GENOTYPE BY ENVIRONMENT INTERACTIONS FOR FITNESS IN HYBRID GENOTYPES OF AVENA BARBATA

Authors


Abstract

We examined genotype (G) by environment (E) interactions for fitness in mesic and xeric ecotypes of the self-fertilizing annual grass, Avena barbata and their recombinant inbred hybrid progeny. Fitness was assayed (1) in experimental water and nutrient treatments in the greenhouse and (2) in common gardens in each ecotype's native habitat. G × E interactions were significant in the greenhouse. Nevertheless, the same recombinant genotypes tended to have high fitness across all water and nutrient treatments. G × E interactions were less pronounced in the field, and were driven by the contrast between the uniformly low survivorship at the mesic site in 2004 and genetic variation in fitness at the other years/site combinations. Moreover, the mesic ecotype consistently outperformed the xeric in both field and greenhouse. Several of the recombinant genotypes outperformed the parents in the novel greenhouse treatments, but these genotypes did not outperform the mesic parent in field trials. Indeed, it is only in the comparison between field and greenhouse environments that there was a noticeable change in the identity of the most-fit genotype. The results provide evidence that hybridization can create genotypes that are better adapted to newer environments such as those imposed in our greenhouse experiments.

The study of hybrid crosses between genetically diverged populations from contrasting habitats can yield important insights into the niche breadth of genotypes. If there are trade-offs between adaptation to different environments (Futuyma and Moreno 1988; Kawecki and Ebert 2004), then hybrids between genetically distinct lineages will have low fitness, because they will contain some alleles that are maladaptive in each environment (Jordan 1991; Dudley 1996; Nagy 1997; Lexer et al. 2003a). In some cases, the hybrids can occur within intermediate habitats between those of the parents (termed bounded hybrid superiority Moore 1977). Wild F1 and backcross generations often associate with the environmental gradient separating the parental habitats (Emms and Arnold 1997; Wang et al. 1997; Fritsche and Kaltz 2000; Johnston et al. 2001; Campbell and Waser 2001). Alternatively, if such a trade-off is absent, then hybridization, along with the attendant recombination can produce genotypes that are broadly adapted across both environments (Clausen 1949; Allard 1965; Kawecki et al. 1997). This occurs because a subset of the recombinant hybrid genotypes combines the adaptations to one environment with the (separate) adaptations to another. A further possibility is that the novel gene combinations that hybridization creates can contain genotypes adapted to novel environments allowing the colonization of new habitats (Allard 1965; Ellstrand and Schierenbeck 2000), and in some cases leading to speciation (Anderson and Stebbins 1954; Grant 1981; Arnold 1997; Rieseberg et al. 1999; Siehausen 2004). Indeed, adaptation to novel environmental conditions is one of the leading hypotheses for the selective advantage of recombination (Bell 1982; LeNormand and Otto 2000).

Which of these outcomes occurs is an expression of the genotype by environment (G × E) interactions for fitness among the recombinant hybrid genotypes. G × E occurs when the effect of the environment on a trait varies among genotypes, and indicates that the genotypes have different reaction norms (Schlicting and Pigliucci 1998). G × E is widely studied to measure genetic variation in the degree of phenotypic plasticity (plasticity being a change in a trait induced by environmental variation), and by extension the potential of plasticity to respond to natural selection (Via 1994). However, it seems inappropriate to speak of plasticity in fitness. Rather, G × E for fitness may indicate ecological specialization if the reaction norms for fitness cross (Clausen et al. 1941; Via 1991; Bennington and McGraw 1995)—that is if the rank order of fitness changes across environments, such that the highest fitness genotype is different in different environments. Alternatively, if the reaction norms do not cross, then G × E simply represents a change among environments in the intensity of selection.

In this article, we assess the G × E for fitness in a series of hybrid recombinants derived from an intraspecific cross between ecotypes. We examine reaction norms across both field environments with which the ecotypes are associated as well as in novel environments created in the greenhouse. Our goals are to determine (1) whether there are trade-offs between fitness in different environments that would lead to ecological specialization; (2) whether hybrid recombination produces broadly adapted genotypes; and (3) whether hybrid recombinants are adapted to the novel conditions of the greenhouse.

STUDY SYSTEM

We have chosen a study organism that has a well-defined association of genotypes with contrasting environments. Avena barbata is a diploidized tetraploid (2n= 4×= 28) annual grass with a high level (greater than 95%) of self-fertilization (Allard 1996). Originally from the Mediterranean, it was first introduced to California approximately 200 years ago (Allard 1996), where two ecotypes have since been recognized based on five allozyme loci (Allard et al. 1972). The pattern of association between these allozyme loci and the environment is nonrandom, with plants that occur in mesic regions tending to be fixed for one set of alleles whereas plants from the xeric regions tend to be fixed for the opposite set of alleles. This pattern of association has been found on a regional (Clegg and Allard 1972) as well as a very local scale (Hamrick and Allard 1972; Hamrick and Holden 1979). Patches of polymorphism for the xeric and mesic genotypes can be found where the two habitats meet, with the two genotypes being correlated with the level of moisture (Hamrick and Holden 1979). The two allozyme genotypes also show fixed differences for 129 AFLP markers (Gardner and Latta 2006) as well as being strongly differentiated for a suite of quantitative characters (Hamrick and Allard 1975; Hutchinson 1982; Latta et al. 2004). Heritabilities for quantitative traits within each ecotype are minimal (Gardner 2004). The strong genetic and ecological divergence between the allozyme genotypes originally identified by Allard et al (1972), as well as the limited genetic variation within genotypes suggests that they represent true ecotypes.

In a previous study (Johansen-Morris and Latta 2006), we conducted a line cross analysis (Lynch and Walsh 1998; DeMuth and Wade 2006) of a hybrid cross between the ecotypes. Recombinant inbred lines that were derived from the cross show hybrid breakdown (i.e., the mean fitness of the recombinant inbred lines (RILs) is significantly lower than that of the mid-parent), confirming the existence of coadapted gene complexes in the naturally occurring ecotypes. This finding was repeated across both native and greenhouse environments, although hybrid breakdown was more pronounced in the field than in the greenhouse. Transgressive segregation also occurred (i.e., the extremes of the recombinants exceeded the range between the parents) such that the highest fitness recombinant genotypes outperformed the parents, especially in the greenhouse environment, but this was far less pronounced in the field. There was significant variation among families within generation classes (P1, P2, F2, F6), when each environment was analyzed separately, but across environments, there was no overall family effect, but rather a strong family by trial interaction. (We use the term “family” to refer to the selfed progeny of a single parent—because A. barbata is highly selfing, families of the P1, P2, and F6 generations are genetically uniform). Thus the relative fitness of the different families appears to vary among environments.

Methods

CROSSES

Seeds of the mesic and xeric genotypes were kindly provided by Dr. Pedro Garcia. These seeds were collected from a variety of mesic and xeric populations in Northern California between 1984 and 1989. A series of RILs were created following a cross between the mesic and xeric ecotypes of A. barbata, for use in QTL mapping studies (Gardner and Latta 2006). Crossing methods are described in Johansen-Morris and Latta (2006). Lines were propagated by a single seed descent to prevent selection among RILs prior to our fitness measurements. At the F6 generation, RILs are on average 96.75% homozygous, and genetically uniform among individuals within each RIL. Each RIL contains a random combination of alleles from the parents (i.e., for parent genotypes AABBCC and aabbcc, different RILs will be fixed for genotypes AAbbCC, AABBcc etc.). The F6 RILs thus disrupt any coadapted gene complexes present in the parental ecotypes, but create a large pool of variation on which selection may act.

GREENHOUSE EXPERIMENT

This experiment examined 12 F6 families (i.e., RILs) in four different environments to ascertain whether G × E interactions occur and whether any of these families outperform the parental ecotypes in any of the environments. Twelve of the 25 F6 families used in Johansen-Morris and Latta (2006) were chosen to represent the entire range of fitness values found in the greenhouse, with four families representing high-fitness values, four representing low-fitness values, and four representing those families with fitness values around the average for the F6 generation.

Eighty mesic individuals, 80 xeric individuals, and 64 individuals from each of the 12 families were divided equally among eight plots and their position within each plot was randomized. Two plots were assigned at random to each of four environments, made up of all four combinations of two different watering regimes and two different fertilization regimes. Seeds were germinated by the method of Latta et al. (2004), planted in cell packs, and placed in a growth chamber with a 12 h light:dark cycle. The temperature within the growth chamber was 20°C during the light cycle and 15°C during the dark cycle. After two weeks the plants were transferred to pots in the greenhouse and the pots were placed in trays with eight pots per tray. The plants were allowed to become established in the pots for two weeks at which point the environmental regimes were initiated.

The placement of the environments within the greenhouse was random. The high-water treatment was watered every other day and the low-water treatment once a week. The high-nutrient treatment was fertilized once a week and the low-nutrient treatment received fertilizer once at the midpoint of the experiment. The fertilizer used contained nitrogen, phosphorous, and potassium at a ratio of 15:15:18 (Plant Products Ltd., Brampton, Ontario, Canada) and was added directly to the water that was used for the watering regime. The plants were watered by adding water to the trays, which was then absorbed by the soil in the pots until fully saturated. During a period of extremely hot weather in July and August, the low-water treatments were augmented to receive water twice a week, to prevent excessive mortality in these treatments.

Flowering time was recorded throughout the experiment. After seven and a half months the plants were senescent, and the experiment was harvested. We counted the number of spikelets on each plant, and collected the aboveground portion of the plant for dry weight measurements. As A. barbata is an annual plant, the number of seeds produced by each plant will give a measure of the lifetime reproductive success of each plant, and because it is predominantly selfing, this fitness measure includes both male and female success. There are two seeds per spikelet.

FIELD EXPERIMENT

Johansen-Morris and Latta (2006) present fitness data for 25 RILs in two native field sites in each of two growing seasons, and these data were used here to study G × E interactions and local adaptation in the native habitats. Detailed field methods are given in Johansen-Morris and Latta (2006). Briefly, Sierra Foothills Research and Extension Centre is considered a dry site, where the native A. barbata are of the xeric genotype (Hutchinson 1982) whereas Hopland Research and Extension Centre is moister, and falls within the range of the mesic genotype. A total of 50 mesic and 50 xeric individuals, and 400 F6 individuals (16 from each of the 25 families, including all of the 12 families used in the greenhouse experiment) were planted at each site in November 2002 and 2003. Each site was divided into three blocks, with the plants divided equally between the blocks, and the position of the plants within each block was randomized. The plants were allowed to grow until senescence had been reached in early June 2003 (May 2004) at which point fitness, quantified as the number of spikelets produced, was recorded. Aboveground biomass was measured by cutting the plants off at soil level, placing them in paper bags, drying them in a drying oven for four days at 50°C, and obtaining their dry weights. As it was not possible to be in the field during the growing season, flowering time was not recorded.

STATISTICAL ANALYSIS

Analysis of variance (ANOVA) was performed first on a full model in each experiment. In the greenhouse, water (W) and nutrient (N) treatments were treated as fixed effects, with plot (P) as a random effect nested within water/nutrient combination. Family (F) was treated as a random effect, and all interactions were tested.

image

In this analysis, the first four terms represent different aspects of environmental variation, (cf. Bell 1990), family represents genetic variance, and the final four terms represent interaction between genotype and the different aspects of environment. There appear to be differences between plots subjected to the same water by nutrient treatment. Thus a simpler model was also analyzed in which each plot was treated as a separate environment, and a two-way ANOVA was conducted with both family and environment treated as random effects. In essence, the simplified model conflates the four environmental terms [W, N, W×N, and P(W×N)] into a single factor: environment, which we treated as a random factor because it contains a source of random variation, P(W×N). Variance components were calculated for the F6 generation for genotype, environment, and the G × E interaction terms using restricted estimation maximum likelihood.

Analysis of the field data used a similar model with site (S) and year (Y) as fixed effects, and block (B) nested within site/year combination.

image

Although this model is similar to that analyzed in the greenhouse, note that block differs from plot in that blocks represent subdivisions of a single Y×S treatment, whereas W×N treatments were assigned at random to whole plots in the greenhouse. Again family was treated as a random effect, and all interactions were tested. Block effects were not significant and were dropped from the model. As in the greenhouse, a simplified model was analyzed, in which each of the four year/site combinations were treated as a separate environment. All analyses were conducted in SPSS 13.0.

In most cases, plants that died, or that did not flower could confidently be assigned a fitness of zero. However some of the field trials (notably Hopland in 2003) resulted in plants that had been broken off near the base of the reproductive tillers. It was not possible to tell whether these plants had been broken before or after seeds had been matured, so it was not possible to tell their fitness. We reanalyzed the field data both with and without these individuals to determine whether this breakage had any bearing on our estimates of G × E. In addition survivorship was extremely low at Hopland in 2004, giving essentially zero genetic variance. Because this could induce an inflated G × E interaction, the field data were analyzed both with and without Hopland in 2004.

Results

GREENHOUSE EXPERIMENT

There were significant environmental main effects on all three traits, as overall means for each trait differed significantly among the eight plots (Table 1a). In the case of fitness, the only significant environmental main effect in the ANOVA was that among plots within treatments. For mass and flowering time the environmental treatments significantly affected overall means over and above the variation among plots, which was also significant. This shows, therefore, that despite attempts to hold the environment constant within the water and nutrient treatments, there are other variable environmental factors in the greenhouse, which affect the growth of the plants. All three traits showed significant variation among families (Table 1a). However, with the reduced power to detect family effects in the full model (family was tested over family by plot interaction, giving few degrees of freedom), the main effects of family were not significant in the full model (Table 1a).

Table 1.  Analysis of variance (ANOVA) on fitness (spikelet number), above ground dry mass and flowering time of mesic and xeric ecotypes and 12 recombinant families of A. barbata, under artificially varied environmental conditions in the greenhouse. (a) Mean square and statistical tests for the complete model analyzing each environmental factor separately, and for the condensed model treating each block as a separate environment. (b) Components of variance attributable to genotype, environment and G×E interaction for the condensed model.
(a) SourcedfSpikelet numberDry massFlowering time
MS  FMS  FMS   F
  1. *P<0.05; **P<0.01; ***P<0.001; ****P<0.0001.

 Complete model
  Water (W)15,993,455.535.2240,718.3029.43**83,946.208.49*
  Nutrient (N)17,432,112.236.31129,287.0192.08***274,793.2724.89***
  Water×Nutrient15,670,669.864.9430,174.1622.20 **92,594.8013.76*
  Plot (W×N)41,123,233.5544.94****1296.1762.20****6900.569.47****
  Family13122,843.521.52149.490.9754,593.776.74****
  Family×Water1350,979.501.01108.771.293751.886.62***
  Family×Nutrient1380,322.371.59129.681.544916.898.68***
  Family×W×N1350,468.842.01*84.294.06****566.420.77
  Plot (W×N)×Family5225,127.152.11****20.720.60733.531.68**
  Error77211,889.44 34.57 436.41 
 Condensed model
  Environment73,436,367.23289.03 ****28,326.80800.03****70,653.82161.90****
  Family13122,843.5210.33****375.6010.61****54,593.76125.10****
  Environment×Family9141,039.823.45****232.406.56****1716.653.93****
  Error77211,889.44 35.40 436.41 
 
(b) Variance components Estimate  %Estimate  %Estimate   %
 
  Environment 30,298.4264.15264.7887.21637.8930.20
  Genotype 1354.922.871.320.43867.5041.08
  G×E 3693.727.823.091.02169.888.04
  Individuals (G×E) 11,884.3925.1634.4211.34436.6720.68

There were strong interactions between family and at least one aspect of the environment for all three traits, although the nature of the interaction seems to vary among traits (Table 1a). Family variation in spikelet number interacted most significantly not only with plot within treatments, but also with the water by nutrient combination. Flowering time on the other hand, showed the strongest interactions of family with the main effects of water and nutrient separately, and no interaction with the W×N combination. For both fitness and aboveground biomass the majority of the variation was attributable to differences between environments, but the G × E component was considerably larger than the genotype component (Table 1b). In contrast, the majority of the variation for flowering time was attributable to the differences between genotypes, with limited variation due to either environment or G × E.

There is considerable change in rank order of genotypes across greenhouse environments for fitness, and aboveground biomass (Fig. 1) although flowering time is less affected by the environment than either fitness or aboveground biomass (Fig. 2). Family means correlations across environments are generally positive in the greenhouse (Table 2). Although with only 12 families, individual correlations are subject to considerable error, a general pattern emerges that correlations between replicates of the same water/nutrient treatment were generally higher than correlations across treatments that differed in one factor. Correlations between environments that differed in both water and nutrient treatment were generally lowest (Table 2). For example, the fitness of genotypes were strongly correlated across the two W−N− replicates (r= 0.944). Fitness in these treatments showed a weaker correlation with W−N+ and W+N− treatments (r= 0.595 to 0.799) and were weakly correlated with fitness in the W+N+ treatment (r= 0.314 to 0.406). Correlations across environments are high among all environments for flowering time (r= 0.65 to 0.95; Table 3) whereas fitness and aboveground biomass had much lower correlations among greenhouse environments (Table 2).

Figure 1.

Reaction norms for fitness performance (dry mass and spikelet number) across environments in recombinant inbred lines of Avena barbata. Upper panels give absolute family means. Lower panels give relative fitness measures standardized to a mid-parent fitness of 1.0. Within each panel the greenhouse experimental treatments are shown on the left, the field results on the right. Greenhouse data from Johansen-Morris and Latta (2006) are included for comparison. Treatment notation as in Table 2.

Figure 2.

Reaction norms for flowering time across greenhouse water and nutrient treatments in recombinant inbred lines of Avena barbata. Treatment notation as in Table 2.

Table 2.  Cross environment line means correlations of spikelet number (above diagonal) and dry mass (below Diagonal). W+, High-water treatment, W−, low-water treatment; N+, high-nutrient treatment; N−, low-water treatment. Replicates 1 and 2 are indicated following the treatment description. Significant correlations are shown in italics. Correlations significant after Bonferroni correction are highlighted in bold.
 Green house experiment N=14aField experiment N=27b
 W+N+1 W+N+2 W+N−1 W+N−2 W−N−1 W−N−2 W−N+1 W−N+2Green house*Sierra 2003 Sierra  2004Hopland 2003 Hopland  2004
  1. * Greenhouse data from Johansen-Morris and Latta 2006.

  2. aThe a=0.05 significance threshold for N=14 is 0.532. Bonferroni corrected for 78 comparisons the threshold is 0.723.

  3. bThe a=0.05 significance threshold for N=27 is 0.381. Bonferroni corrected for 78 comparisons the threshold is 0.531.

W+N+1 0.643 0.340 0.417 0.361 0.314 0.4700.5880.6770.304 0.2900.324−0.082
W+N+20.796  0.407 0.202 0.406 0.379 0.3750.5380.6510.2160.5400.323−0.026
W+N−1 0.289 0.180 0.8180.9350.922 0.4630.7600.7480.165 0.1180.074−0.078
W+N−2 0.263 0.064 0.280 0.8100.8160.6220.7620.6490.290−0.0870.069−0.120
W−N−1−0.111−0.002 0.104 0.367 0.9440.6170.7990.8190.125 0.1600.121 0.005
W−N−2 0.081 0.045 0.070 0.078 0.450 0.5950.7880.8310.230 0.2020.188 0.216
W−N+1 0.459 0.485−0.128−0.249 0.072 0.165 0.7490.5510.432 0.1470.386 0.091
W−N+2 0.065 0.124 0.320−0.321−0.387−0.150−0.097 0.7110.248−0.0130.134 0.002
Green house* 0.4240.588 0.097−0.079 0.225 0.047 0.279−0.272 0.223 0.3590.259 0.097
Sierra 2003−0.049−0.059 0.260−0.278 0.126 0.408 0.141 0.0060.340 0.5980.697 0.224
Sierra 2004−0.032 0.065−0.366−0.564 0.157 0.310 0.138 0.0320.2600.570 0.458 0.167
Hopland 2003−0.225−0.050−0.017−0.398 0.283 0.388 0.062−0.1420.6250.7740.494  0.279
Hopland 2004−0.276−0.245−0.183−0.141 0.240 0.370−0.141−0.0650.1240.4230.4140.560 
Table 3.  Cross environment line mean correlations of flowering time. Treatment notation as in Table 2.
 W+N+1W+N+2W+N−1W+N−2W−N −1W−N−2W−N+1W−N+2
W+N+1 
W+N+20.960 
W+N−10.7620.842 
W+N−20.6930.7670.930 
W−N−10.6540.7430.9550.868 
W−N−20.6720.7590.9590.9210.952 
W−N+10.8880.9250.8710.7730.8190.853 
W−N+20.8440.9160.8940.8140.8190.8220.920 

G inline image E INTERACTIONS IN THE FIELD

Fitness and mass varied significantly among both sites (Hopland vs. Sierra) and years (Table 4a). Both fitness and aboveground biomass were greater at Sierra than at Hopland and greater in 2003 than 2004. There was no significant block variation, so block was dropped from the model. Family variation interacted significantly with both site and year.

Table 4.  Analysis of variance on fitness (spikelet number) and above ground dry mass of mesic and xeric ecotypes and 25 recombinant families of A. barbata, under naturally occurring environmental conditions in the field. (a) Mean Squares and significance tests for both the complete model analyzing each environmental factor separately, and the condensed model (b) REML variance components for the condensed model. Variance components are also given for both experiments (Field and Greenhouse) pooled.
(a) SourcedfSpikelet numberDry mass
MS  FMS  F
  1. *P<0.05; **P<0.01; ***P<0.001.

 Complete model
  Year (Y)127,938.00213.71***94.54  186.12***
  Site (S)111,299.2188.08***31.64  72.22***
  Family26340.291.471.548 1.76
  Year×Site13252.0857.51***10.81  57.24***
  Year×Family26141.922.85**0.56  3.29**
  Site×Family26138.872.79**0.482 2.79**
  Year×Site×Family2649.720.590.173 0.68
  Error132484.67 0.254  
 Condensed model
  Trial315,547.38136.71***49.75  122.23***
  Family26340.292.82***1.548 3.49***
  Trial×Family78121.431.43**0.447 1.75***
  Error132484.67 0.254  
 Omitting Hopland 2004
  Trial214,492.97115.87***47.35  109.80***
  Family26436.093.39***1.894 4.15***
  Trial×Family52129.191.180.458 1.40*
  Error1018109.57 0.327  
 
(b) Variance components Estimate  PercentEstimate  Percent
 
  Environment 56.1137.950.176438.62
  Genotype 4.092.770.01362.97
  G×E 3.662.480.01533.35
  Individual (G×E) 83.9856.800.251555.06
 Omitting Hopland 2004
  Environment 54.0731.210.173132.99
  Genotype 7.684.440.02364.49
  G×E 2.831.630.01262.40
  Individual (G×E) 108.6962.730.315560.12
 All environments
  Environment 19,99275.66209.5   93.89
  Genotype 3691.40.26  0.12
  G×E 20097.601.98  0.89
  Individual (G×E) 405315.3411.4   4.95

In the reduced model, significant differences were found for the main effects of family and environment for both traits (Table 4). Family by environment interaction was less significant however. The significant G × E term appears to be driven by the low variance for fitness in Hopland in 2004. Most families had near-zero fitness because of the high mortality at Hopland that year. Indeed, G × E interactions were much weaker with Hopland 2004 omitted, and were not significant for fitness. The rank order of families does not change greatly between the environments for either fitness or aboveground biomass (Fig. 1). In contrast to the greenhouse experiment, G × E interactions generally accounted for less of the variance than the main effects of family (Table 4b). Family mean correlations across environments were generally stronger among field sites than among contrasting treatments in the greenhouse (Table 3). At Sierra, fitness and mass in 2003 correlated tightly with that in 2004. Across sites, Hopland performance correlated well with performance at Sierra in 2003. Family mean correlations were lowest between the greenhouse and field experiments (e.g., for fitness, r=−0.09 to 0.54; Table 2). Notably, field mass correlated negatively (although not significantly) with mass under the high-water treatments.

G inline image E IN THE MESIC AND XERIC ECOTYPES, AND NICHE BREADTH

The mesic ecotype outperformed the xeric ecotype at both field sites, as well as in most of the greenhouse environmental treatments (Fig. 1). Although G × E interactions between the mesic and xeric parental genotypes are significant for mass in the greenhouse, they are otherwise not significant, and there is no evidence of G × E in the field (Table 5). Variation among families within each ecotype was not significant (not shown).

Table 5.  Analysis of variance on fitness (spikelet number), above ground dry mass and flowering time of mesic and xeric ecotypes alone, under (a) artificially varied environmental conditions in the greenhouse and (b) natural environmental heterogeneity in the field. The condensed model treating each block as a separate environment is presented, and Hopland 2004 is omitted.
SourcedfSpikelet numberDry massFlower time
MSFMSFMSF
  1. *P<0.05; **P<0.01; ***P<0.001; ****P<0.1.

 Greenhouse
  Mesic vs. Xeric16566.93  0.212 961.39 3.69288,873.5326.551**
  Environment7687,522.17 22.199***6013.8123.060***17,046.43 5.061*
  Genotype×Environment730,970.50  1.772 260.78 5.592*** 3367.90 8.358***
  Error13917,481.20   46.64   402.95 
 Field
  Mesic vs. Xeric1490.93 13.902****  15.1126.080* 
  Environment29521.82283.383**  34.7760.007* 
  Genotype×Environment233.60  0.177   0.57 1.040 
  Error272189.61    0.55 

Indeed, in spite of the significant G × E interactions among F6 families in both the greenhouse and the field, the changes in rank order of fitness generally do not extend to the most-fit genotypes. In both field and greenhouse, there are a few genotypes that consistently have the highest fitness across most of the environments (Fig. 1). In addition to the mesic ecotype consistently outperforming the xeric in the field, a few F6 families consistently outperform the parental types in the greenhouse environments. These recombinant genotypes also showed substantial transgressive segregation (in the greenhouse) in the line cross analysis of Johansen-Morris and Latta (2006). However these genotypes generally do not outperform the mesic ecotype in the field (Fig. 1). Indeed, only one genotype showed any evidence of outperforming the mesic ecotype in the field. Although not included in the greenhouse experiment reported here, this genotype performed poorly in the greenhouse experiment we reported earlier (Johansen-Morris and Latta 2006)

Discussion

The fitness of hybrid recombinants can reveal much about the genetic basis of fitness in heterogeneous environments, and the evolution of niche breadth. If generalist genotypes are possible, then hybrid recombinants may combine the most advantageous alleles from diverged parents into broadly adapted generalist genotypes (Kawecki et al. 1997). Such recombination has been raised as a potentially important mechanism allowing the colonization of novel environments, during either speciation (Grant 1981; Arnold 1997; Rieseberg et al. 1999; Seehausen 2004) or invasion (Allard 1965; Ellstrand and Schierenbeck 2000). Conversely, if novel environments favor new combinations of alleles, then novel environments can be an important selective advantage to recombination (Bell 1982; LeNormand and Otto 2000). The pattern of G × E for fitness in hybrids between geographically distinct genotypes of A. barbata shows little evidence for specialization to the field environments. Although we find G × E interactions are significant in novel greenhouse water and nutrient treatments, here too, a few broadly adapted genotypes were observed. However, it was the novel recombinant genotypes that showed the highest adaptation to these novel greenhouse environments, suggesting an important role for recombination in adopting new niches.

SOURCES OF VARIANCE IN FITNESS

The majority of the total variation in fitness is attributable to differences among environments (Tables 1 and 4), indicating that we have presented the genotypes with a range of environments to which fitness is clearly sensitive. Not surprisingly, those plants grown with abundant water and nutrients had a higher biomass and spikelet production than those grown in the harsher environments. Fitness responded to plot within nutrient by water treatment, as well as to the treatment effects, indicating additional environmental variation that affected fitness. In the field, fitness was affected by both site and year as well as the interaction. There was also a substantial component of variation attributable to the main effect of variation among genotypes (see below).

Nonetheless, it is clear that G × E interactions affect fitness, and this effect is more pronounced in the greenhouse than in the field (Tables 1 and 4). Across greenhouse environments, genotype interacted strongly with environment, primarily the nutrient by water combination, as well as plot. G × E variance in the greenhouse was twice as great as genotype variance. By contrast, although genotype interacted strongly with both site and year to determine fitness in the field, much of this is clearly driven by the extremely low survivorship at Hopland in 2004, which produced a lack of variance among genotypes in Hopland in 2004 field trial. When Hopland, in 2004, is omitted from the results, interactions were significant only for mass, and the interaction accounts for only a small proportion of the variation, and substantially less than the main effect of genotype (Table 4). Thus, rather than indicating a trade-off in fitness across environments, the significant G × E interaction simply reflects the inability of the genotypes to survive to maturity in Hopland in 2004 field trial.

GENERALISTS OR SPECIALISTS?

Regardless of whether we include Hopland 2004 results, the mesic ecotype was consistently the most-fit genotype in the field. This result is somewhat surprising because it has been hypothesized that the xeric and the mesic ecotypes are specialized to contrasting moisture regimes, (Allard et al. 1972; Clegg and Allard 1972; Hamrick and Allard 1972). Indeed, given the strong association between the allozyme genotypes and the environment, A. barbata has been perennially held up as an example of such specialized genotypes (e.g., Lewontin 1974; Grant 1981; Endler 1986; Avise 1994; Linhart and Grant 1996; Cox 2004; Parsons 2005). Under local adaptation, each parental ecotype should exhibit higher fitness in its respective environment (Fry 1996; Kawecki and Ebert 2004), yet no clear evidence for local adaptation was found. Indeed, the fitness advantage of the mesic over the xeric ecotype extended to the artificial greenhouse environments.

Arguably, local adaptation may become apparent if the experiment were to be repeated over more field sites or growing seasons. In particular, if Sierra Foothills is in some way atypical of habitat to which the xeric ecotype is adapted, then it might explain the higher fitness of the mesic genotype in our Sierra Foothills trials. However, the Sierra Foothill site lies squarely within the region occupied by monomorphic xeric populations (Clegg and Allard 1972; Hutchinson 1982). One might alternatively argue that a rare event, such as an extreme drought might be necessary for the xeric genotype to exhibit its adaptation to the dry site. Yet the two seasons examined here contrasted markedly in the amount and timing of rainfall and in the performance of the plants. Additionally, we have explicitly manipulated the water regime (the putative selective agent) in our greenhouse experiment, and the mesic genotype has maintained its advantage throughout. Indeed, the original inference of local adaptation (Allard 1972) was based primarily on the patterns of allozyme variation, whereas experimental evidence (Jain and Rai 1980) was equivocal. Yet, Hutchinson (1982) showed that the xeric ecotype produced smaller plants and fewer spikelets at Sierra Foothills than did the mesic, and in greenhouse experiments (Hamrick and Allard 1975), the mesic genotype again produced more spikelets than the xeric genotype, both results mirrored in our own findings (Table 6). If there are conditions under which the xeric is favored by selection, they appear to be elusive, and it would be reasonable to entertain the possibility that the mesic genotype is favored across a broad range of environments.

Table 6.  Line mean correlations of fitness (spikelet production) with flowering time and mass in each trial.
Site Flower Mass
  1. *Greenhouse data from Johansen-Morris and Latta 2006.

W+N+1−0.76 0.09
W+N+2−0.83−0.13
W+N−1−0.95 0.07
W+N−2−0.87−0.09
W−N−1−0.99−0.13
W−N−2−0.99−0.18
W−N+1−0.79 0.20
W−N+2−0.96 0.38
Green house*−0.91−0.19
Sierra 2003  0.93
Sierra 2004 0.95
Hopland 2003  0.84
Hopland 2004 0.96

Among the recombinant F6 genotypes, there is more change in the rank order of fitness across environments, yet here too, there is a tendency for certain genotypes to consistently outperform the others. There is a significant family main effect in each of the condensed models, and this accounts for a much larger portion of the variance in the field experiment than does the interaction (Table 4b). Families F6–16 and F6–111 both flower particularly early and have particularly high fitness in the greenhouse (Johansen-Morris and Latta 2006). This fitness advantage seems to be consistent in the greenhouse, regardless of the water or nutrient treatment (Fig. 1), and the consistently greater fitness than either of the parents speaks to this being a result of genetic differences between the recombinants and the parents. Other families show consistently low fitness across greenhouse treatments.

We note that although correlations between families are frequently much less than one, there are no cases of negative correlation of fitness across greenhouse environments that would indicate a trade-off in performance between environments. The family means correlations across environments are an indicator of the degree to which the relative fitness of genotypes changes across environments (Bell 1991; Via 1994; Kassen and Bell 2000). These correlations are the strongest for replicate plots of the same treatment, indicating that despite the significant interaction of family with plot, little change in relative fitness is observed between plots of a particular water and nutrient treatment. Similarly, the strong correlations between fitness in one field environment, and another is indicative that the small G × E variance is not enough to substantially alter the rank order of fitness of the genotypes. Correlations are intermediate across water by nutrient treatments in the greenhouse experiment, suggesting a greater reordering of the relative fitness of genotypes. Interestingly, the weakest correlations are between greenhouse environments and field environments. This suggests that between native versus novel environments, the relative fitness of the genotypes changes substantially. When all greenhouse and field environments are pooled, the G × E interaction accounts for over 20 times as much variance in fitness as does the main effect of genotype (Table 4 [part b]).

This pattern of fitness trade-off occurring only between field and greenhouse may be related to the differing targets of selection in different environments. In prior work, we showed that fitness in the greenhouse is strongly correlated with early flowering, but only weakly correlated with aboveground dry mass (Gardner 2004; Johansen-Morris and Latta 2006; Gardner and Latta, unpubl. ms.). This result is confirmed across each of the water and nutrient treatments applied here (Table 6). In all greenhouse environments, fitness was strongly correlated with flowering time (r= 0.8 to 0.9), and generally uncorrelated with mass (Table 6). In no greenhouse environment was the correlation of fitness and mass significant, although there was a weak trend toward a higher correlation in the W−N+ treatments. In the field, spikelet number is strongly correlated with aboveground dry mass. This shift in the traits under selection is accompanied by an equally strong shift in the quantitative trait loci under selection (Gardner and Latta 2006). Loci under selection in the greenhouse have strong effects on flowering time, and are markedly different from those under selection in the field. Across field environments, the same loci are consistently under selection (for the same alleles) in all field trials. Thus the rank order of fitness seems to change most markedly across environments in which the target of selection shifts. This is intuitively appealing, because we would expect the greatest change of relative fitness to occur between the most divergent environments (Kassen and Bell 2000).

G inline image E FOR FLOWERING TIME

Flowering time appears to be much less sensitive to changes in the environment than either fitness or aboveground biomass. Just over half of the total variation is attributable to the main effect of genotype, the largest fraction of any trait, whereas only 36% of the total variation is attributable to environmental differences. There thus appears to be some plasticity of flowering time, and a small (but statistically significant) genetic variance in this plasticity. However, those families that flowered early in one environment tended to flower early in all the environments, and flowering time shows strong correlation between all environments (Table 3). This constancy of flowering time is somewhat surprising, given the correlation between fitness and flowering time (Table 6). One would expect that if there is little change in the rank order of the families across environments for flowering time then fitness would behave in a similar manner, but this was not the case. Thus, some trait(s) apart from flowering time is also affecting the number of spikelets produced, and in our earlier study we showed that fitness variation among families remains significant when flowering time is held constant (Johansen-Morris and Latta 2006). It would appear that these other (as yet unidentified) sources of fitness variation are responsible for much of the G × E interaction for fitness in the greenhouse.

HYBRID FITNESS, RECOMBINATION, AND COLONIZATION

Hybrid recombination has been hypothesized to be an important mechanism of adaptation to novel environmental conditions under a variety of contexts (Anderson and Stebbins 1954; Allard 1965, 1996; Grant 1981; Rieseberg et al. 1999; Arnold 1997; Ellstrand and Schierenbeck 2000; Barton 2001; Lexer et al. 2003b). Similarly, the selective advantage of sexual recombination is also frequently suggested to be a function of novel selective challenges (Bell 1982; Lynch and Gabriel 1983; Lenormand and Otto 2000; Kaltz and Bell 2002). Our results provide concrete evidence of this process. The greenhouse treatments imposed here represent a novel environment, with novel selection pressures (Table 6) to which A. barbata has not previously been exposed. Moreover, it is between field and greenhouse environments, that the largest G × E interaction, and greatest change in the relative fitness of the genotypes is found.

Both here and in an earlier study (Johansen Morris and Latta 2006), we found pronounced transgressive segregation in the F6 recombinants. As a result, although the average fitness of the recombinants was less than the mid-parent (indicating hybrid breakdown through the disruption of beneficial epistatic interactions), certain specific recombinant genotypes consistently outperform both parents. The transgressive segregation for fitness seen in the novel greenhouse environments was less pronounced in the field (Fig. 3). That is, families that consistently outperform the parental ecotypes in the greenhouse show less tendency to do so in the field trials. Few families emerge as superior to the parents in the field, and those that do, typically do so by a very slim margin. Thus the potential for a novel gene combination to outperform the parents appears closely tied to the novelty of the environment. The parental ecotypes are better adapted to the field relative to the novel recombinants, than to the novel environmental conditions with which they were faced in the greenhouse. Thus, hybrid recombination has created novel gene combinations from which the transgressive segregants favored in the novel environment emerge.

Figure 3.

Fitness consequences (relative spikelet production, standardized to a mid-parent mean of 1) of recombination across native and novel environments in A. barbata. Fitness of the average recombinant genotype, and of the most-fit recombinant genotype are given, as indicators of the potential fitness cost and gain following recombination. Greenhouse data from Johansen-Morris and Latta (2006) are included for comparison. Treatment notation as in Table 2.

Intriguingly, not only is this benefit of hybrid recombination greater in the novel greenhouse environment, but also the potential disruption of coadapted gene complexes appears to be less (Fig. 3). The average F6 recombinant is substantially less fit than the mid-parent in the field, imposing a cost on recombination but this cost becomes progressively less as one moves from the field (where the parental genotypes have been subject to selection for many generations) to the more novel environments where they have no prior history of selection. This implies an exogenous component to selection on epistatic interactions, in which it is the interaction between multiple loci and the environment that influences fitness. In our earlier study we found hybrid breakdown in both field and greenhouse, and argued that this reflected an endogenous epistasis (in which the selective value of gene combinations is independent of the environment). As with local adaptation (Kassen and Bell 2000), the relative contribution of endogenous versus exogenous interactions to coadaptation will likely be influenced by the divergence between the lineages being recombined (endogenous) and the difference between the environments being considered (exogenous—Fig. 3).

As A. barbata is predominantly selfing, after an initial hybridization event, recombination will again be restricted by subsequent generations of self-fertilization. This, in turn, will protect any favorable gene combinations that may have formed, resulting ultimately in the production of stabilized transgressive recombinant genotypes (Allard 1965, 1996). The likelihood of this occurring will be enhanced by dominant gene effects, which create substantial hybrid vigor in the early generation hybrids (Johansen-Morris and Latta 2006), increasing the number of recombinant genotypes that may be generated and tested by selection. Both the hybrid breakdown, which is seen in the later generation hybrids, and the transgressive segregation, which results in extreme phenotypes relative to the parents (Johansen-Morris and Latta 2006) are the result of the formation of recombinant genotypes. Our finding that novel environments reduce the costs and increase the benefits of hybrid recombination, provides strong evidence for both the role of hybridization in adapting to novel environments, and the role of novel environments in favoring genetic recombination.

Associate Editor: R. Mauricio

ACKNOWLEDGMENTS

RILs were derived from seeds kindly provided by Dr. P. Garcia. Greenhouse support was provided by C. Mills and a large group of undergraduate students. Field support was generously provided by R. Keiffer, C. Vaughn, M. Connor, and D. Labadie and the staff of Hopland Research and Extension Centre and Sierra Foothills Research and Extension Center of the Division of Agriculture and Natural Resources, University of California. This work was financially supported by the Natural Sciences and Engineering Research Council of Canada through a Post Graduate Scholarship to ADJ and through a Research Grant to RGL.

Ancillary