SEARCH

SEARCH BY CITATION

Keywords:

  • adaptive plasticity;
  • Arabidopsis thaliana;
  • cost of plasticity;
  • flowering time;
  • genetic correlation;
  • genotypic selection;
  • phenotypic plasticity;
  • reaction norm;
  • selection gradients

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Environmental variation in temperature can have dramatic effects on plant morphology, phenology, and fitness, and for this reason it is important to understand the evolutionary dynamics of phenotypic plasticity in response to temperature. We investigated constraints on the evolution of phenotypic plasticity in response to a temperature gradient in the model plant Arabidopsis thaliana by applying modern analytical tools to the classic data of Westerman & Lawrence (1970). We found significant evidence for two types of constraints. First, we detected numerous significant genetic correlations between plastic responses to temperature and the mean value of a trait across all environments, which differed qualitatively in pattern between the set of ecotypes and the set of mutant lines in the original sample. Secondly, we detected significant costs of flowering time plasticity in two of the three experimental environments, and a net pattern of selection against flowering time plasticity in the experiment overall. Thus, when explored with contemporary methods, the prescient work of Westerman & Lawrence (1970) provides new insights about evolutionary constraints on the evolution of plasticity.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Phenotypic plasticity, or the ability of a single genotype to produce multiple phenotypes in response to environmental variation, has been the subject of intense investigation and debate for nearly 40 years (see e.g. Bradshaw, 1965; Schlichting, 1986; Scheiner, 1993; Sultan, 1995, 2000; Via et al., 1995; Schmitt, 1997 for reviews). Phenotypic plasticity is considered adaptive if the pattern of natural selection differs between environments, and the plastic phenotypic response is in the direction favoured by selection (Schmitt et al., 1995; Dudley & Schmitt, 1996; Dorn et al., 2000). There have been several observations of adaptive plasticity as well as maladaptive and nonadaptive plasticity (e.g. Schmitt et al., 1995; Warkentin, 1995; Dudley & Schmitt, 1996; Nunney & Cheung, 1997; Schmitt, 1997; Denver et al., 1998; Donohue et al., 2000a,b, 2001; Dorn et al., 2000; Agrawal et al., 2002), which raises an interesting paradox. That is, given the potential advantages of plasticity, and the observations of adaptive plasticity, why aren't organisms perfectly plastic?

Several potential constraints on the evolution of phenotypic plasticity have been identified. First, significant genetic correlations, either within or across environments (Via & Lande, 1985, 1987; Gomulkiewicz & Kirkpatrick, 1992; van Tienderen & Koelewijn, 1994) are capable of constraining or altering the evolutionary trajectory of the plastic trait. Secondly, it is possible that plasticity is costly (van Tienderen, 1991; DeWitt et al., 1998; Scheiner & Berrigan, 1998). Interestingly, evidence for costs of plasticity in plant morphological/architectural traits is ambiguous or rare at best (e.g. Donohue et al., 2000a; Dorn et al., 2000) while costs of tolerance to herbivore damage (i.e. plasticity in fitness in response to herbivory) have been observed in a handful of cases (Simms & Triplett, 1994; Tiffin & Rausher, 1999; Stinchcombe, 2002; Weinig et al., 2003). In addition, the evolution of phenotypic plasticity is critically dependent on the relative frequency of selective environments (e.g. Gomulkiewicz & Kirkpatrick, 1992). Unfortunately, the frequency of selective environments is rarely quantified, making it difficult to evaluate this potential constraint (but see Weis & Gorman, 1990; Kingsolver et al., 2001; Arnold & Peterson, 2002; Huber et al., in press). Although these potential constraints on the evolution of phenotypic plasticity have been identified, there are still comparatively few data on how frequently they are of sufficient magnitude or prevalence for selection to act against phenotypic plasticity (see Dorn et al., 2000 for some exceptions). In other words, how frequently is a homeostatic response favoured and a plastic response selected against?

Westerman & Lawrence (1970) provided an early example of selection acting against plasticity in their study of plastic responses to temperature in the model plant, Arabidopsis thaliana. Westerman and Lawrence's paper is frequently cited as supporting evidence for the correlation between fruit and seed number in Arabidopsis, but in addition to that oft-cited correlation, Westerman and Lawrence (W&L in our shorthand) proposed and tested a comprehensive framework for understanding the fitness consequences of phenotypic plasticity. According to W&L, a ‘developmentally flexible’ genotype produced different phenotypes in different environments and had high average fitness over all environments, while ‘developmentally inflexible’ genotypes produced the same phenotype in different environments but had low average fitness. These classifications are essentially equivalent to the current concepts of adaptive plasticity and maladaptive homeostasis (Thompson, 1991; DeWitt, 1998; DeWitt et al., 1998; Winn, 1999). W&L proposed that genotypes with low average fitness that produced different phenotypes in different environments were ‘developmentally unstable,’ while genotypes that produced the same phenotype in different environments but had high average fitness were ‘developmentally stable.’ These terms are essentially equivalent to the current concepts of maladaptive plasticity and adaptive homeostasis.

After developing this classification scheme, W&L proceeded to test the developmental flexibility and stability of height at flowering, rosette leaf number at flowering, and days until flowering in Arabidopsis in response to temperature, an important ecological variable for Arabidopsis. [For instance, temperature plays a key role in determining biogeographic range limits for Arabidopsis populations (e.g. Hoffman, 2002), in addition to mediating life history transitions, especially the transition from vegetative growth to flowering (see e.g. Johanson et al., 2000; Blazquez et al., 2003)]. W&L characterized the plasticity of flowering time, height, and leaf number in 33 inbred lines with joint regression analysis (see below). By plotting mean fruit number (over all environments) against a measure of plasticity from the joint regression analysis (their Fig. 3), W&L sought to determine fitness consequences of plasticity in flowering time, height, and leaf number in response to temperature. W&L did not present estimated regression coefficients or their statistical significance for the relationships between plasticity and fitness – although they note that one relationship is ‘not quite significant’, implying that statistical analyses were performed. As such, it is difficult to fully assess the magnitude and statistical significance of the relationships they presented. Moreover, they did not attempt to account for correlations among traits as potential constraints on the response to selection, or to distinguish direct selection from indirect selection on correlated traits (e.g. Lande & Arnold, 1983).

Here we apply modern analytical tools to W&L's data, with the goal of filling in several gaps in their analyses. First, to evaluate genetic constraints on the evolution of plasticity, we estimated genetic correlations between the trait means, estimated over all environments, and plastic responses to temperature (see below), for all combinations of traits. Secondly, to test for costs of plasticity within individual environments in W&L's experiment, we utilized recently developed statistical approaches (e.g. van Tienderen, 1991; DeWitt, 1998; DeWitt et al., 1998; Scheiner & Berrigan, 1998) that were unavailable at the time of W&L's study. In addition, to test for direct selection on plasticity, we estimated the pattern of natural selection on the trait means and plasticities in a multiple regression model that accounts for correlations between traits, taking advantage of the Lande & Arnold (1983) approach for estimating selection on correlated traits. By examining W&L'S remarkably prescient data in a contemporary framework, we were able to obtain new insights into the nature of genetic constraints on the evolution of flowering time plasticity in Arabidopsis.

Westerman and Lawrence's methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Westerman and Lawrence used 33 inbred lines, 12 of which were mutant lines. Four of these lines were termed ‘major mutant’ lines by W&L, and included two that were glabrous (Coimbra-1 from Portugal and Wilna-2 from the U.S.S.R. in their terminology) and two others named stellula-1 and apetala. The remaining eight mutant lines were lines derived from radiation-induced mutants that had been exposed to cobalt-60, as described by Lawrence (1968a,b; note that the mutagenesis experiments were performed by C.W. Lawrence, not the co-author of W&L who was M.J. Lawrence). Most of the 21 nonmutant lines were drawn from the Laibach collection. A modified version of W&L's Table 2 describing the lines is presented in Table A1 in the appendix. Any correspondence between these mutants and inbred lines and current mutants and inbred lines of similar names that are available from Arabidopsis stock centers is uncertain.

Table 2.  Analysis of costs of plasticity within the three temperature environments used by Westerman and Lawrence. For each temperature environment, fitness within that environment was regressed on the terms shown.
TermMutant linesNonmutant lines
βSEPβSEP
  1. Significant terms (P < 0.05) are shown in bold and marginally significant terms (0.05 < P < 0.10) in italics.

15 Degrees
 Flowering time (FT)0.040.660.9490.080.230.733
 Height (HT)−0.040.110.7260.200.090.046
 Leaf number (LN)0.491.170.691−0.050.770.947
 FT plasticity3.6418.670.853−9.876.100.128
 HT plasticity−2.136.230.746−3.753.110.248
 LN plasticity−1.466.310.8272.493.920.536
20 Degrees
 Flowering time (FT)−0.100.240.704−0.210.240.409
 Height (HT)0.150.060.041−0.100.070.157
 Leaf number (LN)−0.370.590.5590.250.590.673
 FT plasticity3.503.670.3839.334.490.0567
 HT plasticity−3.983.930.357−4.652.720.109
 LN plasticity−1.092.750.7070.453.080.886
25 Degrees
 Flowering time (FT)−0.220.230.378−0.090.170.626
 Height (HT)−0.010.040.8550.140.070.0486
 Leaf number (LN)1.280.500.0490.200.550.718
 FT plasticity1.371.760.4728.393.790.044
 HT plasticity−2.372.650.412−0.302.260.897
 LN plasticity0.941.420.5372.222.290.348
Table Appendix.  Table A1 Modified version of Westerman and Lawrence's Table 2, containing information about the inbred lines used in their experiment.
Line numberNameCountry of originSourceMutant
 1EnkheimGermanyLaibach collection0
 2EifelGermanyLaibach collection0
 3S2ELawrence, 1968a,b1
 4A3L2Lawrence, 1968a,b1
 5EstlandUSSRLaibach collection0
 6C3L2Lawrence, 1968a,b1
 7Henley-in-ArdenEnglandNatural populations0
 8MaineFranceNatural populations0
 9S1L2Lawrence, 1968a,b1
10C2L1Lawrence, 1968a,b1
11Langridge0
12C3ELawrence, 1968a,b1
13LimburgGermanyLaibach collection0
14A1ELawrence, 1968a,b1
15S3L1Lawrence, 1968a,b1
16Landsberg-1GermanyLaibach collection0
17Bologna-1ItalyLaibach collection0
18Coimbra-1PortugalLaibach collection1 (glabrous)
20Le Mans-2FranceLaibach collection0
22Palermo-1ItalyLaibach collection0
23BurghhaunGermanyLaibach collection0
24Eifel-6GermanyLaibach collection0
25GückingenGermanyLaibach collection0
26Wilna-2USSRLaibach collection1 (glabrous)
27OysteseNorwayLaibach collection0
28Estland-1USSRLaibach collection0
29Enkheim-2GermanyLaibach collection0
33Pitztal-2GermanyLaibach collection0
34Antwerp-1BelgiumLaibach collection0
35GöttingenGermanyLaibach collection0
37DijonFranceLaibach collection0
39stellula-1 1
40apetala 1

Westerman and Lawrence planted seeds on agar medium in test tubes, and three treatments were applied: growth at 15, 20 and 25 °C (59, 68 and 77 °F, respectively) in ‘environmental cabinets’ with 16 h daylengths and 80% relative humidity. Data available from the International Panel on Climate Change (http://ipcc-ddc.cru.uea.ac.uk/cru_data/visualisation/visual_index.html; for a description see New et al., 1999) indicates that average daily maximum temperatures in the 15–20 °C range are found throughout the geographic range of Arabidopsis for several months of the growing season, although temperatures in the 20–25 °C range are only found in the SE United States and North Africa for a limited number of months. These climatological data suggest the environmental gradient used by W&L was ecologically relevant (cf. Arnold & Peterson, 2002). Westerman and Lawrence placed 10 seeds per line in two completely randomized blocks within each environmental cabinet. The traits W&L measured were flowering time (in days), height at flowering (mm), and rosette leaf number at flowering. To estimate fruit number, W&L counted the number of fruits for five plants in each line, block and cabinet. Westerman and Lawrence also counted seed number in a subset of these lines, and from these data W&L determined the frequently-cited correlation between fruit number and seed number.

Data

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Westerman & Lawrence (1970) reported inbred line means for fruit number, flowering time, height, and leaf number in each environment; these data form the basis of our reanalysis. In addition, they reported the grand mean for each inbred line for all of these traits (i.e. each inbred line's mean over all environments) as well as the three environmental means (i.e. the mean value of these traits within each experimental environment). We refer to the grand mean of a given trait as a ‘trait mean.’ The grand mean of fruit number we considered an estimate of overall fitness; estimating fitness in this manner assumes that these environments are equally encountered by all genotypes.

Measures of plasticity

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Westerman & Lawrence (1970) present convincing evidence of significant inbred line × environment interactions for the traits of interest, indicating that plasticity was genetically variable in their sample (F64,1324 ≥ 7.1, P < 0.0001 for the inbred line × environment interactions for flowering time, height, and leaf number and F64,704 = 2.11, P < 0.0001 for the inbred line × environment interaction for fruit number; F-statistics were calculated from the mean squares reported by W&L in their Table 5). These significant interactions cannot be explained by simple scaling effects or changes in trait variance across environments, because W&L performed their original anova on square-root transformed data to stabilize the variances across all treatments. Moreover, graphical plots of the reaction norms show that inbred lines indeed changed rank order across environments (see Fig 1a–c). The observation of significant inbred line × environment interactions and crossing reaction norms is sufficient evidence to infer statistically significant heterogeneity among the measures of plasticity calculated for each inbred line (Falconer, 1990; Falconer & Mackay, 1996).

image

Figure 1. Reaction norms for flowering time, leaf number, and plant height as functions of the environmental means for these traits. Inbred line means for the traits are portrayed on the vertical axis, while the environmental means for the traits in the three experimental treatments are portrayed on the horizontal axis. For panels (a and b), environmental means increased with decreasing temperature; for panel (c), environmental means increased with increasing temperature. Regression lines show the plastic response of inbred lines across environments. (a) Flowering time, (b) leaf number, (c) plant height.

Download figure to PowerPoint

Following W&L, we estimated plasticity to temperature using joint regression analysis. Briefly, in this procedure an inbred line's mean value of a trait in each environment is regressed against the environmental means for that trait (i.e. the mean value of the trait within an environment over all inbred lines), and the resulting regression coefficient for each inbred line is taken as a measure of its plasticity or environmental sensitivity (Falconer, 1990; Falconer & Mackay, 1996). The joint-regression approach has a long history in traditional and agricultural quantitative genetics (see e.g. Yates & Cochran, 1938; Finlay & Wilkinson, 1963; Perkins & Jinks, 1968a,b; Zuberi & Gale, 1976; Mather & Jinks, 1982; Falconer, 1990; Falconer & Mackay, 1996, pp.133–134; Lynch & Walsh, 1998, pp.672–678; Gurganus et al., 1998; Leips & Mackay, 2000) but has been largely lacking in recent studies seeking to measure selection on phenotypic plasticity.

Although it may appear unorthodox to estimate measures of plasticity from a regression containing three data points, we suggest that this approach is not altogether different from traditional measures of plasticity. Traditionally, plasticity is defined as the difference between an inbred line's mean value in environment ‘A’ and the same inbred line's mean value in environment ‘B.’ (e.g. Dorn et al., 2000). Thus, if we only had data on the 15 and 25 °C environments, plasticity would be calculated as a simple difference in trait values between these environments. However, as the reaction norms to temperature in this study were largely linear, the presence of the intermediate 20 °C environment and estimates of the traits expressed in this environment allows a more accurate estimate of plasticity for each inbred line. These data allow us to fit an estimate of plasticity that reflects the expression of the traits in that environment, rather than simply assuming that it would be the midpoint between 15 and 25 °C environments, as would be the case if we estimated plasticity as a simple arithmetic difference. Using environmental means rather than temperature as the independent variable in these regressions also allows us to estimate an inbred line's plasticity relative to the average population responses to temperature, rather than as a mathematical function of temperature per se (Lynch & Walsh, 1998).

Measuring plasticity in this manner has several advantages. First, it allows a straightforward estimate of overall plasticity that uses all of the data, and produces a single plasticity estimate for cases in which more than two experimental environments are used. Second, the mean plasticity of the population is by definition 1 and plasticity is dimensionless, easing comparisons across traits and species (Falconer, 1990). Finally, as shown by Falconer (1990), the variances and covariances of means and plasticities are easily calculated from the manipulation of variances and covariances, and apply equally to phenotypic and genotypic data. These mathematical expressions allow one to determine if any genetic correlations between means and plasticities are of biological significance rather than numerical/statistical artifact (see appendix B of Tiffin & Rausher (1999) for a discussion of artifactual correlations between means and plasticities when plasticity is defined as the difference between two means).

Estimates of genetic correlations

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

To estimate genetic correlations between trait means and plasticities, we evaluated inbred-line mean correlations between trait means and plasticities for all combinations of traits. We used simple Pearson's correlations of inbred line means as an approximation of genetic correlations, as the inbred line means were the only data available. Given the limited sample sizes and already low power of our analyses, we do not apply a Bonferroni correction, which can be prohibitively conservative (e.g. Moran, 2003).

Estimates of costs of plasticity

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

To estimate costs of plasticity within individual environments, we modified the techniques originally proposed by van Tienderen (1991), and subsequently developed by DeWitt et al. (1998) and Scheiner & Berrigan (1998). To test for costs of plasticity in this framework, an inbred line's mean fitness within an individual environment is regressed on the inbred line means of the trait within that environment and a measure of plasticity, that is:

  • image(1)

where wi,e is the fitness of genotype i in environment e, Xi,e is the mean value of the trait for genotype i in environment e, plXi is a measure of the plasticity of genotype i, and β1 and β2 are estimated regression coefficients. A significant, negative regression coefficient for plasticity (that is, β2 << 0) indicates that more plastic inbred lines have lower fitness in the environment under consideration, even when controlling for the mean value of the trait in this environment – in other words, plasticity is costly (van Tienderen, 1991). To extend this approach to consider multiple traits simultaneously, one simply includes additional terms for the means and plasticities of those traits in the statistical model. In the case of two experimental environments, a genotype's plasticity is usually measured as the difference between its mean in both environments. However, measuring plasticity as a difference between means quickly becomes cumbersome with a large number of environments, as the number of measures of plasticity for each genotype increases quite quickly with the number of experimental environments. Accordingly, to test for costs of plasticity within individual environments, we used the regression coefficients from the joint regression analysis as our estimates of plXi for equation (1). This approach also allowed us to use the same measure of plasticity in our analysis of costs of plasticity in all three environments.

Estimates of selection on plasticity

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

To estimate the net pattern of selection on trait means and plasticities, we used a multiple regression analysis to control for correlations between independent variables (e.g. Lande & Arnold, 1983; Rausher, 1992). In these analyses we used the grand mean of fruit set for each inbred line (calculated over all environments) as our estimate of fitness, and our estimates of trait means and plasticities as the independent variables. In essence, this approach measures the net pattern of selection on plasticity, and is the commonly used method to measure selection on plastic responses to herbivore damage (e.g. Mauricio et al., 1997; Tiffin & Rausher, 1999; Stinchcombe & Rausher, 2002; Weinig et al., 2003). We detected no evidence of quadratic selection and for this reason present only linear selection gradient analyses. Preliminary analyses indicated a significant main effect of whether an inbred line was a mutant or not, and differences in the pattern of selection between mutant and nonmutant lines. Accordingly, we performed separate analyses for mutant and nonmutant lines; in the mutant category we pooled radiation-induced mutants, the named mutants (stellula and apetala), and the glabrous mutants.

Mutant lines

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Genetic correlations between trait means and plasticities in mutant lines were either small and nonsignificant or large and significant (Table 1 above the diagonal). For instance, plasticity in flowering time showed a significant, positive genetic correlation with mean flowering time across all temperatures (r = 0.63, P < 0.05) and plasticity in leaf number also showed a significant, positive genetic correlation with mean leaf number across all temperatures (r = 0.61, P < 0.05).

Table 1.  Genetic correlations between plasticities and trait means in Westerman and Lawrence's experiment. Data from mutant lines (n = 12) are shown above the diagonal while data from nonmutant lines (n = 21) are shown below the diagonal.
 Flowering time plasticityHeight plasticityLeaf number plasticityMean flowering timeMean heightMean leaf number
  1. P ≤ 0.1; **P < 0.05; ***P < 0.005.

Flowering time plasticity0.260.80***0.63**0.0030.48
Height plasticity−0.60***0.60**−0.51†0.18−0.04
Leaf number plasticity0.72***−0.110.19−0.250.61**
Mean flowering time0.18−0.31−0.01−0.160.50†
Mean height0.47**−0.450.210.42†−0.80***
Mean leaf number0.20−0.300.100.88***0.28

Correlations between trait means did not necessarily reflect the correlations among their plasticities. Leaf number and flowering time are usually positively correlated in this species, such that leaf number is often used as a surrogate measure of flowering time (e.g. El-Assal et al., 2001). However, for the mutant lines in this experiment the trait means for flowering time and leaf number over all environments were only marginally correlated, but plasticity in leaf number and plasticity in flowering time were positively correlated (r = 0.80, P < 0.005) – in other words, inbred lines with higher plasticity in leaf number also had higher plasticity in flowering time. Inbred lines with high mean leaf number over all environments tended to have low mean height over all environments, as indicated by the highly significant, negative correlation between these two trait means (r = −0.80, P < 0.01); however, the plasticities of these two traits were positively correlated (r = 0.60, P < 0.05). Plasticity in flowering time and plasticity in height were not significantly correlated with each other.

Nonmutant lines

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Nonmutant lines demonstrated a qualitatively different pattern of genetic correlations (Table 1, below the diagonal). In contrast to the mutant lines, in the nonmutant lines there was only one significant correlation between a trait mean and a trait's plasticity, in this case a negative correlation between average height over all environments and plasticity in height (r = −0.45, P < 0.05) – i.e. taller inbred lines exhibited less height plasticity. In addition, there was a significant positive genetic correlation between average height and plasticity in flowering time (r = 0.47, P < 0.05), the only significant genetic correlation between one trait's mean and another trait's plasticity in the nonmutant lines. In contrast to the mutant lines, the correlation between mean flowering time and rosette leaf number was significant and positive (r = 0.88, P < 0.005), as was the correlation between plasticity in flowering time and plasticity in rosette leaf number (r = 0.72, P < 0.005).

Similar to what was observed in the mutant lines, the correlations between trait means across environments was not necessarily indicative of the correlation between those traits’ plasticities. For instance, average flowering time and average height across all temperatures showed a marginally significant positive correlation (r = 0.42, P < 0.10) while the plasticities of those traits were negatively correlated (r = −0.60, P < 0.005).

Nonmutant lines

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

In the nonmutant lines, we observed similar patterns of selection within the three experimental environments. For instance, there was significant selection against taller inbred lines in both the 15 and 25 °C environments; selection was also acting against height in the 20 °C environment, although this selection gradient was not significant. Inbred lines that exhibited greater plastic flowering time plasticity had significantly lower fitness in the 20 °C environment and a marginally significant trend for lower fitness in the 25 °C environment (P = 0.0567), supporting the hypothesis that flowering time plasticity is costly in those environments. In addition, in the 15 °C environment, there was a nonsignificant trend for inbred lines with greater flowering time plasticity to have lower fitness. Although there was a cost to flowering time plasticity in two environments, we detected no evidence that mean flowering time was under selection in any of the experimental treatments.

Mutant lines

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Natural selection in the mutant lines favoured increased average leaf number across all environments, while marginally significant selection was acting to decrease plasticity in leaf number (Table 3a). These results suggest that selection to increase leaf number in the 25 °C environment was sufficiently strong to lead to selection for increased leaf number over all environments. In addition, selection to increase leaf number in the 25 °C treatment but not the other treatments would have the effect of decreasing plasticity in leaf number, as plants in that treatment produced fewer leaves than did plants in the 15 and 20 °C treatments (that is, the plastic response and the direction of selection are of opposite sign). The evolutionary response to this pattern of natural selection, however, would likely have been constrained because average leaf number across all environments and plasticity of leaf number exhibited a positive genetic correlation (Table 1), yet selection was acting in opposite directions for each of them. In addition, marginally significant selection was acting in opposite directions for plasticity in flowering time and plasticity in leaf number, two traits that were positively correlated with each other.

Table 3.  Selection analysis for trait means and plasticity traits in Westerman and Lawrence's experiments. (a) Mutant Lines, (b) Nonmutant lines.
TraitβSEP-value
  1. Significant terms (P < 0.05) are shown in bold and marginally significant terms (0.05 < P < 0.10) in italics. Error degrees of freedom were five for (a) and 14 for (b).

(a) Mutant lines
 Flowering time plasticity5.172.500.094
 Height plasticity−2.301.810.259
 Leaf number plasticity3.091.330.067
 Mean flowering time−0.290.170.136
 Mean height0.0150.040.721
 Mean leaf number1.610.450.016
(b) Nonmutant lines
 Flowering time plasticity8.033.550.039
 Height plasticity−3.132.050.149
 Leaf number plasticity2.172.400.381
 Mean flowering time−0.00980.160.951
 Mean height0.140.060.032
 Mean leaf number0.0070.500.989

Nonmutant lines

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

In contrast to the pattern seen in the mutant lines, natural selection in the nonmutant lines was acting in same direction on two positively correlated traits – selection acted against height at flowering, and against plasticity in flowering time (Table 3b). These results are entirely consistent with the pattern of selection observed within the individual experimental treatments: within the 15 and 25 °C treatments significant directional selection was acting against height, while within the 20 and 25 °C treatments selection was acting against inbred lines with greater flowering time plasticity. The pattern of selection on plasticity in flowering time in the nonmutant lines is in the opposite direction of the pattern of selection in mutant lines (cf. signs of flowering plasticity terms in Table 3a vs. b). As we observed in the within-environment analyses, selection acted on flowering time plasticity but not mean flowering time.

Constraints on the evolution of plasticity

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

Our reanalysis of W&L's data with current statistical methods for measuring natural selection and estimating costs of phenotypic plasticity allows a rigorous assessment of their work in light of the current context on the evolution of phenotypic plasticity. Several important results emerge from our reanalysis. First, in our reanalysis of W&L's data in a statistical model that accounts for multiple correlated traits, we still detected significant evidence for selection acting against plasticity of flowering time. Secondly, by analysing the mutant and nonmutant lines separately, our reanalysis showed that the negative relationship between flowering time plasticity and fitness observed by W&L is not an artifact of mutant lines having greater plasticity and lower overall fitness: the negative relationship between flowering time plasticity and fitness holds for the nonmutant lines. Thirdly, our analysis of the costs of flowering time plasticity suggested that there is a significant costs of plasticity in the 20 and 25 °C environments. These results illustrate the complementary nature of analyses of the net pattern of selection on phenotypic plasticity (over all environments) and the costs of phenotypic plasticity as expressed in individual environments. Finally, our analysis of the genetic correlations between trait means and plasticities identified several possible genetic constraints on the evolution of these traits.

These data clearly illustrate two potential constraints on the evolution of phenotypic plasticity: first, selection can act against plasticity directly (due to costs of plasticity) and secondly, genetic correlations between plasticity and other traits may constrain the response to selection. The simplest potential constraint on the evolution of plasticity is that selection can act against plasticity directly – as described by W&L originally and confirmed by our reanalysis and our analyses of costs of plasticity within the individual experimental treatments. Interestingly, these costs were most apparent in the high temperature treatments representing the upper end of A. thaliana's climatic range, suggesting a possible role of temperature stress in their expression. It is doubtful that the pattern of selection acting against flowering time plasticity is due to genetic correlations between plasticity in flowering time and other traits which W&L measured and reported. For instance, we detected no significant correlational selection gradients that would indicate that selection was acting against flowering time plasticity and in favour positively correlated traits (or acting against both flowering time plasticity and other negatively correlated traits). The possibility that selection is acting against plasticity in flowering time through a correlation with mean flowering time can be discounted because we included mean flowering time in our selection analyses and failed to detect any selection on it in both the within-environment analyses and in the analysis of the net pattern of selection. As such there is no evidence that strong selection on mean flowering time is creating negative selection on plasticity in flowering time. In addition, we detected no evidence for the negative genetic correlations which would be necessary to support this potential explanation: plasticity in flowering time is positively correlated with mean flowering time in the mutant lines, and not at all in the nonmutant lines.

The potential influence of genetic correlations acting as constraints on the evolution of plasticity can also be seen in the mutant lines – in multiple cases, the evolutionary response to selection would have been constrained because of genetic correlations between plasticity in one trait and plasticity in another trait, or because of genetic correlations between the mean value of a trait and its plasticity. The potential influence of this latter type of constraint has been emphasized recently by Pollard, Pigliucci, and Cruzan (e.g. Pollard et al., 2001; Pigliucci et al., 2003). The consequences of these correlations have been clearly described by Lynch & Walsh (1998), pp. 675–676): if selection favours an increase in the trait mean (e.g. mean leaf number across all environments), a correlated response to that selection will be increased phenotypic plasticity, in this example, of leaf number. Any subsequent deterioration in environmental conditions, because of the increased plasticity to environmental conditions, will often lead to populations with lower mean values of the focal trait (e.g. Simmonds, 1981; Lynch & Walsh, 1998).

Insight from mutant lines

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

The analysis of the mutant lines presents some evidence about the potential for random mutation to potentially change the genetic variances and co-variances governing the evolution of plasticity and quantitative traits (also see Camara & Pigliucci, 1999; Pigliucci & Schmitt, 1999). Although the use of mutagens is not without drawbacks (see Pigliucci, 2003 for a review), C.W. Lawrence's original papers (Lawrence, 1968a,b) provide some detail that suggests the mutant lines analysed by W&L were not unusually aberrant. First, C.W. Lawrence discarded lines in which crosses suggested a single gene of major effect on flowering time. Secondly, as originally noted, the flowering times of the mutagenized lines were well within the typical flowering times of ecotypes in the Laibach collection at the time. Given these caveats and mitigating factors, it is interesting that the mutant lines used by W&L showed dramatically different patterns of genetic correlations between means and plasticities than the nonmutant lines. Despite the low sample size in these analyses, we were able to detect several significant genetic correlations, perhaps due to the increased variance and because of the mutagenized lines. In like fashion, we also detected significant and marginally significant selection on trait means and plasticities in the mutant lines despite these small sample sizes. Interestingly, the signs of the significant and marginally significant selection gradients differed between the mutant and nonmutant lines in two of three cases. Although the observation that mutation creates new genetic variation in populations is a truism, we find it striking that mutation in these lines created novel patterns of genetic covariances between quantitative traits and plasticities.

Prescient features of Westerman and Lawrence

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

As has been noted before, (Pigliucci et al., 1995), the growing conditions used by W&L – especially the agar test tubes – were less than ideal approximations of natural growing conditions. Nevertheless, their study was conceptually sophisticated and remarkably prescient in several ways. First, W&L's analysis of the relationship between mean fruit number and plasticity and their classification scheme for developmental flexibility/stability essentially represents a very early approach to testing the adaptive plasticity hypothesis, 15 years before Via & Lande's (1985, 1987) pioneering papers, which in turn inspired the experimental tests of the adaptive plasticity hypothesis usually cited in the current literature (e.g. Dudley & Schmitt, 1996). Secondly, the analysis on line means performed by W&L to test the adaptive plasticity hypothesis can be thought of as an early use of genotypic selection analysis (Rausher, 1992; Stinchcombe et al., 2002), approximately 19 years before this technique was first performed in its current context (Rausher & Simms, 1989). Thirdly, W&L helped to pioneer the use of Arabidopsis mutants and ecotypes for addressing questions in ecological and evolutionary genetics approximately 25 years before the current renaissance (e.g. Pigliucci et al., 1995; Mauricio & Rausher, 1997). Their prescient approach produced a classic data set that yields novel insights of contemporary interest when explored with modern analytical methods, and more contributions than a simple correlation between fruit number and seed number.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References

We thank Eric von Wettberg and Robert J. Baker (University of Saskatchewan) for informative discussions of joint regression analysis, although any errors are entirely our own. Two anonymous reviewers and D. Fairbairn made numerous helpful comments on an earlier version of the manuscript. This work has been supported by NSF grants DEB 0079932, DEB 9976997, and DEB 0129018.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Westerman and Lawrence's methods
  6. Data
  7. Reanalysis
  8. Measures of plasticity
  9. Estimates of genetic correlations
  10. Estimates of costs of plasticity
  11. Estimates of selection on plasticity
  12. Results
  13. Genetic correlations between Trait Means and Plasticity
  14. Mutant lines
  15. Nonmutant lines
  16. Costs of plasticity
  17. Mutant lines
  18. Nonmutant lines
  19. Net patterns of selection on plasticity and trait means
  20. Mutant lines
  21. Nonmutant lines
  22. Discussion
  23. Constraints on the evolution of plasticity
  24. Insight from mutant lines
  25. Prescient features of Westerman and Lawrence
  26. Acknowledgments
  27. References