SEARCH

SEARCH BY CITATION

Keywords:

  • adaptation;
  • I × E;
  • natural selection;
  • random regression;
  • sliding-window

Summary

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

1. Phenotypic plasticity, the response of individual phenotypes to their environment, can allow organisms to cope with spatio-temporal variation in environmental conditions. Recent studies have shown that variation exists among individuals in their capacity to adjust their traits to environmental changes and that this individual plasticity can be under strong selection. Yet, little is known on the extent and ultimate causes of variation between populations and individuals in plasticity patterns.

2. In passerines, timing of breeding is a key life-history trait strongly related to fitness and is known to vary with the environment, but few studies have investigated the within-species variation in individual plasticity.

3. Here, we studied between- and within-population variation in breeding time, phenotypic plasticity and selection patterns for this trait in four Mediterranean populations of blue tits (Cyanistes caeruleus) breeding in habitats varying in structure and quality.

4. Although there was no significant warming over the course of the study, we found evidence for earlier onset of breeding in warmer years in all populations, with reduced plasticity in the less predictable environment. In two of four populations, there was significant inter-individual variation in plasticity for laying date. Interestingly, selection for earlier laying date was significant only in populations where there was no inter-individual differences in plasticity.

5. Our results show that generalization of plasticity patterns among populations of the same species might be challenging even at a small spatial scale and that the amount of within-individual variation in phenotypic plasticity may be linked to selective pressures acting on these phenotypic traits.


Introduction

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

Organisms in the wild continually face spatio-temporal variation in abiotic and biotic conditions. Phenotypic plasticity, the capacity of a genotype to change the expression of a trait in response to environmental variation, is a powerful means by which individuals can cope with changing environmental conditions (Stearns 1989; Schlichting & Pigliucci 1998). Studies of phenotypic plasticity have been conducted in several taxa in the wild, including birds (Nussey et al. 2005c; Charmantier et al. 2008), mammals (Pelletier et al. 2007) and plants (Sultan 2000; Valladares, Gianoli & Gomez 2007). Yet, despite growing interest for phenotypic plasticity in evolutionary ecology, little is known regarding the causes and extent of variation in plasticity displayed among populations of the same species in the wild (Nussey, Wilson & Brommer 2007). As plasticity can be heritable and under selection (Nussey et al. 2005c; Pelletier et al. 2007), knowledge of the extent and ecological causes of variation in individual plasticity patterns among populations is important to determine the potential of natural populations to respond to environmental changes. Deciphering the causes of variation in reaction norm (i.e. the function describing phenotypic change across environments; Stearns 1989) is thus fundamental for understanding the evolutionary and ecological dynamics of populations, particularly in the present context of global environmental change, where plasticity is a key mechanism allowing adaptation (Przybylo, Sheldon & Merilä 2000; Chown & Terblanche 2006).

A major driver of phenotypic plasticity evolution is the degree of environmental heterogeneity experienced by a population (Moran 1992; Sultan & Spencer 2002; Ernande & Dieckmann 2004). A few empirical studies have shown that mean plasticity levels can differ markedly among populations of the same species as a result of spatio-temporal environmental heterogeneity and divergent selective pressures (Lind & Johansson 2007; Liefting, Hoffmann & Ellers 2009; Baythavong & Stanton 2010). However, populations can diverge not only in the average amount of plasticity expressed but also in their patterns of inter-individual variation in plasticity (individual × environment interaction; I × E). For example, larger inter-individual variation in breeding time sensitivity to spring temperature has been found in a Dutch great tit (Parus major) population (Nussey et al. 2005c), than in a great tit population in England (Charmantier et al. 2008; Husby et al. 2010). While the mechanisms underlying such differences in the strength of I × E among populations remain uncertain, population-specific constraints to the expression of phenotypic plasticity, owing to differences in habitat features, climate or environmental conditions (Nussey et al. 2005a,b; Wilson et al. 2007), as well as the degree of spatio-temporal heterogeneity of the environment (de Jong 2005; Ghalambor et al. 2007), are likely candidates. Although theoretical models have mostly focused on explaining the dynamics of plasticity rather than predicting plasticity at equilibrium (Lande 2009), a recurrent finding in quantitative genetic models is that the level of optimal plasticity in a population depends on the predictability of the environment, that is, the covariance between the environment of development and the environment of selection (Gavrilets & Scheiner 1993; Pigliucci 2001). Comparative studies of several populations of the same species in contrasting natural systems, ideally where environment predictability can be measured, are therefore needed to elucidate the mechanisms shaping the observed phenotypic plasticity patterns in nature.

Here, we investigate the interplay between plasticity and selection patterns on timing of breeding (laying date, LD) in a small forest passerine bird, the blue tit (Cyanistes caeruleus), occupying heterogeneous Mediterranean environments. Timing of breeding is a life-history trait strongly related to fitness in several animal species and is particularly sensitive to environmental variation (Both et al. 2004). Flexibility in breeding time allows individuals to synchronize reproductive events with the most favourable period of the year to ensure optimal offspring growth and survival (Visser et al. 1998). This is crucial for insectivorous passerines, which face the challenge of timing their phenology so that the most energetically demanding part of the nesting cycle coincides with the narrow temporal window of food abundance (Thomas et al. 2001). A mistiming of LD therefore usually results in poor breeding success (van Noordwijk, McCleery & Perrins 1995; Tremblay et al. 2003).

The geographic configuration of the Mediterranean landscape, characterized by a heterogeneous, fine-grained mosaic of habitats, provides an exceptional study system to investigate variation in plastic responses and selection pressures of life-history traits in spatially structured populations (Blondel et al. 2006; Blondel 2007). We selected four populations, subject to long-term monitoring, located in either evergreen (holm oak Quercus ilex) or deciduous (downy oak Quercus humilis) forest patches showing large differences in timing and abundance of food (caterpillars) for blue tits (Blondel et al. 2006). The heterogeneity in timing and food abundance is linked to a remarkable phenotypic variation in mean LD across habitats, with differences reaching up to 1 month between deciduous and evergreen populations. Ultimately, these habitat-specific characteristics, and associated selection pressures, may differentially affect phenotypic plasticity for LD both within and among populations. Previous investigations, both in natural habitats and in captivity, have shown that laying date of blue tits in these Mediterranean populations can be plastic, especially but not exclusively in response to photoperiod (Lambrecht & Dias 1993; Lambrechts, Perret & Blondel 1996). These studies also showed that onset of laying differed between populations exposed to the same photoperiodic treatment (Lambrechts et al. 1997), suggesting a divergent adaptation in the response mechanism. To address the issue of the determinism and variation across populations of individual plasticity in timing of reproduction, we used long-term data to test whether (i) there is evidence for phenotypic plasticity in LD in response to temperature variation; (ii) inter-individual differences in the patterns of phenotypic plasticity occur within populations; (iii) plasticity patterns differ among populations and if so whether environment predictability is linked to this variation; and (iv) selection patterns on LD vary among populations.

Materials and methods

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

Study Sites, Population Characteristics and Data Collection

We used long-term data obtained in four heterogeneous Mediterranean blue tit populations breeding in either deciduous or evergreen forests (Fig. 1), and differing in the amount of food available: one in southern France (Deciduous-Rouvière; henceforth named D-Rouvière), and three on the island of Corsica (Evergreen-Pirio, Evergeen-Muro, Deciduous-Muro; henceforth called E-Pirio, E-Muro and D-Muro, respectively). The Corsican sites are located in two study areas isolated by mountains (alt. >1500 m). The E-Pirio site (42°34′N, 08°44′E; 200 m elevation) is dominated by evergreen holm oaks, and data on bird breeding traits have been collected at this location since 1979. The second Corsican study area is located in the Muro valley and separated in two sites: (i) D-Muro (42°32′N, 08°55′E, 350 m), characterized by a habitat of deciduous downy oak and sampled since 1993; (ii) E-Muro (42°35′N, 08°57′E, 100 m), dominated by holm oak at a local scale, and downy oak at a broader scale, and monitored since 1998. The D-Rouvière population (43°40′N, 03°40′E, 300 m) is largely dominated by downy oaks and has been monitored since 1991. In Corsican landscapes, deciduous woodlots (D-Muro) present an early and abundant leaf flush that creates a massive peak in caterpillar availability, translating into earlier LD, larger clutches, and higher fledging success than in evergreen habitats (E-Muro and E-Pirio) (Blondel et al. 2006). On the other hand, the D-Rouvière site has lower caterpillar availability, despite being located in a landscape dominated by downy oak and presenting similar breeding parameters as those of the D-Muro site (see Table 1).

image

Figure 1.  Location of blue tit study sites in the Mediterranean region. The D-Rouvière study site is located in continental southern France, c. 450 km from the Corsican sites. D-Muro, E-Muro and E-Pirio are located in Corsica. D-Muro and E-Muro are separated from E-Pirio by 24 and 29 km, respectively.

Download figure to PowerPoint

Table 1.   Year span, laying date and clutch size (mean ± SD), and sample sizes for plasticity and selection analyses in the four blue tit populations
SiteTime period (years)Mean laying date (1 = January 1st)Mean clutch sizeNumber of observations/number of femalesNumber of recruits
All breeding eventsConsecutive breeding events Selection analyses
D-Rouviere1991–201097·9 ± 7·610·1 ± 1·61168/678735/272763/483602
D-Muro1993–201097·1 ± 8·48·6 ± 1·5781/540371/150690/481336
E-Muro1998–2010106·7 ± 7·47·0 ± 1·2332/173220/74248/13183
E-Pirio1979–2010129·2 ± 6·96·6 ± 1·11295/640928/316781/489275

Monitoring of artificial nest boxes (60–150 per site) allowed us to collect data on LD (date of the first egg laid, January 1st = 1) and CS (number of eggs laid) over several years (see Table 1). Number of recruits for each brood was defined as the number of fledglings subsequently captured as adults in the study site (excluding data from 2010). Adult birds were captured and identified/ringed when nestlings were 9–15 days old (see Blondel et al. 2006 for further details). Only first clutches were included in the analyses, second clutches or repeat clutches only representing c. 1% of total clutches. For selection analyses, we excluded records of females that had their clutch size manipulated during egg laying, as well as data on broods that were experimentally enlarged, reduced or cross-fostered. Table 1 summarizes population characteristics and sample sizes for each population.

Environmental Variables

Temperature during the pre-laying period has been identified as influencing LD in many studies on temperate breeding passerines (Perrins 1965; Nager & Van Noordwijk 1995; Nussey et al. 2005c). However, the exact time period over which birds use temperature cues to adjust their LD remains uncertain. Consequently, we used a sliding-window approach (as in Brommer, Rattiste & Wilson 2008; Husby et al. 2010) to determine the periods during which mean temperature best explained mean annual LD for each site. For each site, mean annual LD was correlated with the average temperature for every period ranging from 10 to 60 days between January 1st and the day preceding the first LD in the population. Mean temperatures for the time window for which the relationship between average temperature and LD had the highest r2 were then used as environmental variables in following plasticity analyses (henceforth called tempLD). Mean daily temperatures (daily maximum temperature + daily minimum temperature/2]) were obtained from local meteorological stations (Muro valley sites: 42°31′N, 08°47′E; E-Pirio: 42°24′N, 08°38′E; D-Rouvière: 43°34′N, 03°57′E). For Muro valley sites, these data were combined with temperatures recorded in situ from dataloggers (iButton thermochrons; model 1922L; Maxim Integrated Products, Sunnyvale, CA, USA) placed in 10 nestboxes per site.

To evaluate the biological significance of the tempLD variable and the predictability of the environment, additional sliding-window analyses were conducted to identify which time periods yielded the best correlations among temperature and date of food abundance peak. Date of food peak was determined using food abundance measures taken every 3 days, from the start of the egg-laying period up to the fledging of the young, by collecting leaf-eating caterpillar frass using 0·25 m2 trays (10–20 trays per site) placed under downy (D-Muro and D-Rouvière) or holm (E-Muro and E-Pirio) oaks (Zandt 1994). Data were collected from 2001 to 2010 for E-Muro, 1993 to 2008 and 2010 for D-Muro, 1987 to 2010 for E-Pirio, and 1991 to 1997 and 2000 to 2002 for D-Rouvière. Concordance between the time periods found for LD and date of food peak (as assessed by the R2 of the linear regression of peak date as a function of tempLD) would suggest that the environment at the time of decision-making (tempLD) is a good predictor of the environment of selection during the food peak, a predictability that is theoretically proportional to the level of plasticity displayed (Gavrilets & Scheiner 1993).

Population and Individual Plastic Responses and their Variation

Average population response of blue tits to environmental variation was determined by performing a cross-sectional analysis where the relationship between annual mean LD and tempLD was assessed for each study site. The slope of each regression was then used to compare the strength of the population-level response to environmental variation. Significant differences in the population response of LD to changes in temperature were tested by pooling each population pair in one dataset and testing the significance of the interaction between population and tempLD in the following linear model: annual mean LD ∼ population + tempLD + population*tempLD.

We then assessed in two ways whether the population response in LD was explained by inter-annual individual variation, that is, plasticity. First, the potential strength of individual plasticity was estimated as the slope of the linear relationship between within-individual change in the trait (ΔLD = LD in year 2 − LD in year 1) and differences in temperature (ΔtempLD = tempLD (°C) year 2 − tempLD (°C) year 1) for females that bred in at least two consecutive years, including age as a fixed effect in a general linear model. Differences in the magnitude of individual plasticity between each population pair were assessed by testing the significance of the population*ΔtempLD interaction term in the following model: ΔLD ∼ age + population + ΔtempLD + population*ΔtempLD.

Second, using data for all females with at least one recorded breeding event, variation in individual phenotypic plasticity was assessed using random regression models, where LD of each individual in each year was modelled as a continuous function of the year-specific tempLD, using Legendre polynomials. For each site, our model included LD as the dependent variable, and age and tempLD as fixed effects. As Legendre polynomials are only defined within the range of −1 to +1, measures of tempLD were standardized to fit this range (Huisman, Veerkamp & Van Arendonk 2002). The random effects structure was hierarchically extended to test patterns of variation in individual phenotypic plasticity, keeping the fixed effects model unchanged. We thus tested whether there was (i) inter-individual differences in average values of the LD (elevation, i.e. intercept of reaction norms); and (ii) inter-individual differences in the response to environmental variation (slopes of reaction norms). Random effects significance was assessed using a Likelihood Ratio Test (LRT; Pinheiro & Bates 2000). The following models were successively compared:

Model 1: LD ∼ Age + tempLD

Model 2: LD ∼ Age + tempLD, random = Year

Model 3: LD ∼ Age + tempLD, random = Year + Female identity (elevation)

Model 4: LD ∼ Age + tempLD, random = Year + Female identity (elevation) + Female identity*tempLD (slope)

For the E-Pirio population, as mean laying date changed through time (see Results section), the above-mentioned models were ran using heterogeneous (decade-specific: 1979–1989; 1990–1999 and 2000–2010) error variances, thus allowing residual error variance to vary through time, while homogeneous error variances were modelled for the other populations. All models were fitted using restricted maximum likelihood (REML) methods implemented in ASReml v3 (VSN International Ltd, Hemel Hempstead, UK). Significance of the correlation between elevation and slope was assessed by comparing the model 4 (above) to a model in which the covariance parameter was fixed to zero using a LRT.

Between-Population Comparison of Plasticity Patterns

To compare plasticity patterns between each pair of populations, we pooled datasets from each population pair and performed two sets of Model 4 (see the above section): one with variance components for each population constrained to be equal, and a second where they were allowed to vary (see Husby et al. 2010). The two sets of models were then compared using a LRT.

Selection on Laying Date and Clutch Size

We first estimated selection differentials on LD (Endler 1986) at each site. However, as LD and CS are often correlated with birds (Sheldon, Kruuk & Merilä 2003; Garant et al. 2008), biases in the estimation of selection on LD may arise if selection on CS is not simultaneously accounted for. Selection gradients (Lande & Arnold 1983) on LD and CS were therefore also estimated at each site (see Table 1 for sample sizes). Individual LD and CS were standardized to mean zero and unit variance within each year, and fitness (annual number of recruits) was converted to relative fitness by dividing individual number of recruits by the mean annual number of recruits for breeding females. Linear selection differentials (i) were obtained by regressing relative number of recruits within each year against standardized values of LD separately within each study site, and quadratic selection differentials (j) were estimated from a model including both linear and quadratic terms. Standardized linear selection gradients (β) were estimated from the following multivariate linear model:

  • image(eqn 1)

where ω = relative fitness, α = intercept, x1 = laying date, x2 = clutch size, and ε = error term. Nonlinear quadratic (γi, γj) and correlational (γij) selection gradients were estimated from the model:

  • image(eqn 2)

Statistical significance of selection differentials/gradients was estimated from generalized linear mixed models with Poisson (D-Rouvière, D-Muro and E-Muro) or quasi-Poisson distribution (correcting for overdispersion – E-Pirio), relating fitness to unstandardized values of the traits, including female identity as a random effect. Selection analyses were performed using Genstat v.10.1.0.72 (VSN international Ltd).

Finally, to tests whether selection patterns on laying date are significantly different among populations, we used a sequential model building approach (as in Draper & John 1998; see also Appendix A in Chenoweth & Blows 2005). For each population pair, we fitted four series of two models (generalized linear models with Poisson distribution) as follows:

Model 1a) Fitness ∼ pop + LD

Model 1b) Fitness ∼ pop + LD + pop*LD

Model 2a) Fitness ∼ pop + LD + pop*LD + LD^2

Model 2b) Fitness ∼ pop + LD + pop*LD + LD^2 + pop*LD^2

Model 3a) Fitness ∼ pop + LD + CS

Model 3b) Fitness ∼ pop + LD + CS + pop*LD

Model 4a) Fitness ∼ pop + LD + CS + pop*LD + LD^2 + CS^2

Model 4b) Fitness ∼ pop + LD + CS + pop*LD + LD^2 + CS^2 + pop*LD^2

where fitness = number of recruits and pop = population. Each model pair was then compared to test for significant differences among populations in linear selection differential (1a vs. 1b), quadratic selection differential (2a vs. 2b), linear selection gradient (3a vs. 3b), and quadratic selection gradient (4a vs. 4b) on laying date. Significance was obtained by comparing models using a LRT.

Results

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

Sliding-Window Analyses

Sliding-window analyses showed that temperature periods explaining most variation in LD varied among populations (Table 2). With the exception of the E-Muro population, the time period where temperature was most correlated with LD was relatively close (<1 month) to the population mean LD (Tables 1 and 2). Annual variation in tempLD explained a relatively high proportion of annual variation in date of food peak in three of four populations (D-Rouvière: R2 = 0·38, P = 0·08; D-Muro: R2 = 0·44, P = 0·004; E-Muro: R2 = 0·34, = 0·08). This suggests that the laying decision-making could rely on temperatures that are good predictors of the caterpillar phenology. In Pirio, however, annual variation in peak date was not strongly related to tempLD (R2 = 0·09, = 0·16), suggesting a low predictability of the selection environment.

Table 2.   Sliding-window results showing time windows (1 = January 1st) for which temperature was most correlated with laying date and date of food peak and associated Pearson’s correlation coefficients
SiteLaying dateFood peak date
Start dateEnd date r StartdateEnd date r
D-Rouviere3973−0·8552146−0·857
D-Muro4273−0·7296271−0·748
E-Muro2050−0·7954456−0·941
E-Pirio86105−0·33642570·383

Population-Level Trends Over Time and Environment

No significant temporal trends were detected for LD in D-Rouvière, D-Muro and E-Muro. Yet, over 32 years of study in E-Pirio, mean LD advanced by 9 days (F1,31 = 12·47, = 0·001). No significant warming during the population-specific tempLD periods was detected in our study sites over the 17 to 32-year spans (linear regressions of tempLD on year: D-Rouvière: slope = −0·056 ± 0·067, t18 = −0·833, P = 0·42; D-Muro: slope = −0·008 ± 0·078, t16 = 0·108, = 0·92; E-Muro: slope = 9·9 × 10−5 ± 0·109, t11 = 0·001, = 1·00; E-Pirio: slope = 0·025 ± 0·017, t30 = 1·457, = 0·16). Linear regression models showed significant negative relationships between annual mean LD and tempLD in D-Rouvière, D-Muro and E-Muro (D-Rouvière: slope = −2·51 ± 0·34, t18 = −6·989, < 0·001, r2 = 0·73; D-Muro: slope = −2·45 ± 0·55, t16 = −4·416, < 0·001, r2 = 0·55; E-Muro: slope = −2·64 ± 0·61, t11 = −4·352, < 0·001, r2 = 0·63), but marginally non-significant in E-Pirio (slope = −1·69 ± 0·87, t30 = −1·951, = 0·060, r2 = 0·11; Fig. 2). Between-population differences in these slopes were not significant (results not shown; all > 0·38).

image

Figure 2.  Population-level plasticity patterns in each study site described as annual mean laying date (±SE) against average mean daily temperature for the time period most correlated with laying date (tempLD). Plain lines represent significant relationships (< 0·05), dashed lines show non-significant relationships.

Download figure to PowerPoint

Variation in Individual Plasticity

Estimated slopes for average individual plastic responses of LD in females breeding over successive years were consistent with population-level responses, with individual females breeding earlier in warmer years (D-Rouvière = −2·51 ± 0·16, t272 = 16·04; E-Pirio = −0·71 ± 0·23, t316 = 3·05; D-Muro = −1·62 ± 0·23, t102 = 7·54; E-Muro = −1·68 ± 0·26, t74 = 6·36; all < 0·01), suggesting that the population-level LD responses to temperature are largely explained by individual plasticity. Between-population comparisons showed significant differences in the magnitude of individual-level plasticity between all population pairs (population × ΔtempLD interaction: D-Rouvière and E-Muro: t597 = 2·50, = 0·013; D-Rouvière and D-Muro: t674 = 3·24, = 0·001; E-Pirio and D-Rouvière: t1057 = 6·45, < 0·001; E-Pirio and E-Muro: t741 = 2·60, = 0·010; E-Pirio and D-Muro: t818 = 2·96, = 0·003) except between E-Muro and D-Muro (t358 = 0·16, = 0·88). Random regression models showed significant inter-individual differences in average values of LD in all populations (Table 3). Furthermore, significant inter-individual variation in the plastic response to temperature (I × E) for LD occurred in two of the four sites (D-Rouvière and E-Muro; Table 3). Covariances between elevation and slope were positive and significant in all populations (see Table S1 for a full description of (co)variance components).

Table 3.   Statistical significance of adding random terms in linear mixed models of laying date against tempLD. TempLD and age were included as fixed effects in each model
 Log-LTestd.f.LRT P-value
D-Rouvière
 Minimal model−2738·41    
 Year−2663·811 vs. 21149·20<0·001
 Year, female−2618·602 vs. 31 90·42<0·001
 Year, I, I × E−2615·113 vs. 42  7·64  0·031
D-Muro
 Minimal model−1971·27    
 Year−1927·761 vs. 21 87·02<0·001
 Year, female−1895·902 vs. 31 63·72<0·001
 Year, I, I × E−1894·673 vs. 42  2·46  0·29
E-Muro
 Minimal model−768·05    
 Year−748·371 vs. 21 39·36<0·001
 Year, female−724·002 vs. 31 48·74<0·001
 Year, I, I × E−716·473 vs. 42 15·06<0·001
E-Pirio
 Minimal model−3102·25    
 Year−2764·381 vs. 21675·74<0·001
 Year, female−2673·512 vs. 31181·74<0·001
 Year, I, I × E−2673·393 vs. 42  0·24  0·89

Comparison of plasticity (co)variance components for LD between populations showed significant differences between D-Rouvière and D-Muro (LRT = 36·02, d.f. = 3, < 0·001) and between D-Muro and E-Pirio (LRT = 56·8, d.f. = 3, < 0·001), and marginally non-significant differences between D-Muro and E-Muro (LRT = 6·84, d.f. = 3, = 0·08) and between E-Muro and E-Pirio (LRT = 6·56, d.f. = 3, = 0·09; see Table 4a).

Table 4.   Between-population divergence in (a) plasticity patterns; and (b) selection differentials (linear: models 1a vs. 1b; nonlinear: models 2a vs. 2b) and gradients (linear: models 3a vs. 3b; nonlinear, models 4a vs. 4b) for laying date
 d.f.D-Rouviere – D-Muro    D-Rouviere –    E-Muro  D-Rouviere –    E-PirioD-Muro – E-MuroD-Muro – E-PirioE-Muro – E-Pirio
Log-Lχ2 P Log-Lχ2 P Log-Lχ2 P Log-Lχ2 P Log-Lχ2 P Log-Lχ2 P
  1. Models :

  2. Plasticity :

  3. 1) (Co)variance components constrained to be equal

  4. 2) (Co)variance components unconstrained

  5. Selection :

  6. 1a) pop + LD

  7. 1b) pop + LD + pop*LD

  8. 2a) pop + LD + pop*LD + LD2

  9. 2b) pop + LD + pop*LD + LD2 + pop*LD2

  10. 3a) pop + LD + CS

  11. 3b) pop + LD + CS + pop*LD

  12. 4a) pop + LD + CS + pop*LD + pop*CS + LD2 + CS2

  13. 4b) pop + LD + CS + pop*LD + pop*CS + LD2 + CS2 + pop*LD2

(a) Plasticity
(1)3−4355·6036·02<0·001−2660·552·880·41−4636·010·820·84−2326·736·840·08−3725·8456·8<0·001−2049·626·560·09
(2)−4337·59−2659·11−4635·60−2323·31−3697·44−2046·34
(b) Selection
(1a)1−1494·018·940·003−826·031·080·30−1346·588·860·003−682·688·410·004−1000·770·240·62−454·125·310·021
(1b)−1489·54−825·49−1342·15−678·47−1000·65−451·46
(2a)1−1489·440·560·45−825·460·780·38−1341·102·430·12−678·251·360·24−1000·655·540·019−449·991·490·22
(2b)−1489·16−825·07−1339·89−677·57−997·88−449·25
(3a)1−1483·096·640·010−823·061·220·27−1335·937·160·008−679·667·930·005−994·580·170·68−452·565·040·025
(3b)−1479·77−822·45−1332·35−675·69−994·50−450·04
(4a)1−1474·800·280·60−819·170·920·34−1326·532·400·12−675·131·080·30−991·725·390·020−445·901·470·22
(4b)−1474·66−818·71−1325·33−674·59−989·03−445·16

Selection on Laying Date and Clutch Size Across Populations

Linear selection differentials and gradients for LD were negative (i.e. number of recruits maximized for early breeding females) and significant in E-Pirio and D-Muro only (Table 5). Annual linear selection differentials were consistent with global selection estimates, with linear selection differentials on laying date being negative in 15/17 years in D-Muro (five significant) and in 24/30 years in E-Pirio (three significant), but only in 8/19 years in D-Rouvière (three significant) and in 5/12 years in E-Muro (none significant; see Fig. 3; Table S2). Annual linear selection gradients followed a similar trend (see Table S2). We also found evidence of nonlinear selection on LD in D-Rouvière, where there was a positive quadratic selection differential (Table 5). Annual linear and nonlinear selection differentials for LD and gradients for LD and CS are presented in Table S2.

Table 5.   Standardized linear (i) and nonlinear (j) selection differentials on laying date and linear (β), nonlinear quadratic (γii), and nonlinear correlational (γij) selection gradients on laying date and clutch size. Values are provided with their standard error. Bold estimates are significant at P < 0.05
SiteSelection differentialsSelection gradients
i LD j LD βLDβCSγLDγCSγLD-CS
D-Rouviere−0·065 ± 0·046 0·082 ± 0·032−0·025 ± 0·050 0·097 ± 0·050 0·093 ± 0·037−0·049 ± 0·034  0·006 ± 0·053
D-Muro0·286 ± 0·058−0·006 ± 0·0410·212 ± 0·071 0·129 ± 0·071 0·019 ± 0·062−0·021 ± 0·060  0·043 ± 0·103
E-Muro−0·135 ± 0·123 0·062 ± 0·101−0·078 ± 0·1370·130 ± 0·1370·002 ± 0·127−0·185 ± 0·123−0·182 ± 0·195
E-Pirio0·247 ± 0·086 0·033 ± 0·0640·220 ± 0·090 0·085 ± 0·090 0·049 ± 0·070−0·100 ± 0·069−0·003 ± 0·095
image

Figure 3.  Annual standardized selection differentials for LD (iLD; ±SE) in each study site. Dashed line refers to a selection differential of zero.

Download figure to PowerPoint

Linear selection differentials and gradients differed significantly among four of six pairwise comparisons, all involving one population with significant linear selection on LD and one without (Linear selection differentials/gradients: D-Rouvière and D-Muro: LRT = 8·94/6·64, d.f. = 1/1, = 0·003/0·010; D-Rouvière and Pirio: LRT = 8·86/7·16, d.f. = 1/1, = 0·003/0·008; D-Muro and E-Muro: LRT = 8·41/7·93, d.f. = 1/1, = 0·004/0·005; E-Muro and E-Pirio: LRT = 5·31/5·04, d.f. = 1/1, = 0·021/0·025; see Table 4b). Nonlinear selection differentials/gradients on LD were significantly different only among D-Muro and E-Pirio populations (nonlinear selection differential/gradient: LRT: 5·54/5·39, d.f. = 1/1, = 0·019/0·020; Table 4b).

Discussion

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

In this study, we documented patterns of phenotypic plasticity and selection on LD in four Mediterranean blue tit populations. We showed that blue tits advanced their LD in years with higher tempLD in all populations studied, yet there was lower individual plasticity in the less predictable habitat. We further showed that I × E for LD occurred in two populations and that linear selection for early LD was present in the other two populations. Our results indicate the presence of variation in phenotypic plasticity over a small geographic scale and a possible interplay between variation in selection and plasticity patterns of breeding time in wild populations.

Population Vs Individual Responses to Environmental Variation

At the population-level, birds adjusted their LD according to temperature, showing earlier mean LD with warmer temperatures in all study sites (although marginally non-significant in E-Pirio). The time period for which mean temperature was most related to LD occurred relatively close (<1 month) to the reproductive period in three of four study sites, which is similar to time lags obtained in recent studies assessing this relationship in avian populations (Brommer, Rattiste & Wilson 2008; Husby et al. 2010). The relationships between annual mean LD and tempLD were strong in D-Muro, E-Muro and D-Rouvière populations, suggesting that temperature has a major influence on seasonal timing of reproduction in these populations, as found in previous studies in birds (Both et al. 2004; Dunn 2004). In the E-Pirio site, however, only 11% of annual variation in mean LD was explained by tempLD, which suggests that other environmental cues, such as an assessment of vegetation phenology, might be used for LD adjustment in this population (Bourgault et al. 2010). We acknowledge that without further experimental investigation, we can only speculate on the causal links between mean temperatures in the time window we used and the egg-laying-date variation. We are aware that such causal links are difficult to establish even in controlled settings, as shown by contrasted results recently obtained in experimental studies (in aviaries) of temperature effects on timing of avian reproduction (see for example Visser, Holleman & Caro 2009; Visser et al. 2011; Schaper et al. 2012). The correlations observed here between tempLD and laying date might thus either be due to a direct use of temperature as cue for the timing of reproduction of blue tits, or, alternatively, to the use of another environmental cue that correlates with the population-specific temperature time-windows.

The population-level plasticity patterns observed could be resulting mainly from individual phenotypic plasticity, as suggested by the similarity observed in the four populations between the slopes describing (i) population-level; and (ii) within-female changes in LD in response to varying temperature (see also Przybylo, Sheldon & Merilä 2000). Previous studies in great tits reported similar trends of intra-individual change in LD, with females breeding earlier in warmer years (Nussey et al. 2005c; Charmantier et al. 2008). The magnitude of the relationship between LD and temperature, however, significantly varied among populations; individual trends in E-Pirio showed the weakest response and D-Rouvière the strongest, with Muro valley sites having intermediate responses. This result may be partly explained by the weak predictability of food phenology in Pirio, where temperature in the time window that explained laying date was unrelated to food peak date. Our results thus confirm the theoretical hypothesis of stronger phenotypic plasticity expressed in more predictable environments (Gavrilets & Scheiner 1993; Tufto 2000).

These differences in the magnitude of plasticity are unlikely related to variation in temperature patterns between sites, because our analyses over 17–32 years show no temporal increase in tempLD in our study sites. This is consistent with a previous study by Visser et al. (2003) that showed no temporal increase in mean temperatures (from January 1st to June 15) at the E-Pirio site (although conducted using fewer years than in our analysis). Our results, however, contrast with multiple studies showing an increase in spring temperatures and related advancement of LD in birds (Root et al. 2003 and references therein; Both et al. 2004; Reed et al. 2006), including in blue tits (Potti 2009). Given that within Europe, the Mediterranean region should be most vulnerable to global change in the coming century, because of combined effects of increased temperatures and reduced precipitation (Schröter et al. 2005), further studies, perhaps conducted over longer time periods, are needed to conclude on the extent and causes of this variation in temperature and LD trends among populations and species in this region.

Between-Population Variation in Plasticity Patterns

Random regression analyses showed significant inter-individual variation in mean LD in all sites, a now-classic result since LD is known to be heritable (e.g. Sheldon, Kruuk & Merilä 2003; Garant et al. 2008). We further found evidence for significant positive correlations between elevation and slope of LD plasticity in all populations, with females breeding earlier therefore having steeper reaction norms for laying date, a result in agreement with previous studies on LD plasticity (e.g. Nussey et al. 2005c). However, I × E for LD was only detected in E-Muro and D-Rouvière. Thus females in these two sites differ in their plasticity patterns, with some females expressing greater LD adjustment than others for a given temperature change. In contrast, females from E-Pirio and D-Muro did not differ in the slope of their reaction norms. Altogether our results indicate the presence of contrasted variation in plasticity patterns between populations at a small spatial scale. Very few studies to date have compared I × E patterns among different populations of the same species. Husby et al. (2010) compared LD plasticity patterns between great tit populations in Hoge Veluwe, Netherlands (Nussey et al. 2005c), and Wytham Woods, UK (Charmantier et al. 2008). They showed that, while I × E for LD was present in both populations, there was more variation in the reaction norm slope in Hoge Veluwe than in Wytham Woods. Our study showed between-population divergence at a much smaller spatial scale, with for example E-Muro and D-Muro being only 6 km apart, and not genetically differentiated at neutral loci (M. Porlier unpublished data). Local environmental conditions are thus likely to affect variation in I × E patterns within populations, with the potential of generating extensive divergence in plasticity patterns even when gene flow between populations is important. An interesting follow-up to our findings would be to test whether variation in plasticity at such small geographic scale has a heritable component, which would suggest evolutionary potential for locally adaptive plasticity.

Selection Pressures on Laying Date and Clutch Size

We found significant directional selection for earlier LD in E-Pirio and D-Muro. Such patterns are in agreement with previous findings in this species and other passerines in temperate regions (van Noordwijk, McCleery & Perrins 1995; Sheldon, Kruuk & Merilä 2003; Garant et al. 2007), where early breeding is an important determinant of offspring recruitment. However, we also show that variation in the strength of selection can be found at a small scale, with for example linear selection estimates of LD being significantly different between E-Muro and D-Muro, located 6 km apart, and between E-Muro and E-Pirio, 24 km apart (see also Garant et al. 2007).

Finally, we show no strong evidence for correlational selection on LD and CS. Few studies to date have documented patterns of correlational selection in these traits. Garant et al. (2007) found that, in a population of great tits in Wytham Woods, UK, early breeding birds with large clutches were favoured; while other studies failed to find such a relationship (e.g. Sheldon, Kruuk & Merilä 2003). In passerine populations, it seems that no constant pattern of correlational selection on LD and CS emerges and that simultaneous optimization of these traits may vary among populations.

Interplay Between Individual Plasticity and Selection Patterns on Laying Date

Interestingly, the two sites where no I × E for LD was found are also the only sites in which we found significant levels of linear selection for earlier LD (Table 6), thus pointing to a possible interaction between selection on mean values of a trait and plasticity patterns associated with the trait. Interestingly, an artificial selection study of Arabidopsis thaliana has recently shown that directional selection on flowering time can reduce the amount of genetic variance in reaction norm slope for this trait (Springate et al. 2011). In the absence of theoretical models explaining variation in I × E patterns across populations as a function of selection, we suggest that the observed patterns of selection and plasticity could be partly explained by habitat heterogeneity and food sources variation among populations. In D-Muro and E-Pirio, habitat is homogeneous over relatively large geographic scale. While Mediterranean landscapes tend to be extremely heterogeneous, c. 95% of the trees belong to a single oak species (deciduous or evergreen) at the scale of local habitat patches, which are larger than the home range and dispersal potential of blue tits, known to exhibit strong philopatry and site fidelity (Blondel et al. 2006). Within those sites, birds are therefore breeding and living in spatio-temporally stable habitats and thus should synchronize their breeding period to the phenology of the same food source, which could reduce inter-individual variation in LD plasticity. Another non-mutually exclusive explanation for this lack of I × E is that constant directional selection pressures on LD might have shaped the observed reaction norms. Annual selection differentials and gradients on LD show constant strength and magnitude across years, thus selecting for constantly earlier LD and possibly for a specific reaction norm, which in turn could have reduced inter-individual variation in the slope. This is especially likely in our populations as we found positive correlations between elevation and slope of plasticity for laying date.

Table 6.   Summary of results for plasticity and selection analyses of laying date. Yes = result significant at < 0·05
SitePopulation-level plasticityIndividual-level plasticityI × ELinear selection
D-RouviereYesYesYesNo
D-MuroYesYesNoYes
E-MuroYesYesYesNo
E-PirioNo (P = 0·06)YesNoYes

On the other hand, the availability of resources is more heterogeneous in D-Rouvière and E-Muro, where we found no evidence for directional selection on LD, but presence of I × E for this trait. While caterpillars are blue tits preferred food source during the reproductive period, when caterpillar abundance falls below 160 mg of frass per m2 per day, blue tits switch to other food prey items, such as spiders and grasshoppers (Banbura et al. 1994). Inter-annual variation in the timing of resources as a function of temperature may differ between different food sources, as vegetation phenology varies differently among species (Sherry et al. 2007). Thus, in D-Rouvière, low caterpillar abundance (mean annual food abundance at food peak = 60 ± 26 mg of frass per m2 per day) and specialization on different food sources could explain the absence of directional selection on LD and the differences among individuals in LD reaction norms. The E-Muro site is located in a small patch of holm oak surrounded by downy oak at a broader scale. Individuals breeding in this site could thus originate from or forage in both habitats, leading to different selection pressures and plasticity patterns among individuals and a lack of directional selection on LD. Further investigation should focus on identifying the habitat of origin of birds breeding in these two populations, to test whether individual plasticity for LD is related to habitat features. More generally, these results offer new insights on the mechanisms that might shape the levels of I × E observed in natural populations. Further theoretical and empirical studies are needed to understand how selection on a life-history trait influences variation in its plasticity. Such studies are crucial to understand what affects the evolutionary potential of plasticity to evolve in natural populations, particularly in the actual context of global environmental changes, where (i) strong directional selection pressures on traits such as breeding time are expected; and (ii) phenotypic plasticity might be the principal short-term mechanism allowing individuals to adapt to such rapid changes (Gienapp et al. 2008).

Conclusion

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

In conclusion, our results (i) suggest a link between the magnitude of plasticity and environmental predictability; (ii) provide evidence for variability in plasticity patterns of breeding traits among populations of the same species at a small geographic scale; and (iii) identify a possible interplay between patterns of plasticity and selection. We suggest that these patterns are caused by differences in spatio-temporal heterogeneity of the habitat among populations. Our study thus emphasizes the need to consider local scale, habitat type, environmental heterogeneity, and selection regimes to get a better insight on the ecological causes of reaction norm variation and contemporary evolution in the wild.

Acknowledgements

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

We are grateful to the late Donald Thomas for his invaluable contribution to the study of blue tits in Corsica and for early discussion on this manuscript. We want to thank the many assistants and graduate students who helped collect data in the field over the years. We also thank Luis-Miguel Chevin as well as reviewers for useful comments on previous versions of this manuscript. M.P. was supported by a postgraduate scholarship from the Natural Sciences and Engineering Research Council (NSERC) of Canada and a Frontenac scholarship from the MRI and the Consulat Général de France à Québec. The blue tit long-term study is currently funded by the OSU OREME, and a grant from the Agence Nationale de la Recherche (ANR-08-JCJC-0041) to A.C., and by a NSERC Discovery Grant to D.G.

References

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

Table S1. (Co)variance components of plasticity for laying date.

Table S2. Annual selection coefficients for laying date and clutch size.

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

FilenameFormatSizeDescription
jane1996_sm_TableS1-S2.xlsx49KSupporting 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.