Pleiotropic effects of environment-specific adaptation in Arabidopsis thaliana

Authors


Author for correspondence:
P. X. Kover
Tel:+44 (0)1612751550
Email: kover@manchester.ac.uk

Summary

  • • Local adaptation may be important for the preservation of genetic diversity and the promotion of speciation. However, local adaptation may also constrain establishment in different environments. The consequences of local adaptation depend strongly on the pleiotropic effects of the genes involved in adaptation.
  • • Here, we investigated the pleiotropic effects of the genetic response to selection in outbred lines of Arabidopsis artificially selected to flower earlier under both winter- and spring-annual simulated conditions. The consequences of adaptation were evaluated by reciprocally transplanting selected and control lines between the two conditions.
  • • Selected lines always flower earlier than their controls, independent of growing conditions. However, selected lines, growing in the same condition in which they were selected, flower earlier than plants selected in the alternative environment. Plants selected to flower earlier in spring produce more fruits than controls when growing in the spring, and less fruits when growing in the winter; indicating that local adaptation has negative pleiotropic effects in another environment.
  • • Our results indicate that local adaptation can arise even when selection targets the same trait in the same direction. Furthermore, it suggests that adaptation under the two different environments can generate fitness trade-offs that can maintain genetic variation for flowering time.

Introduction

All species are confronted with a wide variation in environmental conditions. In plants, the transition to the reproductive phase (flowering time) is under complex genetic control, and responds to a number of environmental cues, including photoperiod, temperature, water availability and shading (Simpson & Dean, 2002; Putterill et al., 2004; Cockram et al., 2007). The timing of reproduction is critical to most plant species, and genotypes that are adapted to local conditions may have an advantage by flowering at a time that maximizes fitness. Accordingly, reciprocal transplant experiments often provide evidence for local adaptation in plants (Clausen et al., 1940; Schluter, 2000; Lexer et al., 2003; Hall & Willis, 2006). However, a fixed phenotype narrowly adapted to local conditions can become costly if the environmental conditions change (for example, in unstable or unseasonable environments, or as a result of human alteration of the environment), or if the genotype finds itself in a new environment through seed dispersal (Van Tienderen, 1997).

For plants to be able to succeed in a range of environmental conditions, evolution must yield environment-specific phenology (Via & Lande, 1985), and a weak correlation in flowering time across environments. At the molecular level, this means that, if a single gene controls flowering in multiple environments, the evolutionary responses to selection occurring in one environment will cause a correlated response in another environment, even if the response is maladaptive. By contrast, if different genes control flowering time in different environments, genotypes can simultaneously adapt to multiple environments, allowing the evolution of generalists (Kassen, 2002). In addition, if flowering time is determined by different genes, the expected erosion in additive genetic variance, as a result of adaptation in one environment, will not affect the future evolutionary potential of an organism's response to climate change (Vergeer et al., 2004).

Environment-specific selection may not only cause local adaptation, but may also play an important role in the maintenance of genetic diversity and in the promotion of speciation (Schluter, 2000). If different environments favour different alleles at the same locus, a trade-off in performance will occur that will accelerate divergence between populations and maintain genetic diversity (Gardner & Latta, 2006). However, if a different set of genes is under selection in different environments, it is possible, through recombination, to create a generalist that can perform well across environments. In the latter case, environment-specific selection will not maintain genetic diversity or lead to speciation. Thus, the importance of environment-specific selection depends on the genetic mechanism through which adaptation occurs, and this is not well understood (Orr, 2005; Gardner & Latta, 2006). Although evolutionary models of adaptation usually assume that adaptive mutations carry costly (negative) pleiotropic effects (Fisher, 1958; Futuyma & Moreno, 1988; Fry, 2003), empirical evidence for negative pleiotropic effects is limited, particularly among plants (but see Weinig et al., 2003).

To investigate the genetic basis of local adaptation and its effects, current studies have focused on the identification of the targets of selection under natural conditions, and on the use of quantitative trait loci (QTL) analysis to determine the QTLs underlying the traits being selected (for example, Fishman et al., 2002; Lexer et al., 2003; Verhoeven et al., 2008). These approaches assume that the current targets of selection are the same as those that generated the local adaptation pattern. A more direct approach would be to compare past and present populations which, although not possible in natural conditions, can be pursued using experimental evolution (Rainey et al., 2000; Ostrowski et al., 2005). A similar approach, which is used here, is to specifically select for a trait thought to be adaptive under different environmental conditions, and investigate the consequences of the adaptive response in alternative environments. This latter approach may not provide information about the dynamics of natural populations, but does provide a clearer understanding of the consequences of environment-specific selection in a specific trait.

Here, we investigate whether selection for early flowering under different environmental conditions can lead to local adaptation in an experimental population of Arabidopsis thaliana. This is an ideal system in which to investigate the consequences of adaptation in flowering time because extensive variation in flowering time is observed across natural populations (Koornneef et al., 2004; Lempe et al., 2005), and A. thaliana has been the primary model used to unveil the genetic basis of flowering time (Putterill et al., 2004; Roux et al., 2006). Furthermore, A. thaliana is thought to be locally adapted (Stinchcombe et al., 2004; but see Callahan & Pigliucci, 2002), and is hypothesized to have been under recent selection for earlier flowering, as it has spread to milder climates and urban environments (Johanson et al., 2000; Le Corre et al., 2002; Hagenblad et al., 2004). However, phylogenetic studies also show weak isolation by distance signals and a star phylogeny, suggesting that contemporary movement must also be common (Sharbel et al., 2000; Jorgensen & Mauricio, 2004; Beck et al., 2008).

In a previous study, we have shown that an outbred population of A. thaliana can rapidly evolve earlier flowering [two standard deviations (SDs) in three generations] under two separate ecologically relevant growing conditions: winter- and spring-annual treatments (Scarcelli & Kover, 2009). Here, we report the results of a reciprocal transplant experiment between these two growing conditions after control and selected lines had been bred for five generations in each of the environments. Using this approach, we address the following questions. First, do plants selected to flower early in one environment also flower early when grown in a different environment? These results should help to determine whether pleiotropic constraints are acting on plants when they move between environments. Second, does selection in one environment affect heritability/evolvabilty in another environment? This will help to determine whether adaptation to one environment constrains adaptation to new environments. Finally, do plants selected to flower early have higher fitness than nonselected plants and, if so, is the relationship constant across environments? These results will determine whether selection specifically on flowering time can generate local adaptation, and may help to explain the maintenance of polymorphism in flowering time.

Materials and Methods

Plant material

We used six lines of Arabidopsis thaliana (L.) Heynh. selected for five generations for early flowering and six lines maintained as controls. Three of the selected and control lines had been bred constantly in growth chambers running simulated winter-annual growth conditions, and three of each under simulated spring-annual growth conditions. The production of the selected and control lines has been described in more detail in Scarcelli & Kover (2009). Briefly, all selection and control lines were derived from the same basal outbred population. An outbred population was used to maximize genetic variation, and allow long-term response to selection, as natural populations of A. thaliana are usually almost completely homozygous (Abbott & Gomes, 1989). The selection lines were produced by manually crossing the 50 earliest flowering plants of each line to each other. Each selected plant was randomly assigned to be the mother or the father in one of 25 crosses. Control lines were started from the same six basal populations as the selected lines, and maintained by manually crossing 50 randomly chosen plants within each line (25 crosses). Crosses between lines with very different flowering times were accomplished with the help of pollen storage (for more details, see Scarcelli & Kover, 2009). Seeds from each cross were planted into eight pots, maintaining the line sizes at 200 plants every generation. This protocol was repeated for five generations.

Growing conditions

Plants were grown in ArabiPatches (Lehle Seeds, Round Rock, TX, USA). Each ArabiPatch contains eight pots (2.5 cm in diameter) and a water reservoir which ensures constant and equally distributed amounts of water. Pots were filled with John Innes #1 soil treated with Intercept 70 WG (Bayer, Leverkusen, Germany) to avoid fungus gnats during the experiment. Each pot was sown with two seeds. If both seeds germinated, one randomly chosen seedling was removed 10 d later.

All plants were grown in Percival growth chambers (model AR-66L, Percival Scientific Inc., Perry, IA, USA). One growth chamber ran a ‘winter treatment’ designed to simulate the light and temperature transitions expected in winter-annual life histories: seeds were placed to germinate under autumn conditions (16°C : 10°C day : night and 8 h : 16 h light : dark), and then transitioned into winter (4°C during day and night, and 6 h : 18 h light : dark), spring (14°C : 10°C day : night and 8 h : 16 h light : dark) and summer (21°C : 18°C day : night and 16 h : 8 h light : dark) conditions through changes in day length and temperature (for more details, see Scarcelli et al., 2007). A separate growth chamber ran a ‘spring treatment’ designed to simulate the light and temperature transitions expected in spring-annual life histories: seeds were placed to germinate in the spring conditions and transitioned into summer conditions (as already described). The temperature and day length for each ‘season’ were chosen on the basis of average daily temperatures and day length in the UK. However, the main goal of the simulated programs was to provide multiple environmental cues (i.e. transition in day length and temperature) akin to what A. thaliana may experience anywhere in the field. To allow selection to be carried out in a timely fashion, the ‘seasons’ were very short. We believe the addition of multiple ecologically relevant cues is preferable to the uniform conditions routinely used in growth chambers. However, we recognize that, under natural conditions, plants would undoubtedly experience a much more complex environment.

Experimental design

From each of the six lines (three control and three selected) that had been growing under the spring-annual conditions, we planted eight pots with two seeds from each of the 25 crosses performed in the fifth generation of selection. Four of these pots were grown under the spring treatment and four under the winter treatment. Simultaneously, we also planted the selected and control lines that had been growing under winter-annual conditions in the same two growth chambers (as above), yielding a reciprocal transplant experiment with 1200 plants in each growing condition. After planting, pots were randomly assigned a patch and, within each growth chamber, all patches were randomly assigned a position and rotated every week.

Pots were inspected daily for germination, and later for the appearance of floral buds (bolting). The flowering time was calculated as the number of days between germination and bolting. After the plants had senesced, the total number of fruits produced by each plant was counted as an estimate of fitness (fruit number and seed set are highly correlated in A. thaliana; Westerman & Lawrence, 1970; Mauricio & Rausher, 1997).

Data analysis

To determine the effects of growing environment and selection treatment on the performance of the selected lines, we analysed flowering time and fruit production with a partially nested, four-way ANOVA using the generalized linear model (GLM) procedure in SAS (version 9.1). Flowering time and fruit production were the response variables and growing environment (GE, spring or winter), environment of origin (OR, spring-annual or winter-annual) and selection treatment (ST, selected or control) were the explaining variables. To determine whether different replicates of the selected and control lines responded differently, we included line nested within the interaction between treatment and environment of origin in the model:

Flowering time or fruit production = GE + OR + ST + GE × ST + GE × OR + OR  × ST + GE × OR × ST + LINE(SE × ST) + GE × LINE(SE × ST)

To further investigate the three-way interactions observed, we performed post hoc contrasts using a least-square means test (LSMEANS/SLICE statement in the GLM procedure in SAS).

Cross-environment genetic correlations in flowering time were determined for each line separately using a mixed model framework as implemented by the Mixed Procedure in SAS (version 9.1) (Fry, 1992). Standard errors were calculated using the jackknife procedure (Roff & Preziosi, 1994; Roff, 2006). Three families were removed from the genetic correlation analysis as a result of insufficient data across environments.

Broad sense heritability for bolting data was calculated for each line separately using the method described in Becker (1992). Standard errors were calculated as for the cross-environment genetic correlations. Changes in heritability can be caused by changes in genetic or environmental variation. Thus, to better estimate the potential evolvability of each line, we also calculated the coefficient of genetic variation (CVg = 100√Vg/) for each line according to Houle (1992), where Vg is the total genetic variation and is the mean phenotypic trait value. Standard errors were calculated using the jackknife procedure.

Results

Flowering time

Growing conditions have a strong overall effect on flowering time, with plants growing in the winter treatment flowering significantly later than plants growing in the spring treatment (Table 1, Fig. 1). As expected, the selection treatment has a significant effect, with selected lines flowering earlier than the controls (Table 1, Fig. 1). Overall, the environment of origin does not have a significant effect on flowering time, suggesting that the maternal environment per se does not affect the flowering time of these plants. However, there is a significant two-way interaction between growing environment and environment of origin, as well as a significant three-way interaction between growing environment, environment of origin and selection treatment (Table 1). This indicates that the effect of selection depends on the environment in which plants are being grown, and in which environment they were selected (that is, the environment of origin). Post hoc mean comparison tests suggest that the effect of original environment depends on whether or not the plants have been selected (Table 2). Plants from selected lines, growing in the environment in which they were originally selected, flower significantly earlier than lines selected under the alternative environment (Table 2, Fig. 1). By contrast, control lines growing in the environment in which they had been bred flower later than control lines bred in a different environment. However, the magnitude of the contrasts for control lines are smaller and, in the winter growth chamber, only marginally significant. All contrasts between selected and control lines are negative (that is, selected lines always flower earlier) and significant, independent of growing and selected environment (Table S1, see Supporting Information), and all contrasts between spring and winter growth chambers are positive and significant (Table S2, see Supporting Information).

Table 1.  ANOVA results for flowering time in Arabidopsis thaliana
SourcedfMean squareFP
  1. GE, growing environment; OR, environment of origin; ST, selection treatment. Factors with a probability of less than 0.05 are shown in bold.

GE1671 426.5320 958.0< 0.0001
OR10.010.000.990
ST149 685.5580.27< 0.001
GE × ST192.522.890.089
GE × OR1929.3729.01< 0.0001
OR × ST11.260.040.843
GE × OR × ST12802.8687.49< 0.0001
LINE(OR × ST)8619.0119.32< 0.0001
GE × LINE(OR × ST)8148.204.63< 0.0001
Figure 1.

Average number of days to flower for selected and control lines of Arabidopsis thaliana separated by environment of origin: (a) averages for plants grown in the spring; (b) results for plants grown in the winter. Control lines are represented by white bars, where lines 1–3 were originally grown in spring and lines 4–6 were originally grown in winter. Selected lines are indicated by grey bars, where lines 7–9 were originally grown in spring and lines 10–12 were originally grown in winter. Error bars represent the standard error for the means.

Table 2. Post hoc contrasts for Arabidopsis thaliana flowering time between lines from different environments of origin grouped within the four possible combinations of growing environment and selection treatment
Growth chamberSelection treatmentdfContrast (winter–spring)FP
  1. The contrast value indicates the mean difference in flowering time between lines of winter and spring origin. Contrasts with a probability of less than 0.05 are shown in bold.

SpringControl1−1.144.30.0383
SpringSelected13.44p52.5< 0.0001
WinterControl10.913.70.0540
WinterSelected1−3.4354.4< 0.0001

Results for the ANOVA on flowering time suggest that the response to selection is specialized to the environment under which plants were selected. However, the genetic factors selected upon for early flowering are not exclusively environment specific, and will cause early flowering in both growing conditions. This is further confirmed by the fact that the cross-environment genetic correlations are always positive (Table 3), although the genetic correlations appear to become weaker in selected lines (only one of the selected lines still having cross-environment genetic correlation significantly different from zero).

Table 3.  Mean trait values, cross-environment genetic correlations (Rg), broad sense heritability (H2) and coefficient of genetic variation (CVg) for flowering time in Arabidopsis thaliana
OriginTreatmentLineMean, spring (SD)Mean, winter (SD)H2, spring (SE)H2, winter (SE)CVg, spring (SE)CVg, winter (SE)Rg (SE)P from 1P from 0
  • Standard deviations (SDs) and standard errors (SEs) are given in parentheses. N, no; Y, yes.

  • 1

    , Unable to calculate reliable standard errors as pseudovalues from the jackknife procedure were not consistent with Rg.

SpringControl 136.966 (8.96)71.163 (3.61)0.988 (± 0.35)0.743 (± 0.25)17.105 (± 5.05)3.113 (± 0.69)0.562 (± 0.09)YY
SpringControl 240.872 (10.52)74.750 (4.07)0.520 (± 0.17)0.697 (± 0.18)13.182 (± 2.23)3.232 (± 0.54)0.552 (± 0.22)YY
SpringControl 336.688 (5.75)68.600 (3.83)0.515 (± 0.25)0.648 (± 0.18)7.982 (± 2.22)3.195 (± 0.58)0.932 (± 0.16)NY
SpringSelected 426.621 (2.76)64.690 (4.70)0.077 (± 0.17)0.370 (± 0.22)2.037 (± 1.80)3.134 (± 0.96)0.215 (N/A1)N/AN/A
SpringSelected 526.302 (3.27)63.970 (3.19)0.521 (± 0.40)0.697 (± 0.24)6.405 (± 3.56)2.963 (± 0.69)0.618 (N/A1)N/AN/A
SpringSelected 626.549 (3.09)65.659 (3.85)0.258 (± 0.30)0.251 (± 0.17)4.194 (± 2.62)2.083 (± 0.71)0.045 (± 0.71)NN
WinterControl 734.302 (5.31)71.466 (5.12)0.940 (± 0.22)0.764 (± 0.20)10.699 (± 1.39)4.460 (± 0.80)0.761 (± 0.20)NY
WinterControl 837.274 (10.47)73.070 (5.30)0.470 (± 0.38)0.846 (± 0.16)13.633 (± 5.70)4.748 (± 0.52)0.902 (± 0.18)NY
WinterControl 941.022 (9.34)72.563 (4.19)0.497 (± 0.29)0.708 (± 0.21)11.445 (± 3.02)3.456 (± 0.61)0.367 (± 0.38)NN
WinterSelected1032.231 (6.33)62.140 (4.61)0.927 (± 0.19)0.691 (± 0.22)13.627 (± 2.30)4.379 (± 0.97)0.431 (± 0.18)YY
WinterSelected1128.747 (3.79)60.515 (2.90)0.468 (± 0.19)0.325 (± 0.22)6.603 (± 1.61)1.936 (± 0.59)0.587 (± 0.40)NN
WinterSelected1229.307 (4.71)61.341 (4.87)1.03 (± 0.25)0.289 (± 0.17)12.210 (± 2.95)3.028 (± 0.89)0.137 (± 0.42)NN

Selection reduces the heritability in flowering time when plants are grown under the same condition in which they were selected (Table 3). However, selected lines have higher heritability when grown in the environment in which they were not selected. This is particularly noticeable for lines selected under the spring treatment. Estimates of evolvability (Table 3) give similar results. In the spring growth chamber, plants selected in the spring have much smaller CVg than plants from any other treatment, including plants selected under winter conditions. In the winter growth chamber, CVg is much smaller than when plants are grown in the spring growth chamber, and differences between treatments are less obvious. Nevertheless, CVg for winter-selected lines tends to be somewhat smaller than for control lines of winter origin, but no clear trend is observed when comparing CVg values for selected and control lines growing in the winter. These results suggest that adaptation to one environment will reduce evolvability in the same environment, but will not necessarily constrain future response to selection in a different environment.

Fruit production (fitness)

There is a significant negative correlation between flowering time and fruit production within all control lines in both growing conditions (Table 4). Thus, both growing conditions favour plants that flower earlier, leading to the expectation that lines selected to flower early should have higher fitness than controls.

Table 4.  Results of the regression analysis of fruit number on flowering time in Arabidopsis thaliana
ChamberLineTreatmentOriginβSDPR2
  1. Slope indicated by the nonstandardized beta coefficient (β), and significance indicated by the probability (P). R2 indicates the proportion of the variance in fruit explained by flowering time. Analysis was run separately for each line in each growth chamber.

Spring
 1ControlSpring−1.280.281.30 × 10−50.182
 2ControlSpring−1.200.261.21 × 10−50.183
 3ControlSpring−0.900.400.0270.051
 7ControlWinter−1.570.540.0050.089
 8ControlWinter−0.9540.330.0040.084
 9ControlWinter−0.8930.215.34 × 10−50.167
Winter
 1ControlSpring−0.530.160.0010.107
 2ControlSpring−0.420.113.9 × 10−40.121
 3ControlSpring−0.850.174.12 × 10−60.199
 7ControlWinter−0.430.130.0010.113
 8ControlWinter−0.200.080.0290.048
 9ControlWinter−0.410.170.0200.054

The ANOVA on fruit number shows that, as with flowering time, growing conditions have a strong overall effect on fruit production. Plants growing in the spring treatment produce significantly more fruits than plants growing in the winter treatment (Table 5, Fig. 2; Table S3, see Supporting Information). Although selected plants tend to produce more fruits, there is much variation between lines (Fig. 2), and the overall effect of the selection treatment is not statistically significant (Table 5). However, the interaction between growing environment and selection treatment, and the three-way interaction, are statistically significant (Table 5), indicating that the effect of selection on fruit production depends on the environment of origin and current growing environment.

Table 5.  Results for the ANOVA on fruit production in Arabidopsis thaliana
VariabledfMean squareFP
  1. GE, growing environment; OR, environment of origin; ST, selection treatment. Factors with a probability of less than 0.05 are shown in bold.

GE11 047 304.552832.17< 0.0001
OR1113.010.310.5805
ST12608.460.90.37
GE × ST11516.994.100.0429
GE × OR119.460.050.8186
OR × ST1495.081.340.2474
GE × OR × ST15540.8114.980.0001
Line(OR × ST)82881.877.79< 0.0001
GE × Line(OR × ST)87855.0621.24< 0.0001
Figure 2.

Average number of fruits produced by selected and control lines of Arabidopsis thaliana separated by environment of origin: (a) averages for plants grown in the spring; (b) results for plants grown in the winter. Control lines are represented by white bars, where lines 1–3 were originally grown in spring and lines 4–6 were originally grown in winter. Selected lines are indicated by grey bars, where lines 7–9 were originally grown in spring and lines 10–12 were originally grown in winter. Error bars represent the standard error for the means.

Specific contrasts indicate that selected lines produce more fruits than controls when growing in the same environment in which they were selected (although the contrast between selected and control lines bred in winter and grown in winter is only marginally significant), as expected given the negative relationship between flowering and fitness (Table 6). By contrast, plants from selected lines produce slightly less fruits than plants from control lines when growing in an environment under which they were not selected, but these differences are not statistically significant. This result is unexpected because selected plants, even when growing in the alternative environment, flower earlier and thus should produce more fruits.

Table 6. Post hoc contrasts for the number of fruits produced by control Arabidopsis thaliana vs selected lines grouped within the four possible combinations of growing environment (GE) and environment of origin (OR)
GEORdfContrast (selection–control)FP
  1. The contrast value indicates the mean difference in fruit number produced by selection and control lines. Contrasts with P < 0.05 are shown in bold.

SpringSpring17.8711.930.0006
SpringWinter1−0.420.010.9038
WinterSpring1−1.6612.980.0003
WinterWinter12.7833.21> 0.0001

The contrasts between plants originally bred in spring vs in winter conditions indicate that plants selected in the spring produce significantly more fruits than plants selected under winter conditions, when grown in spring (Table 7). However, when plants are grown in the winter, the number of fruits produced by plants selected under spring conditions is not significantly different from plants selected under winter conditions. We also observed that, in the spring growth chamber, control lines originally from winter conditions produce significantly more fruits than control lines originally from spring.

Table 7. Post hoc contrasts for the number of fruits produced by Arabidopsis thaliana lines bred under spring or winter conditions grouped within the four possible combinations of growing environment (GE) and selection treatment (ST)
GESTdfContrast (spring–winter)FP
  1. The contrast value indicates the mean difference in fruit number between lines originally grown in the winter and in the spring. Contrasts with P < 0.05 are shown in bold.

SpringControl1−3.734.470.0346
SpringSelect.14.568.070.0045
WinterControl12.542.380.1233
WinterSelect.1−1.91.460.2266

Finally, there is a significant interaction of line with all other factors in the model, further suggesting that, in the different lines, response to selection may have involved different genetic factors.

Discussion

Organisms cannot be adapted simultaneously to all environments. When adaptation to local conditions in one environment reduces the fitness in another environment, there is a genotype by environment effect on fitness which, under some circumstances, can maintain genetic diversity and favor speciation (Levene, 1953; Kassen, 2002). Environmental heterogeneity is hypothesized to maintain genetic variation when selection favours different alleles at the same locus in different environments, or different extremes of a quantitative trait, generating antagonistic pleiotropic effects on fitness (Via & Lande, 1985; Gardner & Latta, 2006). Here, we show that adaptation to early flowering increases the fitness of selected lines when they are grown in the same environment in which they were selected, generating local adaptation. This resulting local adaptation is particularly noteworthy, because plants were selected at the phenotypic level for the same trait and in the same direction in both environments. Under field conditions, where evidence for divergent selection is mainly investigated at the phenotypic level, such a nondivergent phenotypic selection pattern might be dismissed as unimportant in the generation of local adaptation.

Previous evidence that adaptation to one environment causes reduction in fitness in another environment comes from two types of experiment: reciprocal transplants (Ehleringer & Clark, 1988; Angert & Schemske, 2005) and experimental evolution (Travisano & Lenski, 1996; Ostrowski et al., 2005). Studies at the molecular level (mainly using QTLs) have found that selection often operates at different loci (Nuzhdin et al., 1995; Gurganus et al., 1998; Vieira et al., 2000; Weinig et al., 2003) or favors the same allele in a locus (Ostrowski et al., 2005; Gardner & Latta, 2006), although a few QTLs with antagonistic pleiotropic effects have been found (for example, Leips & Mackay, 2000). A possible explanation for the occurrence of a fitness trade-off without antagonistic pleiotropic effects at the locus level is that, during adaptation, there is an accumulation of neutral alleles with deleterious effects in other environments (Kawecki et al., 1997). Under this scenario, local adaptation will not maintain genetic variation or promote speciation, as recombination among local genotypes can lead to genotypes that can succeed across environments (for example, Rieseberg et al., 2003).

It is unlikely that, in our study, the fitness reduction comes from mutation accumulation, as selected lines have only been bred for five generations. It is also unlikely that there is an accumulation of neutral alleles with negative pleiotropic effects across environments, or that there has been unintended selection for the two different growing conditions. This is because control populations experienced the same growing conditions and breeding scheme as the selected lines, and did not experience a reduction in fitness when they were grown in a different environment. Thus, we conclude that the reduction in fitness observed in lines selected in the spring treatment is a result of negative pleiotropic effects of genes for early flowering selected specifically under these conditions. Although specific negative effects of flowering genes on fitness have not been shown previously, a few studies have shown that flowering genes can have pleiotropic effects on a wide range of traits (Van Tienderen et al., 1996; Worland, 1996; Mckay et al., 2003; Scarcelli et al., 2007). Thus, it is quite possible that the combination of pleiotropic effects has a net negative effect in some environments, but not in others.

The observed reduction in fitness is particularly interesting because we also observed a positive genetic correlation in flowering time, indicating that the genes selected under the two growing conditions will cause early flowering independent of the environment in which the plants grow. Furthermore, we also observed that early flowering is associated with higher fitness in both growing conditions. Thus, the negative pleiotropic effects cannot be a result of the flowering time, but must be caused by the effect of these genes on some other unmeasured trait. Although lines selected under winter-annual conditions did not experience a reduction in fitness when compared with their controls, they did not show an increase either – even though they flowered significantly earlier than their controls. Thus, it is possible that, given their flowering time, they experienced some reduction in fitness, which we cannot detect by comparison with controls.

Previous to the selection experiment, the base population was kept at a single constant environment (21°C and 16 h : 8 h light : dark). Therefore, both the winter- and spring-annual growing conditions represent new environments to the population. Thus, we hypothesize that the reduction in genetic correlation across environments indicates that, although some of the selected genes affect flowering time equally under both growing conditions, others are environment specific. Selection probably increases the frequency of environment-specific early-flowering alleles, causing a reduction in correlation. This hypothesis is supported by the fact that the genetic correlations in flowering time across environments are between zero and unity, and that the earliest flowering lines in each growth condition are the lines selected under the same condition. These results are in agreement with our previous analysis of these selected lines (Scarcelli & Kover, 2009), where we investigated changes in allele frequency after three generations of selection at two of the main candidate genes for flowering time in Arabidopsis: FRIGIDA (FRI) and FLOWERING LOCUS C (FLC). We found that, after only three generations, there was a significant reduction in the frequency of functional alleles at FRIGIDA in lines selected in the spring, but not in lines selected in the winter, whereas no significant differences in allele frequency were observed in FLC under either growth conditions.

FRI is part of the vernalization pathway, which detects the passage of winter and synchronizes flowering with the arrival of spring (Michaels & Amasino, 2001). In the absence of cold, plants with functional FRI are late flowerers and plants with nonfunctional FRI flower earlier (Johanson et al., 2000; Lempe et al., 2005). Analysis of the gene sequences of different accessions of A. thaliana suggests that molecular variation in FRI has been shaped by selection (Le Corre et al., 2002; Toomajian et al., 2006), and that nonfunctional alleles of FRI have evolved multiple times from late-flowering functional alleles through independent loss-of-function mutations (Johanson et al., 2000; Le Corre et al., 2002; Gazzani et al., 2003; Shindo et al., 2005). Thus, it is not surprising to find that, in the spring treatment, plants selected to flower earlier under these conditions are the earliest flowering plants, as the frequency of nonfunctional FRI has increased significantly (Scarcelli & Kover, 2009). However, it was unexpected to find that, in the winter treatment, winter-selected plants were the earliest flowerers, as genes that confer early flowering specifically under winter-annual conditions are not yet known.

The ability to adapt to new environments is critical for the long-term survival of many plant species, particularly given the expected change in global climate (Vergeer et al., 2004). Given that the ability to correctly read the local environmental cues and flower at the right time has a strong effect on a plant's fitness, low genetic correlation in flowering time across environments would be expected in weedy species with a wide geographic distribution (Donohue et al., 2005). We found that, although there is positive genetic correlation between flowering in winter- and spring-annual growth conditions, this correlation becomes weaker with adaptation. Furthermore, when plants were moved to an environment different from that in which they were selected, there was an increase in heritability/evolvability. These results suggest that adaptation to one of the environments (particularly spring-annual conditions) may constrain establishment in a different environment, especially when in competition with locally adapted genotypes. However, it may not compromise future response to selection if there is no reduction in genetic variation in another environment, as was the case in our experiment.

The role of environmental heterogeneity in maintaining genetic variation has been the subject of much attention. We found that the genetic basis of adaptation in our experimental population included a mixture of alleles with correlated positive effects and environment-specific alleles. Although, at first sight, these results suggest that selection for early flowering in these environments would not help to explain the maintenance for flowering time variation, we also found that the environment-specific genes have negative pleiotropic fitness effects that would give an advantage to local genotypes. In our simulated environments, there would be a stronger selection against genotypes to move from a spring-annual condition to a winter-annual condition, than vice versa. If our results are translatable to natural conditions, it would suggest that we would find higher genetic variation for flowering time in geographic regions in which A. thaliana has a spring-annual growth habit (hypothesized to be in most of western Europe; Koornneef et al., 2004) than in areas in which it grows as a winter-annual (mostly in northern Europe; Koornneef et al., 2004). A better understanding of the role of adaptation in flowering time to different environmental conditions requires the characterization of the environment-specific alleles, and a test of the selected genotypes under more realistic field conditions. The translation of our results to natural populations also needs to consider that the genetic diversity present in our experimental population is much larger than that observed in natural populations. Nevertheless, we believe that our experiment suggests that adaptation to local environment conditions has the potential to explain the extensive variation in flowering time in Arabidopsis, even if selection always favours early flowering.

Acknowledgements

We are grateful for the technical help of R. George. The manuscript was much improved by helpful discussions with R. Preziosi, D. Rozen and J. Wolf, and by the comments of three anonymous reviewers. This research was supported by a research grant from the Natural and Environmental Research Council (NERC), to P.X.K.

Ancillary