Rapid changes in plasticity across generations within an expanding cedar forest

Authors


François Lefèvre, INRA, UR629 Ecologie des Forêts Méditerranéennes, URFM, Domaine Saint Paul Site Agroparc, F-84914 Avignon Cedex 9, France.
Tel.: +33 4 32 72 29 01; fax: +33 4 32 72 29 02; e-mail: lefevre@avignon.inra.fr

Abstract

We investigated the inter-individual variation of phenotypic plasticity and its evolution across three generations within an expanding forest. Plasticity was assessed in situ from dendrochronological data as the response of radial growth to summer rainfall. A linear mixed model was used to account for spatial effects (environment and stand structure), temporal factors (stand dynamics) and the variation with age. Beyond these effects, our results reveal a significant inter-individual variance of growth and plasticity within each generation. We also show that the mean values and variances of growth and plasticity changed significantly across generations, with different patterns for both traits. The possible environmental and genetic drivers of these changes are discussed. Contrasting with the trade-off between stress tolerance and plasticity generally observed among populations, we detected a positive covariance at the individual level, which does not support the cost of plasticity hypothesis in this case.

Introduction

Rapid climate change represents an evolutionary challenge for species and communities. Long-lived organisms like trees will experience durable change in their environment within a single generation time and significant environmental shift in few generations. Phenotypic plasticity, adaptive capacity and migration potential will determine the response of the forests to these changes. First, phenotypic plasticity (i.e. the environmentally induced phenotypic variation of a given genotype) determines the response of the current tree generations to the experienced changes. Secondly, once acclimatized populations survive and regenerate, further genetic evolution may proceed through selection: the process that turns a plastic response into genetic variation after a permanent environmental shift is referred to as ‘genetic assimilation’ (Waddington, 1953; Pigliucci & Murren, 2003). Evidence now increases that tree populations are capable of rapid adaptation, i.e. genetic evolution, in few generations (Daubree & Kremer, 1993; Skrøppa & Kohmann, 1997; Rehfeldt et al., 2001). Such genetic shifts rely on the genetic variation that is maintained within populations and, among other processes, genetic variation can be maintained thanks to G × E interactions combined with gene flow across environments, either spatially or temporally (Gillespie & Turelli, 1989; Ellner & Hairston, 1994). Thirdly, migration takes several generations to adjust the geographical range of a species following the movement of its fundamental or potential niche (Higgins et al., 2003; McLachlan et al., 2005). Adaptation and migration have specific interactions in the context of rapid environmental change (Davis & Shaw, 2001; Savolainen et al., 2007): rapid environmental change increases the adaptational lag (the departure from optimum of the actual response to local environment) which may reduce the migration load (the chance of an immigrant being less fit in the receiving population).

Much work has been devoted to phenotypic plasticity and its ecological consequences (Miner et al., 2005) but understanding patterns of selection for individual phenotypic plasticity within populations under environmental change still deserves attention (Pigliucci, 2005). Few studies have been devoted to the individual level of phenotypic plasticity within wild populations (Brommer et al., 2005; Nussey et al., 2005, 2007; Pigliucci, 2005; Charmantier et al., 2008) and, to our knowledge, none in forest trees.

Among tree populations, elevational or latitudinal clines are frequently observed, which are interpreted as the result of selection and local adaptation processes (Eriksson et al., 1980; Mikola, 1982; Ducousso et al., 1996). The adaptive traits that show strong clinal variation among populations also show an important genetic variation within populations (Kaufman & Smouse, 2001; Savolainen et al., 2004). Based on common garden experiments, clinal patterns of variation for phenotypic plasticity at the population level were also evidenced in trees: latitudinal and longitudinal clines of reaction norms (Matyas & Yeatman, 1992; Rehfeldt et al., 2001, 2002) and elevational clines of G × E interaction (Modrzynski & Eriksson, 2002). By contrast, in a model-based approach that compared the response to climatic variations in situ, Chuine et al. (2000) hardly detected significant variation among populations for only one species out of five. Schlichting (1986) pointed out the fact that, theoretically, the variation for a quantitative trait and G × E interaction can evolve independently or, in other words, that selection can equally change the height and the slope of a linear reaction norm. However, dependence between traits and plasticity was experimentally observed by Rehfeldt et al. (2001): tested in cold conditions, Pinus contorta populations from higher latitudes grow better than populations from lower latitudes, but they grow worse in a favourable environment where temperature is not limiting and, therefore, are less plastic. Similarly, Modrzynski & Eriksson (2002) observed that Picea abies populations from higher elevation have a better resistance to early frost, in terms of bud-set phenology, and have a lower plasticity because their phenological response does not change when drought stress is applied as it does for lower elevation populations. The trade-off between local adaptation to harsh conditions and plasticity at the population level may result from an intrinsic cost of phenotypic plasticity (Relyea, 2002), or from a selection for higher responsiveness in favourable environments. The first hypothesis would also result in a negative correlation between the response to stressful conditions and plasticity at the individual level within population, which is not necessarily the case for the second hypothesis.

To assess the diversity of phenotypic plasticity at the individual level and investigate its short-term evolution within tree populations, we conducted a retrospective micro-evolutionary study within an introduced population that experienced continuous environmental changes and presents an interesting demographic structure in semi-discrete generations. We studied the variation of phenotypic plasticity under the reaction norm approach (Via et al., 1995), assuming a linear model of the phenotypic response in terms of annual radial growth to a continuous environmental parameter, summer rainfall. Using increment cores we retrospectively obtained numerous (up to 90) observations for each individual in contrasted environmental conditions, leading to individual measures of plasticity. We also estimated the covariance between the height and slope of linear reaction norms at the individual level.

Transplanted forests from the 19th and 20th centuries represent a unique experimental opportunity to perform micro-evolutionary studies in trees through retrospective approaches (Zheng & Ennos, 1999). Indeed, populations that were transplanted out of their natural geographical range can inform us on the impact of evolutionary forces that shape the diversity after rapid environmental change: demo-genetic processes and adaptation. Cedar (Cedrus atlantica Manetti) has been introduced in France during the 19th century using seed material collected in Algeria. The Luberon cedar forest is located in the south-east of France, under Mediterranean climate, at 700 m elevation. Originally planted in 1863 to stop erosion in over-grazed areas (Cointat, 1996), cedar has naturally regenerated without any silvicultural management until the 1990s. The species has been transplanted from southern to northern latitude (from Algeria to France). However, initial conditions in the transplantation site were particularly severe for Cedrus: bare and thin soil, summer drought climate and strong dry northern wind (mistral). Thus, initial mortality was high and the density of surviving founder trees was as low as eight trees per hectare as revealed in an aerial picture dated 1939 obtained from the Institut Géographique National. These founder trees produced many seeds and a closed forest was formed at the next generation, which now provides much more favourable conditions for seed germination and seedling survival during summer (density exceeding 50 seedlings per m2 has been observed in a regeneration plot 8 years after germination; F. Courbet, personal communication). Therefore, we hypothesize that a strong initial selection for drought resistance has shaped the founder gene pool and, for the next generations, that strong selection for competing ability occurred within dense seedling patches. Three generations, or more precisely cohorts, of trees are clearly identified today, including the founder trees. In a previous paper, we showed that population admixture had occurred in the planting stock and that hybridization among initial gene pools started as soon as the first regeneration phase on site (Lefèvre et al., 2004). We observed no severe genetic erosion, neither for neutral nor for non-neutral genes, and we concluded that this population has experienced intensive re-organization of the diversity and not only genetic erosion (Fallour et al., 2001; Lefèvre et al., 2004).

Trees are long-lived sessile organisms and their wood provides a record of the plastic response to the annual change in their environment. In this study, we assessed phenotypic plasticity as the response in terms of radial growth to annual climatic stochasticity through a dendrochronological approach, which retrospectively provides growth data of individual trees at different ages in their life. Temporal changes in environment include annual and long-term climatic variation, as well as the evolution of stand structure. The impact of these temporal factors is partly confounded with an internal age effect. We used a global model of analysis that accounted for these different factors and their interactions: climate, tree age and year, as a surrogate for other climatic variation and evolution of stand structure. Our goals in this study were to: (i) estimate the variance of phenotypic plasticity at the individual level within the forest, (ii) investigate the changes in growth and plasticity among successive generations in various environmental conditions within the forest and (iii) investigate the relationship between growth performance and plasticity at the individual level.

Materials and methods

Tree sampling

The cedar forest on Mount Luberon developed from three initial plantations with a distance of about 1500 m between each of these areas. The population has expanded through natural regeneration and currently forms a closed forest of ca. 300 ha. In 1995, we had sampled 197 trees in three zones of the forest, ca. 8–10 ha each, that correspond to the initial plantation areas. An aerial picture dated 1939 showed that initial mortality has been much lower and density much higher in zone 3 compared with that in zones 1 and 2. Soil conditions are homogeneous on a large scale throughout the forest (calcareous lapiaz), but soil depth varies locally at very small scale (Fallour, 1998).

Sampled trees belonged to three different age classes as described in a previous publication (Lefèvre et al., 2004): G0 (the surviving founder trees issued from seeds collected in the natural range), G1 (the first generation that appeared on site from intercrossing among G0 trees) and G2 (the second generation that appeared when G1 trees became fertile, G0 still there). Sampled trees were individually mapped within each zone: successive generations slightly differ in their spatial distribution with a clearly clustered pattern within generation and repulsion between generations (Fig. 1).

Figure 1.

 Spatial distribution of sampled trees within the three studied zones in the forest (G0 trees appear as ‘+’, G1 trees as open circles, G2 trees as filled dots).

Radial growth and age determination

For each tree, we extracted three increment cores by means of a Swedish increment borer, one at the northern side of the tree and the others oriented at 120° each. Because cedar trunks often had an asymmetric shape due to the strong northern wind, the north side core had the highest probability to reach the pith. Increment cores were taken at 60 cm above the ground level, which was the minimum height possible when using this borer. Increment cores were immediately smoothed with a plane after extraction and stored in dry condition under a press until analysis. Cores were visually cross-dated under a stereomicroscope in order to find missing rings or false rings (Stokes & Smiley, 1968; Schweingruber et al., 1990). Then, annual ring widths were measured using an Eklund measuring device up to 10−2 mm precision. After measuring radial series, cross-dating quality was checked by comparing the visual agreement between plotted ring-width series and by calculating correlation coefficients between ring-width series. Thus, our temporal calibration of ring-width series was not sensitive to the estimation of age, which allowed using all trees simultaneously to estimate the mean effect of each year on radial growth (Year effect in the statistical analysis below).

For most (> 80%) of the trees, at least one of the three cores included the pith. For the others, the number of inner rings missing in the core was estimated from the curvature of the innermost rings according to the geometric method developed by Duncan (1989). To assess the age of each tree, we estimated the number of invisible annual rings at 60 cm height on the basis of the observation of young seedlings grown in different environments in the forest whose age could be precisely estimated from their branching architecture (data not shown). Thus, we added 10 years for the G1 and G2 trees, and 15 years for the founder G0 generation that grew in harsh conditions, which finally gave an estimated date of species introduction that agrees with written historical records.

Climatic data

The climate series were obtained from Météo France. They had to be sufficiently long and had to be situated as close as possible to the study area for the analysis of climate–growth relationships. The best series available were obtained from the meteorological station at Avignon, located at 50 m above sea level, which provided mean monthly maximal and minimal temperatures as well as monthly precipitations for the whole study period that we restricted to the period 1904–1994 in order to avoid missing data. Avignon and the Luberon mountain differ in altitude, but they belong to the same climate sub-unit according to monthly precipitation patterns as defined by Guiot (1986).

Under Mediterranean climate, rainfall is expected to be the main limiting factor for tree growth on thin soils. A previous analysis on the period 1921–1979 (Guibal, 1985) revealed that, among several climatic parameters, rainfall in spring and summer were generally the best predictors of annual ring width for cedars, thus showing that water supply was the limiting factor for radial growth. Therefore, we computed cumulative rainfall from May to August (MJJA) as a yearly climatic parameter. It varied from 16 to 421 mm during the study period. Note that some statistics described below refer to the theoretical value MJJA = 0 which represents stressful conditions that still remain ecologically sound, where no rainfall would occur during this 4-month period regardless of the rainfall during the rest of the year.

Rainfall from May to August was comparable among generations (Fig. 2). Indeed, there was no significant trend for the MJJA climatic variable over the study period 1904–1994 (Pearson’s correlation between MJJA and year was r = 0.016, P = 0.88). Furthermore, there was no difference in average MJJA among the three periods 1904–1920 (G0 trees alone), 1921–1950 (G0 and G1 trees) and 1951–1994 (all three generations; anova test F2,88 = 0.83, P = 0.44). The first period showed a slightly higher variance among years but this was not significant (Levene’s homogeneity of variance test, F2,88 = 2.82, P = 0.064).

Figure 2.

 Annual variation of summer rainfall from May to August (MJJA) during the study period and mean radial growth increment (RGI) for each generation of trees (dotted line for generation G0, solid line for G1, bold line for G2).

Statistical analysis

For each tree, we computed annual radial growth increment (RGI) as the mean of annual ring width by averaging the three core chronologies. Phenotypic plasticity was defined as the linear response (slope) in RGI to summer rainfall. A linear mixed effect model was used where RGI varied among zones, generations and years, depending on the age of trees and summer rainfall (MJJA) as covariates. A random effect of individuals within each zone and generation was also included. The variation of plasticity among zones and generations, but also among individuals within each generation, was directly investigated within the model by the use of interaction effects implying MJJA. Thus, the terms included in the global model were as follows:

  • Zone (α, fixed): large-scale spatial heterogeneity of environmental condition (topography and stand dynamics) in different parts of the forest;
  • Generation (β, fixed): a genetic effect partly confounded with a micro-environmental effect because of different spatial distributions of generations within each zone;
  • Zone × Generation (αβ, fixed): the interaction among previous effects;
  • Age (covariate): we assumed a quadratic effect of age to account for the juvenile and senile phases of RGI, which was not corrected for tree diameter prior to the analysis;
  • MJJA (covariate): a linear effect of summer rainfall on radial growth (it is sufficient to consider a linear rather than a quadratic relationship because water availability is a limiting factor for growth in the Mediterranean);
  • Year (δ, fixed): a categorical variable beside the climatic covariate MJJA to account for other annual climatic variability and all other temporal variations, in particular the evolution in stand structure (tree density);
  • MJJA × Zone and MJJA × Generation (fixed): interaction terms accounting for the difference in plasticity among zones and among generations;
  • MJJA × Generation × Zone (fixed): variation among zones of the generational differences in plasticity;
  • MJJA × Age and MJJA × Age2 (fixed): variation of the response to climate with age, constant among zones and generations;
  • Individual (random): variance of growth among individual trees within each zone and generation; differences among generations of inter-individual variances were allowed;
  • MJJA × Individual (random): variance of phenotypic plasticity among individual trees within each zone and generation; differences among generations of inter-individual variances were allowed;
  • a covariance between the Individual effect and the MJJA × Individual effect was estimated within each generation.

We also investigated the presence of a spatial structure in the Individual effects, and whether the variance among individuals and the spatial structure varied among generations (generation as a GROUP of the RANDOM effect in SAS).

Finally, we included a temporal covariance structure in the residuals to account for the intra-individual correlations among observations. We used a first-order autoregressive model, AR(1), where the correlation between two residuals of a given individuals decreases with the lag t between these observations as a function corr(t) = ρt. The correlation between two residuals of two different individuals equals zero. The auto-correlation parameter ρ was estimated jointly with all other parameters. A positive value of ρ may be due to an ‘after-effect’ of any single event over several consecutive years of growth. A negative auto-correlation ρ is expected when high and low growths alternate in successive years.

The overall model can be written as follows:

image

where z, g, i, y, respectively, stand for Zone, Generation, Individual and Year, and where

  • inline image is the inter-individual variance of annual growth in stressful conditions (RGI, predicted at MJJA = 0), potentially varying among generations;

  • inline image is the inter-individual variance of phenotypic plasticity (PP), potentially varying among generations;

  • cov(Ai(z,g), Bi(z,g)) = σA(g)σB(g)ρAB(g) is the inter-individual covariance between growth and plasticity, this parameter was assumed to vary among generations; this covariance was also used to compute the inter-individual variance of annual growth in favourable conditions (RGI, predicted at MJJA = 350) as follows: var(Ai(z,g)) + 3502 var(Bi(z,g)) + 700 cov(Ai(z,g), Bi(z,g));

  • var(Ez,g,i,y) = σ2E is the residual variance within individual and year, the temporal structure of the residuals being:

    image

Analyses were conducted with proc mixed (SAS Institute, Cary, North Carolina, USA) using the REML method. We systematically used type III tests to eliminate partially confounding fixed effects, except for MJJA for which we used type I test adjusted for Zone, Generation, Zone × Generation, Age and Age2 because this would be totally included within the Year effect in a type III test. In order to achieve a robust test of fixed effects, we used the Satterthwaite correction to compute the degrees of freedom (Littell et al., 2006). Note that tests of fixed effects (such as Generation) refer to the difference among treatments for null values of the covariates and, therefore, to stressful conditions where MJJA = 0.

As trees from different generations both differed in age at a given year and experienced different stand structure or different climatic conditions at a given age, mean values per generation were not directly comparable due to confounding effects. Therefore, for each generation within each zone, we computed mean values of growth and plasticity adjusted for Age, at ages ranging from 10 to 100 years, and for Year, restricting to 1960–1994 that represents a period of stable stand structure shared by all trees from the three generations. We first obtained predicted growth values for each year and each age during this period using the estimated model. Then, we computed adjusted mean values of growth under stressful climate and plasticity, respectively, by the intercept and the slope of the regressions of predicted growth on corresponding values of summer rainfall; we then used these two parameters to compute adjusted mean values of growth in favourable conditions (i.e. value of MJJA = 350). These adjusted mean values were comparable among generations.

Results

Variation patterns for radial growth increment

Although many observations contributed to the analysis (12 901), the global model accounted for a fair amount R2 = 54% of total variation. Differences in RGI among zones was highly significant (P < 0.0001, Table 1). In particular, zone 3 had growth lower than that of zones 1 and 2. Globally, over the whole study period, we detected no difference among generations, neither for the whole forest nor within each zone (Generation and Zone × Generation effects non-significant, Table 1). However, the comparison of growth potential among generations deserves a more detailed analysis because, over the whole study period, they experienced different environmental changes at different ages.

Table 1. anova tests of fixed and random effects on radial growth increment.
Fixed effectsNum d.f.Den d.f.FP-value
Zone214619.06< 0.0001
Generation22602.620.075
Zone × Generation41072.390.056
Age153329.89< 0.0001
Age21127376.58< 0.0001
MJJA12741618.02< 0.0001
Year893818131.62< 0.0001
MJJA × Zone212213.09< 0.0001
MJJA × Generation214950.59< 0.0001
MJJA × Generation × Zone487.42.070.091
MJJA × Age127056.65< 0.0001
MJJA × Age219261256.60< 0.0001
Random effectsLevelVariance/covP-value
  1. Fixed effects were tested following type III tests, except summer rainfall covariate (MJJA) for which we used type I test to avoid complete confounding with Year effect. Variance components were estimated using REML. See text for further details on the model of analysis.

Individual (at MJJA = 0)within G0162.100.259
Individual (at MJJA = 0)within G10
Individual (at MJJA = 0)within G2173.860.404
Individual at MJJA = 350within G03073< 0.0001
Individual at MJJA = 350within G125870.0003
Individual at MJJA = 350within G258470.0006
MJJA × Individual = PPwithin G00.016< 0.0001
MJJA × Individual = PPwithin G10.0090.0003
MJJA × Individual = PPwithin G20.0100.045
cov(Individual, PP)within G01.3290.051
cov(Individual, PP)within G12.1800.016
cov(Individual, PP)within G26.329< 0.0001

Globally, there was a highly significant response of RGI to summer rainfall MJJA (P < 0.0001, Table 1) that confirms the relevance of this climatic variable as a determinant for growth. Besides, the Year effect was highly significant (P < 0.0001, Table 1), which reveals another prevailing environmental factor of temporal variation different from summer rainfall. This temporal Year effect is largely explained by the evolution of stand structure as can be seen from Fig. 2: see the much larger growth of G0 trees alone in the first years compared with juvenile growth of G1 and G2 individuals, see also the decrease in the growth of G0 and G1 individuals, whereas G2 trees emerged during the 1960–1975 period.

The quadratic effect of age on RGI was highly significant and provided a parabolic shape of radial growth with a maximum at ages 67, 61 and 58 for summer rainfall MJJA of 0, 200 and 350 mm respectively.

We estimated the Year and quadratic Age effects from the global model and used these parameters to predict the RGI values that each generation would experience within each zone during any of the years within the period 1960–1994 of stable stand structure and where all three generations were effectively present. The same pattern was observed in all three zones: the G0 generation had the lowest predicted growth in stressful climate (i.e. MJJA = 0), whereas the G2 generation had the highest predicted growth in favourable conditions (i.e. MJJA = 350), the G1 generation was rather similar to G2 in stressful climate and to G0 in favourable conditions (Fig. 3). In zone 3, the difference between generations was larger for growth in stressful conditions but smaller for growth in favourable conditions (Fig. 3).

Figure 3.

 Adjusted mean values of radial growth increment (RGI) under limited summer rainfall (MJJA = 0 mm) or high summer rainfall (MJJA = 350 mm) and phenotypic plasticity (PP) as quadratic functions of age (ranging from 10 to 100 years), for each generation in each zone during the period 1960–1994 of shared and stable stand structure. The model assumes the same quadratic effect of age for all generations and zones (dotted line for generation G0, solid line for G1, bold line for G2).

Inter-individual variance estimates for RGI at MJJA = 0 were not significantly different from 0 within any generation (Table 1). Inter-individual variance estimates for RGI at MJJA = 350 were almost 20 times higher, decreasing from G0 to G1 and increasing from G1 to G2.

No spatial structure of individual effects could be detected. This means that close individuals did not show correlated RGI and implies that the clustering of individuals of the same generation (see Fig. 1) should not be responsible for the detected differences in mean RGI among generations.

Temporal autoregressive structure in the residuals was highly significant and the autocorrelation parameter estimates were positive within each generation (AR(1) = 0.799 for generation G0, AR(1) = 0.867 for G1 and AR(1) = 0.756 for G2 with P < 0.0001 in all cases). This temporal pattern was beyond the temporal trends included at the higher level within the Year and Age effects.

Variation patterns for phenotypic plasticity

The response to summer rainfall varied with age in a quadratic way (MJJA × Age and MJJA × Age2 effects both significant, P < 0.0001, Table 1). Plasticity was maximum at intermediate age, i.e. in a multiplicative effect where growth variation due to climate was maximum when growth was maximum. There was a significant variation of plasticity among zones (different slopes in the different zones depicted by the MJJA × Zone effect, P < 0.0001, Table 1) and zone 3 was characterized by lower predicted values of plasticity for the 1960–1994 period (Fig. 3). Furthermore, the variation of plasticity among generations was highly significant (MJJA × Generation effect, tested for a value of Age = 0, P < 0.0001, Table 1), even greater than variation among zones. This variation among generations was stable over the three zones of the forest (MJJA × Zone × Generation not significant, P = 0.091, Table 1). The difference between generations for predicted values of plasticity during the 1960–1994 period resulted from the differences for RGI both in stressful climate and favourable conditions as previously mentioned. Predicted plasticity values increased from G0 to G2 in zone 1 and zone 2, mainly because of increased growth potential in favourable conditions. The situation was different in zone 3 where growth potential in stressful climate increased even more between G0 and G2, which finally resulted in similar values of plasticity between the two generations. The G1 generation had the lowest predicted plasticity due to a higher RGI in stressful climate than G0 and a lower RGI in favourable conditions than G2 (Fig. 3).

Inter-individual variance of plasticity was highly significant within G0 and G1, and it was marginally significant within G2 (Individual × MJJA variance, Table 1). The inter-individual variance of plasticity slightly decreased from G0 to G1 and remained constant from G1 to G2. The goodness of fit of the regression of growth on summer rainfall for each individual tree varied from R2 = 0.1% to 33% in the raw data and from R2 = 2% to 63% when using growth values predicted by the global model (Fig. 4). As expected from the Individual × MJJA variance parameters, the fit was better for the younger generation G2.

Figure 4.

 Goodness of fit of the individual regressions of growth on summer rainfall within each zone and each generation (a) distribution of the R2 values of individual regressions on raw data, (b) distribution of the R2 values of individual regressions on predicted values from the global model.

Within each generation, the inter-individual covariance between growth (estimated in the model at MJJA = 0) and plasticity was significantly positive and strongly increased from G0 to G2 (Table 1).

Discussion

Over a short period of time (< 5–10 years), the main environmental factor that determines the annual variation of radial growth is the climate. Over a longer period of time, evolution in stand structure and other ecological parameters, as well as the age of the trees must also be taken into account. In this work, we used dendrochronological data as long-term temporal records of the response to annual climate variations. A large data set was produced and a complex model of analysis was required to disentangle the various factors that determine the variation of RGI, especially to adjust the data for the effects of age and large temporal scale stand structure before comparing generations.

Globally, beyond the effect of age, we found a significant response of growth to summer rainfall and longer term changes (MJJA and Year effects respectively). We detected a highly significant response to summer rainfall, although this climate variable was measured in Avignon some kilometres away from the forest, at lower altitude: this reveals the good correlation of seasonal variations across years at this geographical scale. In the statistical model that we used, the discrepancy between summer rainfall in Avignon and the Luberon forest on particular years, e.g. due to local storms, was included in the Year effect. Therefore, although already highly significant, the actual plasticity of trees may be underestimated in our analysis and the differences among groups of individuals even larger than detected here. We found an AR(1) structure in the residuals that could be explained by a lag in the effect of annual climate on growth or by a failure of the model to explain perfectly the effect of age and year on RGI.

Within the forest, the three sampled zones significantly differed for growth and plasticity. The lowest growth was found in zone 3, although the highest survival rate of the initial plantation had occurred in this area. This variation in growth performance among zones relates both to ecological heterogeneity as well as differences in stand structure and stand dynamics. After initial plantation in harsh conditions to which G0 trees had to survive, the environmental conditions have progressively changed since the establishment of the G1 generation: an organic layer has developed, a semi-closed stand structure has developed providing a shelter for young seedlings during hot and dry summer, the competition from herbaceous species has decreased, whereas the seed set and the competition among seedlings have increased. Higher initial survival rate in zone 3 resulted in higher density and higher competition in the area at generation G0, thus reducing both radial growth and its response to climate (plasticity) through a multiplicative effect.

Over the whole study period, there was no significant variation of radial growth under stressful conditions, neither among generations nor among individuals within generation. However, this is not direct evidence of the absence of genetic effects for growth because, during the entire period of time, the different generations did not experience the same stand structures at the same age. Indeed, the adjusted mean values of growth for the period 1960–1994 systematically increased from the G0 to G2 generation in each of the three zones, i.e. in three different environments, both under stressful and favourable climatic conditions (MJJA = 0 and 350 respectively). Furthermore, in medium stress conditions (analysis with centred MJJA, data not shown), the Generation effect on RGI significantly increased from G0 to G2 even for the whole study period. Moreover, individual variance estimates in favourable conditions (MJJA = 350) were significant within each zone and each generation for the whole study period and, finally, there was no spatial structure of the individual effects as would be expected if micro-environmental heterogeneity were the main driver of the inter-individual variance. These results strongly suggest that genetic variation exists for growth potential and that it increased in two generations, although it was masked in the global analysis by major environmental changes that occurred over the study period. The inter-individual variance for growth decreased from G0 to G1 and then increased from G1 to G2: such an evolution was expected given the initial population admixture at G0 (Lefèvre et al., 2004), forming a ‘hybrid generation’ at G1, further segregating at G2.

Individual variance estimates for phenotypic plasticity were significant within each generation over the whole study period. The absence of spatial structure among individual effects is a clear indication that such variation among individuals within generations was not only due to micro-environmental effects but also, at least partly, under genetic control. The existence of genetic diversity for the plasticity of a trait not only has an impact on its response to selection when environment changes (Pigliucci & Murren, 2003; Crispo, 2007), but it also means that plasticity itself can evolve. Previously, by using common garden experiments of controlled crosses in eucalyptus, Bouvet et al. (2005) had demonstrated that plasticity, defined as growth response to stand density, was under genetic control and could therefore respond to selection.

The unbalanced design was an unavoidable challenge for such in situ study of plasticity across several generations of trees: individuals from different generations did not experience the same stand structure at the same age. Therefore, we assumed that each zone and each generation respond similarly to age but can vary in growth and plasticity and we used type III tests to avoid confounding between Age, Year and Generation effects on growth and plasticity. We checked, by simulating data on the same unbalanced experimental design as here (data available as online Supporting Information, see Appendix S1), that the type III tests would not artificially lead to a significant difference in plasticity among generations when such difference does not exist. For example, decreased growth and/or increased plasticity with years (that could be related to the evolution of stand structure) did not lead to confounding between Generation and Year effects. The only simulation where nonexisting difference of plasticity would appear more frequently than 0.05 was in the case of important spatial structure of growth and plasticity (confounded with spatial clustering of the generations), which was not the case here. Thus, based on the assumption of constant response to age, the adjusted mean values for the common period of stable stand structure 1960–1994 computed from the effects estimated over the whole data set could be compared between generations.

We revealed significant changes in phenotypic plasticity across generations both over the whole period and the restricted period of stable stand structure. These changes did not simply result from a multiplicative effect as did the difference among zones: for example, the G1 generation had a higher growth in dry conditions than G0 but a lower plasticity and similar growth in favourable years. Three nonexclusive hypotheses could account for the variation of plasticity across generations: (i) an environmental effect on the response to age inducing a difference of this response among generations (that we assumed to be constant in our analysis), (ii) a genetic response to various selection pressure (selection for tolerance to drought stress in G0 vs. selection for competing ability and higher responsiveness in G2) or (iii) a genetic change due to complex genetic architecture and segregation patterns following initial population admixture. The first hypothesis would represent a long-term record of environmental changes within the lifespan of trees; the other two hypotheses would represent genetic evolution. The first two hypotheses depend on environmental conditions, and we would therefore expect different generational changes among zones, whereas the third hypothesis does not depend on environmental conditions and should have the same impact anywhere. Indeed, in all zones we observed a decrease in plasticity from G0 to G1, followed by an increase from G1 to G2, but the amount of increase from G1 to G2 varied among zones. Therefore, we suspect that phenotypic plasticity has genetically evolved since the original admixture, with or without further response to selection. Contrasting with growth, the inter-individual variance of plasticity did not increase from G1 to G2 as expected from the admixture scenario alone, suggesting a stabilizing selection constraint in G2. This would indicate different selection patterns for growth and plasticity in this forest. The environmental or genetic origins of the generational changes in plasticity still need further investigation, using in situ or common garden approaches.

The evolution of phenotypic plasticity raises the question of costs and limits of plasticity that has been treated theoretically more often than empirically (Relyea, 2002; Steinger et al., 2003; Pigliucci, 2005). In this study, at the individual level within each generation, we detected a positive covariance between plasticity and tolerance to drought stress. Interestingly, despite the positive covariance at the individual level, we observed one case of antagonistic variation across generations: in zone 3, the G1 generation had a better growth in stressful conditions but lower plasticity than G0. More generally, the G1 and G2 generations did not differ for tolerance to drought stress but differed for plasticity. Therefore, the Cedrus case does not support the hypothesis of intrinsic cost of plasticity, it rather shows an increased responsiveness at the G2 generation. If selection has contributed to the change in plasticity across generations, the correlated increase in growth and plasticity from G1 to G2 would suggest a Baldwin effect of adaptive evolution through adaptive phenotypic plasticity rather than a genetic assimilation process with increased canalization (Crispo, 2007). In the absence of cost of plasticity at the individual level, the balance between selection pressure for stress tolerance or competing ability in different environments could explain the trade-off between plasticity and stress tolerance generally observed among forest tree populations.

Acknowledgments

We thank F. Courbet, Ph. Dreyfus, A. Franc, C. Lavigne, O. Ronce, L. Sanchez and the anonymous reviewers for their very helpful comments on an earlier version of this manuscript. We thank B. Jouaud and D. Vauthier for technical assistance. We greatly acknowledge the Office National des Forêts and the Parc Naturel Régional du Luberon for their assistance and, particularly, for helping us to find historical archives. This work was part of the project ‘Adaptation and selection of Mediterranean Pinus and Cedrus for sustainable afforestation of marginal lands’ FAIR CT95-0097 financed by the Commission of the European Communities DGVI.

Ancillary