On January 1, a research article titled “Highly Heritable and Functionally Relevant Breed Differences in Dog Behavior” became publicly available as a preprint. The authors were Evan MacLean (University of Arizona), Noah Snyder-Mackler (University of Washington), Bridgett vonHoldt (Princeton University), and James Serpell (University of Pennsylvania).
Within days, two popular press articles about it were published online—and they went viral. A link to one of those articles posted on the IAABC page had more than 30,000 views. The popular press articles raised a few questions and controversies, as evidenced by their titles: “Dog Breeds Really Do Have Distinct Personalities—and They’re Rooted in DNA,” written by Elizabeth Pennisi for Science magazine, and “Dog Breeds Don’t Have Distinct Personalities,” written by Marc Bekoff for Psychology Today.
Why would a scientific article about breed differences in behavior cause such a dust-up? To help clear the air, I went directly to the source. Noah Snyder-Mackler and Evan MacLean generously agreed to answer a few questions about the research and comment on some of the controversies raised by the popular press articles. In the sections below I briefly summarize (and grossly oversimplify!) a few highlights from the original paper and share the authors’ enlightening responses to my questions.
Breed differences in behavior
To look at differences in behavior across dogs, the researchers tapped into the huge data pool from the Canine Behavioral Assessment and Research Questionnaire (C-BARQ). Between 2005 and 2016, C-BARQ scores for nearly 30,000 dogs were collected by James Serpell, an author of the paper. From this pool, dogs were selected from breeds for which genetic data are also available, resulting in behavior profiles for a grand total of 14,000 dogs from 101 breeds.
To complete the C-BARQ, dog owners and handlers answer a battery of questions about their dog’s typical behavior in various situations. The responses are clustered into 14 different behavior categories, including:
- Stranger-directed aggression
- Owner-directed aggression
- Dog-directed aggression
- Dog rivalry
- Stranger-directed fear
- Nonsocial fear
- Dog-directed fear
- Separation-related behavior
- Attachment and attention-seeking
- Touch sensitivity
- Energy level
More information about the C-BARQ can be found online.
Q: A central controversy that has been raised about your research is the notion of “breed personality.” The original paper uses behavioral data from C-BARQ. Could you clarify what C-BARQ actually measures, as well as the notion of “breed personality”?
A: Noah: We never once mention the word “personality,” and I usually avoid using that work for any nonhuman animals. C-BARQ measures behavioral phenotypes*— as reported by the owners. Dogs may have personalities, but I’m not in the habit of quantifying personalities. I would rather use a much more objective behavioral measure.
*Phenotype refers to the observable expression of a trait.
Q: In the Psychology Today article about your study, Adam Miklósi is quoted as stating: “Dog breeds do not have personalities… this link will cause more harm than gains.” Does your study make conclusions about “breed personality,” and what harm do you think Miklósi is referring to?
A: Noah: We never mentioned or hinted at the word personality, so I assume this is referring to the popular press article (which we were not allowed to read before it was posted). I imagine the harm Adam is referring to has to do with the popular press articles, not our study.
A: Evan: I’d add to this that one concern might be attributing personality to a breed rather than an individual. The concept of personality itself is about individual differences and individual traits, not group differences, so to talk about breed personalities (regardless of what you think about the term) is already a bit of a misuse of the concept. As Noah notes, our paper makes conclusions about broad patterns of behavior across breeds, and this is different than personality.
Breed differences in genetics
Every dog has a unique genetic profile, and the genetic similarities between any two individuals might be small or large. It’s logical that dogs belonging to the same breed would have more similar genetic makeup than dogs from different breeds, but how is this actually measured?
One way scientists measure genetic variation between dogs (or groups of dogs) is to use genetic markers called SNPs (pronounced “snips”). Dogs having a greater number of identical SNPs are more genetically similar. SNP stands for “single nucleotide polymorphism.” This tongue-twister actually refers to something rather simple. Nucleotides are the building blocks of DNA and a SNP is a region of DNA where one small (but potentially important!) building block differs across individuals, so that region might also play a role in traits that differ across individuals.
The genetic information used in this research came from two of the largest dog genetic studies completed to date, each of which included thousands of dogs and hundreds of SNP genetic markers. One study looked at genetic markers for physical traits and diseases in dogs, including dysplasia, epilepsy, and lymphoma (Hayward, 2016). The second study used the genetic information to create a dog family tree, showing relationships between breeds, and their ancestral origins and migration patterns (Parker, 2017).
How does the health and family tree research relate to behavior? As we were taught in grade school biology, traits are influenced by both genes and environment—and behavioral traits are no exception. To get a handle on the relative influence of genes and environment on dog behavior, Evan MacLean, Noah Snyder-Mackler, and their colleagues combined the behavior scores from C-BARQ with information from the two genetic studies and—after applying complex and sophisticated statistical analyses—they found that some differences in behavior across breeds are strongly linked to genetic differences. They also discovered that canine behavior traits are polygenic, meaning that a behavior like “aggression toward strangers” results from the combined effect of several genes.
Behaviors with the highest degree of “heritability” were trainability, stranger-directed aggression, chasing, and attachment and attention seeking. The authors noted that these behaviors have been targeted by breeders over the centuries to improve dogs’ skills at certain jobs, like guarding, hunting, and lap-sitting.
Q: What does heritability mean? And more specifically, what does the .51 heritability reported in your study imply? How does this value compare to heritability estimates for human behavioral traits?
A: Noah: Heritability, in the narrow sense, means the proportion of variation in a specific trait that is explained by genetic similarity among individuals that you measured. The total variation in that trait is equal to Vg (genetic variation) + Ve (environmental variation). So heritability is calculated as Vg/(Vg+Ve). A heritability of 0.51 means that just over 50% of the total variation in a given behavior can be explained by the fact that some animals (or breeds) are more genetically similar to one another. One important thing about any measure of heritability is that it is all relative to the population that you are measuring. So, if we were measuring a population with less variation in a trait (or more environmental variation) then we will likely find different numbers. Compared to humans, these numbers are lower than the heritability of some morphological traits, like height, but higher than that of many behavioral traits (e.g., anxiety).
Q: Why does your study report higher heritability than other research? What do you think accounts for the difference, and in your view what concerns, if any, does the inconsistency raise?
A: Noah: Other studies were mostly conducted within a single breed, so if there is less genetic variation in those individuals (since breeds are not genetically diverse) then you won’t be able to explain much variation in the behavior by genetics alone. If you think about the heritability equation above: if there isn’t any genetic variation, then all the variation in your trait is environmental.
Q: Were there any behaviors for which you expected higher heritability due to past selective breeding in dogs on that trait, but that weren’t found? Given that dog breeds are rarely used for their original purposes, is it possible that a century of relaxed artificial selection on behavioral traits has reduced their heritability?
A: Evan: This is a tricky one. Relaxed selection wouldn’t necessarily reduce heritability, because heritability is just a measure of how much genetic factors explain phenotypic variation. In cases of extremely strong selection, traits can become fixed, in which case there is no heritability (because there is no variance to explain). This is a distinction between heritability and heredity. For example, humans have two legs, which is an entirely hereditary trait, but there is no heritability to speak of because there is no variation to explain. So, if a trait has a genetic basis, and there is relaxed selection allowing for phenotypic variance, we actually have a very good shot at detecting its heritable nature under these conditions! But to answer your question in short, we didn’t identify any traits for which the heritability estimates were surprisingly low.
Q: Your study found that the behavioral traits were polygenic, which was a very interesting result. Could you explain what this means and how you arrived at this conclusion?
A: Noah: We found that most SNPs only explained a small amount of the variance in any given trait. What this means is that many SNPs have small effects and, when added together, can have larger effects. Most complex quantitative traits (including height and behaviors) are polygenic in nature, with single SNPs having only very small effects on the trait. This is in contrast to Mendelian traits, where having a specific dominant genetic variant determines whether or not you exhibit the trait.
A: Evan: I’ll add that it also contrasts with what people find for morphological traits in dogs where a single genetic variant can explain 50% of variation in body size, for example. In humans, height is much more complex and driven by many variants of small effect. So, some have argued that dogs are a relatively simple model where we can often find one or two genetic variants with a large effect. We find this is generally not true for behavior, though.
Q: It was noted in both popular science articles that “correlation does not equal causation,” and the original paper acknowledges that the independent data sets limit the conclusions. What are the limitations of your findings? Can you tell us about any future plans you have to look experimentally at behavior and genes using the same dogs?
A: Noah: We don’t have paired genetic and behavioral data from the same dogs, which would be the gold standard study and allow us to more precisely estimate heritability and identify some more SNPs. We don’t have any studies planned at the moment, but I know that other labs are working on it, and I look forward to seeing what they find!
Linking the behavioral and genetic information
Behavior varies quite a bit from one dog to another. For example, some dogs are extremely fearful of novel objects whereas others are curious; some dogs are gregarious and friendly to strangers whereas others are wary or aggressive. If some of this variation is due to breed differences in behavior, then two patterns should hold up. First, dogs belonging to the same breed should behave more similarly, on average, than dogs belonging to different breeds. Second, a dog “family tree” based on behavioral traits should closely match up with a “family tree” based on overall genetic similarity across breeds. This is exactly what was found when the behavioral and genetic data sets were combined, leading to the conclusion that breed differences in behavior have a significant genetic basis.
Q: The genetic analyses in the original paper are complex and statistically sophisticated. Many readers would like to understand a little more about the techniques you used. How would you summarize the methods in a way that most people could follow without a background in genetics or statistics?
A: Noah: We used one mixed modeling framework to address two questions: 1) How heritable is each trait across breeds, and 2) What specific genetic changes (single nucleotide polymorphisms; SNPs) are most strongly associated with each trait? Both of these things can be estimated simultaneously in the same model, using what is called the “mixed model for association.” To estimate heritability, we essentially compare how similar dogs are genetically (measured as what proportion of the ~100,000 SNPs that were measured are exactly the same between two breeds) and how behaviorally similar breeds are. To identify strongly associated SNPs, you can see if having 0, 1, or 2 copies of a given SNP is correlated with behaviors. Since we didn’t have genotype data and behavioral data from the exact same dogs, we had to calculate the proportion of dogs that had 0, 1, or 2 copies of each SNP in each breed. (When divided by the total number of alleles, this is called the “allele frequency.”)
Q: As you state, heritability estimates can be affected by environmental homogeneity. Despite the very large sample size, it seems that breed stereotypes might result in dogs of the same breed (or breed group) being trained and treated similarly across households and geographical regions. How might this affect your results? Could it possibly create greater environmental homogeneity within and heterogeneity between breeds?
A: Noah: In short, yes. This is entirely possible. Some breeds may be treated differently, which would be a gene by environment correlation. If this is true, then it would suggest that our estimates might be slightly inflated relative to the true heritability of these traits. But I doubt that this gene environment correlation would explain that much of the variation in traits. Especially because we used such a large set of breeds that exhibit some shared and some divergent behaviors across the dog family tree.
A: Evan: To add here, you could make an analogy to height in men and women. Are men taller than women on average? Yes. If you study thousands of men and women will you be able to detect this difference? Yes again. Does this mean than all men are taller than women? Of course not! It’s entirely possible to have statistical and biologically meaningful differences at the population level, while having room for enormous levels of overlapping variation in the population.
Looking at the brain
A compelling finding from this research was that some of the genetic variants—SNPs—that distinguished one breed from another could be traced to specific brain processes. For example, the researchers found that breed differences in trainability were associated with several genes that play a role in human intelligence and information processing, and breed differences in aggression were associated with several genes that play a role in human aggression.
Q: Your original paper looked closely at the link between genes and their expression in the brain. One finding in particular that grabbed my attention was the gene associated with breed differences in aggression, PDE7B, and its possible role in domestication. Can you summarize how you looked at gene expression and elaborate on the link between this gene and domestication?
A: Evan: We didn’t directly measure gene expression in this study, but rather took the genes we identified in GWAS*, and looked at other datasets showing where in the body they tend to be highly expressed. We identified PDE7B through the GWAS, and other data from dogs show that it is highly expressed in the brain. Given that a reduction in aggression was an important phenotypic change in domestication, it’s possible this gene played a role, and other studies have identified it as under selection in dogs.
*GWAS means genome-wide association studies. Scientists can identify the genes involved in disease by looking at SNPs that occur more frequently in people with a given disease than in people without the disease.
Q: Your paper states, “More so than most model organisms, dogs exhibit a suite of cognitive and behavioral traits that make them a unique model for complex aspects of human social behavior and cognition (11,38). These similarities are hypothesized to result from convergent evolution, due to similar selective pressures in human evolution and dog domestication.” The hypothesis of convergent evolution between domestic dogs and people is intriguing. What are some alternative explanations for the similarities? For example, is it possible that some of the genes shared by dogs and people are more broadly conserved, e.g., mammalian?
A: Evan: Yes, it’s certainly possible, in fact I’d say likely, that many of these genes have similar functions across diverse groups of animals. So, it’s not that dogs and humans have special genes that other organisms don’t, but rather that we have certain behavioral and psychological processes that show interesting similarities (between humans and dogs). So, if we can understand genetic variants that contribute to those behavioral/cognitive traits in dogs, this may inform our study of the same traits in humans (and vice versa).
A new client shared the results of her dog’s DNA breed identification test, hoping it might help solve the puzzle about why he is so reactive toward other dogs. The genetic test results revealed that her dog is part Border Collie, Shih Tzu, Staffordshire Terrier, Chihuahua, and Miniature Pinscher. I just threw up my hands and laughed!
Like my client, some readers may have swabbed their dog’s cheek and mailed it to a genetic laboratory to test for breed identification, with the belief that the information will provide insights into behavior. In actuality, the research by Evan MacLean, Noah Snyder-Mackler, and their colleagues is one of only a few studies that have looked at the heritability of breed differences in canine behavior, and it is uniquely powerful by having included information from thousands of dogs representing more than one hundred different breeds.
Q: What are the practical benefits of knowing that genetics, and in particular which genes, may be involved in dog-dog aggression and other behavioral traits?
A: Evan: At a basic research level, a lot of this is just to better understand the biological bases of behavior, and how various genetic variants may contribute to variation in behavior. At an applied level, if we better understand these biological pathways, it informs how we might think about developing therapeutics to address various behavioral problems.
Q: Why do you think a bias exists against the recognition of genetically based breed differences in behavior and cognition, and acceptance of genetically based breed differences in health and morphology?
A: Evan: Some people worry (rightly!) that stereotypes in this area can lead to problematic policies, such as breed specific legislation. We of course see that there are tremendous individual differences within breeds as well, and that’s something that resonates with people intuitively. In general, I think there is also more potential for misuse/sloppy interpretation when you are talking about cognitive and behavioral traits, whereas this doesn’t seem to be as big an issue for morphological or disease-related traits.
Q: A concern voiced about your article was that it was available as a preprint. Could you explain the scientific publication process and your reasons for taking this unconventional step?
A: Noah: The scientific review process can take a really long time. Sometimes it can be years before a publication see the light of day. Preprints are increasingly common in many fields—especially genetics and genomics—and there was a recent study showing that most preprints were published in peer reviewed journals within six months (read more here). This just shows that the preprints are thoughtfully posted by their authors (i.e., are “publication ready” and usually posted at the same time as being submitted to a journal). It also allows for the science to reach the community so they can read it and give feedback sooner. Evan and I posted this to the preprint server bioRxiv and submitted it to a journal for review simultaneously. It’s currently undergoing peer review.
Q: Would you like to share anything else about the study or the popular science articles?
A: Noah: I would encourage critical readers to read our manuscript and not to make judgements based solely on the popular science articles.
A: Evan: I’ll gladly second this!