Introduction

Evolution in wild populations generally depends on natural selection acting on heritable traits leading to adaptation (Endler 1986; Roff 1997). In recent years, there is a renewed interest in describing patterns of natural and sexual selection acting on behavioural traits. Although behavioural traits are considered phenotypically plastic (Dingemanse et al. 2010), their repeatability across temporal or environmental contexts defines them as personality traits (Sih et al. 2004; Réale et al. 2007; Bell et al. 2009; Stamps and Groothuis 2010). As a result, individuals differ from each other in the expression of their personality traits, providing a good substrate for selection (Dingemanse et al. 2004; Dingemanse and Réale 2005; Kingsolver and Pfennig 2007; Smith and Blumstein 2008). Furthermore, a growing number of studies have shown that personality traits have a moderate but significant genetic basis (van Oers et al. 2005a; Réale et al. 2009; Taylor et al. 2012; Poissant et al. 2013; Dochtermann et al. 2014), which is a necessary condition for a trait to respond to selection (Falconer and Mackay 1996; Roff 1997). Yet, to improve our knowledge of the evolution of personality traits, we need more detailed reports of the sources of variation affecting these traits, in combination with estimates of their heritability and of selection patterns acting on them.

Theory predicts that heritable traits under directional or stabilizing selective pressures should display reduced variance (Falconer and Mackay 1996; Kingsolver and Pfennig 2007). However, trait variation is almost ubiquitous in wild populations. A growing number of studies showed that spatially and temporally fluctuating and/or disruptive selection patterns are frequent in nature and are predominant processes allowing the maintenance of phenotypic variance (Kingsolver et al. 2001; Calsbeek and Smith 2008; Hendry et al. 2009; Siepielski et al. 2009; Bell 2010; Siepielski et al. 2013). For example, disruptive and fluctuating selection are acting on beak size in medium ground finch (Geospiza fortis) in Galápagos, leading to the maintenance of two different beak size modes that are specialized for different seed sizes (Grant and Grant 2002; Hendry et al. 2009). Patterns of fluctuating or disruptive selection have also been detected for personality traits, but estimates obtained in the wild are still rare (Dingemanse and Réale 2013; but see Dingemanse and Réale 2005; Quinn et al. 2009; Bergeron et al. 2013; Taylor et al. 2014; Le Coeur et al. 2015).

Variable selection pressures resulting from environmental heterogeneity can also create and maintain variation in personality traits. Factors such as population density and resources availability can affect the expression of personality traits in wild populations (Réale et al. 2010; Webster and Ward 2011). For example in bank voles (Myodes glareolus), variation in population density affects the propensity to infanticide, an aggressive behaviour, with high infanticide at high population density (Korpela et al. 2011). Also, in the mustard leaf beetles (Phaedon cochleariae), boldness differ among nutrient availability conditions, individuals being bolder under low nutrient availability (Tremmel and Müller 2013). Personality can also be affected by sex. For instance, in zebra finches (Taeniopygia guttata), time spent in the presence of a companion at a feeder, a risk-taking behaviour, was different between males and females; females spent more time at the feeder than males (Schuett and Dall 2009).

In this study, we evaluated the repeatability and heritability of docility as well as the ecological and individual factors influencing the expression of this trait and its link with survival in wild eastern chipmunks (Tamias striatus) studied over 10 years. Docility is commonly measured as the animal reaction to the human presence and handling, humans being considered as a possible source of predation or threat, with a potential functional role on survival and fecundity (Réale et al. 2000, 2009; Réale and Festa-Bianchet 2003; Boon et al. 2007; Ferrari et al. 2013; Petelle et al. 2013; Taylor et al. 2014). Docility is part of the coping-style continuum and has been linked to low activity, slow exploration and high stress reactivity (Montiglio et al. 2012; Ferrari et al. 2013). It also reflects the animal reaction towards any risky situation (Réale et al. 2007; van Oers and Sinn 2013). Docility, along with other personality and physiological traits, can be hypothesised to be part of within-population variation in life-history strategies that are maintained by fitness trade-offs between longevity and reproductive success (Wolf et al. 2007; Biro and Stamps 2008; Réale et al. 2009, 2010). In eastern chipmunks, docility is a repeatable trait that is positively correlated with summer hair cortisol concentration (e.g. a physiological trait linked with sympathetic system reactivity) and negatively correlated with exploration behaviour (Martin and Réale 2008; Montiglio et al. 2012; Careau et al. 2015). Studying the factors influencing the expression of docility and the fitness consequences of docility will provide a better understanding of the role of personality in evolution.

Eastern chipmunks experience large fluctuations in food abundance linked with tree masting events (e.g. American beech, Fagus grandifolia) that occur approximately once every 2–3 years (Bergeron et al. 2011a). These events drive chipmunk population density and dynamics and affect the expression of many traits such as torpor and above-ground activity (Landry-Cuerrier et al. 2008; Munro et al. 2008; Bergeron et al. 2011a; Montiglio et al. 2014). We therefore assessed whether the expression of docility and the patterns of selection acting on it could fluctuate with masting events and also if docility is related to survival in this species. In this context, our objectives were to (i) estimate repeatability of docility within and between environmental contexts to confirm its potential as a target of selection, (ii) estimate heritability of docility to assess the extent of its underlying genetic basis, (iii) assess the factors influencing the expression of docility in nature and (iv) describe the patterns of viability selection acting on this trait for adults and juveniles. Based on previous estimations in the population (Montiglio et al. 2012; Careau et al. 2015) and on results from other personality traits in other species (Bell et al. 2009; Taylor et al. 2012; Dochtermann et al. 2014), we predicted that this trait would be both repeatable within and between environmental contexts and heritable. Finally, we predicted that docility would be subject to fluctuating selection associated with the large seasonal and inter-annual variation in resources abundance, or to disruptive selection, as found for another personality trait in the same population (e.g. exploration; Bergeron et al. 2013).

Methods

Study areas and data collection

Eastern chipmunks are small solitary diurnal sciurid rodents that live in burrows where they store seeds from masting trees such as the American beech, the sugar maple (Acer saccharum) and the red maple (Acer rubrum). We monitored wild eastern chipmunks over four different study sites in southern Québec, Canada (45°06′ N, 72°25′ W) from June 2005 to July 2014 (Fig. 1). Individuals were trapped using Longworth traps (Longworth Scientific Instruments, Abingdon, UK) on trapping grids with numbered stakes placed at 20-m intervals and trap set at every two stakes. Site 1 has been trapped from 2005 to 2010, and sites 2, 3 and 4 have been trapped from 2012 to 2014. Site 1 had 228 traps, site 2 and 3 had 84 traps in 2012 and 98 traps in 2013 and 2014, and site 4 had 40 traps in 2012 and 50 traps in 2013 and 2014. All the sites were within 10 km of each other, the habitat was similar between all the sites and there is no significant genetic population structuring at this scale, as previously shown in Chambers and Garant (2010). Sites 2, 3 and 4 were smaller and were not studied the same years as site 1; we therefore combined the data from these sites in all analyses. Chipmunks were trapped from 8:00 am until dusk, in their active period, from April to October each year. Traps were checked every 2 h. Each individual was marked on the first capture with ear tags. Sex, body mass and reproductive status were assessed at each capture. A tissue sample of less than 2-mm diameter was taken from the individual’s ear at first capture and kept in 95% ethanol until extraction, for subsequent DNA analyses.

Fig. 1
figure 1

Location of the four study sites where eastern chipmunks were monitored from 2005 to 2014 in southern Québec, Canada (45°06′ N, 72°25′ W)

Full size image

Following Careau et al. (2010), an individual that was first captured in a reproductive season and that weighed less than 80 g or that weighed more than 80 g but did not show any signs of reproduction (such as developed mammae or scrotum) was considered juvenile. Individuals kept their juvenile status until the next season and were then considered adults. We conducted two trapping periods: ‘early’ trapping period from April to the end of July and ‘late’ trapping period from August 1st to the end of November (Bergeron et al. 2011a, 2013). We defined a reproductive season as a trapping period when there was a reproduction event and emergence of juveniles. Sample sizes for each site are documented in Table s1. It was not possible to record data blind because our study involved focal animals in the field.

Genetic analyses and parentage assignment—site 1

DNA was extracted following a salting out method, and genotypes were obtained using 11 polymorphic microsatellite loci as follows: EACH-1, EACH-3, EACH-4, EACH-6, EACH-8, EACH-10, EACH-11, EACH-12 (Anderson et al. 2007) Chip5, Chip14 and Chip39 (Peters et al. 2007), following the protocol described in Chambers and Garant (2010). Parentage assignments were only performed on site 1 because the data available for the remaining sites did not allow pedigree reconstruction. The software Cervus 3.0.3 (Kalinowski et al. 2007) was used to build a pedigree using assigned dams and sires. We used the same pedigree as in previous studies conducted in the same population (e.g. including 328 individuals, 207 maternities and 172 paternities) and the procedure described in Bergeron et al. (2011b).

Docility measurements

Docility was measured by the handling-bag test which consists in counting the number of seconds spends immobile during 1 min while suspending the chipmunk in a mesh bag in which it was transferred from the capture trap (Martin and Réale 2008; Montiglio et al. 2012). This test is used as a proxy for docility and represents the animal response to human handling (Martin and Réale 2008; Montiglio et al. 2012). An individual that spends most of the time immobile during the test was considered docile. This test was conducted upon most captures for a total of 6392 tests on 664 individuals caught between 2006 and 2010 at site 1, and of 1951 tests on 322 individuals caught between 2012 and 2014 on the other sites (Fig. s1, Table s1). The measurement error on docility was very low, as shown by the high inter-observer correlation = 0.99 (e.g. measures obtained from two or more observers for a given docility score, n = 558 observations).

Data analysis

Repeatability

Repeatability represents an index of individual consistency of a trait over time and provides under certain conditions the highest expected value for heritability (Lessells and Boag 1987; Boake 1989). We estimated repeatability of docility as the ratio of individual-related variance over the sum of individual and residual variance in a model with the handling-bag tests as a response variable and identity of individuals as a random effect (Nakagawa and Schielzeth 2010). We used a Poisson error structure in a Bayesian generalized mixed-effects model using the package MCMCglmm (Hadfield 2010) in the R software 3.0.2 (www.r-project.org). The priors used were the inverse Wishart with an expected variance (v) of 1 and a degree of belief parameter (nu) of 1, which are relatively uninformative priors. We ran each model for 100,000 iterations with a burn-in of 10,000 and a thinning interval (or sampling rate) of 10. Models were inspected for effective sampling size and autocorrelation, and a sensitivity analysis using expanded priors was made (various expanded priors were tested and did not affect the estimates obtained). MCMCglmm allowed us to estimate credibility intervals around the estimate of repeatability. We first estimated adult and juvenile within-year repeatability globally for each site with the exception of juveniles in 2010 where we did not have enough data. Repeatability was then assessed for each sex, trapping period (early or late) and reproductive season. To assess repeatability between environmental conditions at the individual scale, we assessed a global value of repeatability using all data from all individuals, which includes measures taken in different conditions such as mast and non-mast years. Preliminary analysis showed that inter-individual variability in docility becomes null and that docility in general is very high (habituation effects; see Montiglio et al. 2012) after 30 tests, before 8:00 am and after 6:00 pm and for very young juveniles (i.e. <50 g) emerging from their natal burrow. We thus restricted the database to a maximum of 30 handling-bag tests per individual, done between 8:00 am and 6:00 pm and on individuals weighing more than 50 g. Tests with missing data for sex, weight or age were also removed from the analysis. This resulted in the removal of 63 individuals on site 1 and 22 on sites 2, 3 and 4. The final dataset included 4690 tests on 601 individuals for site 1 and 1754 tests on 311 individuals for sites 2, 3 and 4 combined (see Table s1 for details).

Environmental and state effects on docility

Using the same datasets as for repeatability analysis, we ran generalized mixed-effects models in a frequentist framework using the R package lme4 1.1-7 (Bates et al. 2015), to assess the influence of methodological, individual and environmental variables on docility in each site. This frequentist approach allowed us to perform model selections using backward procedures, sequentially removing the least significant term from the model based on its P value (alpha = 0.05) until all remaining effects were significant. We privileged this approach over a Bayesian model selection approach, based on the deviance information criteria (DIC), as this latter method may be problematic for hierarchical models with non-Gaussian data distribution (see Spiegelhalter et al. 2002). Negative binomial error structure models (to correct for slight overdispersion) with the handling-bag tests as dependent variable and individual and observer identities as random effects were used. The significance of each random effect was assessed using likelihood ratio tests (LRT). The full fixed effects part of the model included handling variables (i.e. trial order (lifetime number of test) and time of the day (both linear and quadratic values)), individual state variables (i.e. sex and age class), environmental condition variables (i.e. trapping period, reproductive season, annual population density defined as the scaled number of individuals each year for site 1 and the scaled number of individuals per hectare each year on sites 2–4 (both linear and quadratic values)) and site identity for the analysis of data for sites 2–4. All relevant second-order interactions between environmental and individual states fixed effects were included in the models. Reproduction and population density are closely linked to American beech seed production (Bergeron et al. 2011a), but other resources may affect chipmunk survival. We, therefore, used population density as an indicator of overall resources availability and other environmental conditions, as it captures the actual population response to these conditions.

Docility scores

For the viability selection analysis on docility (see further section ‘Viability selection on docility’), we extracted individual docility scores for each individual from a Bayesian model of docility with the same structure as the one selected in the frequentist framework of the previous section (see section ‘Environmental and state effects on docility’) and by separating adults and juveniles. These Bayesian models allowed us to estimate a 95% credibility interval around each docility score. Due to a lack of statistical power from our capture-mark-recapture dataset on sites 2, 3 and 4, we only tested adult viability selection on site 1. Our dataset contained 3551 handling-bag tests on 441 adults. The adjusted docility score for each individual was extracted from a Bayesian generalized mixed-effects model with a Poisson error distribution that included the chipmunk and observer identities as random effect and with the best fixed structure found previously. Trial order and season were used as fixed effects (age was not included because the dataset was already separated between adults and juveniles). Population density was not included in the model because it was included directly as a covariate in the viability selection models. We ran this model for 6,500,000 iterations with a burn-in of 1,500,000 and a thinning interval (or sampling rate) of 5000. We used the inverse Wishart prior with an expected variance (v) of 1 and a degree of belief parameter (nu) of 1. For juveniles on site 1, docility scores for each individual were also extracted from a Bayesian generalized linear mixed-effects model ran only for measures taken on all juveniles (1139 tests for 308 individuals) and with the same fixed effects as for the adult scores.

Heritability

Narrow-sense heritability was estimated using pedigree information for site 1 for a total of 2657 handling-bag tests for 217 individuals from 2006 to 2009. The package MCMCglmm was used to run a Poisson Bayesian generalized mixed-effects ‘animal’ model with handling-bag test value as the dependent variable. Significant environmental and state effects at site 1 (see section above) were included in the animal model: trial order, age, annual population density (linear and quadratic terms) and trapping period. We partitioned the total phenotypic variance (V P) into additive (V A), permanent environment (V PE), inter-observer (V O) and residual (V R) components. Heritability was estimated as the ratio of V A over V P and repeatability was estimated as the ratio of (V A + V PE) over V P (Wilson et al. 2010). We ran the model for 6,500,000 iterations with a burn-in of 1,500,000 and a thin of 5000. As for repeatability analyses, we used the inverse Wishart prior with an expected variance (v) of 1 and a degree of belief parameter (nu) of 1.

Viability selection on docility

The strength and shape of viability selection on docility was analysed using a capture-mark-recapture framework using the software E-surge v.1.9 (Choquet et al. 2009b) and following an approach similar to Bergeron et al. (2013). We analysed juvenile and adult survival separately. To build up the capture history of each individual, we defined two trapping periods (early (E) or late (L), August 1st being the limit date between them) each year. An individual that was captured at least once during a given period received a score of 1 and an individual that was not captured during that period received a score of 0 (as in Bergeron et al. 2011a, 2013). This allowed us to estimate survival between early and late trapping periods (i.e. 3-month summer period) and survival between late and early trapping periods (i.e. 9 months, including winter). Because time intervals are unequal between trapping periods, estimates were converted to represent 6-month interval. We had a total of nine trapping periods for site 1 (from early 2006 to early 2010) and only ran analysis on that site due to low statistical power on sites 2, 3 and 4. Also, as in Bergeron et al. (2011a, 2013), the best recapture rate model structure was selected before selecting the survival rate structure. The recapture rates were modelled as a function of sex, trapping occasions, linear and quadratic individual docility scores and all possible interactions between these terms. The best recapture model was selected according to the lowest quasi Akaike information criteria adjusted for small samples (QAICc) (Burnham and Anderson 2002). Survival rates were calculated as a function of trapping periods as a multiple-level factor for each transition between trapping p

eriods of each year (time), sex, the linear and quadratic values of individual docility scores and annual population density as an environmental covariate. We ran all model combinations including these variables and all relevant second-order interactions between them. Interactions between time and docility and between population density and docility are used to test the hypothesis that selection fluctuates according to environmental heterogeneity. Again, the best model was selected according to the QAICc.

We tested whether the models respected Cormack-Jolly-Seber’s assumption of detecting possible transience or ‘trap-happiness’ in the data and assessed the amount of overdispersion using the software U-CARE 2.2 (Choquet et al. 2009a). The c-hat was estimated to assess the magnitude of over- or underdispersion and was then used in the models to calculate the QAICc. Overall test was not significant, which means that the models conformed to Cormack-Jolly-Seber’s assumption (see ‘Results’ section).

The effect of docility on juvenile survival was assessed using a generalized linear model with a binomial structure and a Logit link of the survival. The final model was selected comparing deviance and by removing the least significant term from the model until all remaining effects were significant (alpha = 0.05). We therefore opted for the use of a frequentist approach, using ‘glm’ function from R (www.r-project.org). A juvenile was considered to have survived its first year (i.e. survival score = 1) if it was recaptured during at least one of the following trapping periods, and to be dead (i.e. score = 0) if it was never recaptured. Therefore, in our model, mortality could be confounded with dispersal. As for adults, we only conducted these analyses on data from site 1, as dispersal range has been found to be limited on this site (Dubuc-Messier et al. 2012). We had no information on captures of juveniles born in 2010, and they were thus removed from the analysis, leaving us with 306 individuals. We tested whether sex, trapping period (i.e. early and late), population density (linear and quadratic terms), individual global docility scores (linear and quadratic terms) and all two-way interactions between these terms affected survival.

Data availability

The datasets during and/or analysed during the current study are available from the corresponding author on reasonable request.

Results

Repeatability of docility

Over the whole study period, we found a repeatability of 0.41 for adults and 0.31 for juveniles on site 1, and of 0.37 for adults and 0.52 for juveniles on the other sites (see details for sites and years in Table s2). Docility was significantly repeatable for all years, for each sex, trapping period and reproductive season (Tables s2 and s3). Most of the 95% credibility intervals associated with repeatability estimates overlapped indicating that estimates were generally similar (Tables s2 and s3).

Environmental and state effects on docility

On site 1, juveniles were significantly less docile than adults (Table 1). Docility increased significantly between early and late trapping periods, and with trial order (Table 1). Annual population density negatively influenced docility with a significant quadratic effect of population density (Table 1), docility decreased with population density and this effect was more pronounced at low density (Fig. 2). Both chipmunk and observer identities explained a significant part of the total variance in docility (Table 1). Inter-individual effects represented 38% of the total variance in docility. Inter-observer effects represented only 3% of the variance. Full models are presented in Table s4.

Table 1 Final models of environmental and state effects on docility obtained from generalized linear mixed-models. Chipmunk and observer identities were included as random effects (respectively V PE and V O). V R = residual variance. These models were selected by a backward selection with a negative binomial distribution. Adult was the reference age class, female was used as reference sex and late trapping period was the reference trapping period. The model for site 1 included 4690 handling-bag tests performed on 601 chipmunks from early 2006 to early 2010 and the model for sites 2, 3 and 4 included 1754 handling-bag tests performed on 311 chipmunks by a total of 21 observers from early 2012 to early 2014
Full size table
Fig. 2
figure 2

Effect of population density on docility on site 1 for all individuals (n = 601). The predicted values of docility from the handling-bag test as function of the scaled population density (black line) were obtained from the final model presented in Table 1 (which also accounted for trial order, age and trapping period). The grey dashed lines are the lower and upper 95% confidence intervals for the predicted values

Full size image

For sites 2, 3 and 4, docility also increased significantly with trial order (Table 1). Interaction between sex and trapping period had a significant effect on docility (Table 1). Docility in females was more stable between trapping periods than that in males, and was higher in early trapping period than that in late period. Males were more docile in late trapping period than those in early period. Both chipmunk and observer identities explained a significant part of the total variance in docility (Table 1). Here again, inter-observer effects explained only 3%, and inter-individual effects explained 43% of total variance in docility (Table 1).

Heritability

Additive genetic variance for docility on site 1 was significantly greater than zero (credibility interval was not bounded), and translated into a narrow-sense heritability of 0.17 (95% CI 0.07–0.32) (Table 2). Repeatability of docility from this model was 0.38 (95% CI 0.32–0.44) (Table 2). The permanent environment effect (V PE) accounted for 15.8% of the trait variance.

Table 2 Variance components and narrow-sense heritability (h 2) of docility on site 1, obtained from a Bayesian generalized linear mixed ‘animal’ model. CI 95% credibility intervals. Chipmunk, pedigree and observer identities were included as random effects (respectively V PE, V A and V O). V R = residual variance. The fixed effect structure was the one from the best model describing docility in site 1. The model included 2657 handling-bag tests performed on 217 chipmunks by a total of 43 observers from early 2006 to early 2009. Adult was used as the reference age class and late trapping period was used as the reference trapping period
Full size table

Viability selection on docility

Goodness of fit

The goodness of fit for the adult capture-mark-recapture dataset on site 1 conformed to the Cormack-Jolly-Seber assumptions with non-significant results for the overall test (quadratic chi2 = 27.34, df = 31 and P = 0.66). The overdispersion was assessed with the calculation of the c-hat which had a value of 0.88 (df = 31; quadratic chi2 = 27.34). This value of c-hat was included in the models to calculate the QAICc values.

Recapture structure

Recapture structure for site 1 included the interaction between trapping periods as an eight-level factor and sex, based on the model with the lowest value of QAICc (Table 3).

Table 3 Survival rates and recapture rates for each capture occasion on site 1 along with 95% confidence intervals. Individual capture histories were taken for five trapping periods from early 2006 to early 2010 on 457 adult chipmunks. Survival and recapture probability estimates are from the best models selected based on QAICc (Table s5). The survival model included the linear and quadratic values of docility and each transition between trapping periods, and the recapture model included an interaction between trapping periods and sex
Full size table

Adult survival

On site 1, the best model describing the survival probability included the linear and quadratic terms for docility and the additive effect of trapping occasions (Table s5). More specifically, individuals with low or high values of docility had a better survival probability than intermediate individuals (e.g. quadratic estimate of docility = 0.15, 95% CI 0.01–0.29; Fig. 3). Survival varied significantly between trapping occasions (Table 3; Table s5). A model with trapping periods only was nearly as explicative as the best model with a delta QAICc of 0.88 (‘Time’ in Table s5). In that model, values of survival for each trapping period were similar to values from the best model.

Fig. 3
figure 3

Relationship between docility score and the survival probability to the next trapping period for adult eastern chipmunks (n = 441) on site 1. The predicted values of the survival probability as function of docility scores (black line) were obtained from the model with the lowest QAICc presented in Table s5 (which also accounted for trapping period). The grey dashed lines are the lower and upper 95% confidence intervals for the predicted values. The intercept was fixed to the mean survival probability over all trapping periods (0.62)

Full size image

Finally, we tested the effect of population density as a temporal covariate on survival. Models including population density were less explicative than models with trapping periods only and the model with the lower QAICc including that covariate was the 21th with a delta QAICc of 30.75.

Juvenile survival

We found no link between docility (linear and quadratic terms) and juvenile survival on site 1 (coefficients ± SE from the last model including these terms = −0.06 ± 0.15, P = 0.97 and 0.20 ± 0.16, P = 0.21). The linear and quadratic values of annual population density affected juvenile survival, with better survival at high population density (Table 4; Fig. 4).

Table 4 Estimates from the final model for the juvenile survival on site 1. Juvenile survival was based on whether the individual was recaptured as an adult or not, for 306 juveniles caught on site 1. The values of annual population density were scaled
Full size table
Fig. 4
figure 4

Relationship between population density (e.g. the annual number of individuals per hectare) and juvenile survival on site 1 (n = 308). The predicted survival probability as a function of the population density (black line) was obtained from the final model presented in Table 4. The grey dashed lines are the lower and upper 95% confidence intervals for the predicted values

Full size image

Discussion

We studied the effects of environmental conditions and individual state on docility, and estimated its individual consistency (repeatability) and genetic basis (heritability) in a wild population of eastern chipmunks over 10 years. Furthermore, we tested whether among-individual differences in docility translated into differential survival. Docility was repeatable between and within environmental contexts and was also heritable. We also found that seasonal variations, sex and annual population density affected variation in docility. We found evidence of disruptive viability selection for adults. We detected no evidence of viability selection acting on juvenile docility. These results provide a good example of the impact of the environment on the expression of a personality trait and of its genetic basis and selection pattern acting on it in the wild.

Repeatability, heritability and permanent environment effect on docility

We estimated moderate global repeatability for docility with similar values between site 1 and sites 2, 3 and 4. Our estimates of repeatability are close to the average value of 0.37 reported by Bell et al. (2009) for behavioural traits and are higher than the value of 0.29 previously reported by Montiglio et al. (2012) on site 1, but with a slightly different dataset. Repeatability of a trait implies high variability among individuals and consistent expression within individuals (Lessells and Boag 1987). Repeatability estimates did not differ much between age classes, but were more stable in adults. A previous study of docility in yellow-bellied marmots (Marmota flaviventris) by Petelle et al. (2013) found similar results as the repeatability of docility was quite stable in this species.

Repeatability therefore offers the potential for a trait to affect fitness differently according to phenotypes, and repeatability is often seen as the upper limit of heritability (Boake 1989; Dochtermann et al. 2014). Heritability estimated for docility on site 1 was comparable to value reported in Dochtermann et al. (2014) for the mean heritability of personality traits (0.14), but lower than what estimates reported in Stirling et al. (2002) for behaviours in general (0.31). The narrow-sense heritability estimate we obtained for docility is slightly higher than estimates obtained for red squirrels (Tamiasciurus hudsonicus, estimate ± 95% CI 0.09 ± 0.05–0.19; Taylor et al. (2012)) and yellow-bellied marmots (estimate ± 95% CI 0.08 ± 0.05–0.13; Petelle et al. (2015), see also Martin et al. (2017)). In contrast, our estimate is lower than values reported in blue tits (Cyanistes caeruleus, estimate ± 95% CI 0.35 ± 0.21–0.49; Class et al. (2014)) or in bighorn sheep (Ovis canadensis, estimate ± 95% CI 0.65 ± 0.53–0.77; Réale et al. (2009)). Since repeatability was moderate for docility in our population, the relatively low value of heritability we obtained was somewhat expected. Yet, the presence of a genetic basis for this trait indicates that it could respond to selection. Permanent environment effects also explained as much variance as additive genetic effects. As suggested by Dochtermann et al. (2014), this component of variance could reflect long-lasting environmental effects and/or maternal and paternal effects as well as epigenetics effect on docility. The underlying causes of permanent environment variance however remains to be further investigated in this system.

Environmental and individual sources of variation for docility

We found a significant effect of population density on docility on site 1. Docility was greater at low population density and lower at intermediate and high population densities (Fig. 2). Since a low docility is expected to be linked to high aggressiveness and boldness (see Réale et al. 2010), this result may suggest that animals are typically more aggressive and may take more risks at higher population density (Griffiths and Foster 1998; Réale et al. 2007; Korpela et al. 2011; Webster and Ward 2011). Our results are also supportive of those from previous studies that suggested that personality can be partly determined by environmental conditions, such as food abundance or population density (Dingemanse et al. 2004; Schuett and Dall 2009; Webster and Ward 2011; Tremmel and Müller 2013).

Docility was greater during the late trapping season than that in the early season. This temporal variation of docility could be linked to differences at the individual level in activity levels between early and late periods as shown by Munro et al. (2008). Differences among periods could also be linked to physiological differences in cortisol metabolite production that also varies within a year or to habituation to human presence during the summer (Montiglio et al. 2012). Low activity levels are often correlated with high shyness, low boldness and high docility, and high sympathetic reactivity is generally correlated with high boldness and low docility (reviewed in Réale et al. 2010). Also, Montiglio et al. (2012) found that docility in eastern chipmunk varied according to year and date, which is in concordance with the differences in docility we found between trapping periods. Seasonal variation was used as indicator of variation in environmental conditions, such as food abundance that fluctuates in time according to the dynamic of masting trees; thus, the variation in docility we observed could also be linked with environmental variation (Bergeron et al. 2011a).

We found that young individuals were less docile than adults. Differences in personality can depend on the age of the individuals (Stamps and Groothuis 2010). For example, docility in bighorn rams increases with age and is positively linked with reproductive success (Réale et al. 2009). In that case, more-docile males have longer lifespan, due to docility being linked with risk taking, leading them to attain a higher dominance rank and therefore having a better reproductive success (Réale et al. 2009). However, less-docile individuals might have a shorter lifespan, but have a better reproductive success sooner in life (Wolf et al. 2007). Therefore, differences in life-history strategies could possibly explain the differences in the expression of docility depending on age (Wolf et al. 2007; Réale et al. 2009).

Docility was affected by the sex of the individuals and varied depending on the trapping period on sites 2, 3 and 4, males being more docile than females in late trapping period. In great tits (Parus major), van Oers et al. (2005b) found that risk-taking behaviour varies between the sexes depending on the context. Females had higher risk-taking behaviour when alone, while males had higher risk-taking behaviour in presence of a companion (van Oers et al. 2005b). Finally, we found that trial order affected the level of docility, with more docility being detected in later tests. This indicates some level of habituation to the test, which was also previously detected for the handling-bag test in eastern chipmunks (Martin and Réale 2008; Montiglio et al. 2012). Habituation is common in tests assessing personality traits and is often due to the loss of novelty in a repeated test and to the animal remembering the test conditions (Dingemanse 2002; Réale et al. 2007; Montiglio et al. 2010, 2012). These results emphasize the importance of taking environmental, individual and handling conditions into account when studying the expression of personality traits.

Effect of docility on survival

Fluctuation in American beech seed production drives chipmunk activity (Munro et al. 2008), dispersal and genetic structure (Dubuc-Messier et al. 2012), life-history strategies (Montiglio et al. 2014) and the timing of reproduction and population dynamics (Bergeron et al. 2011a). We thus expected that fluctuation in beech seed production would also affect selection pressures on docility. Yet, we found no impact of fluctuation in seed production on viability selection on adult docility. Adult survival on site 1 varied between trapping periods and differed between males and females, but there was no interaction between docility and trapping periods or population density, thus no fluctuating selection. Instead, we found some evidence of a consistent disruptive selection across trapping periods (Fig. 3). Disruptive selection was also reported previously for exploration behaviour in eastern chipmunks where fast and slow explorers had a survival probability nearly twice as high as individuals with intermediate exploration scores (Bergeron et al. 2013). Exploration is usually negatively correlated to docility in eastern chipmunks but the relationships between these traits vary from weak to moderate (coefficient ± SE −0.14 ± 0.06; Montiglio et al., 2012; r ind = −0.48, 95% HPD interval = −0.60–−0.26; Careau et al. 2015). This suggests that the patterns of selection documented here are unlikely to be only resulting from parallel selection on exploration. We were unable to test this possibility directly here, given the limited availability of measures for both traits taken from the same individuals.

Although speculative, a disruptive selection pattern could possibly be related to variation in predator avoidance strategy, as more-docile individuals may take fewer risks and be potentially less likely predated than average individuals. In contrast, less-docile individuals may be more proactive (i.e. bold, aggressive or active; see Réale et al. 2010) and therefore may have higher chance of escaping predators (Réale and Festa-Bianchet 2003; Sih et al. 2004; Smith and Blumstein 2008; Réale et al. 2010). This selection pattern could also be part of a coevolution between docility and other personality and physiological traits within a pace-of-life syndrome where extreme life strategies linked with predator avoidance would be favoured (Réale et al. 2010). The fact that intermediate phenotypes are still abundant however suggests that segregation between extreme docility phenotypes may be limited on a long-term basis. Trade-offs between traits or between fitness components is an example of mechanism by which intermediate phenotypes could be maintained (Wolf et al. 2007; Smith and Blumstein 2008; Réale et al. 2010; Montiglio et al. 2014). For example, exploration behaviour is linked with reproductive success in our system, and it seems to result from a trade-off between reproduction strategies according to timing of birth and age; fast explorers having a better reproductive success sooner in life, and slow explorers having a better reproductive success later in life (Montiglio et al. 2014). Exploring the link between docility and reproductive success could therefore help assess the presence of a trade-off among fitness components for that trait. Finally, the relatively low heritability of docility allows for large differences between parents and offspring for that trait, which could explain the maintenance of population variation and intermediate individuals.

Juvenile docility did not affect survival, and as for exploration behaviour (Bergeron et al. 2013), it is possible that the strength of selection on this trait is too low to be detected and could be masked by environmental forces acting on survival, such as fluctuation in food resources, parasitism and the timing of birth (Careau et al. 2010; Bergeron et al. 2011a, 2013; Careau et al. 2013). The absence of selection on juvenile docility may also result from the difference in the expression of docility between juveniles and adults. Juveniles are less docile than adults, and this may result from an early developmental stage or early-life strategies where juveniles are more active in general for dispersion or competition for burrows, limiting the potential for selection on docility. Instead of temporally fluctuating selection on docility according to environmental variations, we observed that juvenile survival was mainly dictated by population density, which depended mainly on resources fluctuation (Bergeron et al. 2011a). On site 1, survival increased with the increase in the number of individuals and could be explained by seed abundance, where years of high productivity led to a large number of juveniles with better survival rates (Bergeron et al. 2011a, 2013). Because the survival estimate is based on recapture that can be affected by dispersal, our result could also be partly explained by a differential dispersion among years of high and low population density. Like survival, dispersal can be affected by factors such as seasonal variations linked with food abundance (Dubuc-Messier et al. 2012).

Conclusion

While fluctuating selection in nature is widespread, our results suggest that disruptive selection also seems to be a possible cause of maintenance of phenotypic variation in wild populations. Docility showed similar patterns of selection as those reported for exploration behaviour in this species, leading to the hypothesis that selection can act on a set of correlated behavioural, morphological or physiological traits, which could lead to the maintenance of different pace-of-life syndromes linked with life strategies (Réale et al. 2010). The next research steps would therefore require multivariate analyses incorporating a suit of correlated traits to assess the strength and patterns of selection on such pace-of-life syndromes, as well as analyses related to the causes and consequences of plasticity of these traits.