A common feature of domestic animals is tameness—i.e., they tolerate and are unafraid of human presence and handling. To gain insight into the genetic basis of tameness and aggression, we studied an intercross between two lines of rats (Rattus norvegicus) selected over >60 generations for increased tameness and increased aggression against humans, respectively. We measured 45 traits, including tameness and aggression, anxiety-related traits, organ weights, and levels of serum components in >700 rats from an intercross population. Using 201 genetic markers, we identified two significant quantitative trait loci (QTL) for tameness. These loci overlap with QTL for adrenal gland weight and for anxiety-related traits and are part of a five-locus epistatic network influencing tameness. An additional QTL influences the occurrence of white coat spots, but shows no significant effect on tameness. The loci described here are important starting points for finding the genes that cause tameness in these rats and potentially in domestic animals in general.
ANIMAL domestication marked a turning point in human prehistory (Diamond 2002), and domestic animals have been the subject of research for many years (Darwin 1868). Recently, genetic studies have shed light on when, where, and how often a range of animal species were domesticated (Troy et al. 2001; Vila et al. 2001; Savolainen et al. 2002; Larson et al. 2005; Driscoll et al. 2007; Eriksson et al. 2008; Naderi et al. 2008). With the exception of coat color (e.g., Pielberg et al. 2008) and skin pigmentation (Eriksson et al. 2008), little is known about what occurred genetically during animal domestication. At what genes were allelic variants selected for by would-be practitioners of animal husbandry? Although domestic animals differ from each other in many ways, they all share the trait of tameness—i.e., they tolerate and sometimes even seek human presence and handling. Almost nothing is currently known about the genetic basis of tameness.
In a series of studies initiated by D. K. Belyaev, researchers at the Institute for Cytology and Genetics in Novosibirsk (Russia) have subjected several mammalian species to a process of experimental domestication (Trut 1999). These studies, some of them ongoing for several decades, involve selection for tame and aggressive behavior in lines of animals derived from wild populations. They include a fox population that has been “domesticated” to such an extent that the tame foxes are now similar to dogs in some respects (Hare et al. 2005). They also include a population of wild-caught rats (Rattus norvegicus) that was selected for either reduced or enhanced aggression toward humans over >60 generations (Belyaev and Borodin 1982). To select the animals, their response to an approaching human hand was observed, and the rats showing the least and the most aggressive behavior were allowed to mate within the two lines, respectively. The initial response to selection was rapid and then slowed, so that little change in behavior from generation to generation has been observed in the last 10–15 generations, although the selection regime has been continued to the present. Today, the “tame” rats are completely unafraid of humans, they tolerate handling and being picked up, and they sometimes approach a human in a nonaggressive manner. By contrast, the “aggressive” rats ferociously attack or flee from an approaching human hand.
To study the genetic basis of tameness we have established populations of both rat lines in Leipzig. In their new environment, the rats maintained their behavioral differences in response to humans, and these differences were not influenced by postnatal maternal factors (Albert et al. 2008). In addition, the rat lines differ in a number of other behavioral, anatomical, and physiological traits, raising the question whether these traits are influenced by the same loci as tameness and aggression toward humans.
Many domestic animals display conspicuous coat color variations not found in their wild relatives. Prominent examples include the white color variants in dogs, pigs, cows, horses, and chickens. In laboratory rats, it has been proposed that “coat color genes” may account for many of the differences associated with domestication (Keeler and King 1942). It is thus interesting that individuals with white spots appeared in both the tame foxes (Trut 1999) and the tame rats (Trut et al. 2000) at higher frequency than in the corresponding aggressive lines, although they were absent or rare in the founding fox and rat populations, and although they were not selected for. The rat populations studied here provide an excellent opportunity to examine whether tameness is influenced by the same loci as white coat spotting.
In this study, we crossed the two rat lines and bred >700 intercross animals. A broad set of behavioral, anatomical, and physiological traits was measured, and a genomewide set of genetic markers was used to identify genomic regions (quantitative trait loci, QTL) that influence tameness as well as other traits that differ between the lines, including white spots.
MATERIALS AND METHODS
The tame and the aggressive rat (R. norvegicus) lines derive from one population of wild-caught rats, which has been bidirectionally selected at every generation since 1972 at the Institute of Cytology and Genetics at the Siberian Branch of the Russian Academy of Sciences in Novosibirsk, Russia. About 30% of the animals from each generation were selected on the basis of the level of tameness and defensive aggression they displayed in response to humans on a five-point scale (Naumenko et al. 1989; Plyusnina and Oskina 1997). Inbreeding was kept at a minimum by avoiding mating closely related individuals.
The pedigree described in this study was initiated from four tame and four aggressive individuals (2 females each; the “F minus one”, or “F−1”, generation) from the 64th generation of selection. The F−1 animals did not have common parents and at most one common grandparent. They were mated within line to yield 11 (5 tame and 6 aggressive rats, one male each) F0 animals. These were crossed reciprocally, and six hybrid F1 males were repeatedly mated to 37 F1 females to produce 733 F2 rats (383 females). A separate set of 47 F1 animals (25 females) derived from different F0 crosses was used for characterizing the F1 generation in behavioral tests. Phenotypic data from F0 animals discussed in this article are the same as presented in Albert et al. (2008). Animals were maintained under standard laboratory conditions, under an artificial 12-hr light cycle (lights off at 1:00 pm). The light cycle allowed behavioral testing during the dark phase, when rats are more active. Cages were equipped with sliding doors to allow for transfer without handling. During all caretaking procedures and experiments, animals from different lines and generations were treated identically. The study was approved by the regional government of Saxony (TVV Nr. 29/05).
F2 rats were tested in a standardized series of behavioral paradigms. We measured the animals' level of tameness/aggression with the “glove test,” by confronting them with an approaching human hand and attempting to handle them (see Albert et al. 2008 for details on the testing procedure). Beginning 2 weeks after the glove test, rats performed an open-field test, a light–dark test, and a startle response test, which provide various measures of exploratory and anxiety-related behavior. A total of 470 (64%) of the F2 rats performed a second glove test trial. F1 rats were tested once with the glove test at 12–14 weeks of age and then followed the testing schedule described in Albert et al. (2008). All tests were performed with minimal handling following procedures described in Albert et al. (2008). Glove test trials were videotaped and later analyzed by two independent observers (5% of trials only by one observer). Experimenters and observers were blind to the animals' identity and to the further data processing. A set of 11 behaviors (e.g., “attack” or “tolerate handling”) was scored (see Albert et al. 2008 for detailed descriptions of the behaviors), and each score was converted to a numeric measure (e.g., “number of occurrences” or “duration in seconds”; Table 1).
Blood and tissue sampling:
Dissections were performed on 733 F2 (383 females) and 37 F1 animals (16 females). Within 2 weeks after the last behavioral test, animals were killed between 2:00 and 6:00 pm. F2 animals had been starved for 24 hr prior to dissection. Animals were weighed, anesthetized with CO2, and killed by cervical dislocation. Blood was collected rapidly after death by heart puncture and separated into serum and blood cells by centrifugation 30 min after sampling. Serum was stored at −20° and later transferred to −80° until analysis. Liver (small section, not weighed), spleen, kidney, adrenal gland, lung, and heart were removed rapidly, weighed, snap-frozen in liquid N2, and stored at −80°. From F1 animals, kidney and spleen were weighed and stored.
Serum was analyzed in 684 F2 rats (357 females). Electrolytes, metabolites, immunological parameters, enzymes, and hormones were analyzed in serum according to the guidelines of the German Society of Clinical Chemistry and Laboratory Medicine. Measurements for all serum traits but corticosterone were performed using a Hitachi PPE-Modular analyzer (Roche Diagnostics, Mannheim, Germany). Corticosterone was measured using a commercial ELISA assay (IDS, Frankfurt, Germany).
Statistical phenotype analyses:
We sought to control for possible confounding effects, such as observer in the glove test or an animal's litter. We constructed mixed linear models of the phenotypes, estimated effect sizes using restricted maximum likelihood, and adjusted the phenotypes for the respective effects specified in Table S1. Sex and covariates were not adjusted; we instead included them as fixed effects into the QTL analyses. For glove test measures, we separately adjusted observations from different trials and observers and then averaged the available observations from each rat. To summarize a rat's behavior in the glove test, we performed principal component analysis (PCA) on the individual glove test measures. We used only F2 animals in the PCA and calculated scores of F0 and F1 animals on the basis of the obtained regression coefficients.
Comparisons of non-glove test phenotypes between the F0, F1, and F2 generations, as well as tests for sex differences (Wilcoxon's rank test), were performed on phenotypes adjusted for covariates. Correlations between phenotypes were calculated using Pearson's product-moment correlation, on phenotypes adjusted for all effects (including sex and covariates) listed in Table S1. All analyses were performed using the software R (R Development Core Team 2008) and the nlme (Pinheiro et al. 2008) and lme4 (Bates 2007) packages.
Animals were genotyped with 152 microsatellite and 49 single-nucleotide polymorphism (SNP) markers. Markers were selected for maximum allele frequency differences between the outbred rat lines, as determined from preliminary genotyping of a panel of F0 animals. Preliminary genotyping of microsatellites was performed as described below, while SNPs were screened as described in Saar et al. (2008). All markers used in the QTL mapping are listed in Table S2. DNA was isolated either from lung or from spleen tissue. Polymerase chain reaction (PCR) for microsatellite markers was performed using the M13-primer PCR system (Schuelke 2000) and analyzed on an ABI3730 DNA Analyzer (Applied Biosystems, Foster City, CA). Microsatellite genotypes were determined using the software GeneMapper Version 4.0 (Applied Biosystems), and all genotypes were manually double checked. SNPs were genotyped using TaqMan chemistry (Applied Biosystems). SNP genotypes were called automatically as part of the scanning process, and genotype plots were inspected visually.
Genotype data for individual markers were tested for Mendelian pedigree errors, using the program PedCheck (O'Connell and Weeks 1998). Allele calls for inconsistent genotypes were rechecked manually and either they were corrected for obvious genotyping errors or the marker was excluded from further analysis if genotypes could not be determined unambiguously. We found that some rats yielded inconsistent genotypes for several markers, although the respective genotypes appeared to have been called correctly. We interpreted these individuals as having incorrect pedigree records. Initially, genotypes had been obtained from F0 and F2 animals. To clarify the pedigree structure, we typed all markers in all F−1 and F1 animals for which samples were available, as well as in an extended panel of F0 animals. Using the software “Cervus” 3.0 (Kalinowski et al. 2007), we determined the most likely parental pair for all genotyped F0, F1, and F2 individuals, on the basis of a subset of 107 microsatellite markers with unambiguous genotype patterns. Rats for which the inferred parents differed from those in our records were either reassigned to the inferred parents or excluded from further analysis if no unique parental pair could be identified. The final pedigree used in the QTL analyses showed no Mendelian errors and comprised 8 F−1 rats (all genotyped), 11 F0 rats (8 genotyped), 43 F1 rats (30 genotyped), and 706 F2 rats (all genotyped).
Linkage map construction:
We constructed a sex-averaged linkage map, using a version of the program crimap (Green et al. 1990) modified to handle large pedigrees by the University of California Davis, Veterinary Genetics Laboratory. On chromosome 6, we found the markers D6rat213 and D6rat68 to be inverted on our linkage map relative to their physical positions. On chromosome 13, the markers D13rat5 and D13rat64 were found to be inverted and to map very closely (1.2 cM) to each other, despite a physical distance of 21 Mb. These cases may reflect chromosomal rearrangements in our wild-derived rats compared to the genome sequence of the inbred Brown Norway laboratory strain (Gibbs et al. 2004). We used our inferred linkage maps in further analyses (Table S2). We estimated information content at autosomal marker positions on the basis of the fraction of individuals whose alleles could be unambiguously traced to the F−1 generation.
A standard model of a phenotype y influenced by a single QTL can be written aswhere β0 is the population mean, F is a matrix of regression coefficients for fixed effects and covariates (see Table S1 for the effects we included in the QTL models for various phenotypes), Z is an incidence matrix relating observations in F to individual observations, β1j and β2j are the additive and dominance effects at genomic position j, aj and dj are indicator variables relating these genetic effects to individuals, and εj is the residual error. We estimated the parameters β0, β1, and β2 using a variation of the least-squares regression framework (Haley and Knott 1992; Haley et al. 1994). In this framework, F2 animals are grouped at a given genomic position according to whether they carry two, one, or zero alleles originated from the tame (allele “T”) or aggressive (allele “A”) line, forming the genotype classes TT, TA, and AA. Missing F0 genotypes can lead to a loss of power because some alleles in F2 animals might not be reliably traced to parental lines. This limitation can be overcome by including the genotyped parents of F0 animals (the F−1) in the analysis and by tracing alleles back to them. Hence, we inferred missing genotypes of F0 and F1 animals on the basis of their ancestors' and offspring genotypes (File S1). Next, we calculated, in steps of 1 cM throughout the genome, the probability of an F2 animal belonging to the TT, TA, or AA genotype classes on the basis of genotypes of flanking markers, using methods described in Pong-Wong et al. (2001) and Besnier and Carlborg (2007). The genotype probabilities were used to compute the indicator variables aj and dj (Haley et al. 1994). Finally, we fitted the parameters β0, β1, and β2 using least-squares regression. High F-values obtained from the regression point to the presence of a putative QTL at the respective location. The difference in mean between the two homozygous classes TT and AA corresponds to twice the additive effect β1 of the QTL, while the deviation of the heterozygous TA class from the mean of the TT and AA classes measures the degree of dominance β2. We searched for QTL by following a forward selection procedure (Figure S1). After the initial scan, we included the effect and position of the most significant QTL that reached genomewide significance as a fixed effect in the model. The scan was then repeated, and QTL were added until no additional significant QTL were identified. For each phenotype, we performed 1000 permutations of phenotypes with regard to genotypes to determine the F-value threshold that corresponds to a genomewide significance level of P = 0.05 (Churchill and Doerge 1994). We express variance components attributable to QTL as a fraction of the residual phenotypic variance, i.e., the variance in phenotype after fixed effects and covariates have been factored out.
We analyzed the X chromosome using the software QxPak (Perez-Enciso and Misztal 2004) (see File S1 for details). Since the permutation-based significance thresholds derived for the autosomes cannot be directly applied to the X chromosome, we assumed QTL with a nominal P-value <0.001 (0.005) to be significant (suggestive) at a genomewide level, as suggested in the QxPak manual. We tested for linkage to the Y chromosome, using ANOVA as described in File S1.
Mapping of epistatic QTL:
We searched for epistatic QTL using an extension of the least-squares regression model for single QTL, following a search strategy described and applied earlier (Carlborg and Andersson 2002; Carlborg et al. 2003, 2004, 2006; Wright et al. 2006). Here, we first describe the regression model underlying our epistatic analyses and then go through the steps of the search strategy. A schematic representation of the approach is shown in Figure S1.
The standard extension of the model for a single QTL to incorporate two epistatic QTL isThe additional parameters are the additive (β3k) and dominance effects (β4k) of the second QTL at position k, the epistatic effects between the two loci (β5jk–β8jk), and the corresponding indicator variables. The indicator variables for the interaction terms (aajk, adjk, dajk, ddjk) were computed by multiplying the respective additive and dominance indicator variables for the single QTL (Haley and Knott 1992). We estimated the effects β0–β8 at a given pair of loci using least-squares regression.
The search strategy to find epistatic pairs of QTL involves three steps (see Figure S1 and Carlborg et al. 2004 for further details). First, we searched for single QTL as described above. Second, we performed a genomewide search for putative epistatic pairs of loci. Third, each pair is tested for the existence of epistasis.
For each pair proposed by the pairwise search in step two, we assigned significance in one of three ways. If both loci were significant by themselves in the single-QTL analysis, no further testing is necessary, and the pair is declared significant. If one locus in the putative pair was a significant single QTL, we need to test only whether the second locus is significant (i.e., whether its inclusion in the model already containing the first locus improves model fit). To derive the corresponding threshold, we created, for each single QTL, 1000 randomized data sets by permuting only the indicator variables of the second QTL (ak, dk), as well as those of the interaction effects (aajk, adjk, dajk, ddjk), while the first QTL was kept in the model as a fixed effect (permutation test “Type I” in Figure S1). In each permuted data set, we searched for the best fitting pair including the known QTL and a second QTL. We then compared the model fit obtained from the putative pair with the model fits obtained from the permutations. Finally, if neither QTL in the putative pair was significant as a single QTL, we need to test whether the joint inclusion of both loci improves the model fit significantly. We performed 1000 phenotype/genotype permutations and searched each randomized data set for its best fitting pair (permutation test “Type II” in Figure S1). For increased efficiency, this was done using a genetic search algorithm (Carlborg et al. 2000). Significance of the putative pair was assigned by comparing its model fit to the fits obtained from the permutations. Throughout, we used a cutoff of P < 0.05 for “significant” pairs of QTL.
So far, we have detected pairs of QTL, but not yet tested whether there is significant epistasis between the members of a given pair. To do this, we generated 1000 randomized data sets for each pair by permuting only the interaction indicator variables (aajk, adjk, dajk, ddjk), while keeping the additive and dominance effects of the two loci in the pair constant (Carlborg and Andersson 2002) (permutation test “Type III” in Figure S1). Epistasis is assumed if the putative epistatic pair is in the top 5% of model fits obtained from the permutations.
To construct the network influencing tameness, we consider loci (single or as part of a pair) with overlapping confidence intervals to be the same locus. We show these loci as circles in Figure 5, with lines between them indicating significant epistatic interactions for that given pair. We did not fit a model incorporating interactions between more than two loci. To visualize the directions of the epistatic interactions (Figure 5, B–F), we grouped F2 animals according to their genotypes at the respective loci. For each of the resulting nine two-locus genotype groups, we calculated the mean and the standard error of the mean of the respective animals' level of tameness.
Because some loci in the tameness network are part of more than one epistatic interaction, we cannot calculate the residual phenotypic variance explained by the whole network by simply adding up the variances explained by the respective pairs (Table 4). Instead, we used the NOIA model of genetic effects (Alvarez-Castro and Carlborg 2007) as implemented in Le Rouzic and Alvarez-Castro (2008), with analyses restricted to at most pairwise interactions. NOIA is specifically designed to estimate parameters, including genetic variances, in multilocus networks (Alvarez-Castro and Carlborg 2007).
A cross between tame and aggressive rats:
To create an intercross between the tame and aggressive rats, we mated one tame and one aggressive male to 5 aggressive and 4 tame females, respectively. In the resulting F1 generation, we repeatedly mated six males with a total of 37 females to produce an F2 population of 733 animals (362 females). Details of the mating scheme are described in materials and methods.
Analyses of phenotypes:
We recorded a total of 45 phenotypic traits in the F2 animals, including measures from four behavioral tests, anatomical parameters, and serum levels of hormones, enzymes, and other serum components (Table 1). To measure the level of tameness and aggression, we used a paradigm that closely mimics the test used to select the two rat lines over the past 36 years. In this “glove test,” a gloved human hand approaches a rat in an experimental cage and attempts to touch it and to pick it up (Figure 1A). Various aspects of the rat's behavior are recorded (Albert et al. 2008). When testing F1 animals in this test, we found that the extreme levels of tameness and aggression observed in the original F0 lines were absent (Figure S2). By contrast, F2 animals displayed the full range of behaviors found in the original lines (Figure S2). A few F2 animals even exceeded the levels of tameness and aggression observed in the tame and aggressive lines.
A PCA of the behaviors recorded in the glove tests of the F2 animals confirms these observations. The first principal component (PC1) corresponds to behaviors such as attacks, screaming, and (with inverse loading) the toleration of touch or handling (Table 2). PC1 explains 26% of the variance in behavior of F2 rats. Of the PCs explaining >10% of the variance, PC1 most clearly separates the tame from the aggressive F0 animals (Wilcoxon's rank test: PC1, P < 10−15; PC2, P = 0.003; PC3, P = 0.07), although these animals were not included in the PCA (see materials and methods). Of the F1 rats, 94% (44/47) had PC1 scores between the medians of the tame and the aggressive rats. Of the F2 rats, 79% (551/700), fell into this intermediate range, while 6% (43) of the F2 rats had more tame and 16% (109) had more aggressive PC1 scores than defined by the respective F0 medians (Figure 1B). This indicates that there is substantial variation in tameness in the F2 rats and suggests that PC1 is a useful measure of this variation. The glove tests were repeated in 470 F2 rats (materials and methods). The correlation between the PC1 scores obtained for the two trials was 0.44 (Pearson's r, P < 10−15).
We also performed an open-field test, a light–dark test, and a startle response test, which measure traits related to anxiety and fear as well as general activity (Table 1). The values of all these traits in F2 animals overlapped substantially with those in F0 animals (Figure S3). This was also true for body weight, for the weight of six organs, and for 8 of 14 serum traits (Figure S4 and Figure S5). By contrast, >75% of the F2 rats had higher (corticosterone, creatinine, glucose, chloride) or lower (alanine aminotransferase, ALAT; alkaline phosphatase, AP) values in these measures than >75% of the tame and aggressive rats (Figure S5).
Sex differences were apparent for many traits in the F2 animals (Table S3). When males and females were considered separately, most phenotypes were approximately normally distributed in the F2 generation (Figure S6, Figure S7, Figure S8, and Figure S9). By contrast, raw glove test measures had highly skewed distributions, with prominent peaks at zero counts/durations.
Correlations between phenotypes:
Earlier work revealed a multitude of phenotypic differences between the tame and the aggressive rats, including behavioral, anatomical, hormonal, and neurochemical differences (Naumenko et al. 1989; Plyusnina and Oskina 1997; Popova et al. 2005; Albert et al. 2008). If these differences are caused by the same genetic loci, they should be correlated in the F2 animals. We did observe significant correlations among parameters recorded in the same test. For example, correlations between different measures in the glove test are reflected in their contributions to PC1 (Table 2). In contrast, parameters from different tests were generally not, or only weakly, correlated (Figure 2, Table S4). Notably, two measures of “anxiety” (the percentage of time spent in the light compartment of the light–dark test and the percentage of time spent in the center of the open field) were not correlated (r = 0.03, P = 0.48). There were significant but weak correlations between tameness and aspects of behavior in the open-field and light–dark tests. High levels of tameness were also correlated with low corticosterone levels (r = −0.08, P = 0.04), but not with the weight of the adrenal glands (r = −0.06, P = 0.10).
QTL for tameness and associated traits:
We typed the animals in the pedigree for 201 genetic markers (152 microsatelites and 49 single-nucleotide polymorphisms) that were selected to be polymorphic between the parental strains and to provide coverage of most of the genome (materials and methods).
The phenotypic and genetic data were used to identify QTL for the traits measured in the F2 animals. A total of 23 significant and 125 suggestive autosomal QTL, and one significant QTL on the X chromosome, were identified when analyzing both sexes together (Table 3 and Table S5 and Table S6). All but two serum traits (fT3 and calcium levels) showed at least suggestive QTL. In what follows, we focus on QTL for tameness and overlapping QTL, as well as one QTL for coat color.
Two significant QTL for tameness (measured as PC1) were identified (Figure 3A). The strongest of these (termed “Tame-1”) is located at 58 cM on chromosome 1. The difference in tameness between homozygous genotypes at Tame-1 corresponds to ∼20% of the difference between the tame and the aggressive line (1.2 of 6.4 units of PC1). Tame-1 explains 5.1% of the phenotypic variance in tameness. The second locus (“Tame-2”) is located at 78 cM on chromosome 8. Both the tameness difference between homozygous genotypes (∼10% of the line difference) and the portion of residual phenotypic variance in tameness it explains (2.3%, are about half of those of Tame-1.
The region encompassed by Tame-1 also contains significant QTL for rearing in the open field and for adrenal gland weight, as well as a suggestive QTL for the time spent moving in the open field (Figure 3B). The effects of these QTL are in the expected direction—i.e., alleles from the tame line influence the phenotype in the direction expected from the comparison between tame and aggressive animals (e.g., causing higher tameness scores and lower adrenal gland weight). However, Tame-1 also overlaps with a significant QTL that influences spleen weight. This QTL is transgressive—i.e., the alleles from the tame line reduce spleen weight although the rats from the tame line have 30% heavier spleens on average (Albert et al. 2008).
In the proximity of Tame-2 several other QTL are found, two of which are significant. At these QTL, the alleles from the tame line increase body weight and decrease the time an animal spent in the corners of the open field, respectively. These effects are in the expected direction. Among suggestive QTL, those for time spent in the center, time spent moving and the number of rears in the open field, kidney weight, and serum aspartate aminotransferase (ASAT) all have effects in the expected direction. There were also two transgressive suggestive QTL, where the tame allele increases adrenal gland weight and the magnitude of the startle response, respectively.
We found white coat spotting to be linked to a significant QTL on chromosome 14, but no aspect of tameness mapped to this region (Figure 4). White coat spotting did not show association to Tame-1, Tame-2 (Figure S10), or any other QTL for behavior during the glove test. Further, the tameness levels of individuals carrying white ventral spots did not differ significantly from those without them (t-test, P = 0.17, Figure 4A).
To assess whether the QTL we identified might be specific to one sex, we analyzed all traits using only female or male F2 animals, respectively (Table 3 and Table S5 and Table S7). For tameness, Tame-1 reached genomewide significance in females and chromosomewide significance in males, where the F-value (8.6) was close to the genomewide significance threshold (8.8). Tame-2 reached chromosomewide significance only in males. The QTL for adrenal gland weight on chromosome 1 and for white spotting on chromosome 14 were significant in both sexes. Several QTL were suggestive in both sexes (e.g., spleen weight on chromosomes 1 and 10), whereas for several others we found significant or suggestive linkage only in one sex. All individual behaviors in the glove test, but not PC1, fall in the latter category. For example, at Tame-1, only males showed a significant QTL for screaming, whereas only females had significant QTL for flights and toleration of touch. Toleration of touch yielded one additional significant sex-specific QTL in females and males, respectively (Table S5).
Epistatic interactions can have large effects on phenotypic traits. Hence, we searched the genome for interacting pairs of loci for all traits described in this study. Fifteen epistatic pairs affecting 9 traits reached genomewide significance (P < 0.05), exceeding the random expectation of <3 pairs for 48 analyzed traits. Most pairs were found for tameness, forming an interconnected network of 5 loci (Figure 5A). The network explained 14% of the residual phenotypic variance, compared to 7.4% explained by Tame-1 and Tame-2 individually. It is discussed below, while epistatic pairs for the remaining traits are given in Table 4.
The tameness network comprises five pairwise interactions between five loci (Figure 5A). Two loci in the network had significant individual effects (Tame-1 and Tame-2). When considering these loci simultaneously, the tame allele (T) at locus Tame-1 increases tameness regardless of the Tame-2 genotype (Figure 5D). The effect is, however, strongest when Tame-2 is homozygous (AA) for the allele from the aggressive line (A). The effect of the tame allele at Tame-2 is smallest when Tame-1 is homozygous for the allele from the tame line (TT), where the difference between the three Tame-2 genotypes (AA, AT, and TT) is not significant.
Both Tame-1 and Tame-2 interact with a locus on chromosome 19 (denoted Tame-3) that reached only suggestive significance in the scan for single QTL. The tame allele at Tame-1 has a significant effect on tameness only in a heterozygous (AT) background of Tame-3 (Figure 5B). There is no additive effect of Tame-3 on tameness, but strong overdominance for aggression in the Tame-1 AA genetic background and a strong overdominance for tameness in the Tame-1 TT genetic background. This interesting shift in the direction of the dominance effect deserves further investigation. When considering Tame-3 and Tame-2 together, the only genotype with a deviating phenotype is the AAAA double homozygote that significantly increases aggression (Figure 5C). In addition, Tame-1 and Tame-2 interact significantly with one additional locus each. The effect of the tame allele at Tame-1 is strongest when a locus on chromosome 4 (denoted Tame-4) is TT, and the tame allele at Tame-4 has a significant effect on tameness only in the AA Tame-1 genotype (Figure 5E). In this background, it is transgressive, increasing aggression. The Tame-2 genotype also has a major effect on a locus on chromosome 6 (denoted Tame-5), in that Tame-5 affects tameness only when Tame-2 is AA (Figure 5F).
A polygenic basis for tameness:
To uncover the genetic basis for tameness, we analyzed >700 F2 animals produced by cross-breeding two strains of rats that differ drastically in their response to humans. Tameness levels, as measured by the glove test in the F1 and F2 rats, approximate a normal distribution with the mean centered between those of the parental lines (Figure 1), suggesting a polygenic basis for tameness and aggression in the rats (Lynch and Walsh 1998).
The linkage study confirms this. The largest QTL, Tame-1, explains 5.1% of the residual phenotypic variance, while the remaining loci each explain smaller fractions when considered individually. These estimates are in line with the generally small effect sizes reported for QTL for other rodent behaviors (Flint 2003). Across 45 measures and three principal component scores, our study identified 23 significant and 125 suggestive QTL. We note that this greatly exceeds the number of QTL expected to be observed by chance. For 48 traits, we expect <3 significant QTL at a genomewide significance level of 5% and 48 suggestive QTL with a significance level of 5% at each of 20 chromosomes. Given the sample size of >700 F2 animals, we consider it unlikely that other unidentified loci with large individual effects exist in these lines.
Overlap of QTL for tameness-associated traits:
Previous studies have identified a multitude of phenotypic differences between the tame and the aggressive lines of rats (Naumenko et al. 1989; Plyusnina and Oskina 1997; Popova et al. 2005; Albert et al. 2008). However, the behavioral response to humans was the only criterion used during selection. Are the loci influencing the nonselected traits the same as those contributing to the difference in tameness?
If phenotypic traits are influenced by the same loci one would expect them to show some degree of correlation. It is thus noteworthy that correlations between the traits we measured in the F2 population were weak at best, often failing to reach significance in spite of the fact that hundreds of animals were analyzed and that some of the uncorrelated traits were markedly different between the parental strains (Albert et al. 2008). However, the power to detect a correlation caused by shared loci may be limited given that the effect sizes associated with the alleles are small and perhaps obscured by nongenetic influences.
In the QTL analyses, a number of traits mapped to the same regions. This is especially obvious on chromosome 1 where weight of the adrenal gland maps to a region overlapping Tame-1 with virtually identical confidence intervals (Figure 3B). It thus seems plausible that alleles of a single gene with pleiotropic effects underlie both Tame-1 and adrenal gland size variation. Alternatively, a causal relationship might exist between tameness and adrenal gland size. For example, sudden increases of plasma corticosterone, which is produced by the adrenal cortex, promote aggressive behavior in rats (Kruk et al. 2004), while chronically high levels of glucocorticoids seem to inhibit aggressive behavior in several vertebrate species (Summers et al. 2005). However, postmortem corticosterone levels did not map to any locus linked to tameness or aggression. It is thus equally possible that other hormonal activities of the adrenal glands are directly or indirectly linked to the alleles that affect tameness in the rats. The identification of the gene or genes underlying Tame-1 as well as adrenal gland size variation will eventually clarify this.
Tame-1 was obtained by mapping a principal component score summarizing, among other traits, attacks, screaming, flights, and the toleration of touch (Table 2). The fact that, at Tame-1, these individual behaviors yielded QTL apparently specific to one sex (Table 3) may suggest that the causative alleles underlying Tame-1 influence the two sexes differently. However, since the QTL for the principal component at Tame-1 is found in both sexes, it may be the case that the absence of signals for the individual traits reflects lower power due to using half the number of individuals. In addition, individual traits are likely to have less power than the principal component they contribute to. Further work is warranted to clarify whether Tame-1 and other loci truly act in a sex-specific manner.
To our knowledge, this study is the first genetic mapping of tameness and defensive aggression against humans in any species. However, several studies in rats have identified QTL for traits potentially related to those studied here. Tame-1 overlaps with earlier identified QTL influencing several anxiety-related traits (Terenina-Rigaldie et al. 2003), rearing behavior (Fernandez-Teruel et al. 2002), and adrenal gland weight (Solberg et al. 2006). Tame-2 overlaps with two QTL for activity and anxiety-related behaviors (Terenina-Rigaldie et al. 2003; Conti et al. 2004). It is reassuring that phenotypes similar to some of those studied here show linkage to similar genomic locations. However, in the absence of information on the molecular basis of these QTL, it cannot presently be determined whether alleles at the same genes are responsible.
A genetic network for tameness:
Epistasis affects the expression of numerous traits (Phillips 2008). For behavioral quantitative traits, however, epistatic networks identified by genome scans remain rare (for exceptions, see Wright et al. 2006; Bailey et al. 2008). In this cross, a network of Tame-1 and Tame-2 and three additional loci that were identified only as part of these epistatic pairs influence tameness (Figure 5A).
The additive effect of Tame-1 was robust across genetic backgrounds (Figure 5), suggesting that Tame-1 is a major locus influencing tameness. In contrast, Tame-2 seems to act as a capacitor of other loci. When it is homozygous for the allele from the aggressive line, it magnifies the effects of all three loci it interacts with (Figure 5, C, D, and F). This is reminiscent of epistatic loci underlying growth in chicken (Carlborg et al. 2006). In addition, the effect of Tame-2 itself is highly dependent on other loci. In a tame background, i.e., where the other loci are homozygous for the allele from the tame line, Tame-2 has at most a small effect on tameness.
The epistatic network raises interesting questions about the role of Tame-1 and Tame-2 during selection for tameness and aggression. Due to the relatively invariant effect of Tame-1, it can be selected for in many genetic backgrounds, driving alternative alleles rapidly to fixation. On the other hand, the homozygous aggressive genotype at Tame-2 might have had an initial role in selection by magnifying the effects of other loci, allowing them to become more prominent targets of selection, while the allele from the tame line would have decreased the response of other loci to selection. The increased frequency of homozygous tame genotypes at loci other than Tame-2 will, however, decrease the selective advantage of the tame allele at Tame-2, due to its small effect in this background. Given the intricate interactions between Tame-2 and the other loci, it is an intriguing possibility that Tame-2 might harbor multiple alleles in the current tame and aggressive rat lines. Thus, the single tame and aggressive alleles might in fact be average effects across several alleles. A more in-depth analysis of patterns of polymorphism at Tame-2 and other loci might shed light on this.
White coat color and tameness:
Many domestic animals across a wide range of species are distinguished from their wild relatives by conspicuous coat color variants. Possible explanations include direct selection for coat color variants by humans (e.g., Pielberg et al. 2008) and removal of selective pressures for camouflage. It is further conceivable that coat color variation is a pleiotropic effect of alleles influencing other traits and particularly behavior, including the level of tameness (Keeler and King 1942; Cottle and Price 1987; Hayssen 1997).
The F2 rats provide an excellent opportunity to test whether loci influencing tameness also affect white coat spotting. If the same genes are responsible, F2 animals carrying coat spots should be more tame than those without. However, this was not observed (see results). Similarly, the QTL for coat spotting shows no linkage to tameness or any other trait (Figure 4B), and neither Tame-1 nor Tame-2 is linked to coat spotting (Figure S10). Hence, we find no evidence for white spotting being caused by the same loci that contribute to tameness. Pleiotropic effects linking tameness and coat color may occur in other species, or even in other lines of rats, but such scenarios are clearly not strengthened by these results.
It is noteworthy that the QTL for white coat spots on chromosome 14 contains at its center the Kit gene (RefSeq NM_022264), which encodes a tyrosine-kinase receptor involved in melanoblast migration (Yoshida et al. 2001). Allelic variants of homologs of rat Kit, or of the gene for the Kit ligand (Kitl; RefSeq NM_021843), are known to cause white coat color variants in mice (Jackson 1994), pigs (Marklund et al. 1998), horses (Haase et al. 2007), and stickleback fish (Miller et al. 2007). Thus, Kit is an excellent candidate for causing the white coat spots in the rats studied here.
We present a genetic analysis of traits associated with tameness in a rat model of animal domestication. Tameness is found to be influenced by two major loci, which are part of a five-locus epistatic network. A possibility not explored here are epistatic interactions involving more than two loci. Such interactions are, however, very difficult to detect given the sample size limitations in mammals.
The confidence intervals for the two tameness loci contain 744 (Tame-1) and 339 (Tame-2) genes annotated in the Ensembl database, respectively. Since few genes underlying QTL for any behavior have been identified (for two notable exceptions, see Yalcin et al. 2004 and Watanabe et al. 2007), and none of them are located in Tame-1 or Tame-2, it seems premature to speculate about what genes underlie Tame-1 and Tame-2. Rather, fine-mapping approaches such as advanced intercross lines, as well as other approaches, are needed to clarify what genes are involved.
This work greatly benefited from SNP genotyping courteously provided by Ivo Gut and Marc Lathrop at the Centre National de Genotypage, Genome Institute of the French Atomic Energy Commission, Evry, France. We are indebted to Lysann Wagner and Claudia Nestler for experimental assistance; to Sandra Peter, Petra Madaj-Sterba, and Gudrun Lemm for assistance with animal maintenance; and to Inger Jonasson, Jenny Jonsson, Katarina Davidsson, and Ulf Gyllensten at the Uppsala Genome Center for help with genotyping and for being great hosts to F.W.A. in Uppsala. We thank two anonymous reviewers for their constructive comments on the manuscript. This work was funded by the Max Planck Society. The work by Ö.C. and F.B. was funded by the Swedish Foundation for Strategic Research and the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning.
Supporting information is available online at http://www.genetics.org/cgi/content/full/genetics.109.102186/DC1.
Communicating editor: A. D. Long
- Received February 25, 2009.
- Accepted April 3, 2009.
- Copyright © 2009 by the Genetics Society of America