The Genetics of Dog Behavior: What Genes Actually Explain

1. Three Different Questions Called "Genetics"

1.1 Why the Answers Look Contradictory

Two statements from the canine behavioral genetics literature, both from large studies, both correct:

Behavioral differences between dog breeds can show substantial genetic influence at the population level.

Breed explains about 9% of the behavioral variation between individual dogs.

Read as answers to the same question, those are incompatible. They are not answers to the same question. The first concerns breed averages: how much of the difference between what Labradors typically do and what Border Collies typically do tracks genetic difference between the breeds. The second concerns individuals: if you know a dog's breed, how much of what that particular dog does have you accounted for.

Almost every argument about genetics and dog behavior runs on this confusion. Once the levels are separated, the field turns out to be unusually coherent.

1.2 What This Article Covers

This article sets out what heritability actually measures and why one trait has several different heritabilities, the numbers that have been produced at each level, why large breed differences and weak individual prediction are both real, what the gene searches have and have not found, the evidence that genes and environment interact rather than add, where commercial genetic testing sits relative to the research, and what all of it means when a particular dog is standing in front of you.

It is deliberately an article about the architecture rather than about which breed does what. The phenotype level — what breeds actually differ in and how much that predicts — is treated separately (breed and behavior).

2. Heritability: What the Number Means

2.1 A Statistic About Populations, Not About Individuals

Heritability, written h², is the proportion of the variation in a trait within a defined population that is attributable to genetic variation in that population. It is not the proportion of a trait that is genetic in any individual.

The distinction is not pedantic; it is the whole thing. A heritability of 0.40 for noise fear does not mean that 40% of a given dog's fear is genetic and 60% is upbringing. It means that, across the dogs measured, about 40% of how much they differed from one another could be accounted for by how much they differed genetically. Change the population and the number changes, without any gene changing.

2.2 Why One Trait Has Several Heritabilities

Because heritability depends on the variation present in the sample, the same trait yields different figures depending on where you draw the boundary.

Among breeds. Take breed averages as the data points. The genetic differences between breeds are large and structured, so heritability estimated at this level tends to be high.

Within one breed. Take individual Labradors. The genetic variation available is a fraction of what exists across all dogs, so the same trait will show a lower figure.

Within one population, measured directly. Take a cohort of puppies from a breeding program and measure behavior in a standardized task rather than by owner questionnaire. Different measurement, different figure again.

None of those is the "real" heritability. They answer different questions, and the literature contains all three — which is why quoting a single number without saying at which level it was estimated is close to meaningless.

2.3 Four Things Heritability Does Not Tell You

It does not tell you a trait is unchangeable. Heritability describes the sources of variation under the conditions that prevailed in the sample. It says nothing about the effect of a condition nobody in the sample experienced — including a training intervention nobody received.

It does not tell you a trait is inevitable in an individual. A high population figure is compatible with enormous individual variation and with individuals who are nothing like their breed average.

It does not identify a gene. Heritability is a variance estimate. It can be high for a trait whose genetic basis is spread across hundreds of loci of individually negligible effect.

It does not separate genes from what genes did earlier. Maternal effects, litter environment, and prenatal conditions can load onto the genetic term in some designs. This is one reason pedigree-based and genome-based estimates of the same trait sometimes disagree, and why experience that alters gene expression rather than gene sequence belongs in the picture too (epigenetics in dogs).

3. The Numbers, at Three Levels

3.1 Among Breeds: High

MacLean, Snyder-Mackler, vonHoldt, and Serpell combined owner-reported behavior for more than 14,000 dogs across 101 breeds with genetic data at more than 100,000 genomic sites in 5,697 dogs. They reported high among-breed heritability for 14 behavioral traits and identified 131 single-nucleotide polymorphisms associated with breed behavioral differences. The associated genes were enriched for neurobiological functions and developmental processes and were highly expressed in the brain.

The claim being made there is precise and worth restating in its own terms: differences in breed-average behavior were associated with genetic differences between breeds, and the implicated genes are of a kind that could plausibly be doing the work. It is a strong result about breed averages. It is not a result about whether a given dog will do a given thing.

3.2 Within One Breed: Low to Moderate

Ilska and colleagues analyzed 12 owner-reported personality traits in UK Kennel Club-registered Labrador Retrievers, with 1,975 animals in the pedigree analysis and 885 in the genomic analysis. Restricting the question to variation inside a single breed produces a very different picture. The estimates, with standard errors in brackets:

Fetching 0.38 (0.08). Noise fear 0.30 (0.08). Non-owner-directed aggression 0.29 (0.08). Trainability 0.28 (0.07). Unusual behavior 0.25 (0.08). Agitated when ignored 0.22 (0.07). Barking tendency 0.15 (0.07). Attention-seeking 0.14 (0.06). Excitability 0.10 (0.06). Human and object fear 0.08 (0.05). Separation anxiety 0.06 (0.05). Owner-directed aggression 0.03 (0.04).

Several features of that list deserve attention. The highest figure belongs to fetching — a motor-motivational tendency, and in Labradors specifically the one that has been selected on hardest. Noise fear is next (noise sensitivity). And the bottom of the list is occupied by the traits owners most want explained: separation anxiety at 0.06 and owner-directed aggression at 0.03, both with standard errors as large as or larger than the estimate itself, which means neither is meaningfully distinguishable from zero in this sample.

Within a breed, in other words, the problems that bring dogs to a behaviorist showed the least genetic signal of anything measured (separation-related behavior, aggression in dogs).

3.3 Within a Population, Measured Directly

Owner questionnaires and behavioral tests are not interchangeable, and the third level uses tests.

Bray, Gnanadesikan, Horschler, Levy, Kennedy, Famula, and MacLean tested 375 eight-week-old retriever puppies from 117 litters — Labradors, Golden Retrievers, and crosses of the two — on a battery of social-cognitive tasks. The puppies followed human points and attended to human faces successfully from the first trials, with no evidence of within-session learning.

The heritability estimates varied sharply across measures that sound similar: point-following 43%, attention to a human face during speech 43%, sensitivity to a communicative marker 14%, social approach 13%, and gaze during an unsolvable task 8%.

That spread is the useful part. "Responsiveness to people" is not one heritable thing. Following a pointing gesture and looking at a face while spoken to carried substantial genetic signal in this population; looking to a human for help when stuck carried very little (how dogs read human gestures).

Gnanadesikan, Hare, Snyder-Mackler, and MacLean took a different route, estimating heritability across breeds from 11 cognitive measures for 1,508 adult dogs in 36 breeds collected through a citizen science platform. Four cognitive factors emerged, with estimates of h² = 0.70 for inhibitory control, 0.39 for communication, 0.21 for physical reasoning, and 0.17 for memory. The inhibitory control figure remained high at h² = 0.50 after controlling for breed-average body weight, so it is not simply a size effect.

Inhibitory control being the most heritable cognitive factor at the among-breed level is a finding with practical reach, since it is the capacity most often described in training as impulse control (prefrontal control and self-regulation). Memory being the least heritable is equally notable, and matches a breed-comparison result described below.

4. Breed: Large Differences, Weak Individual Prediction

4.1 The Differences Between Breeds Are Real and Large

Nothing in this literature supports the claim that breeds do not differ behaviorally. They differ substantially, and the largest datasets show it most clearly.

Salonen, Sulkama, Mikkola, Puurunen, Hakanen, Tiira, Araujo, and Lohi surveyed owners of 13,715 Finnish pet dogs across 264 breeds on seven anxiety-related traits. Noise sensitivity was the most common at 32% prevalence. Breed differences in prevalence were large across all seven traits: as one example, aggression toward strangers was reported for 10.6% of Miniature Schnauzers and 0.4% of Labrador Retrievers. The authors take the size of these differences as suggesting a strong genetic contribution.

Behavioral testing finds the same thing in cognition, and finds where it stops. Junttila, Valros, Mäki, Väätäjä, Reunanen, and Tiira tested 1,002 dogs from 13 breeds. Breed differences appeared in social cognition, inhibitory control, spatial problem-solving, persistence and human-directed behavior during an unsolvable task, activity level, and exploration of a novel environment. Breed differences did not appear in short-term memory or in inference-based logical reasoning.

So the breed effect is not uniform across the mind. It shows up in how a dog engages with people and with obstacles, and not in the tested aspects of memory and reasoning (cognitive abilities in dogs).

4.2 And Breed Explains 9%

Morrill and colleagues surveyed owners of 18,385 purebred and mixed-breed dogs and genotyped 2,155 of them. Their headline findings, in their own terms:

Most behavioral traits were heritable, with h² above 25%. Admixture patterns in mixed-breed dogs revealed breed propensities — that is, the genetic contribution of a breed to a mixed dog carried detectable behavioral signal. Genome-wide association analysis identified 11 loci significantly associated with behavior. Behavioral loci were not unusually differentiated between breeds. Breed propensities aligned with ancestral function, but only weakly.

And breed explained just 9% of behavioral variation in individuals.

Their interpretation is that behaviors perceived as characteristic of modern breeds derive from thousands of years of polygenic adaptation predating breed formation, and that modern breeds are distinguished primarily by aesthetic traits.

4.3 How Both Results Can Be True

Three things reconcile them.

Within-breed variation dwarfs between-breed variation. A trait can differ substantially on average between two breeds while the ranges overlap almost completely. Knowing which distribution a dog was drawn from shifts the expectation a little; it does not locate the dog within the distribution.

Heritable and breed-predicted are different properties. Morrill found most traits heritable at above 25% and breed accounting for 9%. Both come from the same dataset. A trait can carry real genetic signal that is distributed across dogs rather than sorted neatly by breed — which is exactly what "behavioral loci are not unusually differentiated in breeds" says.

Modern breeds are recent and were largely built on appearance. If the genetic variation underlying behavior is older than the breeds and was never fully partitioned by them, then breed membership is a coarse proxy for it. That is the structural claim in Morrill's conclusion, and it explains the 9% rather than merely restating it.

4.4 What the 9% Does and Does Not Say About Breed-Based Risk Categories

Breed-based restrictions exist in many jurisdictions and the legal frameworks differ too much for a single assessment. What the genetics literature can be asked is narrower: does it support the premise that breed membership identifies dangerous individuals?

The findings bear on that in three ways.

First, breed-average differences in reported aggression are real and can be large. The 10.6% against 0.4% figure for stranger-directed aggression is not a rounding artifact, and any claim that breeds are behaviorally interchangeable is not supported.

Second, breed accounts for roughly a tenth of behavioral variation between individuals. A categorical rule applied to individuals on the basis of breed will therefore misclassify most of the animals it touches in both directions, because the distributions overlap heavily.

Third, and most directly, Zapata and colleagues examined Pit Bull-type dogs specifically in their genotyped sample and reported no unique behavioral genetic signature; the one difference they identified was increased leash-pulling. That is a single exploratory study of 397 dogs and should not be asked to carry more than it can, but it is the finding that speaks most directly to a breed-specific genetic premise, and it is negative.

The defensible summary is that breed carries population-level information and does not identify individuals. What predicts the behavior of a particular dog is the behavior of that particular dog, assessed directly — which is what a functional evaluation does and what a breed label cannot do. Whether a regulatory system should nonetheless work with population averages is a policy question with costs on both sides, and it is not settled by the genetics.

4.5 What That Leaves You With in Front of a Real Dog

Breed is a weak prior. It is not nothing — a 10.6% versus 0.4% difference in reported stranger-directed aggression is a real difference and worth knowing when choosing a puppy. But as a statement about the animal in the room it carries little, and it carries least for exactly the traits an owner asks about.

The practical consequence is that assessment of an individual dog cannot be replaced by pedigree. What this dog does, in which situations, is still something that has to be observed and described (behavioral assessment, operationalizing behavior).

5. What the Gene Searches Have Found

5.1 Why Dogs Should Be an Easy Case

Dogs ought to be a favorable species for behavioral genetics. Breeds are closed populations with high relatedness, long haplotype blocks, and known recent selection on specific working tasks. Where human genetics needs hundreds of thousands of participants to detect small effects, dogs offer structured populations in which effects should be easier to find.

That advantage is also the trap: population structure means breed ancestry is correlated with almost everything, so an association can reflect breed membership rather than the trait. Handling this is the central methodological problem in the field.

5.2 What Has Actually Been Identified

The honest summary is: a small number of loci, mostly of modest effect, rarely replicated across studies.

Morrill's genome-wide analysis found 11 loci significantly associated with behavior across a sample of 2,155 genotyped dogs, and described the characteristic breed behaviors as genetically complex. MacLean's among-breed analysis identified 131 SNPs associated with breed behavioral differences, in genes enriched for neurobiological and developmental functions.

The most specific findings come from single-breed studies. Sarviaho, Hakosalo, Tiira, Sulkama, Salmela, Hytönen, Sillanpää, and Lohi studied fearfulness in German Shepherds — 91 noise-sensitivity cases against 210 controls, and 80 fear cases against 193 controls, 330 dogs in total. They reported a region on chromosome 20 associated with noise sensitivity and a region on chromosome 7 associated with fear of strangers and novel situations. The chromosome 7 region is largely syntenic to the human locus 18p11.2, which has been linked to bipolar disorder and schizophrenia, and the candidate genes in these regions relate to glutamatergic and dopaminergic neurotransmission.

That cross-species overlap is the most interesting thing in the paragraph, and also the one most easily overstated. Syntenic regions containing plausible candidate genes are a reason to look further; they are not a demonstration that the same mechanism produces fear in a German Shepherd and bipolar disorder in a person. The sample was 330 dogs of one breed, and findings at that scale need replication (the neurobiology of anxiety, neurochemistry and behavior).

5.3 Regulatory, Not Coding

Dutrow, Serpell, and Ostrander approached the population-structure problem by reorganizing the question. Rather than comparing breeds, they used genetic data from more than 4,000 domestic, semi-feral, and wild canids together with behavioral survey data for more than 46,000 dogs, and identified ten major canine genetic lineages with their behavioral correlates.

Two of their conclusions matter here. Breed diversification was predominantly driven by non-coding regulatory variation rather than by changes to protein-coding sequence — so the differences are largely in when and how much genes are expressed, not in what proteins they specify. And the lineage-associated genes converged in neurodevelopmental co-expression networks, with a sheepdog-associated enrichment for interrelated axon guidance functions.

Axon guidance is the machinery by which developing neurons find their targets. If selection for herding acted on that, it acted on how a brain gets wired during development rather than on a behavior module in the adult — which is a very different picture from "a gene for herding" (the neurology of dog behavior, the sensitive period).

5.4 Why the Effects Are Small: Polygenicity

The gene lists from this field look thin next to the confidence with which genetic explanations get offered, and there is a structural reason for that rather than a failure of effort.

Behavior is polygenic. A trait can be substantially heritable while the underlying variation is distributed across hundreds or thousands of loci, each contributing an effect too small to reach significance at achievable sample sizes. Morrill's description of characteristic breed behaviors as genetically complex is exactly this observation, and it is what the 11 significant loci from 2,155 genotyped dogs should be read against: not "only 11 genes matter," but "11 are large enough to see at this sample size."

Three consequences follow, and they are worth holding because they predict what future studies will and will not deliver.

Larger samples will yield more loci with smaller effects. This is what happened in human complex-trait genetics, and there is no reason to expect dogs to behave differently. More loci will not mean more predictive power per locus.

No single variant will explain a behavior. The architecture rules it out. Claims of the form "the gene for X" are not premature in this area; they are the wrong shape.

The exceptions are informative precisely because they are exceptions. Single variants of large effect do exist in dogs, and they are concentrated in physiology and morphology rather than behavior — drug metabolism, coat pattern, skeletal structure. Where a behavioral consequence follows from one of those, it usually follows through the body: a conformation that impairs breathing or movement changes what a dog does (brachycephaly), and that is a different causal route from a variant acting on a behavioral tendency.

5.5 The One Consistent Pattern

Across studies with different designs, samples, and methods, the genes implicated are disproportionately neurodevelopmental and brain-expressed. MacLean reported enrichment for neurobiological functions and developmental processes; Dutrow reported convergence in neurodevelopmental co-expression networks; Sarviaho's candidates sit in neurotransmission pathways.

That is the most robust conclusion available: selection on dog behavior has largely been selection on brain development. It is also, notably, a conclusion about mechanism class rather than about any individual gene.

6. Genes and Environment Interact

6.1 The Study That Shows It Directly

Gene–environment interaction is asserted constantly and demonstrated rarely. Espinosa, Zapata, Alvarez, Serpell, Kukekova, and Hecht provide a direct demonstration in dogs.

They collected life-history and environment reports plus C-BARQ behavioral ratings from guardians of 4,497 dogs. Seven types of early adversity were measured: physical punishment, separation from the primary caregiver, physical abuse, attacks by animals, intense fright involving people, serious physical injury, and being chained or tethered outdoors.

Adverse experiences in the first six months of life were significantly associated with increased aggression and fearfulness in adulthood, holding acquisition source, sex, and neuter status constant. And — the finding that makes this an interaction study rather than an environment study — the effects of adversity on fearful and aggressive behavior varied systematically at the breed level.

The named examples are instructive. For aggression, American Eskimo Dogs, American Leopard Hounds, and Siberian Huskies showed markedly larger differences between dogs with and without early adversity, while Golden Retrievers and Labrador Retrievers showed minimal differences. For fear, Airedale Terriers, American Eskimo Dogs, and Golden Retrievers showed stronger associations between adversity and increased fear, while Labrador Retrievers showed very little difference.

Note that Golden Retrievers appear in the resilient group for aggression and the susceptible group for fear. Susceptibility is not a single dimension, and a breed is not simply robust or fragile.

6.2 What Interaction Means

Interaction means the effect of the environment depends on the genotype, and equivalently that the effect of the genotype depends on the environment. It is not the same as both mattering.

The practical translation: the same bad start does not do the same amount of damage to every dog, and the same good start does not confer the same benefit. This is why two littermates with different histories can diverge, and why two dogs with similar histories can end up in different places (temperament and coping styles, fear periods).

It also means that population-level heritability figures are conditional on the range of environments sampled. Add an environment nobody in the sample experienced and the figure can move.

6.3 The Limits of This Design

Two caveats belong with the result. The adversity histories are retrospective owner reports, which is the only feasible method at this scale and is subject to recall and attribution effects — an owner who finds a dog fearful may reconstruct its history differently. And breed-level moderation is a signal that something heritable is involved, not an identification of what; it does not point at a gene, a pathway, or a mechanism.

7. Commercial Genetic Testing for Behavior

7.1 What Research-Grade Prediction Currently Achieves

Zapata, Lilly, Herron, Serpell, and Alvarez tested how far individual-level genotype can predict behavioral diagnoses. Working with 397 dogs — pedigree and mixed-breed, of which 122 carried veterinary behavioral diagnoses and 26 were medicated for behavioral problems — they genotyped 20 markers at 13 behavioral loci.

Eight loci were associated with veterinary behavioral diagnoses, and logistic regression models using marker subsets predicted diagnoses, with five markers on chromosomes 10, 13, and X contributing. The authors characterize the effect sizes as moderate-to-large relative to human complex disease, and describe the work explicitly as exploratory and in need of confirmatory studies with greater statistical power.

Two incidental findings are worth carrying: small body size was associated with problem behaviors and large body size with greater trainability; and Pit Bull-type dogs showed no unique behavioral genetic signature, differing only in increased leash-pulling.

7.2 The Gap Between That and a Consumer Product

The research above represents something close to the current ceiling: an exploratory model, in fewer than 400 dogs, using a small panel, predicting a clinical diagnosis at moderate effect size, and flagged by its own authors as requiring confirmation.

Anything offered commercially as a behavioral prediction from a cheek swab is claiming to be well past that ceiling. A breed-ancestry result is a different matter and is technically sound; what it then licenses about behavior is bounded by the 9% figure. And single-gene tests with genuine clinical value exist for specific pharmacological and health traits — which is a different kind of claim altogether, resting on a known variant with a known mechanism (MDR1 and drug sensitivity, the merle locus).

The distinction to hold is between a variant of large effect on a physiological process and a polygenic behavioral tendency. The first can be tested for usefully. The second currently cannot be, at the individual level, to any standard that would justify a decision about a dog.

8. The Measurement Problem Underneath All of It

8.1 Most of This Literature Is Owner Reports

The heritability figures above rest overwhelmingly on owner questionnaires, principally the C-BARQ. That instrument carries the among-breed estimates, the within-breed Labrador estimates, the 13,715-dog prevalence survey, the 18,385-dog genomics study, and the adversity-interaction study.

Owner questionnaires are not a weak method by default. They aggregate observation across months and across situations a test cannot reproduce, they scale to sample sizes no laboratory can reach, and several of them have been through psychometric development. But what they measure is an owner's description of a dog, and any variance analysis built on them partitions variation in descriptions as well as variation in dogs.

That has a specific consequence for genetics. If owners of a given breed share expectations about how that breed behaves, some of the apparent breed difference is difference in reporting. Nothing in the designs above separates those two, and it is one plausible contributor to breed effects looking larger in questionnaire data than in tests.

8.2 Where Tests and Questionnaires Diverge

The one large study here that used behavioral testing rather than owner report found breed differences in social cognition, inhibitory control, problem-solving, persistence, activity, and exploration — and no breed differences in short-term memory or inference-based reasoning. The among-breed heritability estimates from cognitive testing similarly put memory at the bottom, at h² = 0.17.

Those two results agree with each other, which is reassuring. What they do not do is map cleanly onto questionnaire-derived trait structures, because the constructs are not the same: "trainability" as an owner rates it and inhibitory control as a cylinder task measures it are related but distinct things, and only one of them is behavior observed under controlled conditions (operationalizing behavior for measurement).

Lord, Chen, and Karlsson note that the historical separation between biological and psychological approaches to animal behavior may have impeded progress in this field. The measurement gap is where that separation is most visible: the genetics has scaled on instruments the behavioral side would not treat as the primary measure.

9. What the Field Has Not Looked At

Lord, Chen, and Karlsson make a structural point in their review that reframes everything above: canine behavioral genetics has included almost exclusively dogs from modern breeds, who represent just a small fraction of all dog diversity.

Most dogs in the world are not pedigree pets in Western households. Free-ranging and village dog populations, which make up the majority globally and sit closer to the ancestral condition, are barely represented in the behavioral genetics literature (free-ranging dog social behavior, domestication).

Their recommendation is to stop at correlations less often and move toward understanding the interactions between genes and environment that produce behavior. That is a description of a field that knows its own main limitation.

10. What Follows for Practice

10.1 Treat Breed as a Weak Prior

Breed shifts expectations and does not set them. It is most informative before a dog exists — in choosing a breed for a purpose, where average differences across a population are exactly the relevant thing. It is least informative once a specific dog is present, which is when it is most often invoked.

10.2 Litter and Parents Beat Breed

Within-breed heritabilities in the range of 0.2 to 0.4 for some traits mean that parents carry information that breed does not. For a working or assistance prospect, behavior in the parents and in prior litters from the same pairing is a better guide than the breed standard — with the caveat that puppy tests predict adult behavior weakly (puppy temperament tests).

10.3 Heritable Does Not Mean Fixed

This is the inference to guard against hardest, because it is the one that leads to giving up on a dog. Point-following in eight-week-old puppies was estimated at 43% heritable and is also trainable. Noise fear has a moderate within-breed heritability and responds to systematic desensitization (desensitization and counterconditioning). Heritability describes where variation came from, not what an intervention can do.

10.4 Expect Different Dogs to Need Different Amounts

The interaction finding has a direct consequence for planning: the same protocol will not produce the same change in every dog, and that is not a failure of the protocol or of the handler. Some dogs are more affected by a difficult history and some are more affected by a good one.

10.5 Early Experience Is Where the Leverage Is

Adversity in the first six months was associated with increased aggression and fearfulness in adulthood across 4,497 dogs, after controlling for several obvious confounds. Whatever the genetic contribution, the environmental contribution at that stage is substantial and is the part anyone can act on (the sensitive period, adolescence).

10.6 What to Say When Asked "Is It Genetic?"

The accurate answer is that the question does not have an individual-level answer. Genetic differences between dogs contribute to behavioral differences between dogs; for this dog, the useful questions are what it does, in what situations, what maintains it, and what changes it. Nothing in the genetics literature makes those questions less necessary, and none of it substitutes for them.

11. Summary at a Glance

Heritability is a property of populations. It is the share of variation in a group attributable to genetic variation in that group, not the share of a trait that is genetic in an individual.

The same trait has several heritabilities. Among breeds, within one breed, and within a tested cohort all give different figures, and none is the true one.

Among breeds, high. Across more than 14,000 dogs in 101 breeds with genotypes for 5,697, breed-average behavior showed high among-breed heritability for 14 traits, with 131 associated SNPs in brain-expressed, neurodevelopmental genes.

Within one breed, low to moderate. In Labradors, estimates ran from 0.38 for fetching down to 0.06 for separation anxiety and 0.03 for owner-directed aggression — the last two indistinguishable from zero.

Measured directly in puppies, it depends on the measure. In 375 eight-week-old retrievers: point-following 43%, attention to a human face 43%, marker sensitivity 14%, gaze during an unsolvable task 8%.

Among breeds, inhibitory control is the most heritable cognitive factor. h² = 0.70 across 36 breeds, and 0.50 after controlling for breed-average weight. Memory was lowest at 0.17.

Breed differences in behavior are large. Across 13,715 dogs in 264 breeds, prevalence differences were substantial on all seven anxiety traits — 10.6% versus 0.4% for stranger-directed aggression between two breeds.

And breed explains 9% of individual variation. From 18,385 surveys and 2,155 genotypes, with most traits heritable above 25% and behavioral loci not unusually differentiated between breeds.

Breed differences are not uniform across the mind. Across 13 breeds, differences appeared in social cognition, inhibitory control and problem-solving, but not in short-term memory or inference-based reasoning.

Diversification was mostly regulatory. Across more than 4,000 canids and 46,000 surveys, breed diversification was driven predominantly by non-coding variation, with lineage genes converging on neurodevelopmental networks and sheepdogs enriched for axon guidance functions.

Gene–environment interaction is demonstrated, not assumed. In 4,497 dogs, early adversity was associated with adult aggression and fear, and the size of that association varied systematically by breed.

Behavior is polygenic, so the gene lists are short by construction. Eleven significant loci from 2,155 genotyped dogs means eleven were large enough to detect at that sample size, not that eleven genes matter.

Almost the whole literature rests on owner questionnaires. Variance partitioned from owner reports is partly variance in how owners describe dogs, and no design here separates the two.

Pit Bull-type dogs showed no unique behavioral genetic signature. In one exploratory study of 397 dogs the only difference identified was increased leash-pulling.

Individual behavioral prediction from genotype is exploratory. The most direct attempt used 397 dogs and 20 markers, and its authors call for confirmatory studies.

12. Research Gaps

The sampled population is a small slice of dogs. Modern Western breeds dominate the literature; free-ranging and village dogs, the global majority, are largely absent.

Replication is thin. Single-breed GWAS findings at sample sizes in the low hundreds have not generally been reconfirmed in independent cohorts, and the locus lists from large studies overlap less than one would hope.

Almost everything rests on owner questionnaires. C-BARQ and comparable instruments carry the heritability literature. Where behavioral testing has been used instead, the picture changes in places — which means some of the variation being partitioned is variation in how owners describe dogs.

No mechanism has been traced end to end. The field has loci, enrichment patterns, and a convergence on neurodevelopment. It does not have a single worked path from a variant through a developmental process to a behavior in dogs.

Interaction has been shown at breed level only. That early adversity has different consequences in different breeds is established; which genetic factors moderate it is not.

Maternal and litter effects are poorly separated. Prenatal and early maternal environment can load onto genetic terms, and the designs that would separate them cleanly in dogs are rare.

13. Conclusion

Behavior in dogs is genuinely heritable, the implicated genes are consistently neurodevelopmental, and breed averages differ substantially and in ways that track selection history — weakly, but detectably. None of that is in serious dispute.

What the same literature establishes just as firmly is that this buys very little predictive power over an individual animal. Breed accounts for roughly a tenth of behavioral variation between dogs. Within a breed, the traits people most want explained show the weakest genetic signal of all. The gene searches have produced short lists of modest-effect loci rather than switches. And the effect of a dog's history depends on what the dog brought to it, which means neither term can be read off without the other.

The honest position is therefore not "it's genetic" or "it's how you raise them," and not "a bit of both" either, which sounds balanced and predicts nothing. It is that genetic variation contributes to behavioral variation across populations of dogs, that this fact licenses almost no inference about the dog in front of you, and that the work of finding out what that dog does and what changes it is unaffected by anything genetics has so far discovered.

Key Insights

"Heritable" and "predictable from breed" are different claims. One dataset found most behavioral traits heritable above 25% and breed explaining 9% of individual variation. Both numbers are correct.

Ask at what level a heritability figure was estimated. Among breeds, within a breed, or within a tested cohort — the same trait gives very different answers, and a number quoted without its level says almost nothing.

The traits owners most want explained have the weakest genetic signal. Within Labradors, separation anxiety came out at 0.06 and owner-directed aggression at 0.03, neither distinguishable from zero.

Selection acted on brain development, not on behaviors. Non-coding regulatory variation, neurodevelopmental co-expression networks, axon guidance in sheepdogs. There is no gene for herding.

Breed is most useful before you have a dog. For choosing among populations it is real information. For describing the animal in the room it is close to noise.

Parents tell you more than the breed standard does. Within-breed heritabilities around 0.2 to 0.4 for some traits mean the pairing carries information the breed does not.

Heritable does not mean unchangeable. Point-following was estimated at 43% heritable in eight-week-old puppies and is trainable. Heritability describes the origin of variation, not the ceiling of intervention.

The same start does not have the same effect on every dog. Early adversity was associated with adult fear and aggression, and how strongly varied by breed — and a breed resilient on one measure was susceptible on another.

Behavioral gene tests are not ready. The most direct research attempt used 397 dogs and 20 markers and called itself exploratory. Ancestry testing is sound; what it tells you about behavior is bounded by 9%.

Most dogs in the world have not been studied. The genetics of dog behavior is, so far, the genetics of modern Western breeds.

References

Bray, E. E., Gnanadesikan, G. E., Horschler, D. J., Levy, K. M., Kennedy, B. S., Famula, T. R., & MacLean, E. L. (2021). Early-emerging and highly heritable sensitivity to human communication in dogs. Current Biology, 31(14), 3132–3136.e5. https://doi.org/10.1016/j.cub.2021.04.055

Dutrow, E. V., Serpell, J. A., & Ostrander, E. A. (2022). Domestic dog lineages reveal genetic drivers of behavioral diversification. Cell, 185(25), 4737–4755.e18. https://doi.org/10.1016/j.cell.2022.11.003

Espinosa, J., Zapata, I., Alvarez, C. E., Serpell, J. A., Kukekova, A. V., & Hecht, E. E. (2025). Influence of early life adversity and breed on aggression and fear in dogs. Scientific Reports, 15, 32590. https://doi.org/10.1038/s41598-025-18226-0

Gnanadesikan, G. E., Hare, B., Snyder-Mackler, N., & MacLean, E. L. (2020). Estimating the heritability of cognitive traits across dog breeds reveals highly heritable inhibitory control and communication factors. Animal Cognition, 23(5), 953–964. https://doi.org/10.1007/s10071-020-01400-4

Ilska, J., Haskell, M. J., Blott, S. C., Sánchez-Molano, E., Polgar, Z., Lofgren, S. E., Clements, D., & Wiener, P. (2017). Genetic characterization of dog personality traits. Genetics, 206(2), 1101–1111. https://doi.org/10.1534/genetics.116.192674

Junttila, S., Valros, A., Mäki, K., Väätäjä, H., Reunanen, E., & Tiira, K. (2022). Breed differences in social cognition, inhibitory control, and spatial problem-solving ability in the domestic dog (Canis familiaris). Scientific Reports, 12, 22529. https://doi.org/10.1038/s41598-022-26991-5

Lord, K. A., Chen, F. L., & Karlsson, E. K. (2025). An evolutionary perspective on dog behavioral genetics. Annual Review of Animal Biosciences, 13, 167–188. https://doi.org/10.1146/annurev-animal-111523-101954

MacLean, E. L., Snyder-Mackler, N., vonHoldt, B. M., & Serpell, J. A. (2019). Highly heritable and functionally relevant breed differences in dog behaviour. Proceedings of the Royal Society B: Biological Sciences, 286(1912), 20190716. https://doi.org/10.1098/rspb.2019.0716

Morrill, K., Hekman, J., Li, X., McClure, J., Logan, B., Goodman, L., Gao, M., Dong, Y., Alonso, M., Carmichael, E., Snyder-Mackler, N., Alonso, J., Noh, H. J., Johnson, J., Koltookian, M., Lieu, C., Megquier, K., Swofford, R., Turner-Maier, J., White, M. E., Weng, Z., Colubri, A., Genereux, D. P., Lord, K. A., & Karlsson, E. K. (2022). Ancestry-inclusive dog genomics challenges popular breed stereotypes. Science, 376(6592), eabk0639. https://doi.org/10.1126/science.abk0639

Salonen, M., Sulkama, S., Mikkola, S., Puurunen, J., Hakanen, E., Tiira, K., Araujo, C., & Lohi, H. (2020). Prevalence, comorbidity, and breed differences in canine anxiety in 13,700 Finnish pet dogs. Scientific Reports, 10, 2962. https://doi.org/10.1038/s41598-020-59837-z

Sarviaho, R., Hakosalo, O., Tiira, K., Sulkama, S., Salmela, E., Hytönen, M. K., Sillanpää, M. J., & Lohi, H. (2019). Two novel genomic regions associated with fearfulness in dogs overlap human neuropsychiatric loci. Translational Psychiatry, 9, 18. https://doi.org/10.1038/s41398-018-0361-x

Zapata, I., Lilly, M. L., Herron, M. E., Serpell, J. A., & Alvarez, C. E. (2022). Genetic testing of dogs predicts problem behaviors in clinical and nonclinical samples. BMC Genomics, 23, 102. https://doi.org/10.1186/s12864-022-08351-9