terminus (amino acids 32-37) - a region critical for stabilization of the protein (27) (Fig. 3B). However, while p.H1047R/L in the catalytic domain of PIK3CA was highly recurrent in both species, p.E545K in the accessory domain was far more recurrent in human cancers than feline cancers. In addition, KIT p.D816V was only present in human cancers. Additional shared recurrent mutations may be identified as other cat tumor types are sequenced.

Actionability of feline driver gene mutations

We next sought to use our catalog of feline driver mutations for the benefit of cats by leveraging human tractability and druggability databases. Although this knowledge is absent for cat proteins, candidate treatments may be identified for proteins with a high level of homology with the human counterpart. This reasoning is supported by data from dogs that have tumors with specific genomic alterations and show improved outcomes when treated with human-targeted treatments (13, 14), and evidence of biological activity of the canine-approved tyrosine kinase inhibitor (TKI) toceranib, in cats with KIT-mutated MCTs (28, 29).

We first searched a database of protein druggability predictions (30) and found 6/31 (19%) driver genes encoded proteins for which an approved drug exists (‘Tclin’), with 102/493 (21%) tumors having mutations in these driver genes (Fig. 4A, Table S12A (18)). Next, we searched a database of validated cancer synthetic lethal (SL) targets (31) and found 5/31 driver genes had druggable SL partners, with 181/493 (37%) tumors having mutations in these driver genes (Fig. 4B, Table S12B (18)). Finally, we searched the OncoKB actionability database (32), specifically using ‘humanized’ feline mutations within the driver genes (SNVs only; see materials and methods), and found 67/493 (14%) tumors had an oncogenic/likely oncogenic mutation in at least one of the actionable driver genes (specifically, PIK3CA, MAP2K1, KIT, FBXW7, FGFR2, PTEN, and NF1; Fig. 4C, Table S12C (18)). As a further proof-of-concept we used patient-derived three dimensional (3D) feline MAM tumoroids to investigate whether FBXW7 mutations create a genotype-specific vulnerability (tumoroids details are in Table S2). After WES of tumoroids, six lines were selected; three FBXW7 wildtype lines and three FBXW7 mutant lines (Table S3 (18), Fig. S19). FBXW7-mutant lines were significantly more sensitive to the vinca alkaloids, vincristine and vinorelbine (Fig. 4D, Table S13 (18)), consistent with a previous in vitro study using human HAP1 cells (33); a result recommending further investigation in larger cohorts.

Fig. 4. Actionability of feline driver gene mutations.

(A) The proportion of feline tumors that had a mutation in at least one of the feline orthologs of driver genes present in the Target Central Resource Database (TCRD) and the number of feline tumors with a mutated driver gene where the human ortholog has a Pharos Target Development Ranking (Tclin, Tchem, or Tbio). ‘Tclin’ encode proteins for which an approved drug exists, ‘Tchem’ encode proteins not in Tclin but known to bind small molecules with high potency, and ‘Tbio’ encode proteins with a moderate amount of available data (including literature references, GeneRIF annotations, experimental data). (B) The proportion of feline tumors that have a mutation in at least one driver gene with a SL partner that is druggable, and the frequency of the mutation of these driver genes (TP53, PIK3CA, APC, PTEN, and MSH2). (C) The proportion of feline tumors with an actionable mutation in at least one of the driver genes present in the OncoKB database, and the number of tumors with an actionable mutated driver gene within each therapeutic category (Level 1-4). ‘Level 1’ therapeutic biomarkers represent US Food and Drug Administration (FDA)-recognized biomarkers predicted to respond to FDA-approved drugs, ‘Level 2’ therapeutic biomarkers represent standard care biomarkers that are recommended by professional guidelines to be predictive of response to FDA-approved drugs, ‘Level 3’ therapeutic biomarkers show compelling clinical evidence, standard care or investigational evidence of a response to an FDA-approved or investigational drug, and ‘Level 4’ therapeutic biomarkers show compelling biological evidence of being predictive of response to a drug. (D) Dose–response curves comparing vincristine and vinorelbine sensitivity between FBXW7-mutant and wildtype tumoroids are shown, with error bars representing 95% confidence intervals (CI). Experiments were performed using three independent tumoroid lines (per genotype) representing biological replicates. For each line, two independently cultured experiments were performed, and each condition was assayed in technical triplicate. For vincristine, the estimated IC₅₀ was 0.04226 μM in wildtypes, compared with 0.002895 μM in mutants. For vinorelbine, the IC₅₀ was 0.06163 μM and 0.007452 μM in wildtypes and mutants, respectively. An extra sum-of-squares F test indicated that the dose–response curves differed significantly between wildtype and mutant groups for both drugs (p < 0.0001).

Discussion

Despite cats being the second most frequent companion animal in households across many countries, and neoplastic disease being one of their leading causes of morbidity and mortality, there is a dearth of studies investigating the genetics of feline tumors. Thus, cancer gene mutational profiling of 493 feline tumor-normal pairs from 13 tumor types offers important insights into the feline oncogenome.

TP53 was the most frequently mutated gene in this feline pan-cancer study (33% of all tumors), mirroring reports in human pan-cancer studies (34% of all tumors) (34). Similarly, the most recurrent CNAs identified in the feline tumors were gain of MYC and loss of PTEN/FAS (each observed in 20% tumors). Consistently, human pan-cancer analyses have reported frequent MYC amplification (28% of cancers; (35)) and PTEN hemizygous deletion (25% of cancers; (36)). By contrast, RAS gain-of-function mutations are found in ~25% of human cancers (37), whereas RAS was not a driver in any feline tumor types in our study, and feline hotspot mutation studies have found RAS mutations to be uncommon (38–42).

Consistent with previous accounts of FcaPV2 (DyoPV) and FcaPV3/4/6 (TauPV) DNA being present in feline cSCC (43), and FcaPV3/5 DNA being present in feline BCC (44, 45), we observed DyoPV and TauPV DNA sequence in some cSCC and BCC samples. However, we also observed these sequences in some of the normal samples, and most feline PV infections do not result in neoplasia (43), consistent with human PV infections (46). It is currently poorly understood why PV infections only result in neoplasia in some cats, with additional co-factors proposed to be required to induce transformation, such as UV exposure (43); notably, 25% of our cSCC with PV DNA sequences present also showed a UV mutational signature.

To-date, only one cancer predisposition allele has been reported in cats (a single nucleotide polymorphism in TP53 intron 7 in feline injection-site sarcoma) (47), although this was not replicated in a second study (48). We identified 14 putative pathogenic germline variants that may predispose these cats to cancer. Notably, the limited catalog of feline germline variants that have been collected (49), and the genetic distance between the cats we sequenced and the Abyssinian reference genome (50), will require further validation of these candidate predisposing alleles.

Because several tumor types harbored recurrent, potentially actionable mutations, we considered the potential clinical utility of genomic profiling in feline cancers. FBXW7 is a driver gene of feline MAM (mutated in 53 to 72% of cases) and the combination therapy of lunresertib with camonsertib is in human clinical trials for cancers with specific genomic alterations, including FBXW7 mutations (MYTHIC Study, NCT04855656). PIK3CA is also a feline MAM driver gene (mutated in ~45% of cases) and there are PI3K inhibitors available, such as combinations of alpelisib with fulvestrant or capivasertib with fulvestrant, approved for treating PIK3CA-mutated breast cancer in humans. These drug combinations may hold translational promise in feline oncology pending further evaluation. Finally, because KIT is a driver gene of feline cMCT (mutated in ~40% of cases), screening feline cMCT cases for the presence of oncogenic KIT mutations may be of clinical value, particularly because the canine-approved TKI toceranib, with a wide variety of molecular targets, including KIT, is well tolerated in cats and clinical responses have been reported in feline cMCT (28, 29). Overall, our findings support the value of genomic characterization in feline cancers, particularly MAM and cMCT, both for identifying druggable alterations and for predicting therapeutic responses, as demonstrated through our functional validation using tumoroids.

By sequencing orthologs of ~1000 known human cancer genes, we limited the discovery of driver genes in feline cancer to those in human cancers. Nevertheless, this study offers insight into the feline oncogenome, identifying strong similarities to the driver events in human cancers and providing several potentially actionable mutations for further investigation. This should serve as a valuable resource for veterinarians and comparative oncologists, providing a step towards precision oncology for our feline companions.

Data and materials availability

The raw sequencing data for each cohort are available for download from the European Nucleotide Archive (ENA; https://www.ebi.ac.uk/ena/browser/home) under study accessions ERP143204 (feline BCC), ERP145946 (feline CCA), ERP141603 (feline CRC), ERP133248 (feline cMCT), ERP133244 (feline cSCC), ERP141601 (feline GLIO), ERP165333 (feline LUCA), ERP141159 (feline LYM), ERP141158 (feline MAM), ERP136468 (feline MEN), ERP133245 (feline oSCC), ERP143495 (feline osteosarcoma), ERP137019 (feline PANC), and ERP162973 (WES of feline MAM and tumoroids). All code is archived within Zenodo (as detailed in the Materials and Methods), and links to individual analyses are collated in GitHub (51). Requests for the feline mammary carcinoma tumoroids can be addressed to Prof. Sven Rottenberg.