Genomic ancestry, behavior, and the domestication of the dog

It has been exactly 25 years since the first dog genetic maps were published and 20 years since the first dog genome sequence, a boxer, was released. The domestication of dogs was a landmark in human history, and the details of that process are key to understanding fundamental principles of evolution, domestication, and behavior. Dogs also provide the foundation for studies aimed at using canine phenotypes for studies of human health. While many millions of dogs exist as more than 350 recognized breeds, there are also an untold number of nonbreed, globally distributed village dog populations that are not under intense human selection.

This special feature comprises a total of eight articles, more than half of which explore the extent and effects of genomic introgression between wolves and dogs, offering new insights and intriguing directions for future research. In addition, four studies investigate the genetic basis for different dog behaviors. The studies included in this special feature address a kaleidoscope of important issues in the evolution of dogs, the development of breeds, and the challenges faced in exploiting the novelties of canines. Each article is intended to be a launching point for discussion and experimentation. The issue highlights the convergence of multiple fields, will likely generate some controversy, and presents data that will advance the study of canines.

Dog–Wolf Hybridization

The development of rapid, economical, and longer-read sequencing has revolutionized biology by generating millions of whole genomes from a broad diversity of organisms. One of the most significant insights from the analysis of these data has been the realization that gene flow between populations and species is significantly more common than previously believed. Early studies of Anopheles mosquitoes (1) revealed that species were able to maintain their phenotypic integrity despite exchanging up to 60% of their genomes. Gene flow has proven so common in studies across the Tree of Life that species boundaries are typically considered semipermeable (2).

The generation of genome-wide data from ancient populations, and even extinct species, has also shown that many modern lineages possess a hybrid origin and are best reflected as a reticulation on an evolutionary network. For example, the butterfly Heliconius elevatus is a hybrid species, with 99% of its genome replaced by introgression from a close relative, Hyperolius pardalis, and the remaining 1% from a second species (3). A hybrid origin has also been proposed for Columbian mammoths, which may have resulted from an equal contribution of two distinct Middle Pleistocene mammoth lineages that independently gave rise to the Eurasian and North American mammoth lineages (4).

Genomic studies of modern and ancient domestic animals have also uncovered an intriguing pattern in the sources and direction of introgressive gene flow. Virtually all populations of domestic animals possess introgressed genetic variation from wild populations from which the domesticates were not derived, and even species closely related to their immediate ancestors. The frequency and extent of genetic replacement, however, vary considerably across domestic taxa (5). For example, domestic pigs derived from Near Eastern wild boar populations experienced a near-complete genomic turnover following 3,000 years of introgression with genomically distinct European wild boar after the first domestic pigs were introduced to Europe alongside Neolithic farmers (6). Dogs are positioned at the opposite end of the spectrum from pigs. Despite frequent opportunities for gene flow given the distribution of wolves across Eurasia and North America, studies of ancient and modern dogs have generally failed to uncover evidence of gene flow from wolves into dogs above the 1 to 2% minimum level of detection (7).

One exception to this rule is deliberate crossings of dogs and wolves, including breeds such as Sarloos and Czechoslovakian Wolfdogs that were developed in the 20th century. Wolf ancestry within all other modern breeds is negligible to nonexistent, even in cases where disputed claims have been made that wolves were included during the initial development of, for example, the German Shepherd. Scarsbrook et al. (8) directly assessed this hypothesis by analyzing nine genomes derived from museum specimens of German Shepherds accessioned between 1906 and 1993. The first individual in this sequence, named Lord Quirin, possessed ~10% of his nuclear ancestry and his mitochondrial genome from a Eurasian wolf. This result should perhaps not be a surprise since one of the putative parents was named Wolfi von Wolfnest. Despite this early evidence for the use of wolves in the development of German Shepherds, the lack of detectable wolf ancestry in any other historic or modern German Shepherd populations suggests that dogs possessing the wolf ancestry were deliberately excluded from further breeding programs.

Imputation of genotypes from low-coverage shotgun sequencing data has become an increasingly common technique in human population studies, including the incorporation of ancient genomes. One lingering concern about imputation is how well it will work in species where many fewer genomes are available. Bougiouri et al. (9) showed that genotypes can be imputed from sequence coverage as low as 0.5×. The authors applied this approach to a large dataset of ancient and modern wolves and dogs, assessing changes in inbreeding over time. One notable observation was the depletion of Runs of Homozygosity in genomic regions enriched with immunity and olfactory receptor genes, implying selective retention of adaptive variation in loci crucial for survival. These new approaches presage a future where large populations of modern and ancient canids can be reliably genotyped and assessed for signatures of selection and/or admixture.

Accurate detection of gene flow is predicated upon the availability of unadmixed reference sequences. For example, the presence of Neanderthal and Denisovan ancestry in modern humans was impossible to detect until high-coverage Neanderthal and Denisovan genome sequences became available (10). These same issues confound the detection of specific wolf or dog population ancestry in modern populations. In Australia, for example, controversy remains surrounding the degree to which European dogs (first introduced in 1788) may have hybridized with dingo populations that were present in Australia by 3,500 years ago (11).

The only way to unambiguously test this hypothesis is to generate whole genomes derived from directly radiocarbon-dated, precontact dingoes that can be used as reference sequences contrasted against modern, potentially admixed populations. Doing so allowed Scarsbrook et al. (12) to demonstrate that European dogs were admixed with dingoes almost immediately upon their introduction. The extent of this gene flow varies spatially across Australian dingo populations. Interestingly, this admixture may have had a positive effect by offsetting the deleterious effects of inbreeding by increasing genomic diversity in some dingo populations. The authors also speculate that a portion of the European introgression in dingoes may be adaptive, similar to the finding that domestic cat introgression may be providing benefits to Scottish wildcat populations (13).

Even armed with high-coverage modern and ancient genomes, the standard detection threshold for genetic admixture, using D and F₄ statistics, has been ~1 to 2%. Lin et al. (14) employed ~2,000 dog and wolf genomes and phylogenomic local ancestry inference methods (15) to detect admixture proportions less than 0.1%. Their results demonstrated that 100% of 280 village dogs and ~80% of breed dogs possessed at least 250 kilobase pair (~0.005%) of wolf ancestry. The authors suggest that this signature could result from a long-term legacy of admixture from wolves into dogs that was undetectable using previous D and F₄ methodologies. The authors also ran simulations to ensure that the trace amounts of ancestry were not the result of false positives due to incomplete lineage sorting. The conclusion that the majority of the modern global dog population possesses a detectable level of wolf ancestry may spark controversy. If this approach and these findings withstand scrutiny, it will spur a new era in the detection of trace amounts of hybrid ancestry in scores of other domestic and wild taxa.

Finally, Girdland-Flink et al. (16) obtained low-coverage genome sequence data from two ancient canids found in archaeological contexts on a 2.5-square-kilometer island in the Baltic Sea. Their population genetic analyses demonstrated not only that their ancestry matched Eurasian wolves but that the maximum potential amount of dog ancestry in these individuals could not exceed ~4.4%. Given the island’s lack of a terrestrial connection to either Gotland or Mainland Sweden, the authors propose that the presence of these wolves on the island was the result of deliberate human translocation. Stable isotope analyses also suggested that the diet of both wolves was dominated by fish and marine mammals, which was likely mirrored by the diets of the people. The authors conclude that these lines of evidence suggest a close association of wolves with people and, more generally, that close relationships between people and canids do not necessarily imply that those canids are dogs.

The Genetic Basis of Dog Behavior

While there is still a great deal of controversy surrounding where, when, and how dogs evolved, they did so as a result of a shift in the relationship between humans and wolves (17, 18). The emergence of dogs within human environments makes them excellent models for studying how behavior contributes to evolution, and the specific genetic changes that underlie those behavioral shifts. The more recent advent of dogs that share our home environments also makes them an intriguing model system for the genetic underpinnings of human psychological disorders. Although datasets are far below the size required to determine the causal genetic underpinning of behaviors in dogs or humans, advances in genomic technologies and the advent of ever larger datasets, have allowed for the application of Genome Wide Association Studies (GWAS) of behavioral traits.

Alex et al. (19) explored whether regions associated with problem behaviors in dogs have also been found in association with human psychiatric phenotypes. The authors leverage 13 years of data collected by the Golden Retriever Lifetime Study (GRLS) (20) which includes 1,187 dogs. They perform the largest single-breed GWAS to date, based on behavioral traits captured in the Canine Behavioral Assessment and Research Questionnaire (21), and genotype data from leukocytes. The authors identified 12 significant associations for eight behavioral traits. They then performed a Phenome-Wide Association Study and found that seven of these associations were previously associated with human psychiatric traits. The authors suggest that their findings identify shared biological mechanisms underlying broader emotional states or behavioral regulation, rather than specific behaviors. Based on this, they conclude that candidate genes identified in this and similar studies could help prioritize genes for deeper studies that explore human psychiatric disorders. The authors also suggest the potential for genetic association studies of anomalous behavioral traits in dogs could form the basis of clinical treatments.

The GRLS study is certain to reignite the debate of single versus cross-breed studies. While association studies are necessary first steps in identifying genes, they generally lack the power to identify causal variants. By performing their analysis in a single breed, the authors avoid many of the population structure complexities associated with GWAS across breeds which often lead to spurious associations (22). Any associations found within a single breed, however, may only be causative within that breed or in the specific genetic background of that breed.

Lin et al. (14) conclude that small proportions of wolf ancestry are detectable in all village dogs as well as a large number of breeds and in the same study also investigate the connection between ancestry and behavior. They then tested whether these limited contributions of wolf ancestry could be associated with breed behavior stereotypes which they gleaned from kennel club descriptors and web queries. This analysis suggested that breeds with smaller proportions of wolf ancestry (≤500,000 sites) were described as “friendly, easy to train, and affectionate.” In contrast, breeds described more routinely with the descriptors “suspicious of strangers, reserved and territorial” possessed larger proportions (≥5 million genomic sites) of wolf ancestry. The authors concede that breed descriptors are not ideal for genetic studies since they are based on breed club assumptions and not on empirical measurements. They also point to the fact that previous studies (23) have shown (at a general level) that behavior cannot be predicated from breed type. As a result, Lin et al. call for follow-up studies in a wide variety of dogs to test the correlation between their demonstration of low-level wolf ancestry and both behavioral and morphological traits.

To test the robustness of previous studies that have used breed-average phenotypes (averaged measurements across a breed from dogs that have not been genotyped) to identify behavioral-genetic associations, Lord et al. (24) made use of data from Darwin’s Ark (23), a community science project that has generated individual owner-reported behavior assessments and genetic data from >3,000 single and mixed-breed dogs. Their GWAS did not corroborate any of the 135 candidate behavior SNPs that had previously been reported. Instead, they showed that the majority of previously reported behaviorally associated candidate SNPs are highly correlated with breed and breed ancestry and that, when a model did not account for adequate covariates, previous behavioral association appeared.

They did, however, find four significant associations for breed-defining aesthetic traits, including leg length and ear shape. Lord et al. (24) conclude that studies using breed-average behavioral phenotypes are likely to be confounded by an inability to control for population structure. While breed dogs have been important in advancing our understanding of canine genomics in general, by omitting data from dogs with multiple breeds or nonbreed ancestry, these studies lose a large portion of available dog genomic variation, which likely leads to the omission of key findings (25).

Car et al. (26) made use of nonbreed dogs to investigate one of the biggest behavioral shifts during the domestication process, the move away from monogamous mating strategies seen in all members of the genus Canis to polygamous mating systems suspected in dogs (18), though polygamy has not yet been confirmed using genomic data on a large scale across multiple free-living dog populations. To assess whether a shift to a primarily polygamous mating could have been an adaptation to the human-created environment, Car et al. (26) first characterized the mating system of free-living dogs by pairing genomic data (consisting of ~150,000 SNPs) derived from three free-living Moroccan dog populations, with four years of detailed behavioral observations.

A reconstruction of the genealogies of each population revealed a frequent occurrence of half-siblings and examples of multiple paternity within litters, both of which are consistent with a polygamous mating system. Simulations of genealogies based on the demographic characteristics, however, showed that the mating patterns were not random. Instead, repeated matings between partners and observational data on interactions between individuals showed that dogs have a preference for familiar mates. The authors conclude that free-living dogs are primarily polygamous. This result supports the hypothesis that a polygamous mating system may have been an adaptation to a shift toward more predictable and accessible food sources that reduce the need for paternal care within a monogamous pairing (18). Further, preference for familiar mates within this system may have reduced breeding with ancestral wolves outside of this niche, allowing for the persistence of selection within the new human niche.

Conclusion

In assembling this special feature, we asked ourselves whether the promise of the dog genome as a genetic system for studies of evolution, domestication, disease risk, and behavior had been fulfilled. At the onset, each of these lines of inquiry seemed like relatively straightforward topics. What we have learned in the ensuing years is that none of the above are easy topics, and as this issue highlights, the questions underlying each topic are far more complex than we could have imagined at the start. However, the two words that describe the canine research community are “persistent” and “collaborative.”

We have learned a great deal about the prevalence of admixture during canid evolution. Yet we still do not fully understand its role in the genetic architecture of diseases and traits and the limits of our ability to detect small traces of admixture within modern genomes. The dog has become a poster child for studies of domestication, but we are still working to understand where, when, how, and how many times the process took place. Canine disease genes, many of relevance for human health, have been found and, as promised, these studies remove many of the complexities associated with human disease mapping studies, such as locus heterogeneity. But we also have found that even within single breeds, disease phenotypes show striking variation, and a gold standard for assigning disease status is still needed in most breeds. Finally, as the papers here demonstrate, behavioral studies of dogs are very much a work in progress and significant controversy exists regarding approaches and results.

However, the collaborative nature of the community is also evident in the papers presented here. The development of a dataset for imputation of ancient genomes, the creation of the GRLS cohort, and the extraordinary number of genomes that had to be sequenced and are now public because of this issue make that clear. Dog domestication occurred at least 16,000 years ago, a fraction of an evolutionary second. We argue that the studies presented here, while highlighting the fact that a great deal remains to be done, also speaks to equally rapid advances by a collaborative and curious field that approaches questions with originality and rigor. We look forward to the next 25 years when our answers will present ever more questions. Please enjoy!