Quantitative genetic variation for thermal performance curves within and among natural populations of Drosophila serrata

Authors


Stephen F. Chenoweth, School of Biological Sciences, The University of Queensland, St Lucia, QLD 4072, Australia.
Tel.: +61 7 3365 2188; fax: +61 7 3365 1655; e-mail: s.chenoweth@uq.edu.au

Abstract

Thermal performance curves (TPCs) provide a powerful framework for studying the evolution of continuous reaction norms and for testing hypotheses of thermal adaptation. Although featured heavily in comparative studies, the framework has been comparatively underutilized for quantitative genetic tests of thermal adaptation. We assayed the distribution of genetic (co)variance for TPC (locomotor activity) within and among three natural populations of Drosophila serrata and performed replicated tests of two hypotheses of thermal adaptation – that ‘hotter is better’ and that a generalist-specialist trade-off underpins the evolution of thermal sensitivity. We detected significant genetic variance within, and divergence among, populations. The ‘hotter is better’ hypothesis was not supported as the genetic correlations between optimal temperature (Topt) and maximum performance (zmax) were consistently negative. A pattern of variation consistent with a generalist-specialist trade-off was detected within populations and divergence among populations indicated that performance curves were narrower and had higher optimal temperatures in the warmer, but less variable tropical population.

Introduction

Thermal adaptation has been an enduring focus for evolutionary biologists interested in the relationship between environmental heterogeneity and ecological specialization (Levins, 1968; Lynch & Gabriel, 1987; Gilchrist, 1995; Angilletta, 2009). Ambient temperature varies in both time and space, and for ectotherms, is recognized as a strong determinant of individual fitness (Kingsolver & Watt, 1983; Lynch & Gabriel, 1987) through its effects on physiological processes that in turn mediate locomotion (e.g. Bennett, 1980; Weinstein, 1998; Lyon et al., 2008), growth (e.g. Kingsolver, 2000; Yamahira et al., 2007), resource acquisition (e.g. Greenwald, 1974; Ayers & Shine, 1997) and survival (e.g. Domenici & Blake, 1993; Ahnesjo & Forsman, 2006).

Thermal performance curves (TPCs) provide a powerful framework for testing the hypotheses of thermal adaptation and, in particular, illuminating evolutionary constraints on the evolution of thermal sensitivity (Levins, 1968; Huey & Kingsolver, 1989). A TPC is a continuous nonlinear reaction norm describing the relationship between performance and temperature (Huey & Stevenson, 1979; Izem & Kingsolver, 2005). TPCs have a characteristic shape; performance gradually increases with temperature, reaches a maximum and then falls sharply (Huey & Stevenson, 1979; Huey & Kingsolver, 1989). Because TPCs exhibit a characteristic shape, their variation among species, populations or genotypes can be described in terms of three different modes of variation, each deviating from a common template curve in a particular direction. These modes of variation are overall height (vertical shift), position of the thermal optimum (Topt) (horizontal shift) and the thermal breadth (width shift) (Huey & Slatkin, 1976; Izem & Kingsolver, 2005). Height variation describes variance in performance that is independent of temperature and involves vertical shifts of the whole curve with no change in shape. The remaining two components of variation describe temperature-dependent variation; Topt variation captures variation at the peak of the curve, where performance is maximal, while a combination of variation in the breadth of the curve and the performance level at the optimal temperature, zmax, characterizes a generalist-specialist mode of variation, representing a width shift of a TPC.

The ‘hotter is better’ hypothesis of thermal adaptation posits that organismal performance at the optimal temperature scales positively with optimal temperature (Hamillton, 1973; Bennett, 1987; Huey & Kingsolver, 1989; Savage et al., 2004) and seeks to explain why cold-adapted species or populations might perform more poorly than their warm-adapted counterparts and also remains a leading hypothesis for the evolution of thermoregulation (Angilletta, 2009). The hypothesis is based on thermodynamic constraints stemming from metabolic theory showing that rates of chemical reactions, and therefore maximal performance, scale with temperature (Gillooly et al., 2001, 2002; Savage et al., 2004). Empirical observations have indeed shown that many biochemical reactions are more efficient at higher temperatures (Hochachka & Somero, 2002). Within a TPC framework, the hotter is better hypothesis predicts a positive correlation between maximum performance level (zmax) and Topt (Huey & Kingsolver, 1989). Comparative studies of population growth rates in insects (Frazier et al., 2006), trees (Rehfeldt et al., 2002) and locomotion in lizards (Bauwens et al., 1995) have found a positive correlation between Topt and zmax which is consistent with hotter is better. While such correlations are suggestive, they fall short of a conclusive demonstration as comparative data are always subject to alternative explanations.

If hotter is better can explain these macroevolutionary correlations, then there must exist axes of genetic variation at the within-population level that are capable of producing such a response to selection. Therefore, a positive genetic correlation between Topt and zmax may be expected within populations of a single species (Angilletta et al., 2010). To date, however, the very few microevolutionary tests of hotter is better have provided limited support. Gilchrist’s (1996) pioneering quantitative genetic study of TPC shape in aphid wasps failed to detect a positive genetic correlation between Topt and zmax. Similarly, no genetic correlation between these two components was found for growth rate of TPCs in caterpillars (Izem & Kingsolver, 2005). However, a recent quantitative genetic study of natural isolates of a bacteriophage did detect a positive correlation between Topt and overall performance level (Knies et al., 2009). Thus, the generality of the hotter is better hypothesis is unresolved (Angilletta et al., 2010).

A second type of evolutionary constraint that may affect thermal adaptation are trade-offs. Building on Levins’ (1968) principle of allocation, Huey & Slatkin (1976) suggested that the evolution of thermal specialization would involve a trade-off between performance level and the range of temperatures across which an organism can perform; the higher the overall performance level the narrower the performance range. This idea is an example of the classic generalist-specialist trade-off that is the basis of much life history evolutionary theory (Futuyma & Moreno, 1988; Stearns et al., 1991) and underpins different optimality models of thermal adaptation (Lynch & Gabriel, 1987; Gilchrist, 1995). The potential contribution from trade-offs to thermal adaptation can also be tested using quantitative genetic analysis of TPC variation. A generalist-specialist trade-off predicts a negative genetic correlation between zmax and thermal performance breadth (Huey & Hertz, 1984), which is supported by studies of male wasps (Gilchrist, 1996). Subsequent genetic studies of TPC variation in growth rate for medaka fish (Yamahira et al., 2007), caterpillars (Izem & Kingsolver, 2005) and a bateriophage (Knies et al., 2006), however, have provided mixed support.

While quantitative genetic studies to date have illuminated genetic constraints on thermal adaptation, patterns remain to a large extent inconsistent across species and clearly further tests are needed (Angilletta, 2009). Thus far, quantitative genetic tests of hypotheses of thermal adaptation have been performed on single populations (but see Yamahira et al., 2007), which may lack generality within a species. In this study, we set out to address this issue by performing a quantitative genetic analysis of TPC variation within and among multiple natural populations.

Drosophila species are ideal models to explore the evolutionary genetics of TPCs; they are amenable to quantitative genetic analysis, are often widespread and inhabit varied thermal environments. Drosophila serrata are a member of the montium subgroup endemic to the Australasian region (Lemeunier et al., 1986; Kellett et al., 2005). The geographic range for this species spans seasonally stable tropical thermal environments (Papua New Guinea, 6.31°S, 143.95°E) to more seasonally variable temperate environments (Wollongong, 34.42°S, 150.89°E eastern Australia) (Jenkins & Hoffmann, 1999, 2001). Over their range, D. serrata thus experience a wide range of thermal regimes that may lead to spatially divergent selection for thermal adaptation. For example, previous studies have demonstrated latitudinal variation in both chill-coma recovery (Hallas et al., 2002) and cold resistance (Jenkins & Hoffmann, 1999).

In this study, we perform a quantitative genetic analysis of locomotor activity TPCs within and among three natural populations of D. serrata. We use these data to perform a replicated test of the ‘hotter is better’ hypothesis of thermal adaptation by testing for the predicted positive genetic correlation between optimal temperature (Topt) and maximal performance (zmax) for males and females within each population. Then by isolating a component of TPC shape variation corresponding to genetically correlated changes in thermal breadth and performance, we tested for patterns consistent with a generalist-specialist trade-off; an assumption of many optimality models for the evolution of thermal specialization (Lynch & Gabriel, 1987; Huey & Kingsolver, 1993; Gilchrist, 1995). Our results are interpreted in the light of the available microevolutionary tests of thermal adaptation. Locomotor performance is a thermally dependent trait in ectotherms that often features in studies of thermal adaptation (Huey & Kingsolver, 1993; Gilchrist, 1996; Wilson et al., 2001; Izem & Kingsolver, 2005; Yamahira et al., 2007). We used TPCs of locomotor activity as our measured trait of performance as this trait is likely associated with fitness in Drosophila because of its links with reproductive success, dispersal, predator avoidance and foraging (Gilchrist, 1996; Roberts et al., 2003). Adult locomotor activity covaries with fitness in Drosophila melanogaster under laboratory conditions (Long & Rice, 2007).

Materials and methods

Field collections and husbandry

Drosophila serrata were collected from three natural populations along the east coast of Australia: Cooktown (15.45°S, 145.19°E), Yeppoon (23.14°S, 150.75°E) and Brisbane (27.39°S, 153.13°E) that vary in mean annual temperature (Commonwealth of Australia, 2010, Frentiu & Chenoweth, 2010) span 11.94° of latitude and represents 28% of the natural range of the species (Jenkins & Hoffmann, 2001). We founded mass-bred populations for each of the three collections using an equal number of isofemale lines founded from wild-caught inseminated females (eight lines, 12 males and 12 females per line). Flies were maintained in the laboratory for a minimum of three generations on a yeast–sucrose–agar medium as isofemale lines before establishing single mass-bred populations at 25 ± 0.5 °C in a 12:12-h light cycle to remove common environmental and maternal effects. Sixteen full-sib families were established for each of the three populations using progeny from the respective mass-bred populations. Single mating pairs were transferred across three replicate vials every 32 h for egg laying. Phenotyping was conducted on 5- to 7-day-old virgin males and females. From each of the three vials, four males and females were sexed and held singly in vials prior to phenotyping, giving a total of 12 males and 12 females per family.

Thermal dependence assay

Individual thermal dependence curves were estimated by measuring locomotor activity for 20 min at seven temperatures in the order of 30, 25, 33, 36, 35, 37 and 38 ± 0.5 °C, after a brief exposure of 5 min. Temperature was measured in this mixed order to account for daily activity variation and had small as possible changes to prevent overly sudden changes in temperature. We measured the highest temperature last to prevent any detrimental effects interfering with any subsequent measurements (Knoppien et al., 2000). Between each temperature treatment, flies were rested at 25 ± 0.5 °C for 40 min to prevent stress and acclimation to the temperature treatment. Individual flies were exposed to all seven temperatures in 1 day, and all measurements were conducted in one single temperature cabinet. Although it is commonplace to use a fixed testing order when estimating TPCs (Gilchrist, 1996; Wilson, 2001; Wilson et al., 2001; Angilletta et al., 2002; Ben-Ezra et al., 2008), we conducted a pilot experiment to assess the consequences of using a fixed rather than random testing order. The average vector correlation among TPCs between six randomly selected testing orders was high for both sexes (females: = 0.9216; males: = 0.9346). Given this high repeatability, we favoured higher throughput with a fixed testing order over the use of random testing orders, which would have dramatically reduced the sample size of the experiment owing to a limited logistical capacity.

Thermal dependence curve assays were conducted in 12 blocks performed over six consecutive days. Two blocks were processed per day, and each contained one male and one female per family from each of the three populations, resulting in complete data for 96 individuals. In total, 12 individuals × 2 sexes × 16 families × 3 populations were measured; data were collected for a total of 1152 individual flies.

Locomotor activity was measured using Drosophila Activity Monitors (DAM; TriKinetics, Waltham, MA, USA). A DAM comprises 32 × 5 mm holding tubes that are bisected by an infrared beam. When a fly crosses the beam, a connected computer records the number of intersections. We measured activity between 4.15 am and 1.30 pm, to correspond with the light : dark cycle of our flies (lights on between 4 am and 4 pm) and, more importantly, to avoid the major activity peaks that occur with lights on and off in D. melanogaster (Rieger et al., 2007), and in a preliminary study we had conducted for D. serrata. Flies were placed in the 5-mm measurement tubes the afternoon prior to testing to allow the flies to adjust to the new physical environment. Each 5-mm vial contained 1.5 cm of a 3-day-old agar–sucrose medium that was capped at one end and stoppered with 0.2 mm of foam on the opposing end.

Our specific measure of locomotor activity differs from some previous measures of locomotor activity that have been used as proxies for thermal performance in flies such as walking speed which usually involves measurement after reaction to a stimulus (Crill et al., 1996; Gilchrist et al., 1997). Although all activity metrics will include both physiological and motivational aspects of physical movement, it could be argued that our metric includes more motivational variation (i.e. the willingness to move) than other measures (e.g. sprint speed) used in previous studies (Gilchrist et al., 1997; Lachenicht et al., 2010). However, we note that in D. melanogaster, basal activity is strongly positively correlated with walking speed (Burnet et al., 1988).

Statistical analysis

We used a multivariate linear mixed effects model to test for genetic variance within populations and divergence among them. The counts for locomotor activity were square root transformed to improve normality (Quinn & Keough, 2002). We treated family within population and vial within family as random effects and block and population as fixed effects; males and females were analysed separately as early analyses indicated that locomotor activity was sexually dimorphic (e.g. see Fig. 1b, d & f). Population was treated as a fixed effect as our primary interest was in population mean differences rather than estimating among-population genetic variance, which would have required many more populations to be sampled. Block was treated as fixed because blocks are temporal and refer to twelve successive runs of the experiment conducted over the 6-day test period rather than a random sample of times (Quinn & Keough, 2002). Significance tests for fixed effects were performed using F tests, and likelihood ratio tests were used for random effects. Models were fitted using the MIXED procedure of SAS using Restricted Maximum Likelihood (ver. 9.2; SAS Institute, Cary, NC, USA). The corresponding linear model was:

image(1)

where l is a vector of locomotor activity scores at the seven temperatures, pop is the vector for population and μ is a vector of means for each of the seven tested temperatures. The variance–covariance matrix corresponding to the family(pop) term provides an estimate of the average within-population (broad-sense) genetic variance–covariance matrix. It was not possible to fit an unstructured covariance matrix (i.e. SAS PROC MIXED, type = UN) owing to the large number of parameters that need to be estimated in such a model. We therefore fitted a reduced rank (Kirkpatrick & Meyer, 2004), (factor analytic) covariance structure for the family(pop) and vial(family(pop)) terms (SAS Proc MIXED type = FA0(n) where n is the number of factors considered in the model).

Figure 1.

 (a) Representation of a typical thermal performance curve showing the width, height, optimum temperature (Topt) and maximum performance (zmax) components, and (b–f), the population and sex mean (±2 × SE) locomotor activity scores taken at seven test temperatures for Drosophila serrata collected from three natural populations. Males are in grey, females in black, Cooktown is represented by dashed lines, Yeppoon is represented by dot-dash lines and Brisbane is represented by solid lines. The square root of raw locomotor activity is plotted. Units are number of infrared beam dissections that occurred during a 20-min testing period for each temperature.

While the analysis above was useful for providing a test for significant genetic variation within and divergence among the three populations, our primary goal was to understand how genetic variation was partitioned across the TPCs components. To achieve this, we used Izem & Kingsolver’s (2005) functional analysis, Template Mode of Variation (TMV). TMV analyses genetic variation in TPCs by estimating variation in three modes, vertical and horizontal shift and the width (generalist-specialist) as deviations from a common ‘template’ curve. Vertical shift captures changes in the overall curve height and is independent of temperature. Horizontal shift refers to a decrease or increase in Topt because of a shift in the TPC to the left or right. Finally, a change in the width incorporates the generalist-specialist trade-off by constraining the area under the curve to remain constant, resulting in wider curve having a lower zmax. These three simultaneous modes of variation were modelled by the following three-parameter shape-invariant model (Lawton et al., 1972):

image(2)

where zi (t) is the continuous function of temperature, t, that represents the locomotor activity of family i. The common shape template, consistent for all families, is represented by z. Each family’s curve is described by the three parameters (overall height; hi, Topt; mi and width; wi) as a deviation from the common curve z, a cubic polynomial, along the modes of interest. Height hi parameterizes the vertical shift mode of variation, location of maximum temperature mi parameterizes the horizontal shift mode of variation and width wi parameterizes the generalist-specialist mode of variation.

We estimated the thermal dependence curves and the template curve from the data using the semi-parametric method recommended by Izem & Kingsolver (2005) that is currently implemented in a matlab package. This produced for all families three modes of variation, m, h and w for each group. A fourth mode zmax (the performance at Topt) was calculated separately using the following equation implemented in the TMV package in matlab ver. 7.6.0.324 for Windows (The Mathworks Inc., Natick, MA, USA):

image(3)

where intercept is the intercept for the template curve which is in our case a cubic polynomial, wi is the width mode of variation for the ith family and hi is the height mode of variation for the ith family. A ratio sum of squares was calculated for the m, h and w modes of variation to estimate the amount of among-family variation for each. For the analysis of within-population variation, we analysed each sex and population separately while maintaining a cubic polynomial as the common template function. We estimated the broad-sense genetic correlation between modes using Pearson’s product moment correlations on family level parameter estimates. All TMV analyses we performed on residual activity scores after correcting for block effects using the MIXED procedure in SAS.

We also attempted to compare TPC component divergence between sexes and populations. As the present implementation of TMV fitting code is capable only of a one-way analysis, we used TMV to estimate m, h and w values for each family in each population for males and females, and then tested for differences between the populations and sexes using the two-way fixed effects model:

image(4)

where μ is the population grand mean. Family m, h, and w values were estimated by conducting TMV on 96 families (16 full-sib families × 3 populations × 2 sexes), where we included both sexes so the same template curve was assumed, enabling the modes of variation to be compared simultaneously among sexes and populations.

Results

Genetic variation within populations

Mean locomotor activity across the seven temperatures resembled the typical shape of a thermal dependence curve. All curves increased between 25 and 36 °C, reached a peak around 36 and 37 °C and then declined rapidly (Fig. 1b–f). Males generally were more active than females (Fig. 1b, d & f). The mixed effects models indicated significant genetic variance within populations for both sexes (Table 1). Factor analytic modelling showed that only one factor, or genetic principal component, was required to explain the average within-population genetic variance–covariance matrices for both sexes. We note that the effect of block was also significant, because blocks are effectively temporal runs here, suggesting some day-to-day variation in the thermal dependence of locomotor activity. This is not unexpected for behavioural traits.

Table 1.   Results for multivariate mixed effects linear models testing for genetic differences in thermal performance within and among three natural populations of Drosophila serrata. F-tests are provided for tests of fixed effects and likelihood ratio tests for random effects.
EffectMalesFemales
Test statisticP-valueTest statisticP-value
  1. Degrees of freedom for LRTs in the family term correspond to a reduced-rank model where a single factor (type = FA0(1) in SAS PROC MIXED) was fitted. For the vial term, it was possible to fit an unstructured variance–covariance matrix that has 28 parameters. Error degrees of freedom for fixed effects were estimated using Sattherwaite’s approximation (ddfm = SAT in SAS PROC MIXED).

BlockF66,383 = 9.82< 0.0001F66,384 = 9.19< 0.0001
PopulationF12,95.3 = 3.90< 0.0001F12,88.8 = 4.80< 0.0001
Block × PopulationF132,472 = 0.960.604F132,402 = 1.110.2143
Family(Population)inline image = 17.100.0168inline image = 14.500.0429
Vial(Population(Family))inline image = 32.900.2395inline image = 33.600.2144

We performed separate TMV analyses for each sex and population combination to partition the within-population genetic variance into the three modes of variation; Topt, height and generalist-specialist. Consistently, more variation was explained in females than in males. Most genetic variance explained by the TMV model occurred in the generalist-specialist component in males and females with averages of 13% and 18%, respectively (Table 2). There was consistently very little among-family variance in the height component for either sex, which describes performance across all temperatures (Table 2).

Table 2.   Variation in the three components of TPC variation estimated via TMV for each sex–population combination. Values are ratio sums of squares (RSS) expressed as a percentage of total variation. For all analyses, the common template function was a cubic polynomial.
Mode of variationCooktown RSS%Yeppoon RSS%Brisbane RSS%Average RSS%
  1. TMV, template mode of variation; TPC, thermal performance curves.

Males
 Vertical shift (overall height)0.450.320.460.41
 Horizontal shift (Topt)5.776.171.554.49
 Generalist-specialist (width)7.6421.3310.2013.05
 Total explained by model13.8627.8212.2117.95
 Error86.1572.1887.8082.04
Females
 Vertical shift (overall height)0.260.360.440.38
 Horizontal shift (Topt)4.7610.845.409.03
 Generalist-specialist (width)27.5922.6711.4718.94
 Total explained by model32.6133.8717.3128.35
 Error67.4066.1482.6871.65

The hotter is better hypothesis of thermal adaptation predicts a positive genetic correlation between Topt and zmax. We calculated the broad-sense genetic correlation between estimated family level parameters for males and females within each of the populations. For males, there was a positive but nonsignificant correlation in the Brisbane population (= 0.199, = 16, P = 0.46) whereas the relationship was significant, but negative, in both Yeppoon (= −0.896, = 16, P ≤ 0.0001) and Cooktown (= −0.642, = 16, P = 0.007). For females, the genetic correlation between Topt and zmax was significant and negative in all three populations (Brisbane: = −0.859, = 16, P ≤ 0.0001; Yeppoon: = −0.923, = 16, P ≤ 0.0001; Cooktown; = −0.690, = 16, P = 0.003).

Genetic divergence between populations

Significant genetic divergence among populations was indicated in the manovas for both sexes (Table 1, Fig. 1b, d & f). Follow-up pairwise comparisons (Table S1) indicated that for males, population divergence was chiefly driven by differences between Cooktown and the two southern populations. Brisbane and Yeppoon did not differ for any temperature whereas Cooktown males had significantly lower activity for all tested temperatures between 30 and 36 °C (Fig. 1c). For females, population divergence also occurred in the central temperatures, but with a slightly different pattern (Fig. 1e). As with males, activity was lower for Cooktown at 33 and 36 °C than for Brisbane and Yeppoon, which did not differ from each other (Fig. 1c). Yeppoon females differed from both Brisbane and Cooktown at 30 and 35 °C, having higher activity than Cooktown and Brisbane at these temperatures (Fig. 1e).

To explore population divergence in TPC shape components, we performed a second round of TMV analyses with populations and sexes pooled. Univariate anova conducted on the family mean TMV component values indicated significant differences among populations and sexes in TPC shape. The generalist-specialist mode significantly differed among populations (population effect: F2,90 = 7.46, P = 0.0093) and between the sexes (sex effect: F1,90 = 7.07, P = 0.001). Males had wider curves than females. Cooktown had narrower curves, followed by Yeppoon and Brisbane, which did not differ from each other (Fig. 2a). The family mean Topt differed significantly between sexes (F1,90 = 63.81, P < 0.0001) and populations (F2,90 = 14.46, P < 0.0001) but there was no significant interaction. Females had higher Topt values than males, and Topt tended to decrease from north to south. Females had higher optimal temperatures than males. Cooktown flies had higher mean Topt values than Yeppoon and Brisbane flies, which did not differ from each other (Fig. 2b). The family mean height differed among populations in a sex-dependent manner (sex × population interaction: F2,90 = 2.46, P < 0.0001). A TMV analysis was also conducted for each sex separately but with populations again pooled to account for a potential violation of the assumption of a common template curve, no major differences in the patterns were found (Fig. S1).

Figure 2.

 The mean for each mode variation (±2 × SE) for each population and sex. Females are represented by closed shapes, and males are represented by open shapes with dotted lines; (a) generalist-specialist or width mode (w), (b) Topt or horizontal shift (m) and (c) zmax (maximum performance at the Topt).

Discussion

Although well supported by comparative studies, the hotter is better hypothesis of thermal adaptation has been poorly supported by quantitative genetic studies (Angilletta et al., 2010). However, there are very few quantitative genetic tests available and these typically have involved estimates from a single population, which may lack generality. We have addressed this issue by performing a replicated test in three natural populations of D. serrata. A key prediction of the ‘hotter is better’ hypothesis of thermal adaptation is a positive correlation between maximum performance level (zmax) and temperature of maximal performance (Topt) (Huey & Kingsolver, 1989). Within an evolutionary genetic context, genotypes should therefore have similar effects on Topt and zmax leading to positive genetic covariance between the two. Considering tests in both sexes, five of the six genetic correlations between Topt and zmax were significant. However, in all cases, these correlations were negative, which suggests that selection for increased optimal temperature will lead to a correlated reduction in performance. These results provide no support for the hotter is better hypothesis.

While our results were consistent with other quantitative genetic studies in failing to support the hotter is better hypothesis (Gilchrist, 1996; Izem & Kingsolver, 2005), they are unusual in that the genetic correlations between Topt and zmax were consistently strong and negative. One explanation for these negative genetic correlations is that they may have been shaped by past selection. If the genetic correlation between Topt and zmax, because of pleiotropic mutations, is in fact positive, but historical selection for increased performance at Topt has fixed alleles with positive effects and purged alleles with negative effects, a negative genetic correlation could potentially result (Falconer & Mackay, 1996). As our design permitted among, as well as within-population comparisons, we can test this explanation. If the pattern of genetic covariance between Topt and zmax within populations reflects past selection, we would not expect to see the same pattern across populations from differing thermal environments. Instead, we would expect the warmest population (in our case Cooktown) to have both the highest mean Topt and also zmax. This was not the case, Cooktown males did indeed have the highest mean Topt but had the lowest mean zmax value (Fig. 2c). A similar pattern was evident in females, although results were not statistically significant. Neither this pattern of genetic divergence nor genetic covariance within populations match the hotter is better hypothesis. With the data at hand, it therefore seems unlikely that historical selection is purely responsible for the negative genetic correlations we have observed between Topt and zmax. A more powerful test of this explanation could be performed using a mutation accumulation experiment where the genetic correlation because of purely mutational effects could be estimated.

It remains difficult to understand why results from comparative studies (e.g. Eppley, 1972; Wilson, 2001; Frazier et al., 2006) tend to support the hotter is better hypothesis whereas most quantitative genetic studies do not. The consistency of our results across three natural populations suggests reasons beyond a large variance of parameter estimates drawn from a single population. One possible source of uncontrolled variance in interspecific comparative datasets is that Topt and zmax are seldom estimated under common garden conditions free from environmental effects. It is also noteworthy that support for hotter is better, including some from quantitative genetics (Knies et al., 2009), is strongest in studies of population growth rate (e.g. Frazier et al., 2006). To date, most quantitative genetic studies have analysed other performance proxies, which may not be as reliable as population growth rate as indicators of fitness (Angilletta et al., 2010). Quantitative genetic analyses of the thermal sensitivity of population growth rates remain badly needed.

A striking aspect of the TMV analyses was that genetic variation in the height component, which describes overall performance, independent of temperature, was consistently limited in all populations. Minimal height variation has been found in some studies (Izem & Kingsolver, 2005; Knies et al., 2006) but not others, (Gilchrist, 1996; reanalysed by Izem, 2004; Yamahira et al., 2007). There may be persistent positive directional selection for increased overall performance, which will tend to deplete genetic variance (Izem & Kingsolver, 2005). However, if locomotor activity in D. serrata is subject to sexually antagonistic selection as is the case in D. melanogaster (Long & Rice, 2007), we would not necessarily expect genetic variance to be depleted as readily, because alleles conferring higher activity and favoured in males, would have lower fitness when expressed in females. Thus, a depletion of genetic variance hypothesis assumes either sexually concordant selection under natural conditions or an independent genetic basis to locomotor activity in males and females. The latter is not supported in D. melanogaster where the intersexual genetic correlation of locomotor activity is positive and cannot be distinguished from unity (Long & Rice, 2007). Further work will be required to understand the intersexual genetic architecture of locomotor activity in D. serrata, and sexual differences remain a potentially important but overlooked aspect of genetic variation in thermal sensitivity (Huey & Pianka, 2007; Lailvaux, 2007).

The second aim of our study was to explore to what extent the standing variation for TPCs reflected a possible generalist-specialist trade-off. This type of trade-off is central to optimality models of thermal adaptation (Lynch & Gabriel, 1987; Huey & Kingsolver, 1993; Lynch & Lande, 1993; Gilchrist, 1995) but its existence has seldom been validated empirically. In both sexes, the width component, which describes a generalist-specialist axis of variation, comprised the greatest proportion of genetic variance explained by the TMV model, which is consistent with growth rate of TPCs for caterpillars (Izem & Kingsolver, 2005). Building on the idea that spatial and temporal heterogeneity in an environmental stress factor may select for varied degrees of specialization (Levins, 1968), Lynch & Gabriel (1987) developed an optimality model for the evolution of environmental tolerance. Within the context of temperature heterogeneity, this model predicts that environments with lower within-generation temperature fluctuations should favour thermal specialization. That is, evolve narrower but taller TPCs whereas greater variability favours generalists with broader but shorter TPCs. The pattern of among-population divergence in D. serrata also suggested the involvement of a generalist-specialist trade-off. In males, the significant increase in temperature of maximal performance (Topt) in the Cooktown population which is on average warmer, but thermally less variable than the two southern populations (see Table S2 for climate data for each population), was accompanied by a decrease in the width TMV component (Fig. 2). Note that the reduction in activity between 30  and 36 °C in this population (Fig. 2c) is not necessarily inconsistent with this interpretation because of the way that the TMV model partitions variation. The overall reduction in zmax in Cooktown is because of contribution from the vertical shift component, which is statistically independent of any change in curve shape.

Although the pattern of genetic divergence we detected among populations is suggestive of local adaptation, divergence per se is insufficient to demonstrate adaptation unless genetic drift can be excluded as a cause of differentiation (Spitze, 1993; Whitlock, 2008). Although we did not test for it directly, genetic divergence at neutral loci is very weak across the range sampled here (average FST = 0.026, Chenoweth & Blows, 2008), suggesting that genetic drift is unlikely the major factor driving TPC divergence among these populations. We acknowledge that a greater number of populations than we were able to sample here are required to permit a robust quantitative genetic test of adaptive differentiation such as a QSTFST comparison (O’Hara & Merila, 2005) and more powerful comparative analyses of among-population variation and environmental conditions (Hoffmann et al., 2003).

While the TPCs measured in our study are likely to mainly reflect basal activity variation, they may also vary because of other aspects of fly physiology that could affect performance. For example, in D. melanogaster, it has been noted that activity may covary with heat resistance (Kjaersgaard et al., 2010). Although, D. serrata does exhibit variation in heat resistance (Berrigan, 2000; Hercus et al., 2000), we do not yet know whether it covaries with activity. Also, Drosophila flies, including D. serrata (Berrigan & Hoffmann, 1998), exhibit a hardening response when exposed acutely to a high but nonlethal temperature (Hoffmann et al., 2003) that dramatically increases their thermotolerance. While no study to our knowledge has investigated the relationship between hardening and activity, it remains possible that in our experiment, a hardening response was triggered following exposure to one of the higher test temperatures. We are unable to determine whether this has occurred, as the threshold temperature and required exposure times for inducing the response in D. serrata are not clear. Moreover, the time generally allocated to induce hardening in experiments is longer than the 20-min test period we have used here (Hoffmann et al., 2003). It will be interesting to determine how hardening affects TPC shape variation in future targeted experiments.

In conclusion, we found significant standing variation and population divergence in the TPCs for D. serrata. At least within the context of broad-sense genetic estimates as we have obtained here, thermal adaptation appears constrained by a lack of genetic variance in some components (e.g. height) and by the genetic covariance structure between others (e.g. Topt and zmax). In contrast to the remarkable consistency of the general form of TPC across species and trait types, it appears that the distribution of genetic variance within and among populations in these types of continuous reaction norms is highly variable. More quantitative genetic studies will be needed to determine to what extent consistencies in TPC genetic architecture emerge. The estimation of mutational contributions to TPC covariance structures remains an important, but as yet unstudied aspect of the evolution of thermal sensitivity in eukaryotes.

Acknowledgments

We thank R. Huey for stimulating comments and suggestions that improved an earlier version of this manuscript. We also thank C. Condon, T. Gosden and members of the Chenoweth lab for useful discussion. This research was funded with the support of an Australian Postgraduate Award to CL and an Australian Research Fellowship to SFC. Additional funding was provided by The University of Queensland.

Ancillary