Sociality genes are associated with human-directed social behaviour in golden and Labrador retriever dogs

Introduction

Social behaviors are complex traits affected by environmental factors as well as by many genes, each with small effects. The genetics of complex traits such as social behavior is difficult to study in humans as it requires a standardized environment, dense genotyping as well as phenotyping of a large number of individuals. The dog (Canis familiaris), on the other hand, has a genome with large haplotype blocks and is thus more convenient for genetic mapping (Lindblad-Toh et al., 2005). In addition, their human-like social skills could make them a suitable model species for human social behavior and disorders.

Through sharing our ecological niche for thousands of years, dogs have developed social talents that, in some cases, have been reported to surpass the skills of our closest relative, the chimpanzee (Pan troglodytes), as well as their own wolf ancestor (Canis lupus) (Hare & Tomasello, 1999, 2005). For example, dogs are able to comprehend human ostensive cues and referential gestures such as pointing and gazing (Lakatos et al., 2012; Soproni et al., 2001). They are also able to communicate with humans through intentional communicative referential gestures involving both attention-seeking and directional-showing behavior (Marshall-Pescini et al., 2013; Miklosi et al., 2000; Passalacqua et al., 2011). Furthermore, dogs have been demonstrated to discriminate between human emotions when viewing facial expressions (Muller et al., 2015). Not even socialized wolves are as prone as dogs to seek human attention (Gacsi et al., 2009; Topal et al., 2005), or to communicate with humans through mutual gazing (Nagasawa et al., 2015). Different hypotheses for how these differences between the wolf and the dog evolved has been proposed. For example, Hare & Tomasello (2005) suggested that selection against fear and aggression toward humans also mediated the evolution of dogs’ social skills. More recently, however, Range & Viranyi (2015) showed that wolves are as attentive to both human and conspecific actions as dogs are. They propose the canine cooperation hypothesis, suggesting that wolf–wolf cooperation established the basis for the evolution of dog–human cooperation.

Human-directed contact-seeking behaviors have specifically been studied using a problem-solving paradigm. Hare & Tomasello (2005) and Miklósi et al. (2003) showed that dogs and socialized wolves differ in their human-directed contact-seeking behavior when faced with an unsolvable task. Whereas wolves were more oriented toward the task, dogs quickly gave up and turned to a nearby human in a help-seeking manner. Hence, behavioral tests involving unsolvable tasks that stimulate communication and attention-seeking toward humans can be used to study dog–human social interactions, including variation between and within breeds (Persson et al., 2015, 2017; Sundman et al., 2018).

When studying dogs’ human-directed social behavior it is important to take into account that dogs form strong attachment bonds with their owners (Topal et al., 1998). It has been suggested that owners function as a secure base influencing persistence in cognitive tasks, because the presence of the owner affects dogs’ duration of task manipulation (Horn, Huber & Range, 2013). Additionally, these authors found that dogs spent more time in proximity of their owner and the presence or absence of the owner affected dogs’ interactions toward the unfamiliar experimenter. It is therefore relevant to analyze social interactions toward an unfamiliar experimenter and the familiar owner separately if both are present during a cognitive task. Consistent with this, we previously found that dogs’ social behaviors directed at owners and those directed at unfamiliar experimenter were separated into different components in a principal component analysis (Sundman et al., 2018).

The dog is not only well-suited for studies on social behavior, the species is also well-suited for studies of the genetics of both simple and complex traits such as social behaviors. The structure of the dog genome with long regions of linkage disequilibrium is particularly suitable for genome-wide studies identifying candidate regions for traits of interest (Lindblad-Toh et al., 2005; Sutter et al., 2004). Regarding human-directed social behavior, heritability estimates show a significant genetic component underlying variation in behaviors as measured in an unsolvable problem task (Persson et al., 2015). Furthermore, genome-wide association studies have identified two candidate regions on chromosome 26 associated with human-directed social behavior in laboratory beagles, with single nucleotide polymorphisms (SNPs) located within the SEZ6L and ARVCF genes (Persson et al., 2016). Interestingly, these genes have previously been associated with human social disorders such as autism for SEZ6L (Chapman et al., 2015) and schizophrenia for ARVCF (Sim et al., 2012). Additionally, Persson et al. (2016) found that the SNP associated with the gene ARVCF is in linkage with three other genes of interest for dogs’ sociability, for example, the COMT gene that has previously been associated with mood regulation in humans (Qayyum et al., 2015).

These earlier findings suggest a possible cross-species genetic basis for social behavior shared between dogs and humans. However, further studies are necessary to verify that the same SNP markers are associated with human-directed contact seeking in other dog breeds. In the present study, we utilized behavioral data collected in two different experiments in two other breeds, each experiment using similar methods for studying dog–human interactions in a standardized behavioral test. In both breeds, we collected DNA and analyzed associations with the previously reported candidate SNPs. The aim of the study was thus to investigate whether there are associations between human-directed social behavior, as measured in an unsolvable problem task, and two candidate SNP markers in groups of golden and Labrador retrievers. It is possible that genetic variants affecting social behavior have been under selection during domestication and more recent breed formations. To provide some tentative data in relation to this, we genotyped wolves and included a dog breed that has undergone recent selection for its cooperative bond with humans, the Labrador retriever. The Labrador retriever has recently been split into a common and a hunting type differing in many aspects of social behavior (Sundman et al., 2016).

Materials and Methods

Unsolvable problem task

Testing was carried out at ten different locations in Sweden during the autumn of 2014 (Labrador retrievers) and autumn of 2015 (golden retrievers). Results from the behavior test have been previously published in Sundman et al. (2018) (Labrador retrievers) and in Persson et al. (2017) (golden retrievers). The unsolvable problem task was performed in the same way in both breeds. To have a uniform setting, testing took place in a 3 × 3 m marquee tent without the presence of any other dogs. The tent had three walls and no flooring. A mesh fence was placed at the open side to keep the dogs within the tent (testing area). An HD camcorder (Canon Legria HF G25) was placed on a camera stand approximately three m from the testing area to record the behavior of each dog.

Dogs were tested with the unsolvable problem task thoroughly described in Persson et al. (2015) with the addition of the presence of the owner. The device used consists of a plastic tray (55 × 25 cm) with three identical circular wells (seven cm in diameters) covered with plexiglass lids with odor ports (Fig. 1). Three quarter-pieces of Frolic© dog treats were placed underneath each lid. The dogs could easily access the treats in two of the three wells by sliding the lids to the side. However, the lid in the middle could not be opened hence making the task unsolvable. The experimenter cleaned, prepared and placed the unsolvable problem task on the ground approximately 15–30 cm from the middle of the back wall prior to the arrival of each dog.

Upon arrival at the testing arena, owners were reminded to stand passively immediately close to the fence at the front right corner inside the tent, facing the problem task. The experimenter was standing in the same position but on the opposite side of the tent (in the front left corner). After closing the fence gate, the owner was asked to unleash the dog that then could freely move around inside the testing area from this point onward. The owner had been instructed to not interact with the dog unless it was attempting to escape. If the dog tried to leave the tent, the owner was allowed to interrupt and call the dog back. Behaviors were not recorded during this interruption. If the dog had not opened any of the lids within 60 s, the experimenter opened both solvable lids halfway and immediately went back to her original position. The duration of the behavior test was 3 min.

The behaviors scored from the behavior test were the human-directed social behaviors proximity, physical contact, and gazing in relation to owner and experimenter. Duration and frequency were scored for each of these behaviors as described in the ethogram (Table 1). In addition to the social behaviors, we scored the duration of the time the dog spent in close proximity of the test-setup. The behaviors were scored from the video recordings using the Observer XT 10, Noldus software (https://www.noldus.com/knowledge-base/observer-xt-10).

Table 1:

Ethogram of the behaviors analyzed in the unsolvable problem task.

Behavior	Description
Experimenter zone	Dog’s head is within one body length of the experimenter
Owner zone	Dog’s head is within one body length of the owner
Experimenter gaze	The dog directs its eyes towards the face of the experimenter
Owner gaze	The dog directs its eyes towards the face of the owner
Experimenter physical contact	The dog is in physical contact with the experimenter
Owner physical contact	The dog is in physical contact with the owner
Duration test-setup	The duration of time (s) the dog’s head is within one body length of the test-setup

DOI: 10.7717/peerj.5889/table-1

Note:

Duration and frequency of the behaviours were scored. Zone behaviours were mutually exclusive.

Results

Variation was found in SNP1 in golden retrievers (HWE: p = 0.769) and in Labrador retrievers (HWE: p = 0.153) but all wolves were fixed for the C allele (Fig. 2A). For SNP2, variation was found in both golden retrievers (HWE: p = 1), in Labrador retrievers (HWE: p = 1) and in wolves (HWE: p = 1) (Fig. 2B). Additionally, when looking at the two types of Labrador retriever separately, there was a variation in both SNPs for both the common and the field type and the HWE was not significant for any of them (for SNP1 p = 0.580 and p = 1 and for SNP2 p = 1 and p = 1 for common and field, respectively) (Fig. 2).

The Fisher’s exact test showed a significant difference in genotype frequencies between the common and field Labrador retrievers for both SNP1 (p < 0.001) and SNP2 (p = 0.031) (Fig. 2). For SNP1, AC was the most common genotype in common type Labrador retrievers whereas CC was the most common in the field type. In SNP2, AA was the most frequent genotype for both types, but the proportion between AA and AG differed between the types. In the common type a larger proportion was of the AG genotype than in the field type.

In the unsolvable problem task, there were no associations between the SNPs and the time spent close to the test-setup. However, we found effects of both SNPs on behavioral variables related to social behavior in both breeds. The final models for each social behavior variable as well as F and p statistics for all behaviors can be found in Tables 2 and 3. Figures of all genotype-behavior associations are presented in Fig. S1. In the golden retriever, the genotype of SNP1 was significantly associated with physical contact, both on the frequency with the experimenter (Fig. 3A, F_1, 56 = 4.339, p = 0.018) and on frequency (Fig. 3B, F_2, 56 = 6.996, p = 0.002) and duration (Fig. 3C, F_2, 56 = 4.389, p = 0.017) with the owner. Golden retrievers with the AA genotype had a higher frequency of physical contact with the experimenter than individuals with AC. The AC genotype had instead more frequent physical contact with their owners than both AA and CC, and genotypes AA and AC had contact for a longer duration than individuals with the CC genotype.

In the Labrador retrievers, we found a significant interaction between genotype of SNP1 and breed type for the duration of gazing at owner (Fig. 4, F_1, 90 = 10.394, p = 0.002). In the field type, dogs with CC genotype gazed longer at their owner than those with AC genotype, whereas there were no differences between the genotypes among the common type.

SNP2 was significantly associated with several human-directed social behaviors in both breeds. In the golden retrievers, AG individuals spent less time in the experimenter zone than AA and GG dogs (Fig. 5A, F_2, 55 = 3.304, p = 0.044) and AA dogs gazed at the experimenter more frequently than GG dogs (Fig. 5B, F_2, 55 = 3.243, p = 0.047). Individuals carrying the AA and AG genotype gazed at their owners more frequently (Fig. 5C, F_2, 55 = 6.330, p = 0.003) and with a longer duration (Fig. 5D, F_1, 55 = 4.477, p = 0.016) than GG dogs. Also, golden retrievers with the AG genotype had longer duration of physical contact with their owner than those with AA (Fig. 5E, F_2, 55 = 14.809, p < 0.001).

In the Labrador retrievers, AA individuals at SNP2 spent more time in the experimenter zone (Fig. 6A, F_1, 94 = 6.252, p = 0.014) and visited it more frequently (Fig. 6B, F_1, 94 = 6.860, p = 0.010), as well as spent more time in the owner zone (Fig. 6C, F_1, 94 = 5.945, p = 0.017) and visited it more frequently (Fig. 6D, F_1, 94 = 13.846, p < 0.001). AA individuals also gazed more often at their owner (Fig. 6E, F_1, 93 = 5.209, p = 0.025) whereas AG individuals instead gazed at their owner for a longer duration (Fig. 6F, F_1, 93 = 7.209, p = 0.009). For the duration of gazing towards experimenter (Fig. 7A) and the frequency of gazing towards the owner (Fig. 7B) there was an interaction between genotype and type in the Labrador retrievers (F_1, 93 = 7.796, p = 0.006 and F_1, 93 = 3.947, p = 0.05, respectively). In the field type, AA individuals gazed longer at the experimenter and more often at their owner than AG individuals whereas there were no differences in the common type.

In golden retrievers, sex and intranasal oxytocin treatment (part of the experiment for which the dogs were originally recruited) were included in the models. In the analyses for both SNP1 and SNP2, sex had an effect on the duration dogs spent in the owner zone where males spent significantly more time with their owner than females (F_1, 56 = 6.590, p = 0.013 for SNP1 model and F_1, 55 = 7.374, p = 0.009 for SNP2 model; males 35.02 ± 7.986 vs. females 16.72 ± 3.921). Additionally, in analyses for both SNPs, intranasal oxytocin treatment significantly decreased the duration (F_1, 56 = 15.797, p < 0.001 and F_1, 55 = 22.377, p < 0.001; oxytocin 0.24 ± 0.109 vs. control 1.75 ± 0.591) and frequency (F_1, 56 = 9.705, p = 0.003 and F_1, 55 = 12.305, p = 0.001; oxytocin 0.97 ± 0.481 vs. control 4.20 ± 1.603) of experimenter physical contact seeking as well as the frequency of owner physical contact (F_1, 56 = 8.137, p = 0.006 and F_1, 55 = 5.437, p = 0.023; oxytocin 0.71 ± 0.377 vs. control 1.67 ± 0.615).

For the Labrador retrievers, sex and type were included in the models. In the models for both SNP1 and SNP2, the difference between males and females for the duration of gazing at their owner was significant where males looked for a longer time (F_1, 90 = 7.667, p = 0.004 and F_1, 93 = 13.995, p < 0.001; 10.19 ± 2.72 vs. 4.42 ± 0.73 s for males and females, respectively). There were many differences between the types. In the models for both SNPs, the frequency of owner and experimenter zone were significant as well as the duration of gazing at the owner and frequency and duration of owner physical contact. Field type Labradors visited the experimenter zone at a higher frequency (F_1, 92 = 9.046, p = 0.003 and F_1, 94 = 4.821, p = 0.031; 3.27 ± 0.37 vs. 2.12 ± 0.22) as well as the owner zone (F_1, 92 = 18.153, p < 0.001 and F_1, 94 = 10.559, p = 0.002; 4.21 ± 0.43 vs. 2.51 ± 0.23 times). The field type gazed longer at their owner (F_1, 91 = 10.475, p = 0.002 and F_1, 93 = 25.596, p < 0.001; 5.74 ± 0.92 vs. 3.16 ± 0.50 s) and were in physical contact with their owner both more often (F_1, 92 = 8.065, p = 0.006 and F_1, 94 = 20.862, p < 0.001; 0.77 ± 0.17 vs. 0.29 ± 0.10) and for a longer time (F_1, 90 = 15.910, p < 0.001 and F_1, 94 = 9.025, p = 0.003; 0.70 ± 0.19 vs. 0.09 ± 0.04). Additionally, in the model for SNP1, there were significant differences between the types in frequency and duration of gazing at the experimenter as well as frequency of owner gazing. The field type gazed both longer and more often at the experimenter (F_1, 91 = 10.475, p = 0.002 and F_1, 91 = 7.679, p = 0.007) and more often at their owner (F_1, 91 = 20.208, p > 0.001). The means of the duration and frequency for looking at experimenter were 5.74 ± 0.92 s and 4.02 ± 0.49 times in field type vs. 3.16 ± 0.50 s and 2.63 ± 0.34 in the common type. For the frequency of looking at owner, the means were 6.52 ± 0.82 vs. 3.00 ± 0.33 times.

Discussion

In a previous genome-wide association study on beagles tested in the unsolvable problem paradigm, Persson et al. (2016) identified two SNPs on chromosome 26, BICF2G630798942 (SNP1) and BICF2S23712114 (SNP2), associated with social interactions directed toward humans. Here, we show that both SNPs are also associated with human-directed social behavior in two additional dog breeds, golden and Labrador retriever. We also show that genotype frequencies for the two SNPs differ between wolves and dogs, between breeds and between recently selected breed types. Thus, these loci could have been affected by selection during domestication as well as during breed formations. This suggests that selection of favorable alleles in the genomic region of the SNPs may have been an important part of dog domestication.

In a population of laboratory beagles, Persson et al. (2016) found a significant association between SNP1 and the duration of physical contact and duration of human proximity. Additionally, a suggestive association was found between SNP2 and the duration of human proximity. It is important to stress that these SNPs are not causative for the behavior differences but rather linked to the specific causal loci. SNP1 and SNP2 are both located on chromosome 26, and within the linkage disequilibrium regions five possible associated genes are present. The marker SNP1 is located in an intron of the gene SEZ6L and there are no other genes present in the same linkage block. SNP2 is located in an intron of the gene ARVCF and three additional genes are in linkage: COMT, TXNRD2, and TANGO2. Previously, SEZ6L, ARVCF, COMT, and TXNRD2 have been associated with social disorders and schizophrenia in humans (Chapman et al., 2015; Mas et al., 2010; Sanders et al., 2005; Xu et al., 2013).

In the present study, we verify that the association between these genomic regions and human-directed social behavior are not specific to the previously studied beagles. Similar effects were found in two other breeds, Labrador and golden retrievers, tested in the same unsolvable problem task as the beagles. The genotype of SNP1 had an effect on physical contact seeking with both experimenter and owner in the golden retriever and was associated with a difference in owner gazing between Labrador types. The genotype of SNP2 was primarily associated with eye contact seeking with the experimenter as well as owner in both golden and Labrador retrievers. In the Labradors this effect differed between types where only field type individuals carrying the AA-genotype were gazing more. Additionally, SNP2 was associated with experimenter proximity seeking in both breeds and owner proximity seeking in only Labradors. Finally, an association was found between SNP2 and owner physical contact seeking in golden retrievers but not in Labradors. Genotype and sex interactions were identified in the beagles for SNP1 (Persson et al., 2016) but this was not present in the retrievers.

In addition to SEZ6L, ARVCF, and COMT, there are other genes suggested to be associated with dogs’ human-directed social behavior. Previous research has found associations between polymorphisms in the oxytocin receptor gene and social behavior both during problem solving and in additional situations (Kis et al., 2014; Persson et al., 2017). The oxytocin receptor gene has also been associated with successful training of detection dogs (Konno et al., 2018). Another gene of interest is the dopamine receptor D4 gene that has been associated with gazing toward humans (Hori et al., 2013). Additionally, VonHoldt et al. (2017) investigated a candidate region associated to Williams–Beuren syndrome (WBS) in humans, which is causing hyper-social behavior amongst other effects. It was found that structural variants in this region were also associated with extreme sociability in dogs.

During the course of domestication, dogs seem to have evolved impressive interspecific cooperation skills with humans (Jensen et al., 2016; Miklósi & Topál, 2013). For example, comparative studies have shown that dogs have a higher sociability in general than wolves (Bentosela et al., 2016) and, specifically, that dogs seek more human contact when faced with a problem than wolves do (Heberlein et al., 2016; Miklósi et al., 2003; Udell, 2015). It seems that genetic variants contributing to human-directed social abilities have been selected during domestication and, thus, the two genomic regions investigated in the present study may have been targeted. Variation in sociability toward humans has a significant genetic component, which has been shown by heritability estimates (Persson et al., 2015; Sundman et al., 2016; Van Der Waaij, Wilsson & Strandberg, 2008; Wilsson & Sundgren, 1998) and this is a requirement for selection. Wolves were found to be fixed for the C-allele on SNP1 whereas there was a variation in SNP2. The former polymorphism may thus not exist or is rare among wolves and could even have appeared after the split with dogs, whereas the latter is present in both species. However, as this current study has investigated the SNP genotypes in only a very limited sample of Scandinavian wolves, these results should be cautiously interpreted. A recent study describing the genetic architecture of a dog x wolf crossbreed, the Czechoslovakian Wolfdog, found regions containing excess of wolf and dog ancestry genes respectively (Caniglia et al., 2018). The Czechoslovakian Wolfdog derives from a cross between Carpathian wolves and German shepherd dogs and, even though it shares many morphological features with the wolf, it shows mostly dog-like behavioral phenotypes. Interestingly, the SEZ6L, ARVCF, and COMT genes were all detected within regions of excess dog ancestry.

When using an unsolvable problem paradigm to study human-directed social behavior we should also consider the dogs’ persistence to solve the problem and their food motivation. Previous studies have, for example, found a negative correlation between persistence and eye-contact duration (Brubaker et al., 2017), and that more persistent dogs show a longer latency until they gaze at the present human (Marshall-Pescini et al., 2017). On the other hand, in both Persson et al. (2015) and Sundman et al. (2018), test-setup interactions form a component separate from social behaviors in a principal component analysis. It is difficult to disentangle sociability from persistence and food motivation when using the unsolvable problem paradigm. However, neither of the two SNPs in the present study were associated with the duration the dogs spent in proximity of the test-setup, and thus, they seem to be associated with human-directed social behavior rather than persistence.

Previous studies have reported breed differences in behaviors related to human communication (Passalacqua et al., 2011; Sundman et al., 2018; Udell et al., 2014). We can study recent selection by examining established breeds recently diverged into two types due to different breeding goals; for example, dogs suitable for field-work vs. pet and conformation dogs. The Labrador retriever is a breed clearly divided into two types. Pedigrees as well as morphology and behavior distinguish the common type from the field type (Sundman et al., 2016, 2018). Specifically, Sundman et al. (2018) compared the types in the unsolvable problem task and found several behavioral differences. During recent selection, it seems that the field type has increased its human-contact seeking behavior in comparison to the common type, although environmental causes cannot be discarded. Likewise, in the present study, we found differences in allele frequencies between the types for both SNP1 and SNP2. These genetic markers have thus been affected by recent selection in two selection lines of dogs that also differ in social abilities, lending further support to the association between the genomic region and human-directed social behavior.

Conclusions

Our results verify the associations between human-directed social behavior and the SNPs BICF2G630798942 and BICF2S23712114 on canine chromosome 26. We suggest that these loci could have been affected by domestication and selection for sociality in dogs and that genetic variants linked to the SNPs may have been targeted during domestication. Hence, genes within the linkage disequilibrium of these genetic markers are of high interest for further investigation of the genetics behind the impressive social skills of dogs.

Supplemental Information