Introduction

Classically traits are split up in more or less strict categories: physiological, morphological, behavioural and life-history traits. Morphological and life history traits have had most attention, since they are easily measurable and their variation is very obvious. Morphological traits comprise characteristics that influence the appearance of an individual (eg colour or bone size) where life-history traits are traits that play a direct part in reproduction and survival (eg sex ratio or clutch size; Stearns, 1992).

The strength of the relation between fitness and a particular trait category is often assumed to be negatively correlated with its heritability (Merilä and Sheldon, 2000) since selection is assumed to erode additive genetic variation (Houle, 1992; Stirling et al, 2002). Empirical studies have shown that life-history traits, which one assumes are closely related to fitness, have lower heritabilities than for example, morphological traits (Mousseau and Roff, 1987; Houle, 1992; Merilä and Sheldon, 2000), which seems to confirm this hypothesis. Although behavioural ecologists consider many behavioural traits to be closely related to fitness, the link between fitness and behavioural traits is often unclear (Houle, 1992). A recent study of Stirling et al (2002) showed that heritabilities of behavioural traits are not different from heritabilities of life-history traits, but smaller than heritabilities of morphological traits, suggesting that behavioural traits are as closely related to fitness as life-history traits. One major problem is that these low heritabilities could be caused by an erosion of genetic variation, by selection (elimination–selection hypothesis Houle, 1992) or by a disproportional increase in residual variation (disproportional compounding hypothesis, Houle, 1992; Merilä and Sheldon, 2000; Stirling et al, 2002). The residual variance equals the remaining variance that cannot be explained by the regression when calculating the heritability (Lynch and Walsh, 1998). In most studies where estimates of the magnitude of components exist, lower heritability is not due to lower genetic variance, but due to a high residual variation.

To get a better understanding of the link between heritabilities and fitness consequences and therefore the evolution of a trait, a good knowledge of the structure of its genetic variation is needed (Van Noordwijk, 1990; Merilä and Sheldon, 1999, 2001; Réale and Festa-Bianchet, 2000). A major advantage of using behavioural traits for these kind of studies, is the possibility of measuring them relatively early in life. Whereas life-history traits are often measurable only later during life, many differences in behavioural traits arise already soon after birth.

Individuals within populations differ consistently in how they react to mildly stressful challenges (Gosling, 2001). Although dependent on the environmental context, the same range of reactions is found independent of sex, age or social status (Wilson et al, 1994). Such behavioural differences are quantified on axes such as the ‘big five’ (openness to experience, conscientiousness, extraversion, agreeableness and neuroticism) in humans (John, 1990), and aggressiveness (Hessing et al, 1993), reactivity (Benus et al, 1991), boldness/shyness (Wilson et al, 1994), temperament (Réale et al, 2000), neophobia (Greenberg and Mettke-Hofmann, 2001) and exploration (Benus et al, 1987) in other animals. Different behavioural, physiological and pharmacological reactions are correlated, indicating that these are fundamental aspects of variation in behavioural organisation. In humans, this is referred to as variation in human personality, in other taxa also as behavioural tendencies, temperaments, syndromes, constructs, styles or strategies (Wilson et al, 1994).

Personality traits are known to influence reproduction, survival and dispersal and therefore fitness (Réale and Festa-Bianchet, 2003; Armitage and Van Vuren, 2003; Dingemanse, 2003). Traditionally, it has been assumed that there is little or no additive genetic variation in traits that influence fitness, as they are supposed to be under strong directional selection (eg Jones, 1987). Quantitative genetic traits are most likely influenced by many loci, with each locus having a small effect on the trait. Therefore, it is important to not consider only additive but also nonadditive sources of phenotypic variation (Falconer and Mackay, 1996), like dominance, genetic maternal effects and sex-dependent expression (Lynch and Walsh, 1998).

In contrast to additive genetic variation, quantitative genetic theory predicts that the relative amount of genetic dominance variation should increase under selection (Mousseau and Roff, 1987; Roff, 1997; Merilä and Sheldon, 1999). This was confirmed in the study of Crnokrak and Roff (1995), who found that levels of dominance variance in life-history traits were higher than those in morphological traits. In contrast, Stirling et al (2002) found in a comparative study that relatively high dominance variation was common in domestic and semi-domestic, but not in natural, populations. Maternal effects are known to be able to play an important role in evolution (Bernardo, 1996; Mousseau and Fox, 1998). Studies on maternal hormones (Schwabl et al, 1997; Eising et al, 2001), on the relation between maternal environment and antibodies (Heeb et al, 1998) and on the influence of females on their offspring sex ratio (Komdeur et al, 1997; Sheldon et al, 1999) show that female birds may control a surprisingly wide range of characteristics of their offspring. Recent theoretical, laboratory work and work in natural populations (McAdam et al, 2002) has suggested that heritable maternal effects can have important influences on the potential of evolution (Wolf et al, 1998; McAdam et al, 2002). Both in humans and in other animals, sex differences in personality traits are reported (Buirski et al, 1978; Budaev, 1999; Benus, 2001). It is therefore quite possible that the expression of the genes is dependant on the gender of the individual.

In our model species, the great tit Parus major hand-reared individuals of both sexes consistently differ in the way they explore a novel environment, and these differences are strongly correlated with differences in behaviour towards novel objects (Verbeek et al, 1994; Drent and Marchetti, 1999; Drent et al, 2002). A heritable component was shown to exist for exploration and boldness in a directional selection experiment (Drent et al, 2003) and in a natural population (Dingemanse et al, 2002; Drent et al, 2003). These individual differences in exploration and boldness have predictive value for differences in risk-taking behaviour (Van Oers et al, 2003), aggressiveness (Verbeek et al, 1996), recovery time and behaviour after lost contests (Verbeek et al, 1999), foraging behaviour (Drent and Marchetti, 1999; Marchetti and Drent, 2000) and reactions to stress (Carere et al, 2001; Carere, 2003; Carere and Van Oers, 2003; Van Oers, 2003). With this system, we are able to conduct a quantitative genetic study on natural variation in several correlated behavioural traits.

In this study, we performed a crossing experiment to estimate additive and nonadditive genetic components, maternal effects and sex-dependent expression of exploration and boldness. Great tits of two lines resulting from a two-directional artificial selection experiment for the extremes of the combination of these traits (‘fast’ and ‘slow’ explorers; Drent et al, 2003) were crossed to produce hybrid F1 and their first backcross generations. By using the two original lines (two groups) and the reciprocals of the F1 (two groups) and first-generation backcrosses (four groups), we have phenotypic means of eight groups. This provides enough data to test the adequacy of genetic models of expected group means containing additive, dominance, maternal effects and sex-dependent expression (Mather and Jinks, 1971; Houle, 1991).

Our aims are (i) to get a better insight into the structure of inheritance of exploration and boldness in a wild bird species, and (ii) to see whether the expression of exploration and boldness depends on offspring sex. We will discuss how our results fit the current theories of the genetic structure and the maintenance of genetic variation in life-history traits.

Materials and methods

Study system

The great tit is a very common monogamous territorial passerine, which breeds in secondary holes and artificial nest boxes in all types of wooded areas throughout Europe and parts of Asia and North Africa (Perrins, 1965). From September of the year of fledging onwards, young males start to claim a territory or individual dominance area on vacant ground between the still existing territories of adult males or on less attractive parts of large territories. Early territory ownership is strongly related to survival, reproduction and thus fitness (Drent, 1983). Males are territorial throughout the annual cycle. During autumn and winter, the spatial intolerance is often replaced by hierarchical intolerance during flocking behaviour with other neighbouring territory owners and their mates and nonterritorial birds, particularly when food is locally unpredictable, scarce or difficult to find.

We breed great tits in semi-open aviaries of 2.0 × 4.0 × 2.5 m3. Birds are paired up in December and breeding pairs are kept in aviaries from December until the end of the breeding season. From September until December, birds are kept in groups of six to eight individuals per aviary, to mimic natural winter flocking. Juveniles are housed individually in standard cages of 0.9 × 0.4 × 0.5 m3 with a wooden bottom, top, sides and rear walls, a wire-mesh front and three perches, as soon as they reach independence. All birds are kept under natural light conditions and have auditory and visual contact with other individuals. We feed the captive great tits with a protein rich mixture, and a commercial seed mixture, supplemented daily with mealworms (Tenebrio molitor) or sunflower seeds, while water is provided ad libitum.

Lines and crosses

All genetic groups (lines and crosses) and their sources used in the analyses are shown in Table 1. The parental groups P1 and P2 were birds from the fourth generation of selection lines for ‘fast’ and ‘slow’ early exploration, respectively (Drent et al, 2003). During these four generations of selection no full or half mate or first cousin matings took place, to keep the level of inbreeding low. To obtain the reciprocal F1 crosses (F1 and F1R in the table), we mated birds from both lines (P1 × P2 and P2 × P1; in which the female is always the first in the combination) with a total of nine pairs each. Of the available F1 offspring, 36 birds were mated with both P lines, forming two backcross combinations (B1 and B2) and their reciprocals (B1R and B2R). For the analyses of sex-dependent expression, the F1 and F1R crosses were split up into males (F1m and F1Rm) and females (F1f and F1Rf). Since we did not have the sex of all individuals, F1m and F1f, and F1Rm and F1Rf, do not add up to F1 and F1R, respectively.

Table 1 Sources of the groups (genetic lines and crosses) used in the analyses

To avoid the effects caused by the parental environment as much as possible, eggs were collected daily before 0900 h, and replaced with dummy eggs. Eggs were stored in a separate room, in a machine that turned the eggs every 2 h. Full clutches were exchanged with clutches of wild females in natural field populations. Breeding in the aviaries is synchronous to breeding in the natural populations. Nestlings were collected from the foster nests at an age of 10 days and then hand reared until independence in the lab (for details of hand rearing, see Drent et al, 2003). We tested birds of all groups 35 days after hatching, as described below.

Tests

To measure the early exploratory behaviour score (exploration and boldness), we performed two types of behavioural tests: A novel environment test, conducted in a standard observation room (analogous to an open field test Walsh & Cummins, 1976) was followed by two tests of the reaction to different novel objects conducted in the home cage (Drent et al, 2003). The combination of the novel environment score (further referred to as exploration) and the novel object test score (further referred to as boldness) is referred to as early exploratory behaviour (EEB). EEB was used as the selection criterion in the bi-directional selection experiment of Drent et al (2003). The exploration test was carried out between 30 and 35 days after hatching, the boldness tests 10 and 12 days later.

For the exploration test, five tree-like models (further referred to as trees) were placed in an observation room of 4.0 × 2.4 × 2.3 m3 (Dingemanse et al, 2002; Drent et al, 2003). The time a bird needed to visit the fourth tree was converted linearly to a scale of 0–10. Birds who reached the fourth tree within 1 MIN were given a score of 10, birds that reached the fourth tree within 2 min were given a score of 9, etc. Birds that did not reach the fourth tree within 10 min, received a score of zero. The result of each boldness test was converted linearly to a 0–5 scale, with a score of five when a bird pecked the object and a score of zero when the bird did not reach the perch on which the object was placed within 120 s. The scores for the two novel objects were summed giving a total score of 0–10. The sum of the exploration and the boldness test scores gives the EEB score (Drent et al, 2003; for more details on the tests see Verbeek et al, 1994).

Scaling

To study the relative levels of variation, it is necessary to know whether any differences are simply a consequence of scale (eg Houle, 1992). Our exploration test is measured in a score that has been converted from a time axis, while our boldness score is converted from a combined time and proximity axis. To be reliable, a scale should be chosen, where the variance is independent of the mean, in which case significant differences in observed levels of variance between samples (ie groups) must be attributable to other factors than differences in the mean (Falconer and Mackay, 1996; Lynch and Walsh, 1998). Since the variance increased with the mean in the exploration test and decreased with the mean in the boldness test, the data had to be rescaled. A simple log transformation, as often used in behavioural characters (Falconer and Mackay, 1996; Stirling et al, 2002), is therefore insufficient.

For optimal scaling we used the procedure CATREG, version 1.0 by DTSS, which is available in the statistical package SPSS 10.1 for Windows. CATREG uses categorical regression with optimal scaling, which quantifies categorical data by assigning numerical values to the categories (ie scores), resulting in an optimal linear regression equation for the transformed variables. The procedure treats quantified categorical variables in the same way as numerical variables. Using nonlinear transformations allow variables to be analysed at a variety of levels to find the best-fitting model. CATREG was applied only on the scores of the original lines. Applying it on all groups would artificially lower all variance components other than additive variance. The original scores of all birds (original lines and crosses and backcrosses) were then replaced by the computed scores. Although the analysis with the scaled data changed the exact values of the model parameters, the overall conclusions would have been the same had the analysis been carried out on the original data.

Analysis of group means

The observed group means were analysed following the methods of Mather and Jinks (1971). The observed exploration, boldness and EEB means of the groups were used to estimate parameters, errors and χ2 values of an initial model, using weighted least-squares methods (for details, see Mather and Jinks, 1971; Kearsey and Pooni, 1996; (Starmer et al, 1998 or Gilchrist and Partridge, 1999)). This initial model consisted of an overall mean m and additive [a] and dominance [d] genetic effects (following the notation of Mather and Jinks, 1971; Kearsey and Pooni, 1996). The estimated parameters were then used to calculate expected group means, For each group, the difference between the observed and the expected group means together with the weight values of the group means, was used to calculate a contribution value for the χ2. All contribution values added up to the χ2 value, with the number of group means minus the number of estimated parameters, as the number of degrees of freedom. A significant χ2 would indicate that the expected group means, generated through the model, significantly deviate from the observed group means. This would imply that the model insufficiently describes the observed means.

Adding the parameters of interest to the initial model produces an extended model. We calculated parameter estimates (with standard error) and expected group means as described above. To test whether added parameters in the extended model increased the fit of the model significantly compared to the initial model, we used a likelihood-ratio test (Lynch and Walsh, 1998). A t-test was used to test the significance of the parameter estimates, with the degrees of freedom being the total number of offspring used in the model minus one (Zar, 1999). The significance of the added parameters indicated which parameters could be omitted to simplify the model, where we started with the least significant parameter. When parameters were omitted, the goodness of fit was recalculated. We repeated omitting parameters until the goodness of fit decreased significantly by omitting one more parameter. The model that results from this is referred to as the minimal adequate model.

Three sets of two different models were considered. In the first model (model A) we used the original lines and the reciprocal F1, where the male and female F1 offspring were treated as two groups, to calculate whether or not the expression of EEB, exploration and boldness is sex-dependent. In the second model we used the original parental lines, the reciprocal F1 and the reciprocal backcrosses to test whether an additive maternal effect ([a]m} and a dominance maternal effect ([d]m) are involved in the inheritance of EEB, exploration and boldness. Each model (A and B) was constructed for exploration, boldness and EEB separately, giving a total of six models. The parameter coefficients that we used for the two models are given in Table 2.

Table 2 The parameter coefficients used in model A and B

To test whether parameter estimates differed significantly from each other, t-tests were used (Zar, 1999, p. 124). We performed several t-tests, so a Bonferroni correction would be appropriate for hypothesis testing, as the chance of a significant result increases with the number of tests. Since we did not formally test hypotheses, we did not perform a Bonferroni correction and present original P-values.

Results

The observed group means (scaled) used in both models are shown in Table 3. For all traits (exploration, boldness, EEB) separately, expected group means were calculated from a simple genetic model containing a grand mean (m) and an additive component ([a]) only (maximum-likelihood additive model). The observed group means and the regression lines on the expected group means derived from this maximum-likelihood additive model for all traits are plotted in Figure 1. In no case did this model describe the observed means adequately (minimum χ2=8.20; P<0.05).

Table 3 Mean scaled test scores for the parental, cross and backcross groups, with their SEM and with N=number of individuals for EEB, exploration and boldness
Figure 1
figure 1

Mean observed test score of the lines and crosses, for EEB (a), exploration (b) and boldness (c). Cytoplasm origin of the lines are indicated with: ○=P1 cytoplasm; •=P2 cytoplasm; =group is a mix of individuals with P1 and individuals with P2 cytoplasm. For reasons of clarity, mean values are plotted without standard errors. The lines are the regression lines based on the expected means.

In a first extended model (model A), means were calculated for the separate sexes of the reciprocal F1. An initial model was made in the form TRAIT=m+[a]+[d]. To test whether the expression of exploration, boldness and EEB is sex dependent, we used group means and their standard errors of the original ‘fast’ and ‘slow’ lines and the reciprocal F1. This model described the observed means adequately in the cases of exploration and EEB, but not boldness (Table 4, model A). Both [a] and [d] contributed significantly in all models. As m was scaled around 0, the grand mean is expected to be, and was, equal to zero in all cases. To test sex dependence, this parameter was added to the model. In all cases this was done together with a maternal additive parameter ([a]m), since the observed mean of the F1 differed from the mean of the F1R cross, in boldness (t42.6=2.674, P<0.05). The means did not differ significantly from each other in exploration and EEB, respectively (t50=0.281, P=0.78; t42=0.971, P=0.34). The parameter estimate for [a]m was not significant in either exploration or EEB when running the models in the form TRAIT=m+[a]+[d]+[a]m+[sde], but it was significant in boldness (Table 4, model A). So to test sex-dependent expression, [a]m was included in the model of boldness, but not in the models of exploration and EEB. In neither of the tests was sex-dependent expression found to be significantly different from zero, and the fit of the models did not increase significantly when sex-dependent expression was added (boldness: Δχ12=0.66; P=0.42; exploration: Δχ12=0.02; P=0.88; EEB: Δχ12=0.04; P=0.85). None of the minimal adequate models were significantly different from the observed means.

Table 4 Estimates of composite genetic effects underlying difference in EEB, exploration and boldness

In a second model (model B), the observed means of all available groups were used to estimate the additive ([a]m) and dominant-maternal effect ([d]m). A new base model was made, in the form TRAIT=m+[a]+[d], with the means of all available lines and crosses (P1, P2, F1, F1R, B1, B2, B2R). Since sex-dependent expression was not significant in model A, all groups were combined for sexes. Again, the initial model adequately described the observed means in the cases of exploration and EEB, but this was not the case for boldness (Table 4, model B). The additive and the genetic dominance parameter were significant in all models. To these models, the maternal parameters were added, so a model was formed, in the form TEST=m+[a]+[d]+[am]+[dm]. In all cases dm was the least significant parameter and the fit did not increase in comparison to the same models without dm (boldness: Δχ12=0.55, P=0.46; exploration: Δχ12=0.19, P=0.66; EEB: Δχ12=0.13, P=0.72). After removing dm, all other parameters were significant in the boldness model. Removing am from this model would significantly decrease the fit of the model (Δχ12=9.91, P<0.005), and the maternal effect was relatively small, being about 1/4 of the additive effect. Removing am from this model did not significantly decrease the fit of the model for both exploration or EEB models (exploration: Δχ12=0.27, P=0.60; EEB: Δχ12=0.40, P=0.53). This does not automatically mean that the results for the two tests and the combined test are different. The additive maternal parameter of boldness does not differ significantly from either that of exploration (t137=0.87, P=0.39) or of EEB (t137=0.78, P=0.44), which indicates that the difference was rather in power than in the strength of the effect. Removing any of the other (all significant) parameters would decrease the fit significantly in these models. All expected means generated through the minimal adequate models were not significantly different from the observed means (Table 4, model B). The dominance effect is best seen in the F1 and F1R crossings in Figure 1a, b and c. The mean values all lie beneath the regression line of the additive model, which indicates a dominance effect in the direction of the ‘slow’ line. The strength of the dominance effect lies between one and 1/3 times the additive effect. The maternal effect is best seen in the difference between F1 and F1R, and B1 and B1R. A difference between these means, where the white has the highest value, would indicate that the scores of the offspring of a particular pair are more dependent on the score of the female than on the score of the male.

Discussion

Our results demonstrate that substantial additive and dominance effects are present in both the exploration test and the boldness test, and in the combination of the test scores. In the last decades, the assumption that there is little or no additive genetic variation in traits that influence fitness, has been under discussion (Frank and Slatkin, 1992). Controversy has arisen about implicit assumptions and interpretations of Fisher's ‘Fundamental Theorem of Natural Selection’ (Fisher, 1930). Price and Schluter (1991) and Houle (1992) showed that low heritabilities in fitness-related traits are not automatically caused by low amounts of additive genetic variation, but rather by a high residual variance. Estimates of the additive genetic component turned out not to be different from those in morphological traits, when expressed as a fraction of the mean value. Moreover, it is perhaps unreasonable to assume that selection will constantly act in one direction in variable environments (Roff, 1997). The net selection pressure over a longer time might therefore be low. Existing genetic variation available for adaptation may be protected from selection by fluctuating selection pressures. Examples of this are antagonistic pleiotropy or frequency-dependent selection (Wilson et al, 1994). But alternative theories like selection-mutation equilibrium may also be plausible causes for the maintenance of additive genetic variation in avian personalities (Mousseau and Roff, 1987; see Roff, 1997).

Merilä and Sheldon (1999) pointed out that dominance variance is an important variance component in selection studies. Our results show a substantial dominance effect. Quantitative genetic theory predicts that the relative amount of genetic dominance variation should increase under selection (see Introduction). The combination of the presence of both additive and nonadditive variation in our study could thereby be explained through the existence of substantial selection pressures on these traits, but in variable directions. A strong dominance effect is more likely to occur in traits where variation is due to a relatively low number of variable loci. However, our results have to be seen in the right context. The lines used for the crosses have been selected for four generations with a small population size, and our results are therefore dependent on the animals chosen for the selection experiment. This implies that extrapolations from these results could be unreliable (Hill, 1977).

Significant heritabilities of personalities in great tits are found both in the laboratory (Drent et al, 2003; Van Oers et al, 2003) and in natural populations (Dingemanse et al, 2002). Since h2 represents the ratio of additive genetic variance to total phenotypic variance, and environmental variance is smaller in the laboratory than in field populations, laboratory estimates of heritabilities possibly overestimate natural heritabilities (Riska et al, 1989). Studies show, however, that laboratory estimates provide reasonable estimations of magnitude and significance of heritabilities in the wild (Riska et al, 1989; Weigensberg and Roff, 1996), but any difference may depend on the maternal and dominance effects (Blanckenhorn, 2002). The fact that the laboratory estimate of the realised heritability derived from a regression on the cumulative response to selection as a function of the cumulative selection differential (h2=0.54 Drent et al, 2003) is about twice the estimate of heritability derived from the parent-offspring regressions in natural populations (h2=0.34 Dingemanse et al, 2002) could possibly be caused by the large dominance effect found in this study (Blanckenhorn, 2002). These heritabilities, however, did not differ significantly.

We found no evidence for sex-dependent expression of either of the traits, which is a surprising finding since sex dependant expression is reported in extraversion in humans (Costa et al, 2001). Extraversion is classified as boldness in non-human animals (Budaev, 1999). Other personality traits like aggression in mice are clearly differently expressed in both sexes (Benus, 2001; Sluyter, 1994). In humans, sex differences are found in the main personality axes Agreeableness, Neuroticism and Extraversion. Both biological and social psychological theories try to explain the existence of gender differences in personalities. Some biological theories predict that sex-dependent expression in personality traits arise from innate temperamental differences between the sexes, evolved by natural selection (Costa et al, 2001). Evolutionary psychology predicts that sexes will differ in domains in which they have faced different adaptive problems throughout evolutionary history (Buss, 1995). These are confirmed by some other biological theories, which point to hormonal differences and their effects on personality. Studies on human personalities confirm that the sex differences in androgens during development, cause differences in interests, activities and aggression (Berenbaum and Resnick, 1997; Berenbaum, 1999).

The presence of an additive maternal component in boldness and not in exploration was the only difference we found in the analysed traits. The calculated estimated additive maternal parameters for the different traits are not, however, significantly different from each other, which indicates an effect of the small sample size, and we recognise that, with including the additive maternal parameter, we have reached the limits of the detection ability of the experimental design. The significant maternal component in boldness suggests that maternal effects are likely to play a role in both exploration and boldness. Since we collected the eggs just after laying, the most plausible maternal influence would be through the deposition of substances (eg maternal hormones) in the egg. Our results show that phenotypic expression is likely to be influenced by maternal hormones, but that this is independent of the offspring sex.

Our sample size was relatively small, which is a general problem in these kinds of experiments. Since the betweenline variance is highly dependent on the withinline variance, we would expect interpretation problems when the withinline variances increase rapidly. This increase in withinline variance depends mainly on the number of chromosomes and the map length (Hill, 1977) and less on the population size. Since the number of chromosomes in birds is relatively high, we expect that our initial population size of 18 birds gave us enough power to detect effects that exceed 1/5 of the additive effect.

Although there is no evidence yet for a direct relationship between fitness and personalities, evidence is accumulating that personality traits the affect time of breeding, reproduction, survival and dispersal (Armitage, 1986; Réale and Festa-Bianchet, 2003; Dingemanse et al, 2003; Van Oers, 2003). Our study on the genetic structure of avian personality traits show that these traits have a substantial amount of additive genetic variance, a considerable dominance variance, and that sex-dependent expression is absent.