SEARCH

SEARCH BY CITATION

Keywords:

  • cost of canalization;
  • cost of plasticity;
  • environmental canalization;
  • meta-analysis;
  • natural selection;
  • phenotypic plasticity;
  • reaction norm evolution

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Organisms are capable of an astonishing repertoire of phenotypic responses to the environment, and these often define important adaptive solutions to heterogeneous and unpredictable conditions. The terms ‘phenotypic plasticity’ and ‘canalization’ indicate whether environmental variation has a large or small effect on the phenotype. The evolution of canalization and plasticity is influenced by optimizing selection-targeting traits within environments, but inherent fitness costs of plasticity may also be important. We present a meta-analysis of 27 studies (of 16 species of plant and 7 animals) that have measured selection on the degree of plasticity independent of the characters expressed within environments. Costs of plasticity and canalization were equally frequent and usually mild; large costs were observed only in studies with low sample size. We tested the importance of several covariates, but only the degree of environmental stress was marginally positively related to the cost of plasticity. These findings suggest that costs of plasticity are often weak, and may influence phenotypic evolution only under stressful conditions.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Environmental canalization and phenotypic plasticity refer to changes in the development of the phenotype as the environment changes, and are inferred from the slope of the phenotypic norm of reaction (Woltereck, 1909; Waddington, 1942; Flatt, 2005). A relatively strong response to environmental variation (i.e. a steep reaction norm) indicates phenotypic plasticity, and a weak response (a flat reaction norm) represents environmental canalization. The slope of the norm of reaction is modified by natural selection in two ways. One arises from simple costs and benefits associated with phenotypic expression (Via & Lande, 1985; Gomulkiewicz & Kirkpatrick, 1992; Moran, 1992). For example, individuals with chemical or behavioural defences against predators or herbivores may enjoy high fitness in risky environments compared with undefended individuals, whereas the undefended phenotype may do better in the absence of risk (De Meester et al., 1995; Van Buskirk et al., 1997; Agrawal, 2000; van Hulten et al., 2006). This occurs because defences themselves are costly and therefore enhance fitness only when enemies are present (Lively, 1986; Moran, 1992). Depending on the locations of fitness optima in the different environments, this kind of selection can favour the evolution of adaptive plasticity or canalization (Via & Lande, 1985; Gomulkiewicz & Kirkpatrick, 1992; de Jong, 1995).

However, selection acting purely within environments is not the sole adaptive evolutionary process influencing reaction norms. This study is concerned with a second kind of selection, which targets the capacity to exhibit phenotypic plasticity or canalization even if the individual organism never expresses that capacity. In other words, selection can act on the shape of the reaction norm if that shape affects fitness independently of the character values expressed within single environments (DeWitt et al., 1998). Of course, parameters describing reaction norm shape are not actually visible to selection (Kingsolver et al., 2001a, b), but there may be fitness consequences of the genetic, sensory or behavioural mechanisms that underlie the capacity to produce a canalized or plastic phenotype (Via et al., 1995; DeWitt et al., 1998). When this kind of selection favours genotypes with a reduced capacity for phenotypic plasticity, there exists what is called a ‘fitness cost of plasticity’ (van Tienderen, 1991; DeWitt et al., 1998; Callahan et al., 2008). Conversely, a ‘cost of environmental canalization’ occurs when the selection coefficient acting on the capacity for plasticity is positive (Tucic et al., 2005). The two are simply flip sides of the same coin, in both cases referring to natural selection acting on reaction norm slope. Plasticity costs help explain the limits of plasticity, and costs of canalization may constrain the ability of organisms to maintain phenotypic constancy in the face of environmental variation. These costs are expected to be common because developmental responses to environmental changes, like most other characters, show ample evidence of imperfect adaptation (van Tienderen, 1991; Travis, 1994).

Several factors may influence the costs of plasticity and canalization. For example, costs may be greater or more easily detected under stressful conditions, especially if canalization or plasticity utilizes resources that become scarce under stress (Dorn et al., 2000; Steiner, 2007; Steiner & Van Buskirk, 2008). Secondly, fitness costs are easier to detect using fitness measures that are closely associated with lifetime reproductive success (van Kleunen & Fischer, 2005). Thus, we should see the clearest costs of canalization and plasticity when fitness is represented by components such as survival or fecundity. Third, the adaptive value of canalization or plasticity may influence its cost. DeWitt et al. (1998) argue that only adaptive plasticity should be costly because selection quickly eliminates genotypes that express plasticity with costs but without any fitness benefits. If true, this same argument would also apply to adaptive canalization.

The way in which phenotypic plasticity is scored might also influence conclusions about its fitness consequences. Previous studies characterize plasticity as either the absolute value of the response to an environmental manipulation, so that genotypes with either ‘positive’ or ‘negative’ responses are judged to exhibit plasticity, or with signed values, so that only those genotypes responding in the adaptive direction have high plasticity. Authors favouring the signed approach correctly point out that plasticity should be costly only in the adaptive direction (van Kleunen & Fischer, 2005; Weinig et al., 2006). On the other hand, working with signed values produces the peculiar situation in which genotypes that exhibit a strong nonadaptive response (opposite to the slope of the adaptive reaction norm) are assumed to have higher fitness than genotypes showing no plasticity at all (see panel A in Appendix S1). This is a simple consequence of the regression method used to test for plasticity costs: if genotypes with high plasticity have reduced fitness, and genotypes with no plasticity have higher fitness, then those with negative plasticity are forced to have the highest fitness of all. This assumption is difficult to justify, and is clearly violated in some data sets (e.g. fig. 3 in Dechaine et al., 2007). In the end, the adaptive value of plasticity is usually uncertain (see results below; Sultan, 2004), and in that case it is necessary to treat plasticity as unsigned (panel C in Appendix S1). Thus, neither approach is ideal under all circumstances, and it therefore seems worthwhile to assess whether estimates of cost depend on which method is used.

Many recent studies have estimated costs of phenotypic plasticity and environmental canalization, all employing broadly similar methodology. It is now possible to make a quantitative assessment of whether costs are sufficiently frequent to influence the evolution of reaction norms. Here we report a meta-analysis of 27 studies published by September 2008, focusing on the importance of the four factors mentioned above: the level of environmental stress, the type of fitness component, evidence for adaptive plasticity and whether plasticity is signed. Our results have important implications for the widely held view that phenotypic evolution is constrained by plasticity costs (Agrawal, 2001) and for many influential models that explore that assumption (van Tienderen, 1991; Gomulkiewicz & Kirkpatrick, 1992; Moran, 1992; Sultan & Spencer, 2002).

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

The data set

We quantified the evidence for costs of canalization and plasticity by compiling all studies that have used the selection gradient method proposed by van Tienderen (1991; DeWitt et al., 1998; Scheiner & Berrigan, 1998) or a derivative of it. At minimum, this method involves a multiple regression of relative fitness measured within environment e (we) against trait values expressed within e (Xe) and trait plasticities across all environments (plX):

  • image(1)

The βs are estimated partial regression coefficients, trait values and plasticities are expressed in standard deviation units and the observations are the mean values of sibships, clonal genotypes or inbred lines. β2 and β3 are interpreted as selection gradients acting on trait expression within environment e directly and on the slope of the reaction norm across environments, respectively. The analysis can also include interactions or terms to estimate nonlinear selection (Scheiner & Berrigan, 1998).

The data set included 19 publications on 16 species of plant, and 8 publications on 7 species of animal (Table 1). The data set reflected some taxonomic bias in the organisms that have been studied: 25% of the plant species belonged to the family Brassicaceae, and 43% of the animal species were amphibians. Most studies reared organisms in two environments with high or low stress (judged from mean individual fitness; Hoffmann & Parsons, 1991). In these cases, plasticity was the difference in trait values between treatments. If three environments were included, reaction norms were estimated by regression (Stinchcombe et al., 2004) or by differences between pairs of treatments (Weijschede et al., 2006; Avramov et al., 2007). It would clearly be desirable to characterize the plasticity of each genotype based on phenotypes expressed across a large number of environments, but this was never done in the original studies. The data set was better suited for estimating costs of plasticity than costs of canalization, because the original investigators usually measured traits for which plasticity was known to be present and thought to be adaptive.

Table 1.   List of studies contributing to the meta-analysis.
TaxonLinesExperimental manipulationType of plastic traitFitness componentsSigned plasticity?Source
  1. ‘Lines’ is the number of sibships, clonal genotypes or inbred lines for which both fitness and plasticity were assessed. The covariates are defined in Table 2. We could not include Tucic et al. (1998) because that study reported only significant selection coefficients.

Plants
 Arabidopsis thaliana36Density, lightdev, morfecNoDorn et al. (2000)
21 DecTemperaturedev, morfecYesStinchcombe et al. (2004)
72Vernalizationdev, morfecYesCallahan et al. (2005)
360Densitydev, morfecYesWeinig et al. (2006)
 Brassica rapa24Densitydev, morfecYesPoulton & Winn (2002)
150Densitydev, morfecYesDechaine et al. (2007)
 Geranium carolinianum60LightmorfecNoBell & Galloway (2008)
 Impatiens capensis35Lightdev, morfecYesDonohue et al. (2000)
 Iris pumila13–15Lightmor, physizeNoAvramov et al. (2007)
 Lobelia cardinalis29 NovMoisturedev, morsizeYesCaruso et al. (2006)
 Lobelia siphilitica12Moisturedev, morsizeYesCaruso et al. (2006)
 Marchantia inflexa14–16Lightdev, morsizeYesFuselier & McLetchie (2002)
 Picea omorika117Densitydev, grow, morsizeNoTucic & Stojkovic (2001)
21LightdevsizeNoTucic et al. (2005)
 Plantago coronopus30Salinitymor, phyfec, sizeYesSmekens & van Tienderen (2001)
 Polygonum hydropiper22DensitymorfecYesGriffith & Sultan (2006)
 Polygonum persicaria24DensitymorfecYesGriffith & Sultan (2006)
 Ranunculus reptans102Competitionmorfec, sizeYesvan Kleunen et al. (2000)
 Raphanus raphanistrum28Herbivore damagemor, phyfecYesAgrawal et al. (2002)
 Sinapis arvensis31Lightdev, morfecNoSteinger et al. (2003)
 Trifolium repens34LightmorsizeYesWeijschede et al. (2006)
Animals
 Daphnia pulex50Predation riskmordev, fecNoScheiner & Berrigan (1998)
 Drosophila melanogaster20TemperaturephysurvNoKrebs & Feder (1997)
 Physa heterostropha243–253Predation riskgrow, morfec, sizeNoDeWitt (1998)
 Poecilia reticulata37Food levelphysizeYesBashey (2006)
 Rana sylvatica21Predation riskbehav, mordev, sizeYesRelyea (2002)
 Rana temporaria40Habitat durationdevsizeYesMeriläet al. (2004)
40Predation riskbehav, mordev, sizeNoSteiner & Van Buskirk (2008)
 Scaphiopus couchii12Food, temperaturedevsizeNoNewman (1994)

We recorded two response variables: the cost of plasticity or canalization (which is the selection coefficient β3 in eqn 1) and a measure of its variation. The values of β3 were directly comparable among studies and traits because they were standardized (expressed in units of relative fitness per 1 SD change in plasticity). The second response was the 95% confidence interval of β3 divided by the value of β3 itself, which is a measure of the detectability of plasticity costs (Steiner & Van Buskirk, 2008). A narrow confidence interval reflects either a large sample size or a clearly defined relationship between fitness and plasticity. This response was not available for the 44% of tests that did not report a measure of variation in β3.

We recorded from each study eight covariates that may influence selection on reaction norms (Table 2). These include the four covariates discussed in the Introduction: the level of environmental stress, the type of fitness component, whether plasticity was signed or positive and whether plasticity was judged to be adaptive. Either of two conditions was taken as evidence for adaptive plasticity: (a) selection in the two environments occurred in opposite directions and was consistent with plasticity, regardless of its significance or (b) selection was significant in the correct direction in one environment (α = 0.05) and was not different from zero in the other environment. We also recorded four additional covariates that were biologically relevant and available for all studies: the taxon (plant/animal), the trait exhibiting plasticity, whether the trait showed significant plasticity and the kind of experimental manipulation.

Table 2.   Results of generalized mixed linear models testing for effects of eight covariates on the cost of phenotypic plasticity (β3 in eqn  1).
CovariateLevel% of testsCoefficientSEFd.f.P-value
  1. All models included study as a random subject. The levels exhibited by the covariates and the distribution of the 536 tests among levels are shown. The environment with higher stress was characterized by lower mean fitness. ‘Signed plasticity’ records whether genotypes with negative values of plasticity were expected to have lower costs than genotypes with no plasticity (yes; Appendix S1, panel A) or whether costs were assumed to vary with the absolute magnitude of plasticity (no; Appendix S1, panel C). Density and competition manipulations differed according to whether competitors were conspecific (density) or heterospecific.

TaxonPlant72.40.03230.07860.171, 22.90.6853
Animal27.60
Environmental stressHigh52.1−0.08480.04713.251, 511.800.0722
Low47.90
Fitness componentFecundity, seed mass50.9−0.15940.15541.233, 94.40.3027
Size, growth rate37.5−0.0850.1529
Development, phenology8.20.07060.1754
Survival3.40
Plasticity adaptiveNo70.9−0.03680.05270.491, 460.400.4852
Yes29.10
Signed plasticityNo49.60.03880.05640.471, 105.900.4934
Yes48.90
Plasticity trait typeMorphology68.7−0.05160.16860.084, 192.600.9949
Development, phenology16−0.08520.1765
Physiology10.1−0.04610.1871
Behaviour3.4−0.04570.2325
Size, growth rate1.90
ExperimentalLight35.10.34810.19661.584, 26.70.209
manipulationPredation, herbivory26.90.32730.1979
Density, competition23.10.3630.1951
Physiological stress12.70.1860.2045
Food1.90
Plasticity significantNo20.50.02250.0580.151, 438.800.6979
Yes79.50

Statistical analysis

We used generalized mixed effects linear models to test whether the cost of plasticity was related to the covariates, in eight separate analyses, with study included as a random subject. It was not possible to evaluate interactions among covariates because the available studies often did not include the necessary combinations of factors. Observations were weighted by the natural logarithm of the number of lines; weighting did not employ a measure of variation in β3 because none was available for many tests. A normal distribution was assumed (and confirmed), and Satterthwaite’s approximation gave the degrees of freedom for testing fixed effects. We used similar models, but assumed a lognormal distribution, to determine whether the scaled 95% CI of the cost depended on the covariates. The models were implemented in sas proc glimmix (SAS, 2006).

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Costs of plasticity and canalization

The 27 studies reported 536 separate selection analyses involving different combinations of species, plastic traits, fitness components and treatments. An initial vote-counting approach to these many tests suggested that costs of plasticity and canalization were detected, although neither was consistently more frequent than the other. Overall, 28.6% of the 262 negative selection coefficients were significant at α = 0.05, reflecting costs of plasticity, and 21.8% of 262 positive coefficients were significant, reflecting costs of canalization (Fig. 1). More tests were significant than expected by chance (< 0.0001, binomial test), but there was no difference between positive and negative coefficients in the probability of significance (odds ratio = 0.69, P = 0.09, Fisher’s exact test). This was surprising, because most investigators measured traits known to be plastic, and most expected plasticity to be adaptive (although the data did not suggest that plasticity was adaptive for 71% of traits). Therefore, for reasons discussed in the Introduction, this data set should contain more instances of costly plasticity than costly canalization.

image

Figure 1.  Summary of the number of comparisons reporting positive, zero and negative estimates of β3 from eqn  1 (selection acting on phenotypic plasticity). Negative coefficients correspond to a cost of plasticity. Significance was assessed at α = 0.05. The number of tests (= 536) is greater than the number of publications (= 27) because each study included multiple tests for costs of plasticity in different traits, fitness measures and environments.

Download figure to PowerPoint

A funnel plot illustrated that studies with low sample size contained more extreme estimates of costs (Fig. 2a). The increase in variance with decreasing sample size is typical in meta-analyses, but the preponderance of highly negative estimates suggests a modest reporting bias against costs of canalization in small studies (Palmer, 2000). This should be kept in mind when interpreting the results.

image

Figure 2.  Costs of phenotypic plasticity and environmental canalization reported by the 27 studies included in the meta-analysis. Panel (a) illustrates selection on plasticity (β3) in relation to the number of lines (sibships, clonal genotypes or inbred lines for which plasticity and fitness were estimated). A modest reporting bias is indicated by the absence of high costs of canalization in studies with low sample size. Panel (b) shows that most costs were small (69% were between −0.2 and 0.2). Median coefficient of the 71 tests with > 50 lines was −0.01.

Download figure to PowerPoint

There was a mild cost of plasticity across all studies because negative coefficients (β3 in eqn  1) tended to be larger than positive coefficients (Fig. 2). The study-level mean cost came from the intercept of a mixed effects linear model (β3 = −0.074, SE = 0.034, P = 0.037). Thus, a 1 SD increase in phenotypic plasticity was associated with, on average, a 7.4% reduction in fitness. In the same model, there was significant heterogeneity in β3 among studies, reflecting variation in selection on plasticity among taxa or methodologies (likelihood ratio = 5.59, d.f. = 1, < 0.025).

Effects of covariates on costs

None of the covariates was significantly related to costs, although the impact of environmental stress was nearly significant (Table 2). Plasticity tended to be more costly in treatments with greater stress, but this was only true in studies of animals (Fig. 3, Table 2). Overall, β3 was not strongly related to the taxon, the kind of fitness trait or plastic trait, whether plasticity was adaptive or signed, the experimental manipulation used to induce plasticity, or the statistical significance of plasticity itself. Moreover, β3 was no different in three studies that used recombinant inbred lines (RILs) than in other studies (Callahan et al., 2005; Weinig et al., 2006; Dechaine et al., 2007) (F1,27.4 = 0.08, P = 0.771), although those studies were highly replicated and plasticity costs are said to be higher in populations of RILs (Weinig et al., 2006). Hence, of the expectations outlined in the Introduction, only one (more costly plasticity under stressful conditions) was weakly supported by the data.

image

Figure 3.  Relationship between selection acting on plasticity (β3) and the level of environmental stress for animals and plants. High stress was defined as the treatment in which fitness was the lowest. Symbols represent least square mean values (± 1 SE) from a mixed effects model that included taxon and stress as fixed effects and study as a random subject.

Download figure to PowerPoint

Detectability

Most studies had low power to detect costs of canalization or plasticity. Our measure of detectability, the 95% CI of β3 in eqn  1 divided by the value of β3, averaged 2.13 (95% CI 1.70–2.68). This indicates that the confidence intervals around β3 averaged more than twice as large as β3 itself. There was considerable variation among studies in detectability, which may partly reflect differences in sample size (likelihood ratio = 25.2, d.f. = 1, < 0.001). Detectability of costs was unrelated to any covariate that we recorded, and detectability was no better in the three studies of RILs than in other studies (F1,14.8 = 1.09, P = 0.314).

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Environmental variation experienced by individual organisms during development presents them with a challenge. On the one hand, the developing phenotype must be buffered against environmental insults (adaptive environmental canalization). However, other kinds of environmental variation are better met with appropriate phenotypic modification (adaptive phenotypic plasticity). Most organisms exhibit a combination of canalization and plasticity during development. Factors that limit the capability to express plastic or canalized phenotypes are interesting because they force organisms to evolve into either ecological specialists or generalists, respectively (Futuyma & Moreno, 1988; van Tienderen, 1991). Quantitative genetic methods for estimating these costs have been available for about 15 years, but no systematic review of empirical results has been conducted. Our meta-analysis of all empirical estimates of plasticity costs revealed that, surprisingly, fitness costs were relatively mild and acted about equally often in favour of and against plasticity. Plasticity costs were slightly larger in stressful environments (Fig. 3), but the other covariates that we measured were unrelated to costs or their detectability. These results will encourage a reconsideration of constraints on the evolution of plasticity and canalization, and may stimulate the development of alternative methods for estimating the potential importance of plasticity costs.

The magnitude of selection coefficients acting on plasticity (β3) was similar to that reported by Kingsolver et al. (2001b; Knapczyk & Conner 2007) in a review of natural selection-targeting morphological and life-history characters. This means that costs of plasticity and canalization are usually weak (Fig. 2b). The mean absolute value of β3 in our study (0.25 ± 0.49 SD) was somewhat larger than the mean directional selection coefficient observed by Kingsolver et al. (0.22), but the median was smaller in our study (0.08 vs. 0.16). A key difference, of course, is that the traits Kingsolver et al. studied were expected to undergo positive and negative selection equally often, whereas the investigators contributing data to our study expected plasticity to undergo primarily negative selection (reflecting a cost of plasticity). Our meta-analysis firmly rejected the latter expectation. In fact, the modest publication bias that we discovered created an over-representation of large negative coefficients in studies with small sample size (Fig. 2a), so that actual costs of plasticity are probably lower than suggested by our analyses.

Why are costs apparently so low?

There are three possible explanations for these results, each with different implications for understanding the evolution of developmental reaction norms. One explanation proposes that costs are in fact low or absent most of the time, another suggests that costs simply appear to be low because they are uniformly present in all genotypes and a third states that costs occur only under special conditions.

When natural selection favours the evolution of either plastic or canalized responses to the environment, any inherent costs of these responses are also exposed to selection. Hence, one explanation for low costs observed in nature is that selection has already removed genotypes with high costs, especially what DeWitt et al. (1998) refer to as genetic costs (plasticity loci interact with loci affecting fitness). This explanation implies that fitness costs and limits associated with adaptive reaction norms are themselves evolutionarily labile, and therefore are not of long-term importance for explaining observed levels of plasticity and canalization.

One way to address this idea involves comparison of fitness among newly created RILs that have not yet been exposed to selection, and therefore may exhibit reaction norms not observed in nature (Callahan et al., 2005; Weinig et al., 2006; Dechaine et al., 2007). Some of these reaction norms could be associated with plasticity costs that exceed those occurring in nature, and therefore costs may be more apparent in populations synthesized of RILs than in natural populations. We found no evidence to support this suggestion, although rather few studies are available. Weinig et al. (2006) report that they are currently testing whether selection can modify costs by rearing a sample of RILs, some of which show detectable plasticity costs, under natural conditions over several generations. If lines having costly plasticity decline in frequency, the results will provide experimental confirmation of this explanation for our findings: costs of canalization and plasticity are low because they have been purged by selection.

A second possibility is that costs only seem to be low because they are uniformly present in populations. Costs of plasticity are proposed to arise from unavoidable consequences of possessing the capacity to respond adaptively to environmental variation (DeWitt, 1998; DeWitt et al., 1998). If all genotypes have that capacity for at least some traits, then all genotypes will bear some measure of cost. However, if there is little genetic variation in plasticity costs, then current methods have no hope of detecting them. This explanation implies that costs of plasticity are widespread, not easily eliminated by selection, and potentially important for setting the level of adaptive plasticity within natural populations. The observed reaction norm therefore represents a compromise between the fitness benefits of greater plasticity and the increasing costs associated with it, as envisioned by quantitative genetic models (van Tienderen, 1991; Gomulkiewicz & Kirkpatrick, 1992). This argument applies equally to canalization (although the precise mechanisms underlying costs may differ), because it emphasizes fitness consequences of either steep or flat adaptive reaction norms that cannot easily be purged by selection. One prediction of this explanation has been confirmed: directional selection acting within environments sometimes favours even more extreme phenotypes than those produced by an adaptively plastic reaction norm (Kingsolver, 1995; Van Buskirk et al., 1997). The conclusion is that, in the absence of plasticity costs, even greater plasticity would evolve.

A final explanation is that phenotypic plasticity and canalization are costly only under ecological contexts not replicated in experiments (e.g. so-called ‘ecological costs’; Agrawal, 2001). Our finding that plasticity costs were slightly higher in stressful treatments may indicate that ecological costs increase in the context of competition, predation risk or resource limitation. Other possibilities have been suggested such as delayed phenotypic expression experienced only by plastic (or canalized) genotypes or limits imposed by the reliability of information about the environmental state (Padilla & Adolph, 1996; DeWitt et al., 1998; Tufto, 2000). These explanations salvage an important role for costs in reaction norm evolution, and some of them might not be detected by existing studies of plasticity costs.

Alternative approaches

Other methods for measuring selection on canalization and plasticity are available, based on estimating trade-offs associated with novel reaction norms in artificial populations. In one approach, experimental populations of micro-organisms are allowed to evolve under constant and variable conditions. Generalists usually evolve in variable environments, and specialists evolve in constant environments. If plasticity is costly, then the fitness of generalists averaged across environments should be lower than that of specialists. Costs of plasticity, defined in this way, are usually not detected (reviewed in Kassen, 2002). However, the interpretation of these studies is complicated because they measure only fitness, rather than the underlying traits responsible for fitness variation. The argument is that canalized fitness across environments implies that physiological plasticity has evolved at a more mechanistic level (DeWitt et al., 1998). However, if such a study observed that generalists pay fitness costs, it would be unclear whether these should be considered costs associated with canalization (of traits closely related to fitness) or with plasticity (of unknown underlying traits).

In the end, the most compelling evidence for costs of canalization or plasticity may come from studies of trade-offs that appear when reaction norm parameters are subjected to artificial selection. Numerous studies have imposed selection on developmental reaction norms (e.g. Waddington, 1960; Scheiner & Lyman, 1991; Callahan & Pigliucci, 2005; Suzuki & Nijhout, 2006), but few have checked whether fitness variation among the resulting lines is associated with canalization or plasticity. Mery & Kawecki (2002, 2003, 2005) found that Drosophila populations selected for increased capacity to learn suffered reductions in stress resistance and competitive ability, implicating fitness costs of the neurological plasticity required for learning. The cost was associated not with the process of learning, but rather with the capacity to learn. Experiments such as this could be extended over realistic time frames to determine whether plasticity costs represent a long-term constraint on phenotypic evolution.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Thanks to F. Bashey, J. Dechaine, T. Griffith, T. Steinger, B. Tucic and M. van Kleunen for sharing data or answering questions about their publications. For helpful comments on the manuscript, we thank T. DeWitt, K. Donohue, M. Pigliucci, M. van Kleunen and Y. Willi. We were supported by the Swiss Nationalfonds, the Australian Research Council, and the Zoological Institute of the University of Zürich.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Appendix S1 Illustration of methods for detecting selection on plasticity using the signed and unsigned approaches.

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
JEB_1685_sm_2008-00556 appendix.doc291KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.