Phenotypic Plasticity of Growth Trajectory and Ontogenic Allometry in Response to Density for Eucalyptus Hybrid Clones and Families

Abstract

• Background and Aims Response to density is a crucial aspect of the ecology of trees in forests and plantations. Few studies have investigated the genetics of plasticity in response to density for growth traits such as height and circumference through development.

• Methods Two experiments were carried out in the field, the first with full-sib families of Eucalyptus urophylla × E. grandis hybrids, and the second with clones of E. tereticornis × E. grandis hybrids planted across a range of densities (625, 1111 and 2500 trees ha⁻¹). Height, circumference and stem taper were measured through development in both experiments. Variance components were estimated and a repeated measure approach for plasticity and three different methods were used to compare the variance–covariance matrix across densities.

• Key Results Genetic variance was significantly different from zero but the density × genotype interaction was significant only for clone experiments at the adult stage. Significant plasticity for three traits in both experiments was found. In the clone experiments, a significant clone × time × density interaction was found, suggesting that plasticity for growth and stem form is under genetic control. In both experiments, density did not affect environmental correlation, which remained high throughout tree development. The impact of density on genetic correlation was marked in the clone experiment, with a reduced value at lower density, but was not observed in the family trial. The differences between clones and family are mainly explained by the distribution of genetic variation within and among genotypes.

• Conclusions The results suggest that plasticity for growth traits and form of tropical Eucalyptus species is under genetic control and that the environment changes genetic co-variation through ontogeny. The findings confirm that a tree population with a narrow genetic basis (represented by clones) is sensitive to a changing environment, whereas a population with a broader genetic basis (full-sib family here) exhibits a more stable reaction.

INTRODUCTION

Environmentally induced phenotypic change, phenotypic plasticity, has been recognized as an important strategy for plants to maximize or maintain fitness in variable biotic and abiotic environments (e.g. Via and Lande, 1995; Schlichting and Pigliucci, 1998; Wright and McConnaughay, 2002). In an organism, phenotypic plasticity generally concerns different traits that can be related by an allometric relationship which varies through development (e.g. Archibald and Bond, 2003; Bonser and Aarssen, 2003). Although these issues have been addressed in several plant species (Schlichting and Pigliucci, 1998; Preston and Ackerly, 2004), very few studies exist for tree species, especially those addressing the genetic basis of phenotypic plasticity and ontogenic allometry.

Being long-lived organisms, trees experience fluctuations in their competitive neighbourhood both spatially and through development (Coomes and Grubb, 1998; Alves and Santos, 2002; Martinez and Lopez-Portillo, 2003). Competition due to increased density is among the most crucial factors that influence growth and trait expression in both natural forest tree stands (Archibald and Bond, 2003) and in forestry plantations (Evert, 1971; Bouvet et al., 2003). Higher tree density translates into a reduction of light, water and nutrient resources which, in turn, can have manifold phenotypic effects on an individual. Density generally reduces tree circumference and to a certain extent tree height (e.g. Dhote 1997). Higher density also affects overall plant vigour through reduction in canopy size, and fruit production, both fitness-related traits (Archibald and Bond, 2003). In the most widely planted forestry species, including pine (Kusnandar et al., 1998), temperate eucalyptus (Wei and Borralho, 1998) and tropical eucalyptus hybrids (Bouvet and Vigneron, 1995), it has been recognized that genetic and environmental variances change with age. Our temporal understanding of the genetic variation of traits through development comes primarily from models that compare the magnitude of variances of growth traits at different stages of stand development (Balocchi et al., 1993), but very few studies have taken a developmental reaction norm perspective to examine trait variation across changing environments (Brouard and John, 1999; Bouvet et al., 2003). Here, the impact of inter-plant spacing on the plasticity of growth trajectories and on ontogenic allometry is analysed from a genetic point of view.

Tropical Eucalyptus and their hybrids are among the species most used in plantations. Because of their economic value, species of the genus Eucalyptus have been the subject of many research programmes in genetics and silviculture (e.g. Vigneron and Bouvet, 2000; Henson and Smith, 2004; McRae et al., 2004). In addition, the rapid growth, short period until reproductive maturation (adult size is reached after 5–7 years), and their ability to be cloned easily facilitate their use in forest genetic experiments. Numerous studies have demonstrated significant additive genetic components in growth traits through development (Bouvet, 1995; Bouvet and Vigneron, 1996) and have shown that several quantitative trait loci (QTL) explain variation (Verhaegen et al., 1997; Gion, 2001), but no studies have addressed the genetic basis of ontogenic plasticity.

In this study, based on two experiments including clones and full-sib families of eucalyptus hybrids, several specific questions are asked: (1) Is there genetic variation for plasticity of growth trajectory for height, stem circumference, and allometry between those two traits across tree densities? (2) Is there variation through ontogeny for the genetic and environmental correlations between height and stem circumference with changing tree densities? (3) Are there differences in patterns of plasticity between two sets of hybrid populations (one population of clones and the other of families)? The experiments were conducted with two populations that differ in their within-genotype variation (no genetic variation within each clone, high within-family variation). Although these hybrid populations have a single parent species in common, it is possible to gain some insight into whether the population with a broader genetic diversity exhibits a more stable reaction to changing environments.

MATERIALS AND METHODS

Study organisms and experimental design

Data were obtained from two field trials established in the Republic of Congo at the experimental research centre — Unité de Recherche sur les Plantations Industrielles du Congo (04°45′S, 12°00′E, 50 m a.s.l.). The climate is tropical humid with a mean annual temperature of 24 °C and a mean annual rainfall of 1200 mm. The dry season lasts from May to October. The first experiment, R91-1, used a set of 12 hybrid clones of Eucalyptus tereticornis × Eucalyptus grandis planted in 1992 at three densities 2 m × 2 m (2500 trees ha⁻¹), 3 m × 3 m (1111 trees ha⁻¹) and 4 m × 4 m (625 trees ha⁻¹). The 12 clones were originally selected from a naturalized hybrid population from the Republic of Congo. The E. tereticornis parent trees represent the progeny of a single tree introduced to the Congo in the 1960s. Thus, the 12 clones were likely to be closely related, but there is not enough information to determine kinship coefficients. The experiment used a split-plot design, density as the main effect and clone as the secondary effect. Rectangular plots of 30 trees were replicated across three blocks. To minimize the impact of competition between plots, a buffer zone of one row of trees was planted between blocks and only the 12 inner trees of the 30 were measured.

The second experiment, R95-10, used a set of 16 hybrid families of Eucalyptus urophylla × Eucalyptus grandis. The hybrid families were created by controlled pollinations in a mating design that included 16 unrelated parents for each species. The experiment was planted in 1995 at the same three densities 2 m × 2 m (2500 trees ha⁻¹), 3 m × 3 m (1111 trees ha⁻¹) and 4 m × 4 m (625 trees ha⁻¹). The experiment was a split-plot design with three blocks, with density as the main effect and family the secondary effect. Each square plot was composed of 36 trees. The total number of trees per family was 324. To minimize the impact of competition between plots, in each plot only the 16 inner trees were measured.

Circumference at breast height (C) and total height (H) were measured at several points during ontogeny. For the clone experiment, measurements were done monthly between months 9 and 24, biannually during the juvenile and the mature phase which spanned from 24 to 74 months. In the family experiment, measurements were done biannually from 7 to 39 months and yearly thereafter. Allometry was assessed by examining the form of the stem, which was quantified from the relative rate of change in stem diameter with increasing tree height (hereafter called the stem taper). Stem taper (DM) was estimated using the formula DM = C/(H – 1·3), cm m⁻¹ (developed in forestry science; Larson, 1963). Mortality was assessed at the time of each measurement. The percent mortality was calculated within each plot as the ratio of dead individuals to the total number of planted individuals.

There are two non-independent interpretations that can be made when contrasting the two experiments. First, the hybrids of the two experiments share one parent species, but the other two parent species are different. Therefore differences between the experiments could be attributed to differences in genetic background of the species that are not shared. Second, since the hybrids share one-half of their genetic background, it could be assumed that the observed differences in variance trends with age are explained by the different distributions of genetic variation between and within the two designs (family and clones).

In experiment R91-1, as the clones were selected from a hybrid population, the variance among clones is an estimation of the total genetic variance including additive, dominance, and epistatic effects. The within-clone variance estimates the variance due to environmental effects. This is a classical application of the quantitative genetic model (Gallais, 1991). On the other hand, in experiment R95-10, the variance among families is an estimate of half of the additive genetic variance and a quarter of the dominance variance and some percentage of epistatic variance. The within-family variance includes both genetic and environmental variance (Gallais, 1991; Falconer and MacKay, 1996).

Data analyses

Analysis of interaction (model 1)

The analysis of plasticity through development was first examined through classical analysis of variance in SAS version 8, GLM procedure (SAS Institute, 1988). In order to fulfil the assumptions of ANOVA, height and circumference were log transformed. In models testing the interaction of principal factors, the residual variances are assumed to be identical for all spacings, and Bartlett's test was used to verify homogeneity of variances (Snedecor and Cochran, 1989). At each age, the GLM procedure tested the density, genotype, the block and the genotype × density interaction effects. All terms were tested with the appropriate error variance. Density, block and family or clone effects were tested over the density by family or clone interaction. The density by genotype interaction was tested over the full model error term. Although the sample size is limited for both clone and family experiments (12 and 16 genotypes, respectively), a random model considering the clone and family effects as random and the interaction with density as random was used.

Repeated measure analysis of variance (model 2)

Plasticity through development was analysed through repeated measure analyses of variance in SAS version 8, GLM procedure (SAS Institute, 1988) following the methodologies of Pigliucci et al. (1997). The model included time (indicating growth in traits over time), block (indicating micro-environmental variation among blocks), family or clone (indicating genetic variation), time × density interaction (indicating plasticity in the growth trajectory), time × family or clone (indicating genetic variation in growth trajectory) and time × density × family or clone interaction (indicating genetic variation for plasticity of growth) over time.

All terms were tested with the appropriate error variance; density, block and family or clone effects were tested over the density by family or clone interaction. The density by genotype interaction was tested over the full model error term. The multivariate Wilk's λ test was used to account for the reduction in the number of degrees of freedom due to the non-independence of the variance–covariance matrices of the time intervals. Wilk's λ is a conservative test and appropriate for unbalanced designs, which is the case in both experiments.

Estimation of variance–covariance matrix within each density (model 3)

Covariance matrices between height and circumference were estimated utilizing standard quantitative genetics techniques (Falconer and MacKay, 1996). Analyses were conducted separately for each experiment. For each density, the mixed model used included block (fixed effect to account for micro-environmental variation), family or clone (a random effect with variance of and mean 0), and the interaction between family or clone and block (random effect with variance of and mean 0). Genetic and residual variance and co-variance components between height (H) and circumference (C) were calculated using the SAS GLM procedure (MANOVA option) and the Type III sums of squares (SAS Institute, 1988). The variance and covariance components were estimated by equating the expected mean square (calculated by the random option) to the mean square given by type III of analysis of variance (Becker, 1984). The correlation coefficients between H and circumference C were calculated using the estimates of variance and covariance given by the mixed model. Two correlation coefficients of correlation can be estimated, σ_{(H, C)} being the covariance between traits H and C, and being the variance of height and circumference. The genetic correlation coefficient can be calculated using: ρ_g = σ_g(H,C)/σ_g(H) × σ_g(C). The residual within-population coefficients of correlation ρ_r = σ_r(H,C)/σ_r(H) × σ_r(C) are interpreted differently for the two experiments. The residual correlation coefficient estimates environmental correlation for the clone experiment and estimates a part of both genetic and environmental correlation for the family experiment (see also Gallais, 1991).

Comparison of variance–covariance and correlation matrix

There have been several attempts to compare the covariance matrix or correlation matrix of populations in variable environments. Roff (2002) suggested that using complementary methods to examine covariance matrices is the most appropriate methodology. Here three approaches were used, two based on correlation coefficients and one based on the covariance matrix.

Comparisons of correlation coefficients: method 1

First, the two correlations were examined to see if they have different strengths (Papoulis, 1990). The null hypothesis for this test is that both samples show the same correlation strength, i.e. ρ_g1 = ρ_g2. The test requires the assumption that the values of both members of both pairs of samples conform to a normal bivariate distribution. To approximate a normal distribution, the correlation coefficients are transformed using the Fisher z-transformation (Sokal and Rholf, 1969). The z statistic is used to determine the level of significance. The first method is the least sensitive of all the tests used.

Comparisons of correlation coefficients: method 2

The correlation coefficients were compared by using a jackknife approach (e.g. Knapp et al., 1989; Roff, 1997). A t statistic is constructed where ρ_g is the genetic correlation for the ith population (i = 1, 2), n_i is the number of pseudovalues generated in the jackknife procedure for the ith population and S_i is the estimated standard deviation of the pseudovalues. In the cases of experiments R91-1 and R95-10, n_i is the number of families (16) and clones (12). The comparison of coefficient is done using the t-statistic formula:

This is a usual formulation for t-statistics and its significance is tested with appropriate tables.

Comparisons of variance–covariance: method 3

The method developed by Roff (2002) is based on both the jackknife procedure and multivariate analysis of variance (MANOVA), following a log transformation to minimize scaling effects. In each experiment the genetic variance–covariance matrix is estimated after deleting in turn one genetic unit, a clone in R91-1 or a family in R95-10, and the pseudovalues of the matrix are calculated according to the jackknife procedure. These estimates are then used as data points in MANOVA to compare experimental tree densities. MANOVA has the advantage of taking into account the correlation between the variance and the co-variance.

RESULTS

Impact of density on mortality, growth and form

The two trials were differently affected by mortality, with greater overall mortality in the family experiment (Table 1). In the clone experiment R91-1, a low percentage of dead trees across all three densities was found (Table 1), although there was a slight (but non-significant) increase in percentage with increasing age and density. In the family experiment R95-10, the mortality was fairly high, especially at the final measurement for both the 625 trees ha⁻¹ and 2500 trees ha⁻¹ densities (Table 1). The increase in mortality over time was greater for the family than the clone experiment. Mortality in the family experiment was mainly due to termite-generated damage during the first 18 months, and is not family- or treatment-specific.

Table 1.

Trends in mortality (%) with age (months) in the clone and family experiments, for each density (625, 1111 and 2500 trees ha⁻¹)

Clone trial: R91-1							Family trial: R95-10
	Trees ha−1						Trees ha−1
Months	625	1111	2500	Months	625	1111	2500
12	1·6	2·3	3·4	11	5·4	3·4	7·5
24	1·6	2·3	3·5	20	9·7	4·6	9·4
48	2·1	2·5	3·9	39	11·0	5·1	10·4
74	2·1	2·5	3·9	59	12·0	6·2	11·3

Percentages are cumulative.

Tropical eucalyptus species generally exhibit rapid growth during the first 2 or 3 years and then the growth rate progressively declines (Bouvet, 1991). This pattern (illustrated in Figs 1 and 2) was observed in both experiments. By later life stages there is evidence of strong competition, demonstrated by a significant reduction in circumference and height as experimental density increases (Table 2). Tree height decreased by <30 % from 625 to 2500 tree ha⁻¹, while circumference decreased by 46 %. The coefficient of variation increased markedly with increased density for height and circumference.

Fig. 1.

Growth curves for height (A) and circumference (B) of a representative sample of four genotypes planted in the three densities in the clone experiment R91-1.

Fig. 2.

Growth curves for height (A) and circumference (B) of a representative sample of four genotypes according to the three densities in the family experiment R95-10.

Table 2.

Impact of density (trees ha⁻¹) on height (m), circumference (cm) and stem taper (cm/m) means and coefficients of variation for the clone (at 74 months) and family experiments (at 59 months)

		Height			Circumference			Stem taper
	Density	Mean	CV	Mean	CV	Mean	CV
Clone experiment	625	21·2	7	48·9	14	2·45	10
	1111	18·2	14	38	22	2·24	11
	2500	14·6	24	26·2	30	1·96	9
Family experiment	625	21·6	19	51·0	22	2·51	12
	1111	19·6	22	41·6	25	2·27	12
	2500	16·6	28	30·6	33	2·00	15

Stem taper decreased with age, with most reduction occurring before 3 years of age, suggesting that, architecturally, trees changed from a conical to a more cylindrical form (Fig. 3). Density also affected stem taper, as trees having a more cylindrical form were in the higher density treatments (Table 2). The coefficient of variation for this trait did not vary strongly among density treatments yet the magnitude was reduced in comparison to height and circumference (Table 2).

Fig. 3.

Change in stem taper with age and density using four clones (A) and four families (B) as representative examples of observed patterns.

Variation of growth trajectory and stem taper

A significant density effect was apparent from the results of ANOVA model 1 (see Appendix I), which included density, clone or family, and the density × clone (or family) interaction effects, for each of the three traits. Clone and family variance were significantly different from zero and the density × genotype interaction effect was significant for height and circumference at some ages and for stem taper across all ages only for the clone experiment.

Table 3 presents the developmental reaction norms using analysis of variance with repeated measurement (model 2). Examination of the results shows that time was a highly significant effect for all three traits, which demonstrates the increase in size and change in architectural form with age. A significant time × density interaction was also found which indicates an overall plasticity for growth, represented by increases in height and circumference, and for stem taper over time. Also a highly significant time × genotype interaction was found which suggests variation among clones (Figs 1 and 3A) or families (Figs 2 and 3B) in their growth trajectories and stem taper. The three-way interaction time × genotype × density was significant for the clone experiment (R91-1), indicating plasticity of growth that differed among clones (genetic variation for plasticity of growth). This pattern was not found for the family experiment (R95-10) but was marginally significant for circumference (P = 0·007). In the case of the clone experiment the latter result suggests that some clones initially grew rapidly but ceased their growth early in low density, whereas the same clones initially grew more slowly and grew for a longer period in high density (see Fig. 1).

Table 3.

Repeated measure analysis of variance for the experiments R91-1 clone experiment and R95-10 family experiment and the three traits (model 2)

	Height			Circumference			Stem taper
Source of variation	Wilk's λ	F	Wilk's λ	F	Wilk's λ	F
Clone experiment
Time	0·001	3817·2	0·003	1369·1	5 × 10−4	2319·0
Time × density	0·0001	46·83	0·0006	27·47	9 × 10−4	43·82
Time × clone	6 × 10−6	2·89	8 × 10−6	3·5	8 × 10−7	5·95
Time × density × clone	0·002	2·22	0·004	2·19	5 × 10−6	1·71
Family experiment
Time	0·002	3989·4	0·0043	2264·9	0·006	1440·5
Time × density	0·008	24·7	0·0038	38·19	0·044	9·46
Time × family	0·0001	4·05	0·0004	2·85	0·0004	2·73
Time × density × family	0·083	1·09	0·072	1·16	0·077	1·12

Multivariate test of no effect for each source of variation.

Bold font indicates P < 0·0001.

Genetic and environmental co-variation between height and circumference over time

Although sample size was small in both experiments, the phenotypic correlation between height and circumference was significantly different from zero. Using a power analysis (Kendall and Stuart, 1963) to calculate which true correlation produces an 80 % chance of being distinguished from zero, a one-sided test of a sample of 16 or 12 with a 0·05 significance level was used: r = 0·607 for the family test and r = 0·689 were found—values which were smaller than those obtained in both experiments (results not shown).

For both experiments, the within-genotype coefficient of correlation was high (ρ_r > 0·9) and was not affected by density or age (Figs 4A and 5A). The between-clone or family genetic correlation coefficient, however, exhibited different patterns of variation with age and density (Figs 4B and 5B).

Fig. 4.

Trends in patterns of correlation between height and circumference at the three densities for the clone experiment R91-1: (A) within-clone correlations and (B) between-clone correlations.

Fig. 5.

Trends in patterns of correlation between height and circumference at the three densities for the family experiment R95-10: (A) within-family correlations and (B) between-family correlations.

In the clone experiment (Fig. 4B), the between-clone coefficient of correlation remained constant and close to 1 for 2500 trees ha⁻¹ but decreased slowly for 625 trees ha⁻¹ from age 20 months (ρ_g = 0·9) up to 74 months (ρ_g = 0·57). For 1111 trees ha⁻¹ it dropped markedly from age 10 (ρ_g = 0·9) up to 30 months (ρ_g = 0·12) and then increased to a plateau at 50 months (ρ_g = 0·70). Comparison of the coefficients using the method described by Papoulis (1990) (method 1 above) revealed significant stage-specific differences (Table 4), but the jackknife method (Knapp et al., 1989; method 2) and MANOVA (Roff, 2002; method 3) did not uncover any significant density effects (illustrated in Table 5).

Table 4.

Test of comparison of the genetic correlation coefficients (method 1) estimated at each age in each density giving the probability associated with the test

	Age (months)
Clone experiment R91-1	12	24	30	36	48	62	74
Density compared
625/1111	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.	n.s.
625/2500	n.s.	n.s.	n.s.	n.s.	0·062	0·037	0·047
2500/1111	n.s.	n.s.	0·030	n.s.	0·046	0·078	0·095

n.s., non-significant at 10 % level.

Table 5.

Parameters associated with MANOVA examining the influence of the three densities of the variance–covariance matrix (G matrix) and with ANOVA examining the influence of the three densities of the genetic coefficient of correlation in the clone experiment R91-1

Age (months)	Wilk's λ	F	d.f.	Prob > F
Comparison of coefficient of correlation ANOVA (method 2)
30	–	0·46	2, 27	0·63
74	–	0·54	2, 27	0·59
Comparison of G matrix with MANOVA (method 3)
30	0·82	0·86	6, 50	0·53
74	0·64	2·08	6, 50	0·07

In the family experiment R95-10, the between-family coefficient of correlations was slightly affected by density (Fig. 5B), and no significant differences among densities were observed with any of the three methods. However, the overall patterns of variation with age were different from that of the clone experiment. The three coefficients of genetic correlation were reduced between 12 and 24 months (ρ_g = 0·90 and ρ_g = 0·50, respectively) and then increased to fluctuate around ρ_g = 0·7.

DISCUSSION

Experiments that examine genetic variation and phenotypic plasticity in trees are often limited in replication or restricted to early seedling stages (Bonser and Aarsen, 1994; Chen and Klinka, 1998; Spinnler et al., 2003). However, understanding developmental reaction norms in the field can lead to novel insights into the patterns of growth for long-lived perennial taxa. Experimentally manipulating the effects of density in a silviculture setting allows the effect of tree density to be isolated while minimizing other environmental variation present in natural forest systems (Chen and Klinka, 1998). For the study two crosses of Eucalyptus in the field at multiple life stages were chosen to examine how patterns of plasticity vary with density and how patterns of plasticity change over time.

The present strategy based on field experiment led to the analysis of a small sample. The sample size effect on errors in the parameter estimates have been considered by numerous authors in genetic experiments (Sales and Hill, 1976). The precision of variance components is reduced when sample size is small. The error is higher when considering the coefficient of correlation because three parameters are required for the calculation. To reduce the effect of small sample size in the experiments, the clones and families were selected to represent the entire variation of their respective population. In addition, the number of individual trees used to calculate the effect was high: 108 and 36 per density for each family and clone, respectively. Although there is no optimal design from which to draw general conclusions, the results constitute a first step towards understanding phenotypic plasticity with increased competition in a field experiment.

Plasticity in growth trajectory and allometry

The impact of density on plant size characters is often size-asymmetric, i.e. slightly larger individuals receive more resources than smaller trees, and such size inequalities increase with density (Evert, 1971; Pierik et al., 2004). Higher densities also increase the phenotypic variation, confirming the size asymmetry hypothesis, a pattern which has been observed in other tree species such as Douglas fir (Campbell et al., 1986; Dreyfus 1990). In the case of the two hybrids of Eucalyptus, in the poor soil conditions of the field trials, the impact of closer spacing is dramatic in that it strongly reduces circumference, height and consequently the growth trajectories.

Previous results have shown that height was higher in close spacing during the first months (Bouvet, 1997), which can be explained by the morphological response of plants to the presence of neighbours with enhanced shoot elongation, the so-called shade-avoidance response (Pierik et al., 2004). The rapid growth leads to strong competition affecting circumference but also height in later life stages. In temperate species, Parde and Bouchon (1988) have demonstrated that density reduces the growth in circumference but does not affect the height. However, here we confirm the pattern that others have observed in tropical Eucalyptus in which density influences both height and circumference (Patinot-Valera and Kageyama, 1995; Brouard and John, 1999).

The impact of tree density on the allometric relationship has demonstrated that species adapt to crowding in different ways, e.g. greater elongation of branches for Abies compared with Picea (Takahashi, 1996), more flexibility in relative allocation of energy in growth for rainforest species (Alves and Santos, 2002). However, the impact of density on allometric relationships through ontogeny for tree species remains poorly documented, especially using controlled experimental approaches (Saint-André et al., 2005). The relationship between height and circumference, the stem taper, was found to decrease with age and with close spacing. This result has been explained by various authors (Larson, 1963; Bouillet and Lefevre, 1996). It has been observed that in open-grown trees with long vigorous crowns, stem taper continues down the branch-free bole. In lower density, the size of the crown is proportionally greater than in closer density where natural pruning eliminates under shaded branches; as the crown base recedes and the clear boles elongates with increasing tree age and stand closure, the stem become more cylindrical because photosynthates are regularly distributed and the increment in basal area is constant along the bole.

Genetic control of ontogenic plasticity in eucalyptus

Genetic variation for plastic responses of different traits has been widely observed for different non-perennial plant species (Schlichting and Pigliucci, 1998). Pigliucci and Schlichting (1995) and Pigliucci (1997) suggested that inclusion of ontogenic data in plasticity studies will broaden our understanding of how phenotypes evolve in response to selection. Although genetic variation for growth trajectory has been demonstrated for several tree species (Namkoog and Conckle, 1976; Bouvet, 1991; Danjon, 1994), as far as is known, few studies have examined the genetic basis of ontogenic plasticity for growth and form in trees, except in Populus (Wu, 1997; Wu and Stettler, 1998).

Here, high genetic variation among clones and among families for growth trajectory and stem form was observed. Although this result was based on a limited number of genotypes (12 clones and 16 families), other studies on eucalyptus hybrid populations introduced in the Congo, using a significant sample, have shown high genetic variation within these two hybrid species (Bouvet and Vigneron, 1996). These results demonstrate that these populations maintain significant quantitative genetic variation that could be used for breeding programmes. In the clone experiment R91-1, a significant time × clone × density interaction effect was observed, implying that the plasticity of growth trajectory for height, circumference and ontogenic allometry (stem taper) has a genetic component. The interaction was not significant for the other experiment (R95-10) which used a family rather than a clonal structure. Two explanations are proposed to examine the different results of these two experiments.

The first explanation relates to the fact that clones are generally more unstable when the environment changes. Broader within-family genetic variation allows a more stable response of the genotype to a changing environment. These results have often been observed in plants, the higher the within-genotype (clone, family) variation, the smaller the genotype × environment interaction (Gallais, 1991). Although the time × family × density interaction was not significant for the three traits, it should be stressed that the probability was close to the threshold of significance for circumference (F = 1·16, P = 0·0862), whereas for height F = 1·09 and P = 0·217. This point can be explained by greater sensitivity of circumference to competition and confirms previous results showing that height is less sensitive to density.

The second explanation could be related to the higher preponderance of epistatic effects involved in between-clone compared with between-family genetic variance. Several authors have suggested that epistatic mechanisms could explain genetic plasticity. Pigliucci (2003) has suggested that regulatory elements could play an important role in the genetics of development. Blows and Sokolowski (1995) have demonstrated that expression of non-additive genetic variation is increased at both extremes of the environmental range. In a Populus clone study, Wu (1997) suggested that the genetic difference in response to different environments might be mainly due to epistasis between regulatory loci for plasticity and loci for trait means. As epistatic effects are preponderant in the ‘genetic variance between clones’ compared with the ‘genetic variance between families’ (Gallais, 1991), genetic control of plasticity could be greater in clone populations.

Impact of density on ontogenic allometry

Ontogenic allometry was analysed through the relationship of stem taper with age but also through the genetic and environmental correlations between height and circumference. According to several authors (e.g. Parson, 1987; Bennington and McGraw, 1996) both phenotypic and genetic variation are enhanced under stressful conditions but few studies have investigated the impact of density on environmental and genetic correlations in plants. Examples in rice (Kawano and Hara, 1995) and in an annual plant Tagetes patula (Weiner and Thomas, 1992) have shown that with increasing density the number of significant phenotypic correlations between traits increases and that the allometric relationships between height and diameter change with higher density. For tree species, the genetic basis of ontogenic allometry has rarely been investigated. Poplar is one exception. Wu and Stettler (1998) demonstrated higher genetic correlation between crown and growth traits in a less favourable environment, but their study follow-up was only 2 years. In the present study, the genetic and environmental correlations were analysed from juvenile up to adult size and cast new light on the genetic and environmental basis of the relationship between traits in a changing environment.

The within-genotype correlation was positive and varied little with the density and age of the stand (Figs 4A and 5A). In the case of clones, the origin of within-clone variation and co-variation is purely environmental because genetic variation does not exist among ramets. The high environmental correlation means that changes in the environment are likely to influence the genetics underlying circumference and height in similar ways. The present finding is consistent with data on other forest trees where favourable situations lead to a simultaneous increase of height and circumference (Bouvet, 1995). In the case of the family experiment, a similar result is observed, although within-family variance and co-variance are composed of genetic and environmental variance. This is explained by the greater magnitude of environmental variance compared with genetic variance, a result which is generally observed in forest trees and especially in eucalyptus (Bouvet et al., 2003).

The genetic correlations between circumference and height were positive in both experiments. This positive genetic co-variation is consistent with experiments on other tree species (Zobel and Talbert, 1984) and especially in eucalyptus (Bouvet, 1995). Most of these investigations have concluded that height and circumference are likely to be governed by similar genes and that pleiotropy could be an explanatory factor for these relationships. Studies of molecular genetics have demonstrated that the same set of QTL explains a part of genetic variation in height and circumference (Verhaegen et al., 1997; Gion, 2001). This result, which has been observed for other species (Wu et al., 2003), suggests the maintenance of integration within module, i.e. between functionally related traits in plants (Murren, 2002; Preston and Ackerly, 2003), and agrees with quantitative genetic models suggesting that selection favours positive correlation between functionally related traits (Hartl and Clark, 1997).

The trend in genetic correlation with density presents different patterns in the clone and family experiments (Figs 4B and 5B). This difference can be explained by multiple factors (sampling, species) but, as for genetic control of plasticity, the distribution of genetic effects between and within each genotype (family vs. clone) is also an explanatory factor. With respect to the clone experiment, which represents phenotypic expression of individual trees, these results suggest that genetic control of ontogeny depends on environmental conditions. Although the genetic mechanisms behind plasticity of trait correlation are poorly elucidated (Murren and Kover, 2004), one hypothesis is that regulatory genes with environmentally dependent expression might explain variation in genetic correlation in changing environments. Epistatic variance within genetic variance between clones (Gallais, 1991) might explain the marked impact of density on genetic correlation.

The variation in genetic correlation with age can be interpreted for both clone and family experiments, as if the genetic control of height and circumference is progressively dissociated during establishment of the competition phase, and then, once competition is established, the correlated genetic control of the two traits is re-established. One gene action model can be proposed to explain this observation. During the progressive competition phase, genes controlling height growth are very active because a rapid height growth facilitates light capture. When trees reach a critical height which threatens their stability, genes controlling circumference growth act to ensure they remain stable. Another complementary explanation of the variation in genetic correlation with age is age-related changes in genes controlling growth. Genes involved in height and circumference may change during tree development thus changing the co-variation between those traits. This assumption is supported by data on QTL of height and circumference in some eucalyptus experiments. With the same hybrid as in experiment R95-10, Gion (2001) has shown that the QTL of height and circumference change between 14 and 59 months. More research is needed to clarify the molecular basis of ontogenic allometry in the tropical eucalyptus and the impact of a changing environment. A first step would be to apply to this population the statistical model for QTL analysis developed by Wu et al. (2003).

Summary and conclusion

Although the present results are based on a small sample, thus affecting the precision of variance components and the power of the experiment, they suggest that tropical Eucalyptus species tested in the present study are plastic for growth traits and form and present a genetic system able to respond to a changing environment. The present clone and family experiments demonstrate that competition due to close spacing induces different patterns for the environmental and genetic correlations, suggesting that a simple phenotypic approach is not the most appropriate way to analyse the functional basis of plasticity and allometry. The present results also suggest that the environment changes the genetic expression of growth trait co-variation through ontogeny, which is a new result in forest trees which must be confirmed by further research. The present finding confirms that a tree population with a narrow genetic basis (represented by clones in the experiments) is sensitive to a changing environment, whereas a population with a broader genetic basis (full-sib family here) exhibits a more stable reaction.

APPENDIX

Probability value associated with the Fisher test of the analysis of variance testing the density (D), the genotype (G) (family or clone), and the density × genotype interaction (D × G), for the clone and family experiments (model 1)

	Height					Circumference					Stem taper
Age (months)	D	G	D × G	D	G	D × G	D	G	D × G
Clone experiment R91-1
12	0·003	10−4	0·05	10−4	10−4	0·27	10−4	10−4	0·00
24	10−4	10−4	0·30	10−4	10−4	0·25	10−4	10−4	0·01
48	10−4	10−4	0·11	10−4	10−4	0·02	10−4	10−4	0·00
74	10−4	10−4	0·002	10−4	10−4	0·003	10−4	10−4	0·00
Family experiment R95-10
11	0·58	10−4	0·95	10−4	10−4	0·96	10−4	10−4	0·40
20	10−4	10−4	0·78	10−4	10−4	0·95	10−4	10−4	0·35
39	10−4	10−4	0·72	10−4	10−4	0·84	10−4	10−4	0·37
59	10−4	10−4	0·89	10−4	10−4	0·89	10−4	10−4	0·68