Environmental quality and evolutionary potential: lessons from wild populations

Anne Charmantier, Dany Garant

Abstract

An essential requirement to determine a population's potential for evolutionary change is to quantify the amount of genetic variability expressed for traits under selection. Early investigations in laboratory conditions showed that the magnitude of the genetic and environmental components of phenotypic variation can change with environmental conditions. However, there is no consensus as to how the expression of genetic variation is sensitive to different environmental conditions. Recently, the study of quantitative genetics in the wild has been revitalized by new pedigree analyses based on restricted maximum likelihood, resulting in a number of studies investigating these questions in wild populations. Experimental manipulation of environmental quality in the wild, as well as the use of naturally occurring favourable or stressful environments, has broadened the treatment of different taxa and traits. Here, we conduct a meta-analysis on recent studies comparing heritability in favourable versus unfavourable conditions in non-domestic and non-laboratory animals. The results provide evidence for increased heritability in more favourable conditions, significantly so for morphometric traits but not for traits more closely related to fitness. We discuss how these results are explained by underlying changes in variance components, and how they represent a major step in our understanding of evolutionary processes in wild populations. We also show how these trends contrast with the prevailing view resulting mainly from laboratory experiments on Drosophila. Finally, we underline the importance of taking into account the environmental variation in models predicting quantitative trait evolution.

Keywords:

1. Introduction

In the evolution of continuous traits, phenotypic changes over time are dependent on two major factors: the intensity of the selection and the heritable fraction of the trait (Falconer & Mackay 1996). Selection on quantitative traits is routinely estimated by relating individual measurements to individual relative fitness (e.g. the Lande–Arnold approach for correlated traits; Lande & Arnold 1983). Description of selection occurring in natural populations has long been a central focus in the study of evolutionary change and mechanisms of natural selection (Endler 1986). However, even when selection is known to occur, an essential requirement for evolutionary change is the amount of genetic variability expressed for the trait under selection. One of the early goals in quantitative genetics was to quantify levels of heritable variation in order to distinguish how much of the standing variation in populations was due to genetic or environmental causes (Lynch & Walsh 1998). For this purpose, evolutionary genetics has focused on a measure first introduced in animal breeding: heritability (h2). Heritability in the strict sense is defined as the fraction of the total phenotypic variance of a trait (VP) which is due to the additive effects of genes (VA): h2=VA/VP.

Early investigations of heritability in laboratory conditions showed that the magnitude of genetic and environmental components in total phenotypic variation for specific traits can change with environmental conditions (Hoffmann & Parsons 1991). This result was of great importance as it implied that the amount of additive genetic variation on which evolution could act was variable according to the environment. Understanding the amplitude and direction of this change depending on environmental conditions would greatly improve the extent to which one could predict evolutionary change.

The study of environmental effects on the genetics of quantitative traits was at first restricted to laboratory studies under severe stresses such as γ-irradiation exposure (Westerman & Parsons 1973), desiccation (Hoffmann & Parsons 1989) or extreme temperatures (Clark & Fucito 1998; Imasheva et al. 1998). Many of these studies were conducted on Drosophila species, and the general trend was for increasing heritability of size-related and life-history traits with increasing stress levels, because of higher additive genetic effects (for an extensive review, see Hoffmann & Parsons 1991). However, investigations in other taxa show that this trend is not universal, and can even be reversed, raising the questions of how and why expression of genetic variation is sensitive to different environmental conditions (Hoffmann & Merilä 1999). For example, several early brood size manipulation (BSM) experiments on passerine bird species have shown that an increase in brood size, triggering poorer feeding regimes for offspring, results in lower additive genetic variance and higher environmental variance for size-related traits in birds (Gebhardt-Henrich & Van Noordwijk 1991; Merilä 1997; Merilä & Fry 1998). Furthermore, a review of 19 studies in natural bird populations published before 2000 is consistent with the conclusion of lower heritable variation in unfavourable conditions (see table 5 in Merilä & Sheldon 2001). The proximate causes for this trend have been shown to be alternatively or simultaneously lower additive genetic variance in stressful conditions, higher environmental variance, or low cross-environment genetic correlation. Similar changes have been noted in the animal and plant controlled breeding literature (Hoffmann & Parsons 1991; Geber & Griffen 2003), although the types of stress studied were argued to be very different from Drosophila experiments (Hoffmann & Parsons 1991). Several other differences have been proposed as potential explanations for the discrepancies between the different studies (Hoffmann & Merilä 1999) but there is as yet no consensus. Hence, the inconsistency of genetic variation and other quantitative genetic estimates across environments is still debated.

The study of quantitative genetics in the wild has recently been revitalized by the application of maximum likelihood (ML) and restricted maximum likelihood (REML) procedures to estimate variance components (Lynch & Walsh 1998). These methods, such as the so-called ‘animal model’ (see Kruuk 2004 for a review), are more powerful than traditional analyses (e.g. parent–offspring regressions) because they take into account every relationship link in a pedigree, they are less sensitive to departure from normality, and they make efficient use of complex and unbalanced datasets collected from wild populations (Knott et al. 1995). Moreover, the use of mixed models allows the quantification and decomposition of different environmental components of variance, such as maternal effects, which were previously difficult, if not impossible, to estimate.

As a consequence of these methodological improvements and the growing interest in the effect of environmental changes on evolutionary potential, a number of studies of environmental effects on genetic variance in wild populations have recently been published. These studies provide a broader treatment of different taxa and traits that now make it possible to draw more general conclusions. The environments under which traits and their variation are measured have also evolved from the BSM experiments performed in birds, to the study of various other experimental manipulation of environmental quality in the wild. In addition, naturally occurring favourable or unfavourable environments, defined as variation in climate, population density or habitat quality, have also been exploited. These approaches provide a new perspective on the interpretation of the role of ecological factors in the evolution of heritable variation and its expression. We illustrate here how these recent contributions represent a step forward in our understanding of evolutionary process in wild populations, focusing on non-laboratory and non-domestic animal studies. In particular, we will investigate whether these studies provide evidence for a specific relationship between heritability estimates and environmental quality, and examine the underlying changes in different variance components. We believe this is a central theme in evolutionary ecology that deserves further emphasis in the context of evolutionary processes in natural heterogeneous environments, where the potential rate of evolution may be reduced as a consequence of lowered heritability.

2. Trends in heritability across environments

Here, we review studies investigating the sensitivity of quantitative genetic parameters to environmental conditions, excluding laboratory and domestic animal studies. These include studies using natural variation in environmental quality (table 1a, e.g. Réale et al. 1999), experimental manipulation of environmental quality in the wild (excluding BSM, table 1b, e.g. Kause & Morin 2001) or studies in laboratory conditions with animals originating from the wild, less than five generations before study (e.g. Pakkasmaa et al. 2003). In all studies included here, heritability was estimated in two conditions of different quality (if more than two, we used the extreme conditions). The ranking of favourable versus unfavourable was provided by the authors, on the basis of either seasonal or annual climatic conditions (e.g. Garant et al. 2004, poor conditions when NAO index <median), food abundance differences (e.g. McAdam & Boutin 2003; De Neve et al. 2004) or effects on fitness factors (e.g. Qvarnstrom 1999, poor conditions when mean fledging condition was low).

View this table:
Table 1

Trends in additive genetic variation and heritability, estimated in non-domestic and non-laboratory animal populations comparing favourable and unfavourable environmental conditions.

The reliability of laboratory estimations for quantitative genetic parameters has been repeatedly debated (e.g. Weigensberg & Roff 1996), and several studies have shown important discrepancies between field and laboratory estimations of heritability (e.g. Sgro & Hoffmann 1998b; Hermida et al. 2002). Hence, our choice was to focus on animal studies that were not potentially biased by homogeneous laboratory conditions and several generations of breeding in artificial conditions. In addition, we do not include studies performing brood-size manipulation in birds as they have been reviewed extensively elsewhere (Hoffmann & Merilä 1999; Merilä & Sheldon 2001). We also exclude studies that compared different populations evolving under different environmental conditions (e.g. Charmantier et al. 2004a; Merilä et al. 2004) as their purpose is to compare the levels of additive genetic variation in different stress conditions among populations adapted to different environments, whereas the studies presented here compare levels of VA within populations. When different populations are involved, the interpretation of differences between levels of VA and h2 could be dependent on the selection history because genetically isolated populations can evolve differences in their genetic variances due to selection or drift. Finally, we focus on recent results published between January 1999 and January 2005.

Of the 46 traits presented in table 1a,b, the majority (38) did not show significant dependence of the heritability estimates on the environmental conditions. However, the eight cases where a significant difference was found among conditions all pointed to an increase of heritability under favourable conditions. We performed a meta-analysis using Metawin 2.0 (Rosenberg et al. 2000) to test whether the overall trend was significant for the studies in table 1. To conduct this analysis, we used heritability estimates in both environmental conditions along with a statistical metric, which could be converted to an estimate of effect size. Two studies from table 1 were removed from the meta-analysis (Pakkasmaa et al. 2003; Gomez-Mestre & Tejedo 2004) because of the lack of information on their sample sizes and standard errors that prevented us from estimating an effect size. For each studied trait, we estimated the correlation coefficient (r) from the t-value of the difference between favourable and unfavourable environments (equation 7.5 in Rosenberg et al. 2000, Appendix 1). Effect sizes were Z-transformed in all analyses and they were weighted by the inverse of the variance associated with the effect size estimate (for more details on a similar analysis, see Sheldon & West 2004). All the analyses were performed using random effects models. The overall estimated mean effect size was positive and significantly different from zero (r=0.0221, obtained from 10 000 bootstraps, n=40, see table 2 for CI) and no difference was found among taxa (p=0.966). There was a trend for a publication bias against negative values of effect sizes, especially for small sample sizes (figure 1). Because most of these studies had no a priori expectation for the direction of the size effect, this bias was not expected. However, the correlation between estimated effect size and sample size showed that the bias was non-significant (rs=−0.234, p=0.147).

View this table:
Table 2

Meta-analysis results.

Figure 1

Effect size as a function of sample size for studies using natural variation in environmental conditions (black diamonds) or experimentally induced environmental variation (white squares) to compare heritability estimates in favourable versus unfavourable conditions. The dotted line indicates the overall mean effect size of 0.0221.

Overall, these studies provide evidence for a significant increase of the estimated heritability under favourable conditions as compared with unfavourable conditions, although the absolute value of the mean effect size remains small. This result is thus consistent with the conclusions of the two previous reviews on size-related traits in birds (Hoffmann & Merilä 1999; Merilä & Sheldon 2001) and it is in marked contrast to results on Drosophila (Hoffmann & Parsons 1991).

There has previously been some evidence for trait-specific responses of genetic variation to environmental variation. For example, some avian studies have shown differences in h2 for different growth conditions in body mass but not in tarsus length (e.g. Gebhardt-Henrich & Van Noordwijk 1991; Charmantier et al. 2004a). We tested here for a difference in the response of morphometric traits compared with non-morphometric traits, by repeating the meta-analysis described above with a random categorical variable distinguishing between the two types of traits (see Electronic Appendix 1). The mean effect size for studies performed on morphometric traits (0.0313, n=19, table 2) was almost four times higher than the effect size for non-morphometric traits, which was in fact not significantly different from zero (0.0084, n=21), although the difference between the two was not significant (table 2). As classical population genetics theory predicts (Mousseau & Roff 1987; Falconer & Mackay 1996), life-history traits, which are more closely related to fitness than morphometric traits, have lower heritabilities (but see Price & Schluter 1991; Kruuk et al. 2000; Merilä & Sheldon 2000). Hence, the absence of variation in heritability between different ecological conditions for these traits could be partly a result of the fact that any variation will be harder to detect for smaller values. In addition, in stressful conditions such as food shortage, resource allocation to various physiological functions may favour traits that directly affect survival to the detriment of less vital functions such as morphological characters. Hence, this could explain that heritability of morphometric traits tends to decrease more in unfavourable growth conditions than heritability of traits more closely related to fitness.

Interestingly, mean effect size for studies using natural variation in environment quality (n=16, r=0.0169, table 2) was nearly half the effect size of studies using experimentally induced variation (n=24, r=0.0294), although the two estimates did not differ significantly. This could result from the fact that experiments manipulating environmental quality can induce a greater among-habitat difference in quality than what is usually found in nature.

3. Why are genetic variances and covariances variable?

There are three main explanations to why heritability decreases in unfavourable conditions (Hoffmann & Merilä 1999). The first is decreased additive genetic variation (VA). The second reason may be higher environmental variance (VE), especially higher variance in maternal contribution (VM). The third proposed explanation for differences in heritability is a low genetic correlation between the expressions of the trait in the different environments. Because the evolution of a trait depends not only on its genetic variation, but also on its correlation with other selected traits, we will also summarize what is documented about differences in variance–covariance matrices across environments.

(a) Expression of VAacross environments

The simplest partitioning of the phenotypic variance of a trait is as the sum of an additive genetic variance and a residual variance (VR):Embedded Image

In this equality, VR includes all non-additive genetic effects such as dominance or epistatis, and all effects due to differences in the environment. Whereas many quantitative genetic studies cannot identify specific environmental effects, most of them provide information on the differences in VA between environments. The first explanation offered by these studies for a decrease of heritability in unfavourable conditions is a decrease in VA triggered by constraining growth conditions which limit the genetic potential (Gebhardt-Henrich & Van Noordwijk 1991). In table 1, 65% (26 out of 40) of the traits studied showed such a trend in VA, although no meta-analysis could be conducted on VAs because of lack of standard errors or required statistical tests. However, when restricting to non-morphometric traits, there was no evidence for trends in VA since only half the traits studied showed an increase in favourable conditions (11 out of 21 traits).

It is important to underline that an absence of difference in heritability does not necessarily imply that the underlying variance components are not variable. For example, even though overall heritability of parasite resistance, measured as faecal egg count (FEC) in the Soay sheep (Ovis aries), did not differ significantly between stressful conditions (rut in autumn for males, parturition and lactation in spring for females) and favourable ones (summer in both cases), Coltman et al. (2001) still found higher VA in spring for females. The authors suggest that these results imply that energetically stressful conditions during parturition and lactation emphasize the individual differences in resistance to parasites. The difference between seasons in the expression of VA was concealed in the heritability estimation by a simultaneous increase in VR and in the mean of FEC measurements.

(b) Variation in VEand environment-dependent maternal effects

In many cases, differences in heritability between environments were not explained by differences in VA but by differences in residual or environmental variance (e.g. Uller et al. 2002, where environmental variance of tadpole growth rate becomes null in the most favourable conditions). Increased environmental variance (VE) in poor conditions was previously a common finding in the study of avian morphological traits (Merilä & Sheldon 2001). Here, it is supported by 70% (21 out of 30) of the traits for which VE was estimated. A potential explanation for this difference in VE is that when conditions during growth are difficult, variation in local quality, such as variation in food abundance or parental care, will be a greater determinant in final offspring size than any developmental genetic programme (Van Noordwijk & Marks 1998). In contrast, when growth conditions become more favourable, the effects of local environmental differences on growth are diminished compared with the genetic effects.

Maternal effects are special environmental effects which occur when the phenotype and the environment of the mother affect the genotypic expression in the offspring, either for the same trait (thereby increasing mother–offspring resemblance) or for different characters (increasing resemblance between offspring, Falconer & Mackay 1996). Traditionally, maternal effects have been viewed as a nuisance in quantitative genetics and their evolutionary consequences have only recently attracted the attention of biologists (e.g. Mousseau & Fox 1998a,b; Peripato & Cheverud 2003). BSM experiments in birds have previously raised the question of environment-dependence of maternal effects (e.g. Merilä 1997). More recently, further evidence for such a link between the environmental conditions and the influence of maternal effect has been suggested. For example, in Pakkasmaa et al.'s (2003) study of common frog tadpoles, maternal effects estimated over four different tadpole characters was shown to vary according to conditions, although the changes in the expression of maternal effects in stressful conditions are not in the same direction for all traits measured. In theory, the predictions about the way maternal effects should be shaped by differences in environments are similar to those for general environment effects. We would thus expect higher variance due to maternal effects (VM) under unfavourable conditions where individual differences between mothers will be exacerbated by the environmental stress. However, studies in table 1 do not show consistent trends in VM across environments, as there was an increase of VM in unfavourable conditions for only 5 of the 10 traits for which this component was estimated.

(c) Low cross-environment genetic correlations

It is commonly found that genotype-by-environment interactions (GEI; differential genotype response to environmental conditions, Falconer & Mackay 1996) result in heritability differences. GEIs occur either when the genetic correlation between the trait's measures in the two environments is less than one, or due to differences in additive genetic variances between the environments (see above §3a). In the case of low cross-environment genetic correlations, the genetic basis for the trait is different in different environments, i.e. this trait is not governed by the same genes in the different environments (Falconer & Mackay 1996). In this case, VA is bound to be different as it is not underlined by the same set of genes, yet interpretations of the differences in VA become problematic.

One major problem encountered with traditional parent–offspring regressions as a way to estimate heritability is that parents having survived to reproduction are likely to have been reared in conditions that are different from those experienced by their offspring when the environment is variable. It is then impossible to distinguish between decreased VA or low cross-environment genetic correlations to explain differences in heritability (Merilä & Sheldon 2001). Partial cross-fostering in birds combined with different growth conditions for full-sibs have proved to be more efficient in revealing significant GEIs caused by low genetic correlation between parental traits and offspring traits in unfavourable conditions (Merilä 1997). Subsequently, long-term data on successive cohorts brought up in various environmental conditions confirmed the potential importance of low cross-environment genetic correlations. The papers reviewed here illustrate that low cross-environment genetic correlations are commonly found in wild populations studies (e.g. Coltman et al. 2001; Kause & Morin 2001; McAdam & Boutin 2003; Charmantier et al. 2004b) and partly explain the observed differences in h2. For example, in Kause & Morin (2001), families of the sawfly Priophorus pallipes raised in three diet quality treatments show weak cross-treatment genetic correlations of final mass, meaning reversed ranking of families between high and low quality diets. However, even though we can expect VA to change between two environment conditions in the case of low cross-environment correlation, it is to be noted that it cannot explain a consistent direction for this change (e.g. repeatedly low VA in unfavourable conditions).

(d) Genetic correlation structure across environments

Evolution does not depend solely on the selection and genetic variation found in a trait, but also on the genetic correlation among all selected traits (Lande & Arnold 1983). For example, if two traits are under a positive directional selection, a negative genetic correlation between them will slow down the rate of response of both traits (see Roff 1997, equation 5.14). If the correlation between the traits is environment-specific, then so will be the predictions on their evolution. Hence, the G-matrix (the matrix of additive genetic variances and covariances corresponding to a set of diverse traits, Lande 1979) plays a central role in quantitative genetics, and its stability has been the focus of numerous recent theoretical studies (Roff 2000; Jones et al. 2003) as well as empirical studies in animals and plants (e.g. Donohue et al. 2000; Bégin et al. 2004). As such, there is an extensive literature assessing the change in genetic correlations among traits (reviewed in Stearns et al. 1991; Sgro & Hoffmann 2004) and assessing G-matrix stability across species, populations or habitats (e.g. Roff et al. 2004). A brief general overview of the recent comparisons made in animal populations of the same species and that focused on variation in environmental conditions shows that, although a number of studies have assessed the G-matrices from different populations (e.g. Brodie 1993; Roff & Mousseau 1999), very few of them studied variation in free-ranging populations (but see Arnold & Phillips 1999; Roff et al. 2004). Furthermore, even if a few studies found evidence for variation in G-matrices among isolated populations in the wild, no specific reference (or predictions) to variation with respect to the quality of the environment are made. As a result, the degree to which G-matrices vary in nature remains largely unknown (Steppan et al. 2002) and many more studies are needed across different environments before we can conclude about the degree of its stability.

Some conclusions could still be drawn from studies performed under laboratory conditions where environmental variability is taken into account for differences in G-matrices (Holloway et al. 1990; Guntrip et al. 1997; Bégin & Roff 2001; Bégin et al. 2004). Most of these studies compare ancestral versus novel habitats (see Guntrip et al. 1997 for example), and they show that genetic correlation among traits is stronger in the ancestral environment of a population, thus supporting the general prediction that accuracy of adaptation is related to the frequency of exposure to the different environments (de Jong 1995). For example, Olsson & Uller (2002) have shown that in the common frog (Rana temporaria), the mean growth rate and its variance expressed as the coefficient of variation are negatively correlated, and that this negative correlation is most pronounced at the temperature to which the organism has been adapted through selection. Similarly, for life-history traits, it has been generally observed that there is a breakdown in genetic architecture between traits, when measured in a novel environment (Holloway et al. 1990; Simons & Roff 1996; Guntrip et al. 1997), potentially as a result of novel gene expression. Finally, some studies show evidence for stronger genetic correlations under environments of higher quality and lesser variability. For example, Kause et al. (2001), studying birch feeding insects in relation to seasonal change in foliage quality of the host plant, showed that under predictable environmental quality (birch foliage) genetic correlations between development and mass were strong, whereas they were negative or close to zero when conditions were unpredictable (when leaves were growing or senescing). In the future, insight might thus be gained by conducting more studies of this kind but also by taking into account findings from previous studies on plants where experiments in contrasting habitats have already shown that G-matrices could differ between environments (Donohue & Schmitt 1999; Donohue et al. 2000).

4. Explaining discrepancies between studies

Much of the discussion here has focused around levels of heritable variation in favourable versus unfavourable conditions. However, the differences between environments in empirical studies may go well beyond differences in quality of the environments. We have reviewed studies of non-laboratory and non-domestic animals, which give overall evidence for higher heritability in favourable conditions compared with stressful conditions. We show that some differences between these studies and Drosophila studies in laboratory conditions could explain why the latter find higher heritability in stressful conditions.

In the same line of reasoning as our discussion on different traits (see §2), the nature of the stress studied can itself trigger different responses in genetic variation. While investigating the relationship between inbreeding and environmental stress through laboratory tests on Drosophila melanogaster, Dahlgaard & Hoffmann (2000) illustrated that influences of different types of stresses are not always genetically correlated. Differences between environments in table 1 are variable (e.g. pH, temperature, habitat quality, parasitism) and they also represent milder stresses than those classically used in the Drosophila studies. When discussing differences between studies on agricultural animals (controlled breeding) and studies on Drosophila, Hoffmann & Parsons (1991) argued that extreme environmental stresses in the latter (e.g. ethanol combined with cold shock and low nutrition in Sgro & Hoffmann (1998a) and acetone or desiccation resistance in Dahlgaard & Hoffmann (2000)) could result in drastic changes such as expressions of new sets of genes, whereas mild changes in the latter would trigger different changes.

In Drosophila, increased additive genetic variance under stress has been hypothesized to ensue from direct effects of stress through mutation or recombination, selection being less effective in novel stressful conditions, depletion of phenotypic variation in most common favourable conditions, or finally, emergence of phenotypic differences between genotypes under environmental constraints (Hoffmann & Merilä 1999). It should be noted that the second and third hypotheses imply that unfavourable conditions are usually novel ones. The prediction regarding the evolution of heritability in common versus novel environments has received clear empirical support. Namely, Holloway's (Holloway et al. 1990) hypothesis that additive genetic variance increases in novel environments (independently of whether these are favourable or unfavourable) has been verified repeatedly (Laugen et al. 2002, 2003a,b; Messina & Fry 2003; Rasanen et al. 2003). It is generally explained by the expression of new genes that have not been under selection in the more common environment. Contrary to laboratory studies on Drosophila, unfavourable conditions in the wild are common, and induced novel conditions such as food supplementation (De Neve et al. 2004) or manipulation of parasite load (Charmantier et al. 2004b) are often more favourable than the original state. This difference could partly explain why these two types of studies find opposite results, if heritability persistently increases in novel environments.

Holloway's hypothesis could also explain discrepancies in heritability estimated from laboratory or field studies (Riska et al. 1989; Weigensberg & Roff 1996; Geber & Griffen 2003). In the case of non-domestic and non-laboratory animal species, laboratory conditions are novel compared with the field conditions where they have a history of selection and adaptation. Hence, we would predict an increase in VA in laboratory conditions (see above; Holloway et al. 1990) compounding the overestimation of h2. Indeed, comparative studies on h2 estimations for the field cricket Gryllus pennsylvanicus (Simons & Roff 1994) or Drosophila spp. (Hoffmann 2000) showed increased h2 in laboratory conditions because of both lower VE and higher VA. Similarly, a survey of 74 plant studies published between 1985 and 2002 showed that estimating heritability in controlled laboratory environments overestimates heritabilities in the wild by two- to four-fold (Geber & Griffen 2003). In contrast, in a comparison of six studies estimating heritability of body size for the yellow dung fly, Scathophaga stercoraria, in field and laboratory conditions, Blanckenhorn (2002) showed that the variable field conditions augmented both the genetic and environmental variances, resulting in no change for h2.

Finally, a simple prediction when comparing the evolution of populations in homogeneous conditions versus environments that are heterogeneous in space or time is that VE will be inflated in the more variable environment, hence reducing h2. Substantial reduction in h2 because of higher environmental variation is common in wild populations, as discussed in §3b. The controlled and constant conditions in the laboratory are more homogeneous than natural environmental variation. This will lead systematically to reduced VE and inflated estimates of heritability compared with field conditions for phenotypically plastic traits (e.g. Sgro & Hoffmann 1998b; Hermida et al. 2002).

5. Consequences of variation in heritable variance

There are several major evolutionary implications of variable additive genetic variances, heritabilities and covariances across environments. First, our meta-analysis of empirical data in wild animals has given evidence for smaller h2 in unfavourable natural environments, especially for morphometric traits. In the context of a meta-population, this implies potentially lower rates of evolution in the stressful sink habitat conditions compared with facilitated adaptation in local source habitat. In cases of increased stress in natural environments triggered by human disturbance, habitat fragmentation and climatic changes, the survival of a species depends on its ability to respond to selection ensuing from environmental stress. However, human disturbances and climate change are not only creating unfavourable environments, but also novel conditions, thus complicating the reasoning on the evolutionary potential. In any case, the amount of heritable variation in traits correlated with individual fitness is critical. Our review suggests that these characters might be less prone to decreased h2 than traits less linked with fitness such as morphometric ones. However, it should also be kept in mind that the response to stress in endangered populations might be different from that estimated in ideal situations such as laboratory-based studies.

Second, it is important to note that when differences in heritabilities are a result of low cross-environment correlations, different genotypes will be expressed and favoured under different conditions. In such cases, there is no prediction concerning the direction in which VA should evolve and thus limited implication for lower evolutionary potential in one of the environments. The finding that GEIs commonly occur in the wild might explain the persistence of heritable variation. The mechanism for the maintenance of genetic variation in natural populations has long been debated (Lewontin 1974) and various hypotheses have been suggested. For example, Gillespie & Turelli (1989) modelled GEIs with the assumption that additive contributions of alleles varied with the environment. Without including any variance in fitness of genotypes across environments, their model showed that GEIs could maintain genetic variation in quantitative traits. Following this, GEIs can be regarded as favourable for evolutionary response as they help maintain the heritable variation on which evolution depends. On the other hand, the occurrence of GEIs can also reduce the evolutionary response by weakening selection pressures on certain genotypes.

Finally, if a population's environment is heterogeneous in time or space, then the environmental dependence of genetic variances and covariances reduces the speed of the evolutionary change, and makes it difficult to predict the response to selection (Turelli 1988). Evolutionary models should take into account the possibility that quantitative genetic parameters, and especially additive genetic variance, can be variable. Hence, the evolutionary potential of a population is not static but evolves when environmental conditions differ. This is true when animals are put in experimental conditions such as extreme stress, which they might never experience in nature, but it is also true when looking at natural variation such as variation in temperature or food availability. In addition, if h2 varied randomly across environments, predictions in the evolution of a trait could potentially be made using the mean h2 across all environments. However, if, as we have shown, h2 tends to be lower in harsher conditions, when selection pressures are highest, then explaining past evolution and predicting future evolution becomes very difficult.

6. Future directions

Most of the studies assessing the influence of variable environmental conditions are performed under at most only three different environments. A more constructive approach would be to apply and test the variance and covariance structure under a wider ranger of conditions, ideally using continuous environmental gradients. Theoretical models predicting changes in additive genetic variances and covariances in different environments show that genotype by environment interactions can lead to quadratic trends of additive genetic variance across environments (de Jong 1990). Such nonlinear trends cannot be detected nor understood without investigations of gradients of environmental conditions. Once we understand how the expression of genetic variation varies according to environmental conditions, the models explaining genetic adaptation and response to selection could take into account this variation and become more effective in their predictions.

Furthermore, despite being potentially logistically challenging, more studies of the stability of the G-matrices, ideally performed under natural conditions, are needed before we can generalize about the variability of the genetic variance–covariance in nature depending on environments. Moreover, variability among sex within population and across environmental conditions might need to be also taken into accounts in future studies, because the G-matrix (and genetic correlations) evaluated among sexes has provided some evidences that these parameters might be variable between sexes (but see Simons & Roff 1996; see Guntrip et al. 1997; Jensen et al. 2003; Parker & Garant 2004).

Finally, although the number of studies focusing on the variability of fitness-related traits has increased recently, we still need more of these studies to be able to clearly generalize about the potential lack of variability of estimates for fitness-related traits, as discussed in §2 of this review. Traits linked to reproductive investment such as reproductive timing and propagule size or weight, for which estimates are already available in some long-term population studies (Réale et al. 2003; Sheldon et al. 2003; McCleery et al. 2004), should be easily re-analysed under a range of different conditions to provide further reference.

Acknowledgements

We are grateful to B.C. Sheldon, J. Merilä, W. Hill and two anonymous reviewers for helpful comments and suggestions that greatly improved the manuscript. This work was supported by a Marie Curie Intra-European Fellowship (MEIF-CT-2003-501286) to A.C. D.G. was supported by a Natural Sciences and Engineering Research Council of Canada (NSERC) postdoctoral research fellowship, and by a Biotechnology and Biological Sciences Research Council (BBSRC) grant to B.C. Sheldon and L.E.B. Kruuk.

Footnotes

References

View Abstract