SEARCH

SEARCH BY CITATION

Keywords:

  • environmental heterogeneity;
  • heritability;
  • personality;
  • population density;
  • selection;
  • temperament

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

1.  Temperament traits increasingly provide a focus for investigating the evolutionary ecology of behavioural variation. Here, we examine the underlying causes and selective consequences of individual variation in the temperament trait ‘exploration behaviour in a novel environment’ (EB, based on an 8-min assay) in a free-ranging population of a passerine bird, the great tit Parus major.

2.  First, we conducted a quantitative genetic analysis on EB using a restricted maximum likelihood-based animal model with a long-term pedigree. Although repeatability was relatively high, EB was only moderately heritable and permanent environment (VPE) effects contributed as much to phenotypic variance as additive genetic effects.

3.  We then asked whether heterogeneous selection acted on EB at various temporal and spatial scales. Using estimates of lifetime reproductive success, we found evidence of weak negative directional selection acting on EB amongst females which was driven by selection through recruitment, but not fecundity, in one of the four breeding years. There was no evidence of any selection on EB through survival.

4.  Heterogeneous selection on EB within seasons was also observed amongst males through fecundity along two fine-scale environmental gradients – local breeding density and habitat quality; we are unaware of any previous equivalent demonstrations.

5.  All of these analyses were repeated on a second measure of exploration behaviour (EB2, measured during a 2-min assay) to facilitate comparison with other studies. EB and EB2 were strongly correlated to one another at the genetic level, but were only moderately correlated at the phenotypic level and VPE was undetected in EB2. Selection on EB2 was similar to that on EB; we conclude that both traits are broadly equivalent from an evolutionary perspective.

6.  Our analyses suggest that to the extent that the temperament trait ‘exploration behaviour’ is subject to natural selection in this population, this selection is highly context dependent and most evident along two environmental gradients. Furthermore, the strong VPE effect detected suggests that understanding the causes and consequences of variation in this trait will require studies firmly embedded in an environmental context.


Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Identifying and understanding the mechanisms that contribute to the maintenance of variation within populations is a central aim in evolutionary biology (Roff 1997; Futuyma 1998). For decades, phenotypic variation has been studied intensively in the wild and in captive populations, revealing a variety of genetic and environmental causes of variation. A great majority of these studies have focussed on morphological and life-history traits that are easily quantified and in the field; physiological and behavioural traits, in contrast, are far more difficult to measure under standardized conditions and have not yet received the same level of detailed investigation, particularly in an evolutionary ecological context (Endler 1986; Mazer & Damuth 2001; Merilä & Sheldon 2001). Until the 1990s, there was a tendency to treat individual behavioural variation as noise rather than as an important consequence of evolutionary processes, and to assume that selection acts only with respect to a single behavioural optimum (Wilson 1998; Sih et al. 2004b; Smith & Blumstein 2008). There is now growing acceptance that individual differences in behaviour with respect to traditionally assumed optima can reflect consistent differences between individuals, can have a genetic basis and may be the target of selection in natural environments (Wilson 1998; Sih et al. 2004b; Smith & Blumstein 2008). How this variation arises and is maintained within populations are questions that are especially interesting in the context of behaviour. Lack of individual plasticity in morphological traits can be easily understood, but why natural selection does not always generate unlimited behavioural plasticity remains a contentious issue (Sih, Bell & Johnson 2004a; Sih et al. 2004b).

Heterogeneous selection is one of the main ways in which individual variation can arise and has been demonstrated in most major taxa, for example, in plants (Huber et al. 2004), microbes (Slater et al. 2008) and algae (Jormalainen & Honkanen 2004). In animals, heterogeneous selection in an ecological context has been demonstrated primarily in relation to morphological and life-history traits (McDonald, Fitzpatrick & Woolfenden 1996; Coltman et al. 1999). Associations between behaviour and fitness have been widely reported for decades but in a vast majority of cases neither the repeatability nor the nature of the underlying phenotypic variation have been characterized, restricting the ability to draw firm conclusions about the evolutionary consequences of such variation (Sih et al. 2004a). Applying the ‘phenotypic gambit’, that is, assuming phenotypic variation correlates with genotypic variation, is increasingly viewed as problematic, especially for behavioural and life-history traits that tend to have low heritability (Hadfield et al. 2007).

Intrinsic factors underlying heterogeneous selection commonly include sex, age and ontogeny, whereas extrinsic factors relate to a variety of ecological factors varying along temporal or spatial gradients. Two environmental gradients of general relevance are population density and habitat quality. Negative density dependence is common in nature (Newton 1998), but the importance of population density in the context of selection on individual behavioural variation with a heritable basis is poorly understood. Some studies reveal heterogeneous selection by comparing selection patterns on traits in low and high population density years (Pemberton et al. 1996; Coltman et al. 1999), but behavioural data are generally not collected for sufficient time to generalize about the fluctuating patterns observed. One way to circumvent this constraint is to explore whether heterogeneous selection acts on behavioural variation with respect to individual variation in local breeding density within cohorts or samples, an approach that has been more commonly used on immobile organisms (e.g. Stratton & Bennington 1996, 1998). However, even for mobile organisms, local variation in breeding density may have different effects on individuals within populations (e.g. Wilkin et al. 2006). Habitat quality is another important environmental gradient in nature (Newton 1998) but again evidence for heterogeneous selection acting on behavioural phenotypes with a known genetic basis in this context is absent. Although there is much evidence for fitness variation at the habitat level depending on the behavioural strategies used, the causes of variation in these strategies remain largely untested.

In recent years, temperament traits have been used increasingly as a framework for studying individual behavioural variation (Dingemanse et al. 2004; Sih et al. 2004b; Reale et al. 2007; Biro & Stamps 2008). Temperament traits can be easily measured under standardized conditions, show consistency across an individual’s lifetime, sometimes covary with ecologically important traits and have often been shown to have an additive genetic basis (reviewed in Reale et al. 2007). Suites of correlated behavioural traits in a population are referred to as a behavioural syndrome (Sih et al. 2004b), made up of individuals with different personalities or coping styles of limited plasticity. The terms temperament and personality are frequently used interchangeably but here we use ‘temperament’ to refer to a specific class of behavioural trait – for example boldness, aggressiveness, activity levels or risk aversion (see Reale et al. 2007) – and ‘personality’ to be inclusive of all behavioural traits. Initial theoretical models suggest that the limited plasticity associated with behavioural traits generally may be evolutionarily stable because of heterogeneous selection caused by environmental stochasticity, or variation in individual state, coupled to the proximate costs of maintaining complete plasticity (Tufto 2000; Dall, Houston & McNamara 2004; Sih et al. 2004a; McElreath & Strimling 2006). Alternatively, heterogeneous selection could also arise as a result of spatial heterogeneity in the environment (Wilson 1998; Sih, Kats & Maurer 2003). Of limited evidence from wild populations, one study (of great tits Parus major) showed that selection acts heterogeneously across the sexes and across years (Dingemanse et al. 2004; Both et al. 2005), whereas another reported heterogeneous selection acting on docility across years in bighorn sheep (Réale & Festa-Bianchet 2003). A theoretical modelling approach implies that life-history trade-offs may play a role (Wolf et al. 2007), while a recent review suggests that reproductive selection sometimes acts in the opposite direction to survival (Smith & Blumstein 2008). The degree to which these patterns of selection are widespread in nature, and the temporal and spatial scales at which these patterns occur, are emerging questions in the study of individual behavioural variation.

For phenotypic variation to be evolutionarily relevant it must also be heritable. Heritability estimates for behavioural traits, particularly from wild populations, are comparatively scarce (Merilä & Sheldon 2001) and those available for temperament traits are based predominantly on selection lines, on parent–offspring regression or on cross-breeding experiments (Dingemanse et al. 2002; van Oers et al. 2004a, c, 2005; Reale et al. 2007). Estimates from selection lines are likely to be much higher than those seen in nature (Weigensberg & Roff 1996), while those from parent–offspring relationships are generally also inflated (Knott et al. 1995; Merilä & Sheldon 2001; Kruuk 2004). In the last decade, restricted maximum likelihood (REML) ‘animal models’ have been increasingly used in quantitative genetic analyses of wild populations to overcome many of these limitations. Their strength is due in part to the use of multigenerational pedigree data that allow additive genetic and environmental components of variation to be separated more effectively (Kruuk 2004). Multiple measures from the same individual allows estimation of so-called permanent environment effects, which are otherwise included in the error component of variation (Reale, Festa-Bianchet & Jorgenson 1999; Coltman et al. 2001; Kruuk et al. 2002; McAdam et al. 2002; Kruuk, Merila & Sheldon 2001; Charmantier et al. 2006). Permanent environment effects refer to consistent between-individual differences, over and above those due to additive genetic effects, and include for example, differences caused by conditions experienced in early life, or during the individual’s own reproduction; they also include non-additive genetic effects. To date, most studies that have used animal models have done so on life-history and morphometric traits for which data in long-term studies has been routinely collected for many years, but recent analyses suggest that efficient separation of the underlying variance components is also possible by superimposing data collected latterly on a long-term pedigree (Quinn et al. 2006). Furthermore, errors in pedigree structure are only likely to have significant effects on component estimation in complex models (Morrissey et al. 2007).

Here, we use a natural population of the great tit P. major to quantify the variance components underlying, and selection patterns operating on, a temperament trait, exploration behaviour in a novel environment (henceforth exploration behaviour or EB). Intensive study, on captive and free-ranging populations, has shown that EB in this species can be correlated, across individuals, to a variety of other ecologically relevant behaviours, for example aggressiveness and dominance (Verbeek, Boone & Drent 1996; Dingemanse & de Goede 2004), foraging (Verbeek, Drent & Wiepkema 1994; Marchetti & Drent 2000), stress responses (Carere et al. 2001; Carere & van Oers 2004), risk taking behaviour (van Oers et al. 2004b) and dispersal (Dingemanse et al. 2003). Heterogeneous selection in the wild has been demonstrated across years and sexes, for a single population (Dingemanse et al. 2004; Both et al. 2005). Several experimental studies of other systems suggest that heterogeneous selection along an environmental gradient is likely to be important in maintaining variation in a variety of temperament traits (Sih et al. 2003; Biro, Abrahams & Post 2007; Dingemanse et al. 2007; Brydges et al. 2008a).

In the first part of the study, we conduct a quantitative genetic analysis of EB using a long-term pedigree, estimating both additive genetic and permanent environment components of variation. In the second, we examine whether there is any evidence for directional (linear or nonlinear) selection acting on EB across different temporal and spatial scales, including: (i) over the lifetime of individuals; (ii) across individual breeding seasons, age classes and sexes, the latter scale partially replicates recent work in another population (Dingemanse et al. 2004; Both et al. 2005); and (iii) fine-scale environmental heterogeneity, specifically along local population density and habitat-quality gradients, quantified during reproduction. This has not previously been attempted for any temperament or, apparently, for any other behavioural trait with a known quantitative genetic basis in the wild. As laboratory studies have found correlations with many behavioural traits, and behavioural correlations are likely to vary between populations and over time (Dingemanse et al. 2007), we preferred to make a post hoc assessment of the likely underlying mechanisms in any patterns of selection observed.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

The study population and study area

This work was carried out as part of a long-term study on a population of great tits at Wytham Woods, Oxford, UK. Great tits are socially monogamous birds and, in Wytham, breed almost exclusively in nest boxes distributed throughout the c. 385 ha woodland. Population density has increased markedly over the last 20 years (Garant et al. 2004) and has been relatively high from 2004 to 2007, the period over which most of the data here were collected (271, 396, 342 and 423 breeding pairs respectively). Reproductive success is monitored annually by a team of fieldworkers, and nest boxes checked at least weekly during each spring. The date that the first egg is laid (egg date, where 1 April = 1), breeding adult identity, the number of offspring produced (fecundity) and recruits to the population are all determined using standard methodology (see McCleery et al. 2004 and Supporting Information for further details).

Exploration behaviour assays

A total of 1383 assays of EB were conducted on 1110 individual birds, generating 405 breeding attempts by males and 447 attempts by females of known EB phenotypes at 642 different nests over the four breeding seasons (2004–2007). This represents 23% of all individuals known to have bred over this period, and 45% of all breeding attempts. Assays were conducted during the months of September–March inclusive, from February 2005 to March 2008 (i.e. over four winters). Birds were usually caught with mist-nets at sunflower seed feeders in the woods, weighed, identified and brought into captivity at Wytham field station, c. 2 km from the centre of the wood, for usually not more than 24 h; a minority of birds were caught during nocturnal checks of nest boxes for roosting birds. Birds were individually housed (Appendix S1) and assays were conducted the following morning between 08·00 and 13·00 h. An assay commenced 20 s after a bird was first coaxed from its cage through a trapdoor leading into an adjoining novel environment room and lasted for 8 min, after which the bird was coaxed back into its cage. The assay room was based on the study of Verbeek et al. (1994). The frequency and location of all movements were recorded using a handheld computer, generating 12 behavioural measures (Appendix S1); the first component (PC1) from a principal components analysis had a positive loading for all measures (see also Boon, Reale & Boutin 2007; Martin & Reale 2008), reflecting a combined measure of activity levels and propensity to explore novel objects and areas. PC1 was √ transformed for the quantitative genetic analyses of EB because this led to an approximately normal error distribution. For selection analyses, a GLM of √PC1 was used to generate a single estimate of EB per bird that was unbiased with respect to several fixed effects, including the number of times the bird was assayed (Appendix S1). To facilitate comparison with other studies, Dingemanse et al. (2002) in particular, we repeat several analyses using simply the number of hops and flights in the first 2 min (EB2) of the assay.

Quantitative genetic analyses of exploration behaviour

We used a REML mixed model procedure with a pedigree based on social matings to estimate variance components in EB (and EB2) and their associated standard errors (Knott et al. 1995; Lynch & Walsh 1998; Kruuk 2004). This ‘animal model’ approach separates phenotypes into fixed and random effect variance components by comparing the phenotypes of all known relatives in a pedigree across generations and can accommodate unbalanced data sets (Reale et al. 1999; Milner et al. 2000; Kruuk et al. 2002). We included repeated measures over years to estimate permanent environment effects, which are beyond those due to additive genetic effects, by including individual identity as an additional random effect. Hence, the total phenotypic variance in exploration behaviour was partitioned as follows: VP = VA + VPE + VR, where VA is the additive genetic variance, VPE is the permanent environment effect and VR is the residual variance. The following fixed effects were included in the model: sex, age (juveniles, <1 year old; adults, >1 year old), winter season (e.g. season 1 = October 2004–March 2005) and days since September 1 to account for a within-season temporal trend, which has been described in another population (Dingemanse et al. 2002). To facilitate comparison amongst studies, models were repeated without these fixed effects (see Wilson 2008).

Animal models were run using a full pedigree of great tits (= 46 608 individuals) collected over the period 1985–2006. A total of 1383 assays were undertaken amongst 1110 individuals, of which 713 had at least one generational link in the pedigree. The estimation of variance components, heritabilities in the narrow sense (h2 = VA/VP) and their respective standard errors were performed using the asreml 1.1 software (Gilmour et al. 2002). Statistical significance of variance components with respect to their difference from 0 were assessed by log-likelihood tests, where two times the change in the log-likelihood ratio when the VPE and VA terms were sequentially added to the model was tested against a chi-squared distribution with 1 d.f. Finally, we tested whether EB and EB2 genetically covaried (COVA) or were genetically correlated (rG) in a bivariate animal model. Fixed effects were tested for significance using numerator and denominator d.f. estimated from an algebraic algorithm in asreml 2.0.

Habitat quality and tesselated territory size

Nest boxes varied greatly in terms of their local density and quality of surrounding habitat. In the winter of 2004–2005, the locations of every nest box at Wytham (= 1100) were digitally mapped to ±3 m (Wilkin et al. 2007a). An index of local population density was estimated using a Dirichlet tessellation technique that formed Thiessen polygons around occupied great tit nest boxes in each year (Wilkin et al. 2006; Wilkin, Perrins & Sheldon 2007b), which we call here ‘tesselated territory size’ (or territory size), which is inversely related to breeding density at the level of the individual. Although we do not know how closely correlated this measure is to true territory size, in our population tessellated territory size is significantly related to individual variation in several life-history traits including recruitment rate (see Wilkin et al. 2006). Using the same tessellation technique to form polygons around every potential great tit nest box at Wytham, rather than just those occupied by great tits in a given year, yields a measure of nest box spacing or availability within the environment.

Newly emerging oak tree Quercus spp. foliage supports high densities of the winter moth Operophtera brumata caterpillars. A recent study found that broods near oak trees were provisioned a higher proportion of caterpillars than broods far from oak trees (Wilkin, King & Sheldon 2009) and the degree of synchrony between breeding birds and oak-feeding caterpillars is known to have a significant influence on many fitness components in this species (Betts 1955; van Noordwijk, McCleery & Perrins 1995; Summerville et al. 2003; Wilkin et al. 2007b). Thus, we used the number of oak trees in each tessellated territory as a component of habitat quality (see Wilkin et al. 2007b).

Selection analysis: reproductive fitness

Lifetime reproductive success

Selection analyses based on lifetime reproductive success (LRS) were conducted using two components of fitness: (i) lifetime fecundity – the total number of offspring that survived until 2 weeks of age across all reproductive attempts in the individual’s lifetime; nestling mortality after this stage is negligible and thus this measure is synonymous with the total number of offspring that fledged; and (ii) lifetime recruitment – the total number of offspring produced that survived to join the breeding population across all reproductive attempts in the individual’s lifetime. In order to remove different potential sources of bias, the analyses were repeated on two sets of data – one restricted and the other unrestricted. The former was restricted to individuals who first bred in 2004 or later but did not breed in 2008 and were therefore assumed to have died (= 209 males and 201 females). This removes any bias that might arise if the EB of adults that first bred prior to 2004, but who were still alive in 2004, differed to those from similar cohorts who had died before 2004. It also removes bias resulting from between-year differences in the survival of EB phenotypes born during recent years. We could not test the former source of bias, but the survival analyses below suggested that the second source of bias was unlikely. The second ‘unrestricted’ set of data contained all individuals for which EB was known, irrespective of when they first bred or whether they were still breeding in 2008. This increased the sample size by 45% for males (= 303) and 41% for females (= 284), which we deemed necessary to reduce the likelihood of making a Type II error through limited sample size.

Standardized directional (S′) and nonlinear (c′) selection differentials were estimated using univariate linear and second-order polynomials respectively. We used relative fitness measures (scaled relative to a mean of 1, for annual fecundity and annual recruitment) while EB was standardized to zero mean and unit variance (after Arnold and Wade 1984a, b). Selection estimates and SEs were derived from models with normal errors. The significance of these was tested with an F-statistic against a normal distribution for fecundity but with a Wald (W) statistic against a Poisson error structure for recruits, correcting for underdispersion or overdispersion where necessary.

Within and across years

Initial tests for heterogeneity in selection on EB across years and sexes were conducted on unstandardized data using generalized linear mixed models (GLMMs, with individual as a random effect), using either three-way interactions between year, sex and EB, or two-way interactions between (i) EB and sex across all years or (ii) EB and year within each sex separately. Selection differentials within individual years were estimated and analysed statistically as for LRS (standardizing within years). Heterogeneity in selection along the two environmental gradients (territory size and oak tree abundance, loge and √ transformed respectively) was estimated initially using all data in a GLMM testing for interactions between the gradient, EB and year, with both fecundity (as above) and recruits as fitness measures, and individual as a random effect. Analyses were then conducted on individual years within sexes on standardized measures of fitness and EB. The same EB estimate was used for an individual in all years, as described previously under the ‘Exploration behaviour assays’ section.

Selection analyses: recruitment and survival between breeding seasons

We tested whether selection acted on EB with respect to juvenile survival to the breeding population and survival between breeding seasons for adults. Juvenile survival was established by determining whether birds that were assayed for the first time in their first winter subsequently bred in the population; it does not account for differences in survival immediately post fledging. In the first winter season, birds were assayed late in the year (February–March) and data from this year were excluded. Survival between breeding seasons was conducted for 2005–2006, 2006–2007 and 2007–2008. Use of a mark–recapture model was deemed unjustified because recapture frequency was limited during the winter. Instead we used logistic regression analyses and tested whether dispersal or delayed breeding was likely to have lead to any systematic bias (Appendix S1). Adult great tits rarely move far between breeding attempts (Greenwood, Harvey & Perrins 1978) and breeding dispersal bias is unlikely to affect our analyses of survival between breeding seasons. The occurrence of delayed breeding or gap-years is also unlikely to have biased our results (Appendix S1). Survival analyses for juveniles were conducted using a logistic binary general linear model. For adult survival across breeding seasons, a GLMM was used with a binary response variable. All analyses were conducted using genstat 8.0 (VSN International 2005).

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Components of variation in exploration behaviour

Our animal models detected significant levels of additive genetic variation for both measures of exploration behaviour (EB and EB2) in the population (Table 1), which in turn produced heritability estimates that were significantly different from zero. VPE was equally significant for EB but did not differ from 0 for EB2. EB and EB2 covaried at the genetic level and, although they were relatively strongly phenotypically correlated (= 0·67), at the genetic level they were far more so (rG = 0·91, Table 1). The fixed effects showed that EB and EB2 varied temporally, both within and between seasons. Age had no clear effect but males had faster EB2 scores and had a tendency to have faster EB scores than did females. In summary, these analyses confirm the presence of additive genetic variance in both EB and EB2, which are strongly genetically correlated to one another. Permanent environment and several fixed effects also contributed to individual consistency in behaviour but the relative importance of these varied between EB and EB2.

Table 1.   Components of variation in two different animal models of exploration behaviour (EB) among great tits at Wytham Woods, 2005–2007: EB (√PC1, 8-min assay), EB2 (√hops and flights in 2 min) and the components of covariation COVA between EB and EB2 derived from another model
 EBEB2COVA
Variance ± SE Log L PProportion of total variance ± SE Variance ± SE Log L PProportion of total variance ± SE Covariance ± SE Log L P r
  1. n/a = not applicable. Log L shows the reduction in the log-likelihood ratio when that term was dropped from the model (2 × Log L tested against the chi-square distribution). aVA and VPE in model without fixed effects were 0·0382 ± 0·017 and 0·0519 ± 0·0195 respectively; bheritability; cgenetic correlation; dfemale set to zero; ejuvenile set to zero; B is the parameter estimate; n/p = not provided.

Phenotypic0·197 ± 0·008n/an/a 2·082 ± 0·082n/a  0·431 ± 0·022 n/a0·674 ± 0·015
Random effects 
Error0·113 ± 0·009n/an/a0·573 ± 0·0461·474 ± 0·103n/a 0·708 ± 0·0470·234 ± 0·025 n/a 
PE0·040 ± 0·016a3·5610·0070·201 ± 0·0820·024 ± 0·1640·001·0000·011 ± 0·0790·049 ± 0·0440·6090·270 
Animal0·044 ± 0·015a5·88<0·0010·226 ± 0·075b0·584 ± 0·1608·41<0·0010·281 ± 0·074b0·147 ± 0·0437·895<0·0010·914 ± 0·087c
Fixed effectsFd.f.PB ± SEFd.f.PB ± SE    
Days since 1 September106·51, 1253<0·0010·003 ± 0·000349·551, 1331<0·0010·007 ± 0·001    
Sex2·961, 11750·0860·0423 ± 0·025d7·451, 11630·0070·2109 ± 0·082d    
Age0·011, 13630·917−0·013 ± 0·026e2·691, 13250·1020·090 ± 0·086e    
Season7·752, 1359<0·001n/p5·922, 13330·003n/p    

Selection on exploration behaviour

The numbers of individual great tits that were assayed and recorded breeding in 1, 2, 3 or 4 seasons over their lifetime were 292, 92, 23 and 3 respectively; hence a majority of individuals (71%) bred just once before death (mean ± SD: 1·36 ± 0·62 SD, = 410; using the restricted data set). There was no difference between the sexes in the number of breeding attempts achieved (t1,432 = −0·47, P = 0·638). Most individuals identified as breeding fledged some offspring over their lifetime but only 39·6% of all breeding birds successfully recruited at least one offspring into the breeding population (= 410 individuals).

In the restricted data set, there was no evidence of directional or of nonlinear selection acting on EB, in terms of either lifetime fecundity or lifetime recruitment, for males or for females (Table 2a). In the unrestricted data set, there was evidence of weak nonlinear and weak negative directional selection acting on female EB in terms of lifetime recruitment, but not on male EB (Table 2b). However, the difference between the sexes was not significant (EB × sex, = 2·46, P = 0·117) and overall selection on EB was not significant when both sexes were pooled (EB, = 2·53, P = 0·112). There was no evidence of selection on female or on male EB in terms of lifetime fecundity (Table 2b). Similarly, there was no evidence of any selection on EB2 repeating all of the analyses above (Table S1).

Table 2.   Standardized directional (Si′) and nonlinear (ci′) selection differentials on exploration behaviour (EB) in wild great tits across the lifetime of individuals using (a) restricted and (b) unrestricted data sets
 FecundityRecruitment
MaleFemaleMaleFemale
  1. Two fitness measures were used: the number of offspring produced (fecundity) and the number of offspring recruited (recruitment) to the breeding population. Restricted data set included only birds that were born in 2004 and had died by 2008 (= 201 females and 209 males); Unrestricted included all individuals irrespective of when they were born or whether they had died by 2008 (= 284 females and 303 males). P-values for fecundity were estimated assuming normal error, and for recruitment assuming a Poisson error. aCorrected for overdispersion.

(a) RestrictedSi′ ± SE (P)−0·041 ± 0·036 (0·258)−0·030 ± 0·048 (0·538)−0·073 ± 0·102 (0·450)−0·174 ± 0·121 (0·113)
ci′ ± SE (P)0·001 ± 0·060 (0·985)0·073 ± 0·084 (0·131)−0·216 ± 0·168 (0·155)a0·262 ± 0·210 (0·186)a
(b) UnrestrictedSi′ ± SE (P)−0·022 ± 0·0350 (0·529)−0·057 ± 0·042 (0·174)−0·0003 ± 0·085 (0·997)−0·182 ± 0·098 (0·036)
ci′ ± SE (P)−0·058 ± 0·055 (0·296)0·088 ± 0·037 (0·227)−0·030 ± 0·136 (0·819)a0·309 ± 0·172 (0·048)a

The analysis of fecundity amongst years suggested no heterogeneity in selection acting on EB with respect to year and sex (EB × year × sex, = 2·08, d.f. = 3, P = 0·555); furthermore, there was no evidence for heterogeneity in selection on EB with respect to sex across years (sex × year, = 1·4, d.f. = 1, P = 0·236) or on EB across years within either males (EB × year, = 3·26, d.f. = 3, P = 0·353) or females (EB × year, = 1·99, d.f. = 3, P = 0·574). Standardized selection analyses on the sexes within-individual years did suggest weak nonlinear selection amongst females in terms of fecundity in 2005 (Table 3).

Table 3.   Standardized directional (Si′) and nonlinear (ci′) selection differentials on exploration behaviour in wild great tits within years and sexes using two fitness measures, the number of offspring produced (fecundity) and the number of offspring recruited (recruitment) to the breeding population
 FecundityRecruitment
MaleFemaleMaleFemale
  1. Si′ linear models for directional selection; ci′ quadratic models for nonlinear selection. P-values for fecundity were estimated assuming normal error, and for recruitment assuming a Poisson error. All parameters are from models assuming a normal error. Sample sizes for males were 26, 97, 136 and 142, and 31, 97, 137 and 172 for females in each year respectively. aLog link. All others identity. bCorrected for overdispersion (dispersion parameter varied from 1·55 to 2·9).

2004Si′ ± SE (P)0·070 ± 0·056 (0·220)−0·038 ± 0·058 (0·519)−0·205 ± 0·188 (0·306)0·040 ± 0·155 (0·829)
ci′ ± SE (P)−0·167 ± 0·131 (0·217)0·0574 ± 0·082 (0·488)0·536 ± 0·444 (0·298)−0·138 ± 0·218 (0·597)
2005Si′ ± SE (P)−0·0691 ± 0·045 (0·126)−0·070 ± 0·058 (0·226)−0·196 ± 0·174 (0·221)−0·232 ± 0·187 (0·164)
ci′ ± SE (P)−0·0136 ± 0·071 (0·849)0·191 ± 0·094 (0·044)a−0·116 ± 0·274 (0·497)b0·216 ± 0·356 (0·588)b
2006Si′ ± SE (P)−0·014 ± 0·022 (0·547)0·018 ± 0·033 (0·590)0·065 ± 0·102 (0·546)−0·050 ± 0·110 (0·659)
ci′ ± SE (P)0·026 ± 0·036 (0·481)−0·006 ± 0·058 (0·913)−0·120 ± 0·164 (0·482)b0·118 ± 0·192 (0·553)
2007Si′ ± SE (P)0·141 ± 0·135 (0·294)0·010 ± 0·037 (0·797)0·150 ± 0·167 (0·294)−0·403 ± 0·187 (0·009)
ci′ ± SE (P)0·174 ± 0·190 (0·360)0·074 ± 0·068 (0·277)0·234 ± 0·258 (0·360)b0·936 ± 0·228 (0·001)

The recruitment analysis amongst years suggested heterogeneity in selection acting on EB with respect to both year and sex (EB × year × sex, = 8·47, d.f. = 3, P = 0·037, Poisson error correcting for under-dispersion by a factor of 0·52, GLMM), although the sex × EB interaction was not significant across all years (= 1·93, d.f. = 1, P = 0·165). There was significant heterogeneity in selection between years within females (= 11·58, d.f. = 3, P = 0·009) but not males (year × EB, = 3·91, d.f. = 3, P = 0·271). Standardized selection analyses for individual years and for the sexes separately suggested that this was driven by directional and nonlinear selection acting on female EB in 2007 alone (Table 3; the trend was visualized using binary data, recruited one or more offspring or not, see Fig. 1).

image

Figure 1.  Selection on female exploration behaviour (EB) with respect to the probability that females breeding in 2007 recruited at least one offspring. (a) Raw data with sample sizes and EB categorized into equal percentile groups; (b) nonlinear splines (±1 SE) generated from glmswin 1.0, http://www.zoology.ubc.ca/~schluter/software.html (see Schluter 1988).

Download figure to PowerPoint

Repeating all of the analyses in the previous two paragraphs on EB2 gave slightly different results (linear but not nonlinear selection on females in 2007 using recruits produced, and no evidence of selection at any other stage; see Table S2). In summary, we found limited evidence of heterogeneous selection across the sexes and years. Although there was some suggestion of weak directional selection over the lifetime of females and none over that of males, this was primarily driven by a single selection event on females in just one of the four years.

Heterogeneous selection and habitat amongst males

Analysis of fecundity amongst years provided evidence of heterogeneous selection acting on EB with respect to territory size amongst males (EB × year × loge territory size, = 4·21, d.f. = 3, P = 0·006, log-link function and normal errors, GLMM). Similar analyses performed within individual years suggested that heterogeneous selection occurred in two of the four years (Table 4). To plot these effects, territory size was categorized into three equal-range categories and the relationship between fecundity and EB shown for each category. The plots suggest that, in 2005, fecundity correlated negatively with male EB in large, but not small or medium territories (Fig. 2a). In 2007, similar but weaker effects were observed for males with high EB scores in large and medium, but again not in small territories (Table 4; Fig. 2b).

Table 4.   Heterogeneous selection analyses (gradient ± SE and P) on exploration behaviour (EB) along two environmental gradients: maximum territory size (terr. size, inversely related to local density) and habitat quality (oak trees within territories)
 MalesFemales
FecundityRecruitmentFecundityRecruitment
  1. Parameter estimates for the interaction term between EB and the respective environmental gradients are followed by ±1 SE and by P-values in parentheses. Sample sizes for males were 26, 97, 136 and 142, and 31, 97, 137 and 172 for females in each year respectively. Territory size was loge-transformed.

2004EB × Terr. size−0·134 ± 0·120 (0·278)−0·296 ± 0·390 (0·692)−0·035 ± 0·108 (0·748)−0·474 ± 0·285 (0·107)
EB × oaks−0·078 ± 0·047 (0·114)−0·134 ± 0·163 (0·759)−0·035 ± 0·073 (0·637)0·136 ± 0·197 (0·548)
2005EB × Terr. size−0·239 ± 0·066 (0·001)−0·368 ± 0·272 (0·128)−0·019 ± 0·087 (0·824)−0·265 ± 0·300 (−0·286)
EB × oaks−0·095 ± 0·036 (0·009)−0·093 ± 0·147 (0·376)0·005 ± 0·045 (0·904)−0·089 ± 0·154 (0·482)
2006EB × Terr. size0·016 ± 0·038 (0·669)0·177 ± 0·169 (0·340)0·042 ± 0·057 (0·458)0·107 ± 0·184 (0·488)
EB × oaks0·004 ± 0·018 (0·847)−0·004 ± 0·084 (0·942)−0·053 ± 0·031 (0·084)−0·043 ± 0·100 (0·601)
2007EB × Terr. size−0·123 ± 0·053 (0·023)−0·149 ± 0·309 (0·630)0·016 ± 0·059 (0·783)−0·353 ± 0·302 (0·870)
EB × oaks0·004 ± 0·023 (0·853)−0·106 ± 0·133 (0·428)−0·020 ± 0·030 (0·502)−0·19 ± 0·155 (0·460)
image

Figure 2.  Heterogeneous selection on male exploration behaviour (EB) along a gradient of territory size, which is inversely related to local population density (small territories = higher density) in (a) 2005 and (b) 2007 using the fecundity component of fitness. Territory sizes are categorized on the basis of equal intervals.

Download figure to PowerPoint

Analysing data from all years collectively and controlling for other factors, the EB × loge territory size effect was weaker but occurred across all years (= 3·94, P = 0·047) and was likely to have arisen later in the season when the adults were feeding young (clutch size, = 71·56, P < 0·001; egg date, = 0·39, P = 0·533; age, = 0·00, P = 0·966). It was also independent of habitat quality (oak tree abundance, = 4·56, P = 0·033) and nest-box spacing (distance to the nearest other nest box, = 4·40, P = 0·036), the latter suggesting that lower fecundity amongst relatively fast males did not occur simply because they had to invest more time excluding other birds from adjacent nest boxes.

Turning to habitat quality, a GLMM of fecundity within males suggested heterogeneous selection on EB with respect to habitat quality (EB × year × √oaks, = 11·78, d.f. = 3, P = 0·008, log-link function and normal errors). The within-years analysis showed that the EB × √oaks effect occurred in 2005 alone (Table 4) and Fig. 3 suggests that again males with fast EB phenotypes performed relatively poorly in high-quality territories.

image

Figure 3.  Heterogeneous selection on male exploration behaviour (EB) in 2005 along a gradient of habitat quality (oak tree abundance within territories) using the fecundity component of fitness. Habitat quality was categorized into three levels (low, medium and high oak abundance) on the basis of equal percentiles.

Download figure to PowerPoint

Generalized linear mixed models of recruitment showed no evidence for heterogeneity in selection along either gradient for males (EB × year × √oaks, = 3·5, d.f. = 3, P = 0·321; EB × year × loge territory size, = 5·2, d.f. = 3, P = 0·158; both models with Poisson errors corrected for underdispersion; see also Table 4). All of the above analyses were then repeated using EB2; the only difference was that the EB × loge territory size effect on fecundity seen in 2007 was not present for EB2 × loge territory size (Appendix S1, Table S3).

Heterogeneous selection and habitat amongst females

For females, there was no evidence of heterogeneous selection on EB along any environmental gradient using either of the two fitness measures (using recruits and fecundity respectively: EB × year × √Oaks, = 1·25, d.f. = 3, P = 0·740 and = 1·03, d.f. = 3, P = 0·793; EB × year × loge territory size, = 7·35, d.f. = 3, P = 0·062 and = 0·35, d.f. = 3, P = 0·949). The within-years analyses gave similar results for both EB (Table 4) and EB2 (Table S3). The only exception was that a marginally significant EB2 × oak abundance effect on fecundity was found for females in 2005 but had not been detected for EB.

Juvenile and adult survival between breeding attempts

Forty-four per cent (= 486) of juvenile females and 39·9% (= 451) of juvenile males that were assayed during winter survived to breed at least once. There was no evidence for heterogeneity in selection on EB with respect to juvenile survival to their first breeding season across the sexes or seasons (EB × sex × year, inline image = 1·17, P = 0·318, logistic binary regression). None of the three variables on their own, or in any of the three possible two-way interactions, had any significant effect on juvenile survival. Controlling for when birds were first assayed during the season (days since 1 September) did not affect the results. Similar conclusions were reached from standardized selection analyses on each sex within each year (Table S4).

There was no evidence of heterogeneity in selection acting on EB in terms of survival between breeding attempts (GLMM of EB × year × sex, = 1·06, d.f. = 2, P = 0·587, controlling for age = 4·77, d.f. = 1, P = 0·029, and age2= 6·39, d.f. = 1, P = 0·012). Standardized selection analyses on EB within breeding seasons and sexes gave similar results (Table S5).

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

An increasing number of studies have suggested that temperament traits are both heritable and correlated to a range of ecologically important behaviours, and yet the extent to which they are under natural selection is largely unknown (Gosling & John 1999; Gosling 2001; Sih et al. 2004a, b; Dingemanse & Réale 2005; Bell 2007; Reale et al. 2007; Biro & Stamps 2008). Our results suggested that exploration behaviour is influenced by both additive genetic and permanent environment effects, and is under temporally and spatially heterogeneous selection, although selection occurs sporadically, weakly and is usually highly context dependent.

Components of variation

Behavioural traits typically show modest levels of additive genetic variance in populations leading to relatively low heritabilities (Merilä & Sheldon 2001); our heritability estimate of 0·22 for exploration behaviour is in agreement with this generalization and is at the lower end of the other published heritabilities for exploration behaviour in a wild population (0·22 for mid-parent–offspring regression to 0·61 for sibling analysis; Dingemanse et al. 2002) and from selection lines (0·545; Drent, van Oers & van Noordwijk 2003). Although heritability estimates are population-specific, previous such estimates are likely to be inflated because of environmental covariance between parents and offspring which was at least partially controlled for in our analyses, and because of reduced environmental variance in selection lines. Our analysis also detected significant VPE. Permanent environment effects for morphological and life-history traits generally fall within the range of 4–35% of phenotypic variance (reviewed in Charmantier et al. 2006; Kruuk 2004). Our estimate of 27% for EB lies towards the upper end of the range reported in studies thus far and is perhaps particularly high given the low repeatability of this trait.

Permanent environment effects have not previously been estimated for any behavioural trait in the wild using an animal model and cannot be easily estimated using regression-based methods. Dominance and epistatic effects are encompassed by VPE and although a recent review suggests that they tend to be largely insignificant for complex traits (Hill, Goddard & Visscher 2008), there is some evidence that epistatic effects can be important in behavioural traits, especially under specific environmental conditions (Sambandan et al. 2006). Empirical studies from captive populations suggest that early nutrition may be an important environmental component of variation in behaviour and could therefore contribute significantly to VPE (Eising & Groothuis 2003; Daisley et al. 2005; Groothuis & Carere 2005a; Groothuis et al. 2005b; Arnold et al. 2007). It has even been suggested that effects of early nutrition may provide females with a mechanism to manipulate the phenotype of their offspring by modifying, for example the proportion of Taurine-rich food spiders in the diet of their nestlings (Arnold et al. 2007). Nevertheless VPE can theoretically arise at any time during an animal’s life and it is also likely that genotype × environment (G × E) effects are important, for example those associated with the quality of the breeding environment. Establishing the extent to which these sources of variation occur in temperament traits and behavioural traits generally remains a challenge.

Despite their clear advantages over previous methods of quantitative genetic analysis in wild populations, results from animal models need to be interpreted with care. In particular, failure to correct pedigrees in systems where extra-pair paternity is high can underestimate additive genetic variance, especially with traits of low heritability (Charmantier & Réale 2005). Simulation and empirical work both suggest that the relatively low rate of extra-pair paternity in our population (18%; Patrick 2008) should lead to an underestimate of heritability by no more than 15% (Charmantier & Réale 2005). Rates of other types of errors generally found in pedigrees are also thought to be unlikely to lead to significant biases in relatively simple animal models (Morrissey et al. 2007).

Temporal and sexual heterogeneity in selection

Evidence for heterogeneity in selection on temperament traits in the wild is very scarce (reviewed in Dingemanse & Réale 2005; Smith & Blumstein 2008). Here, we report evidence for such heterogeneity from several sources in our population, although overall the effects were not pronounced. Weak negative selection on exploration behaviour over the lifetime of individual females was detected when recruits to the population was used as a fitness measure. Although the result was only significant in analyses containing all birds and not in the restricted data set, we suggest that this was probably due to the latter’s smaller sample size because effect sizes were similar between analyses on the restricted and non-restricted data sets (Pearson r = 0·868, P = 0·005, d.f. = 7; combining effects for both linear and nonlinear terms for both fitness measures). We know of only one other similar lifetime selection analysis on a temperament trait in the wild (Duckworth 2008), which found negative selection on aggressiveness over the lifetime of western bluebirds Sialia mexicana.

Our within-years analysis using lifetime number of recruits as a fitness measure also found negative selection on female EB, and additionally evidence for nonlinear selection in the same year, although in only one of four years of study. The possibility that these effects are caused by a Type I error due to multiple testing cannot be discounted but seems unlikely as the analysis using data from all years collectively was also significant. These weak selection patterns are similar to those reported in a Dutch population in so far as they were recorded in females and not males, but differ in that there was a tendency for selection to be stabilizing amongst females in all years in the Dutch population (Dingemanse & de Goede 2004). There was only weak evidence of nonlinear selection on EB through fecundity from 1 year’s data (Table 3), although this was not significantly different from all other years when no selection occurred. Standardized estimates were not provided from the Dutch population, but nevertheless they found negative correlation against exploration behaviour, amongst those females that laid early (Both et al. 2005), a pattern that is not observed in our data (Quinn et al. unpublished data). Finally, we found no evidence of heterogeneity in selection in terms of whether juveniles assayed in the winter survived to breed the following spring, or whether breeding adults returned to breed in subsequent years, which contrasts with the annually fluctuating selection patterns reported in the Dutch population (Dingemanse & de Goede 2004).

Although there are some similarities in selection patterns between the Dutch and our population, selection seems to be more pronounced in the former. This could be explained by at least two processes. First, the behavioural traits underlying EB could differ between the populations, so that the presumed correlational selection between EB and the functional (but unmeasured) behavioural trait that led to the selection patterns observed in the Dutch population did not occur at Wytham. Genetic correlations constrain evolutionary trajectories within populations but are reversible (Roff 1997). Several inter-populational studies suggest that behavioural correlations might break down depending on local conditions (Giles & Huntingford 1984; Bell 2005; Dingemanse et al. 2007; Brydges, Heathcote & Braithwaite 2008b). A second possible reason for the difference between the studies is that the within-year selection patterns observed in the Dutch population may be caused by covariance between EB and an environmental factor that does not occur in our population because of differences in habitat between populations. Methodological differences are unlikely to explain contrasting effects between populations for several reasons, which are as follows. (i) Our estimate of EB was highly repeatable and was correlated at the genotypic and phenotypic level to similar measures of EB used in the Dutch study. (ii) The analyses on EB2, which is very similar to the measure used in other studies, gave rather similar patterns to those of EB. (iii) The within-seasonal increases in EB detected in our animal model have also been detected in the Dutch population (Dingemanse et al. 2002).

Therefore, we draw two initial conclusions from our results and those from similar published studies. The first is that predicting selection patterns acting on temperament traits on the basis of those found in other populations is likely to be extremely difficult, although explaining differences between populations will undoubtedly reveal interesting phenomena. Predicting selection patterns accurately is less likely to be a problem where the underlying functional significance of the temperament trait is implicit, as in the case of, for example aggressiveness (Duckworth 2006, 2008; Duckworth & Badyaev 2007), or activity/foraging levels (Biro et al. 2006; Biro & Stamps 2008). The second broad conclusion is that identifying the appropriate temporal or spatial scales of selection is also likely to be problematic. In our population, selection occurred most obviously in the last of four seasons, although incorporating fine-scale environmental data revealed more evidence of selection. Field research programmes are often of shorter duration and may fail to find any selection at all unless an environmental or some other gradient is included in the analysis.

Heterogeneous selection in a variable environment

Variable selection along fine-scale environmental gradients is probably ubiquitous in nature, and direct evidence for its importance in wild animal populations comes primarily from long-term studies on morphological and life-history traits (e.g. Wilson et al. 2006; Charmantier et al. 2008). Previous studies on temperament traits suggest that heterogeneous selection across years is probably linked to yearly changes in environmental conditions, perhaps due to food availability (Dingemanse et al. 2004) or predator abundance (Réale & Festa-Bianchet 2003), although in these studies the number of years for which data are available have been too small to test these hypotheses. Several experimental studies suggest that, in species with indeterminate growth, consistent differences in foraging behaviour arise because of a trade-off between growth and mortality coupled to within-population spatial heterogeneity in either food availability, predation risk or both (Werner & Anholt 1993; Anholt & Werner 1995, 1998). Others have inferred that inter-population variation in temperament phenotypes is caused by differences in these same factors (Giles & Huntingford 1984; Bell 2005; Dingemanse et al. 2007; Brydges et al. 2008b). Our results broadly agree with the general hypothesis that environmental heterogeneity influences selection on temperament traits; heterogeneous selection on EB occurred amongst males with respect to local population density and habitat quality. However, we are unaware of any other study that shows heterogeneous selection at such a fine scale on a behavioural trait with known additive genetic variance.

At least three possible mechanisms could explain selection against exploration behaviour in areas of low density or high habitat quality in our population. The first is that the phenotypic correlation between EB and dominance/aggression seen in other populations of great tits (Verbeek et al. 1996; Dingemanse & de Goede 2004) may play a role in Wytham. If these correlations are strong, this could lead to excessively high levels of territorial defence in large territories, the so-called ‘carry-over’ hypothesis used to explain apparently maladaptive behaviour (Arnqvist & Henriksson 1997). However, when we controlled for nest box availability, and therefore also the likely encounter rate with neighbouring males during territory settlement, it made no difference to the selection gradient suggesting that the amount of time spent being aggressive may not be related to selection, although it remains possible that males may have been defending their nests from non-breeding individuals, the distributions of which are independent of nest box availability. The second potential explanation is that foraging efficiency may have been influenced by an interaction between exploration behaviour and territory size, and specifically that individuals with relatively high EB scores were less capable of finding food in larger, more complex territories. This remains to be tested but is known to occur amongst temperament phenotypes in fish (Hojesjo et al. 2004). It is perhaps noteworthy that selection against fast males in low-density areas occurred to some extent in 2007 but much more strongly in 2005 when breeding success was on average relatively low in the population (mean ± SE’s number of young ringed for 2004–2007, respectively, were 7·12 ± 0·32, 5·96 ± 0·24, 7·22 ± 0·19, 7·29 ± 0·20; F3,639 = 7·8, < 0·001). This raises the suggestion that negative selection on male EB along the environmental gradient occurs primarily in poor breeding years, although clearly data from several more years are needed to generalize. Finally, individual differences in mating strategies and an association between rate of extra-pair social encounters and habitat quality or territory size may also play a role. Rates of extra-pair paternity are linked to great tit temperament traits in another population (Van Oers et al. 2008) and there is evidence that similar behavioural correlations occur in our population (Patrick 2008).

Conclusions

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Three broad conclusions can be drawn from our results. First, given the links between exploration behaviour and fitness, our results support the idea that individual behavioural plasticity is at least to some extent limited within individuals, with respect to either the temperament trait itself or to other correlated traits underlying the fitness patterns observed. Second, our results provide support for the hypothesis that variation in temperament traits might be maintained by heterogeneity in selection. Ultimately, predictive models will need to account for the influence of factors such as migration on mean phenotypic expression of EB in the population, to establish the extent to which selection acts on the genotype as well as the phenotype, and to accommodate the role of G × E effects in determining levels of additive genetic variation in the population. Meeting all of these requirements simultaneously remains a challenge for studies in natural populations generally, and on behavioural traits, especially for studies on temperament traits if the weak selection patterns reported here and elsewhere are typical. It is also possible that frequency-dependent selection plays a role in maintaining variation in EB because, for example, the outcome of competitive interactions between temperament phenotypes depends on their relative frequency (Carere et al. 2001; Dingemanse & de Goede 2004). Finally, our results also have implications for the understanding of how populations adapt to environmental heterogeneity. Temperament phenotypes amongst several fish species are thought to be adapted to discrete niches within populations (reviewed in Wilson 1998) and the spatially heterogeneous selection detected in our study suggests that this may also be the case for EB in our study species.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

This work was supported by a BBSRC studentship to SP, and TAW was supported by a NERC studentship and a NERC grant and by grants from the Schure-Beijerinck-Popping Fund and NWO to SB. Thanks to D. Wilson for bird husbandry and to the following for contributing to fieldwork in various ways: D. Cram, H. Griffiths and the Wytham great tit fieldworkers who over the last 40 or so years have generated the pedigree. Thanks also to A. Charmantier, A. Wilson and J. Hadfield for advice on animal models, to C. Cornwallis and T. Pizzari for helpful discussion and to L. Kruuk for comments on the manuscript.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Conclusions
  8. Acknowledgements
  9. References
  10. Supporting Information

Appendix S1. Further method details.

Table S1. Standardized directional (Si′) and nonlinear (ci′) selection differentials on exploration behaviour EB2 (based only on the number of hops and flights in the first 2 min of the 8-min assay) in great tits across the lifetime of individuals using (a) restricted and (b) unrestricted data sets

Table S2. Standardized directional (Si′) and nonlinear (ci′) selection differentials on a modified form of EB (EB2, based only on the number of hops and flights in the first 2 min of the 8-min assay) in wild great tits within years and sexes using two fitness measures, the number of offspring and the number of recruits to the breeding population

Table S3. Heterogeneous selection analyses on exploration behaviour (EB2) along two environmental gradients: maximum territory size (Terr. size, inversely related to local density) and habitat quality (oaks within territories)

Table S4. Juvenile survival to recruitment and exploration behaviour (EB): results from three generalized linear models relating survival to EB within sexes and across three breeding seasons

Table S5. Adult survival between consecutive breeding seasons

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
JANE_1585_sm_Supmat.doc139KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.