The Genetics of Canine Cancer: A Practical Summary of the Peer-Reviewed Literature

Introduction

Cancer is now the leading cause of death in many purebred dogs, and its incidence varies enormously by breed. Some cancers affect a large share of dogs in one breed while remaining rare in others. This wide variation among breeds is itself evidence of a genetic component. If cancer risk were purely a matter of chance or environment, breed would not matter nearly as much as it does.

Over the past two decades, genome-wide association studies (GWAS) have identified specific chromosomal regions responsible for this inherited risk across several canine cancers and breeds. This document summarizes what the peer-reviewed literature currently shows about the genetics of five of the most consequential canine cancers: hemangiosarcoma, mast cell tumors, lymphoma, osteosarcoma, and histiocytic sarcoma. It also describes a practical, already-proven method for using this genetic information to reduce cancer incidence in future generations through selection. This process works by identifying young dogs to become parents that are at lower risk of producing puppies predisposed to developing cancer.

How Cancer Risk Is Inherited

For more than two decades, veterinary research has focused on finding associations between genes in the canine genome and the development of cancer, in the hope that a DNA test could identify at-risk puppies at birth. Research across the cancers summarized in this document has consistently pointed to the same underlying genetic architecture: cancer risk is influenced by many genes acting together, not by any single gene. When a trait works this way, it is called polygenic, or oligogenic when only a small number of genes are involved.

When a polygenic mode of inheritance is at play, a long-proven method for altering gene frequency in future generations, from the livestock world, is to use statistical modeling to assess each individual's genetic risk. This approach is based on genetic relationships among individuals in a population and on which of those individuals have developed the disease. These statistical estimates of relative genetic merit are called estimated breeding values (EBVs). In humans, a similar health measure is known as a polygenic risk score, calculated from a person's family history of disease and, increasingly, from genetic differences found in their DNA.

Genomically enhanced EBVs, or gEBVs, incorporate DNA information directly and are more accurate than EBVs based on pedigree data alone. Without DNA on a puppy, all puppies from the same two parents share an identical EBV. With DNA information, each puppy's own genetic makeup can be assessed individually, as soon as it is feasible to obtain a DNA sample.

The individual genes that act as modifiers or enhancers within this polygenic architecture can, to some extent, be identified using standard linear modeling techniques applied to SNP array data. This is the statistical basis for a gEBV. Building an accurate gEBV requires genotyping many dogs with a modern SNP array and tracking their health outcomes over their lifetimes. One increasingly practical way to gather this genomic data is low-pass sequencing followed by imputation to a full genome sequence, an approach that both surveys where a breed's genome varies most and supplies the raw material needed to discover new genome locations linked to specific cancers. Realistically, this kind of work requires data from at least several tens of thousands of dogs within a breed before sufficient knowledge can be gained to accurately predict from a single dog's DNA sample its cancer risk as it ages.

Hemangiosarcoma

Hemangiosarcoma is one of the most dreaded and deadly of all canine cancers, and Golden Retrievers are especially prone to it. Cornell University's Baker Institute estimates that 1 in 5 Golden Retrievers will be diagnosed with hemangiosarcoma in their lifetime¹. It is often called a silent killer, because it frequently shows no signs until it has already become untreatable¹.

German Shepherd Dogs are also at high risk. In a 2026 study of more than 130,000 tumors diagnosed in the UK, German Shepherds had the highest odds of splenic hemangiosarcoma of any breed, about seven times that of crossbred dogs¹⁹. A 1988 US study found the same pattern: German Shepherds were the only breed at increased risk, with about 4.7 times the odds of other purebred dogs²⁰.

In 2015, a GWAS of Golden Retrievers identified two loci on chromosome 5 shared between hemangiosarcoma and B-cell lymphoma². Dogs affected by either cancer were analyzed together as a single combined phenotype, since both cancers appear to share much of their genetic risk. The two loci together account for about 20% of the risk of developing these cancers, and all chromosomes combined account for about 43% of the phenotype's variance², indicating that many additional, still-unidentified loci also contribute to risk.

In November 2023, the Morris Animal Foundation funded a follow-on study, still underway, that aims to calculate EBVs describing hemangiosarcoma risk for each of the 3,044 dogs enrolled in the Golden Retriever Lifetime Study (GRLS) and their ancestors. The research team is a collaboration of scientists from the University of Massachusetts Chan School of Medicine, the University of Connecticut, and the International Working Dog Registry (IWDR). Younger dogs related by pedigree to the GRLS dogs are also being included, using relationships obtained from the k9data.com database, and the resulting statistical model will account for pedigree relationships, DNA information, and other contributing factors such as gender and year of birth.

Mast Cell Tumors

The EBV approach being developed for hemangiosarcoma has already worked for a different cancer, mast cell tumors. This was demonstrated in a Labrador Retriever guide dog breeding colony, where the age of onset of mast cell tumors was found to be 28% heritable³. Reducing the risk of mast cell tumors developing at younger ages has been part of the colony's overall selection index since 2010. Its weight in the index was 19% in 2010 and 12% by 2019, with some fluctuation in between, and EBVs improved for every trait in the index over those same years³. By 2019, selection had moved the age of onset of mast cell tumors to older ages, which allowed the weight to be reduced. This progress was made using pedigree-based EBVs alone. No DNA test for mast cell tumor risk was available at the time. This example demonstrates that EBV-based selection can be used to actively manage a polygenic cancer's priority within a real breeding program, not just in theory. It also shows that genetic progress against a cancer does not have to wait for a DNA test in order to reduce its long-term impact on the population.

Lymphoma

Lymphoma is another cancer with a well-documented but complex genetic basis. The same 2015 Golden Retriever GWAS that identified hemangiosarcoma risk loci found those loci shared with B-cell lymphoma². A larger study of histiocytic sarcoma, drawing on Bernese Mountain Dogs, Golden Retrievers, Flat-Coated Retrievers, and Rottweilers, found that its risk loci reflect the additive effect of several risk haplotypes, rather than any single gene, and that these same haplotypes are also involved in lymphoma and mast cell tumors⁴. In Bullmastiffs, a separate GWAS identified two additional risk regions, on chromosomes 13 and 33, reflecting a comparably polygenic background that appears distinct from the loci found in other breeds⁵.

Breed predisposition to lymphoma also varies across countries. A multi-country European study found that Golden Retrievers showed elevated lymphoma risk in the United Kingdom, but not across the wider European population studied. Boxers, in contrast, showed consistently elevated risk across France, Italy, and the UK⁶. These findings suggest that genetic risk factors for lymphoma may differ between regional or national breeding populations of the same breed. This underscores the importance of collecting data from the specific population a breeder is working within, rather than assuming a risk estimate from one country or lineage automatically applies to another.

No heritability estimate for lymphoma alone, separate from the other hematopoietic cancers with which it shares genetic risk, could be found in the published literature; studies to date have analyzed it together with those cancers^2,4. This is a genuine gap in the current literature, not a reflection of lymphoma being any less genetic than the other cancers discussed here.

Osteosarcoma

Osteosarcoma, the most common bone cancer in dogs, disproportionately affects large and giant breeds. A landmark 2013 GWAS across Greyhounds, Rottweilers, and Irish Wolfhounds identified 33 inherited risk loci that together explain 53% to 85% of the phenotype's variance. As with other cancers discussed here, wide variation among breeds was identified: 57% in Greyhounds, 85% in Rottweilers, and 53% in Irish Wolfhounds⁷. The strongest single association, in Greyhounds, lies just upstream of the CDKN2A/B gene region, a region also implicated in canine osteosarcoma tumor tissue itself and as discussed below, in histiocytic sarcoma risk in an entirely different set of breeds.

Pedigree-based heritability estimates for osteosarcoma risk in two additional breeds are similarly high: 0.65 in Irish Wolfhounds⁸ and 0.69 in Scottish Deerhounds⁹. A detailed segregation analysis of a large Scottish Deerhound pedigree, comprising 1,057 dogs and roughly 21% of the breed's AKC registrations from 1975 to 2003, found strong statistical support for a single dominant-acting major gene that raises lifetime risk roughly 15-fold, from a baseline of about 5% to about 75%⁹. No molecular variant has been identified; this is a statistical inference from pedigree and segregation data, not a validated DNA marker. The study estimated the risk allele's frequency in this sample at about 12% to 13%, but the authors caution that this figure is likely higher than the allele's true frequency in the wider Deerhound population, since the dogs studied were deliberately sampled by extending pedigrees outward from five affected dogs, a design that enriches for carriers⁹. The same analysis found that gender significantly modifies penetrance, with estimated lifetime risk substantially higher in females, 90% to 99%, than in males, 72% to 77%, though the authors note this gap could also reflect higher early mortality among male carriers from unrelated causes rather than a true biological effect⁹. A smaller 2017 study of two more recent Deerhound families (54 and 56 dogs) also found inheritance to be explained well by a single gene. In one family the pattern matched the dominant gene of the 2007 analysis. In the other, recessive inheritance was a more plausible fit, though a dominant gene could not be ruled out¹⁸. This remains a notably simpler genetic picture than the many-gene architecture found in most other breeds and cancers discussed in this document, though a contributing role for additional modifying genes cannot be ruled out. This could also indicate that a major gene may be necessary, but not sufficient, to cause elevated osteosarcoma risk in this specific Deerhound population.

A more recent GWAS in Leonbergers found a genome-wide SNP heritability of 20.6%, and independently confirmed the CDKN2A/B locus as a major risk factor in that breed as well¹⁰. A 2019 reanalysis of the original three-breed data concluded that polygenic osteosarcoma risk can, in some dogs who inherit many risk alleles at once, reach a magnitude comparable to a single-gene, or Mendelian, disorder¹¹.

A large UK veterinary-record study found that Scottish Deerhounds, Leonbergers, Great Danes, and Rottweilers all carry substantially elevated odds of osteosarcoma compared to crossbred dogs, with odds ratios ranging from about 27-fold in Rottweilers to as high as 118-fold in Scottish Deerhounds¹².

Histiocytic Sarcoma

Histiocytic sarcoma is most strongly associated with Bernese Mountain Dogs and Flat-Coated Retrievers, and it is unusually lethal. A study of the Dutch populations of both breeds found that it accounts for about 15% of all Bernese Mountain Dog deaths and about 14% of Flat-Coated Retriever deaths¹³, with lifetime risk in Bernese Mountain Dogs reported as high as 25%¹⁴.

A 2009 analysis of Bernese Mountain Dog families found that a fully recessive single-gene model did not explain how the disease was passed on, and favored instead an oligogenic model involving a small number of genes acting together. This built on earlier segregation studies that had already suggested a multigenic mode of inheritance¹⁵. Consistent with that conclusion, a 2021 study combining GWAS in Bernese Mountain Dogs, Rottweilers, and Flat-Coated Retrievers identified three main risk loci, on chromosomes 11 (overlapping CDKN2A, already implicated in osteosarcoma), 5, and 14. Genome-wide, the SNPs explained at least 61.8% of the phenotype's variance, and dogs carrying five or six risk alleles at the three main loci had about five times the odds of developing the disease⁴. A separate study of Flat-Coated Retrievers, in which 20% of the breed develops histiocytic sarcoma, found two different risk loci together explaining 35% to 37% of the phenotype's variance¹⁶. An earlier, more targeted study had already found a single haplotype near the MTAP-CDKN2A region present in 96% of affected Bernese Mountain Dogs studied, a striking concentration of risk around one genomic neighborhood¹⁷.

What These Findings Have in Common

Taken together, these five cancers illustrate a consistent pattern in canine cancer genetics. First, polygenic or oligogenic inheritance, risk shaped by several to many genes acting together, is the norm rather than the exception. The Scottish Deerhound's single dominant-acting osteosarcoma major gene is a notable exception, not the rule. Second, the same genomic neighborhoods keep recurring across unrelated cancers and unrelated breeds. The CDKN2A/B region on chromosome 11 has been implicated in both osteosarcoma and histiocytic sarcoma, and the chromosome 5 loci shared by hemangiosarcoma and lymphoma in Golden Retrievers overlap with a locus also found in histiocytic sarcoma in Flat-Coated Retrievers and Bernese Mountain Dogs. This overlap suggests that a relatively small number of biological pathways may govern cancer susceptibility across many different canine cancers and breeds. Third, no validated DNA test yet exists that can, by itself, reliably identify at-risk puppies for any of these cancers at birth. What does exist, and has already worked, is statistical selection using EBVs built from pedigree and diagnosis data, as demonstrated by the mast cell tumor result described above.

The Deerhound case raises a further question. Why would a single major gene be statistically detectable in this breed but not in Greyhounds, Rottweilers, Irish Wolfhounds, or Leonbergers, where osteosarcoma risk instead appears highly polygenic? Effective population size may be part of the answer. The Deerhound population studied showed a measurable degree of inbreeding, consistent with the breed's small numbers. The original researchers noted that inbreeding tends to concentrate genetic variance onto fewer loci, a process they called trait “Mendelization.”⁹ This can make a major gene easier to detect statistically, even when it is not biologically unique to the breed. A similar risk variant, potentially acting through the same CDKN2A/B pathway already implicated in several other breeds, could exist much more broadly across the species. In a breed with a larger, more genetically diverse population, that same variant would likely remain at low frequency and blend into the polygenic background already described above, rather than standing out as a single dominant gene. This is an interpretation, not a proven fact. No molecular variant was ever identified in the Deerhound study, so the underlying gene cannot yet be directly compared across breeds.

A Practical Path Forward: EBVs and IWDR

Because most of these cancers are late-onset diseases, there has been no tool available for well-intentioned dog breeders to apply meaningful selection pressure against them. Most canine parents will have produced one to three litters by their sixth year of age, so parents who go on to develop cancer later in life will often already have progeny that have themselves become parents, perpetuating elevated risk across generations before anyone notices a problem.

Two practical steps can help in the meantime. First, because these cancers are late-onset and often fatal, a dog that lives well beyond ten years without developing cancer suggests that its family line carries less risk, so replacement breeding stock can be drawn preferentially from longer-lived lines. Traits related to longevity do respond to selection: in a guide dog colony, the length of a dog's working life was heritable, and the risk of it ending early rose with the dog's inbreeding coefficient²¹. Second, in breeds where these cancers are common, such as Bernese Mountain Dogs and Golden Retrievers, breeding stock from outside the local population, including frozen semen from a less related male in another country, can help restore genetic diversity. Selective breeding that reduces genetic diversity can increase rates of disease and shorten lifespans, and outside breeders can restore that diversity, although an imported breeder of lower genetic merit can lower a program's success rate for generations³. Imported stock should therefore be chosen on its own genetic merit, using EBVs where they are available.

EBVs, and gEBVs where DNA information is available, are tools that can be implemented now, without waiting for a single-gene DNA test that may never arrive for most of these polygenic cancers. Calculating them requires that breeders cooperate by contributing pedigree records and diagnosis information to one centralized database. For dog breeders world-wide, the International Working Dog Registry (IWDR) is now a viable option.

IWDR is a world-accessible database widely used by organizations that breed and train working dogs. With about 120 subscribing organizations, it is used by government agencies that breed and train dogs for scent detection work, for police and military work, and by many guide and service dog organizations. Drawing on more than five decades of experience developing record-keeping systems and calculating EBVs for two large US-based guide dog organizations, IWDR is a comprehensive system for managing a canine breeding population, which includes the routine weekly updating of EBVs. As genomic data becomes available on hundreds or thousands of individual dogs, gEBVs will also begin to appear.

Commonly used functionality built into IWDR includes the storage of all medical diagnoses common across working dogs, such as orthopedic, ophthalmic, and cardiac conditions. All medical procedures a dog receives may also be stored. The medical diagnoses and behavioral data are summarized weekly into EBVs for multiple health and behavior traits. Individual trait EBVs are then summarized into an overall index that ranks each dog on relative genetic merit. The relative importance attached to each trait (phenotype) in the overall index is determined by assessing what are the most common reasons preventing dogs from being chosen as replacement breeders. As data on more cancer diagnoses and pedigrees accumulate in centralized systems like IWDR, the same statistical approach can be extended to hemangiosarcoma, lymphoma, osteosarcoma, and histiocytic sarcoma. In one breeding colony it has already improved breeding values for the age of onset of mast cell tumors, meaning that dogs that do develop these tumors are older when the tumors first appear.

Heritability and Genetic Findings at a Glance

Cancer	Breed(s) Studied	Heritability / Variance Explained	Key Loci	Ref
Hemangiosarcoma	Golden Retriever	~20% (2 loci); ~43% all autosomes (combined with lymphoma)	Chromosome 5 (~29, ~33 Mb)	2
Mast Cell Tumors	Labrador Retriever (guide dog colony)	28% heritable (age of onset)	Not locus-mapped in this study	3
Lymphoma	Golden Retriever, Bernese Mtn Dog, Rottweiler, Flat-Coated Retriever, Bullmastiff	Shares loci with hemangiosarcoma/histiocytic sarcoma; no isolated estimate published	Chromosome 5 (shared); CFA13, CFA33 (Bullmastiff)	2, 4, 5
Osteosarcoma	Greyhound, Rottweiler, Irish Wolfhound, Scottish Deerhound, Leonberger	53–85% (3-breed GWAS); h²=0.65 (Irish Wolfhound); h²=0.69 (Deerhound); h²SNP=20.6% (Leonberger)	CDKN2A/B region (chromosome 11)	7, 8, 9, 10, 18
Histiocytic Sarcoma	Bernese Mountain Dog, Flat-Coated Retriever, Rottweiler	≥61.8% (all chromosomes, 3-breed GWAS); 35–37% (Flat-Coated Retriever only)	CDKN2A region (Chr 11); Chr 5, 14, 19	4, 15, 16, 17