Heritable variation and genetic correlation of quantitative traits within and between ecotypes of Avena barbata


Robert G. Latta, Department of Biology, Dalhousie University, 1355 Oxford St Halifax, NS, Canada B3H 4J1.
Tel.: +1 (902)-494-2737; fax: +1 (902)-494-3736;
e-mail: robert.latta@dal.ca


We examined heritable variation for quantitative traits within and between naturally occurring mesic and xeric ecotypes of the slender wild oat (Avena barbata), and in 188 recombinant inbred lines derived from a cross between the ecotypes. We measured a suite of seedling and adult traits in the greenhouse, as well as performance-related traits in field sites native to the two ecotypes. Although the ecotypes were genetically diverged for most traits, few traits showed significant heritable variation within either ecotype. In contrast, considerable heritable variation was released in the recombinant progeny of the cross, and transgressive segregation was apparent in all traits. Heritabilities were substantially greater in the greenhouse than in the field, and this was associated with an increase in environmental variance in the field, rather than a decrease in genetic variance. Strong genetic correlations were evident among the recombinants, such that 22 measured traits could be well represented by only seven underlying factors, which accounted for 80% of the total variation. The primary axis of variation in the greenhouse described a trade-off between vegetative and reproductive allocation, mediated by the date of first flowering, and fitness was strongly correlated with this trade-off. Other factors in the greenhouse described variation in size and in seedling traits. Lack of correlation among these factors represents the release of multivariate trait variation through recombination. In the field, a separate axis of variation in overall performance was found for each year/site combination. Performance was significantly correlated across field environments, but not significantly correlated between greenhouse and field.


The divergence of one population into separate lineages partitions both the genetic variance of traits and the covariance among traits from within-population to between population variance and covariance (Wright, 1952; Ohta, 1982; Barton, 2001). Conversely, the secondary recombination between two such diverged lines can be both a source of novel variation for populations, on which selection can act (Barton, 2001), as well as a source of insight for the researcher. The study of hybrid recombinants can be an important tool in studies of the genetic basis of quantitative traits (Mather & Jinks, 1982; Tanksley, 1993; Mackay, 2001), of genetic correlations and the constraints they might impose (Joshi & Thompson, 1995, Hawthorne & Via, 2001; Steppan et al., 2002), of nonadditive genetic interactions (Fenster & Galloway, 2000; Demuth & Wade, 2005), of natural selection (Jordan, 1991; Nagy, 1997; Lexer et al., 2003) and of the direction of linkage (Rieseberg et al., 2003).

The release of heritable variation following the cross between diverged genotypes has been an important tool in genetic research for decades (Castle, 1921; Mather & Jinks, 1982; Falconer & MacKay, 1996; Lynch & Walsh, 1998). But this release can also influence the evolutionary potential of naturally occurring hybrids between ecotypes or between species (Anderson & Stebbins, 1954; Grant, 1981; Arnold, 1997; Barton, 2001). Depending on the allelic composition of the parental genotypes, it is possible for individual recombinant hybrid progeny to exhibit phenotypes outside the range exhibited by the parents – a phenomenon termed trangressive segregation (Rieseberg et al., 1999). The existence of such transgressive segregants implies that loci affecting the phenotype are associated in repulsion phase in the parents, such that each parent carries both plus and minus alleles for the trait (de Vincente and Tanksley, 1993; Rieseberg et al., 2003). It is these trangressive genotypes that are of interest because they may be able to outperform the parental types within the native habitats and perhaps be able to flourish in a novel environment (Rieseberg et al., 1999; Schwarzbach et al., 2001; Seehausen, 2004).

The degree to which different traits are controlled by the same (or closely linked) set of genes, estimated as the genetic correlation, can dictate how the overall phenotype can respond to selection (Falconer & MacKay, 1996), because genetically correlated traits cannot evolve independently from one another. (Lande & Arnold, 1983; Arnold, 1992). Thus, the genetic relationships among traits in a recombinant population are a guide to what type of phenotypes are possible in natural populations. If traits can freely recombine, then the novel multivariate phenotypes thus created can be an important source of adaptive potential for populations faced with novel environmental challenges (Arnold, 1997; Burger, 1999; Ellstrand & Schierenbeck, 2000; Barton, 2001). Furthermore, if adaptation to contrasting niches occurs through different loci in each niche (Kawecki et al., 1997), then hybrid recombination between diverged populations may be able to produce more broadly adapted genotypes. By contrast, if recombination cannot produce such genotypes, it implies a trade-off that constrains niche breadth (Futuyma & Moreno, 1988; Joshi & Thompson, 1995). As hybrid recombination will break up the contribution of between-population linkage disequilibrium (Ohta, 1982) to genetic correlations (Lynch & Walsh, 1998), studying genetic correlations in a line cross population provides an opportunity to focus on the genetic effects of loci that constitute longer lasting evolutionary constraints. Such constraints include pleiotropy and tight linkage, both of which restrain the production of novel trait combinations over multiple generations (see Gardner & Latta, 2007 for a fuller discussion). It is these loci that are presumably important in shaping the evolutionary trajectory of populations (Schluter, 1996).

In this paper, we investigate the release of genetic variation and the degree of constraint imposed by genetic correlations in recombinant progeny of a cross between contrasting ecotypes of Avena barbata (Pott ex Link, Poaceae). Avena barbata was among the first plants to be studied using electrophoretic techniques, and two ecotypes were identified 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 moister (mesic) regions tending to be fixed for one set of alleles and plants from the drier (xeric) regions tend to be fixed for the opposite set of alleles. This pattern of association has been found throughout California between regions (Clegg & Allard, 1972) as well as on a scale of only a few feet within these regions (Hamrick & Allard, 1972; Hamrick & Holden, 1979). Patches of polymorphism for the xeric and mesic genotypes can be found where the two habitats meet, with the frequencies of the two genotypes being correlated with the level of moisture (Hamrick & Holden, 1979; Garcia et al., 1989).

In a previous work, we have used recombinant inbred lines (RILs) between the A. barbata ecotypes to investigate nonadditive gene action through hybrid vigour and breakdown, along with transgressive segregation (Johansen-Morris & Latta, 2006), to study genotype by environment interactions (Johansen-Morris & Latta, 2008) and to identify quantitative trait loci underlying fitness variation (Gardner & Latta, 2006). This study focuses on the release of univariate and multivariate variation in the recombinant progeny of the cross. We ask first how the parental ecotypes differ in a suite of seedling and adult traits in a greenhouse environment, as well as for performance measures in the field, and whether there is evidence of variation within the ecotypes. We then contrast these results with the heritability in the RILs derived from their cross. Finally, we assess the genetic correlations among traits in the line cross population and seek to identify the underlying dimensions of the multivariate phenotype along which variation segregates. We will focus especially on possible constraints on the combinations of traits that are possible, and on whether constraints on niche breadth are evident in cross-environment correlations in fitness.

Materials and methods

We have developed a set of 188 RILs from a cross between the mesic and xeric ecotypes of Allard et al. (1972). Crossing design and methods are given in detail in Latta et al. (2004) and Gardner & Latta (2006). Self-fertilized progeny from a single hybrid F1 individual were propagated by self-fertilization to the F6 generation. All propagations were conducted in the greenhouse, and employed single-seed descent to eliminate the possibility of greenhouse selection altering the genotypes of the lines. The F6 will be 96.75% homozygous, such that individuals within a line can be treated as replicates of the same genotype. We also have available 15 families (seeds derived from a single wild individual) of the mesic genotype and six families of the xeric ecotype. These accessions were collected by Dr Pedro Garcia over several years during the mid-1980s from locations throughout Northern California (Garcia et al., 1989). Seeds were propagated by self-fertilization from the field collected seed. Because A. barbata is an obligate annual that reproduces primarily by selfing (≈98%, Clegg & Allard, 1973), such propagation created abundant seeds of genetically uniform homozygous families that have the genotype of the field collected individual. Moreover, because of this reproductive system, the number of spikelets (each of which contains two seeds) reflects lifetime reproductive success through both pollen and seed. Thus, the number of spikelets (which are retained on the plant after the seeds drop) provides an accurate measure of fitness.

We conducted common garden experiments to measure the traits and performance of each RIL, and each family of the parental ecotypes. We conducted one common garden in the greenhouse where we were able to make extensive trait measurements, as well as common garden plots located in the field. Our field sites are the same as those reported in Johansen-Morris & Latta (2006). Where that study examined fewer RILs (25) more intensively, the present study uses more extensive sampling of more RILs with fewer individuals each to examine the patterns of variation among genotypes.

Common garden greenhouse experiment

We established three randomized complete blocks. Within blocks, two seeds from each F6 RIL were sown in an individual cell, within a 72-cell plastic tray filled with a 1 : 1 : 1 mixture of peat, black earth and sand. Two replicates of 15 mesic families, containing two seeds per family per replicate (60 seeds in total), and three replicates of six xeric families, two seeds per family per replicate (36 seeds in total), were included within each randomized bock. Trays were watered and placed in a growth chamber set to provide 12 h light, at 20 °C, and 12 h dark, at 15 °C, per day. Each block was examined daily during the first week for signs of shoot emergence to assess germination time.

At 20 days, the height of each seedling was measured, and the pair of seedlings from each cell was transplanted into a 5-inch (12.5 cm) standard pot filled with 1 : 1 : 1 mixture of peat, black earth and sand and placed in one room of the Dalhousie University greenhouse in November 2001. Lighting in the greenhouse was augmented by high pressure sodium bulbs and metal halide bulbs set to illuminate the pots for 12 h from approximately 7 am to 7 pm. After 60 days, one seedling from each pot was harvested and dried (at 55 °C for 14 days) to assess early biomass accumulation. The remaining plants were watered and fertilized every 3 and 14 days, respectively, and allowed to grow to senescence. Each plant was examined daily for the emergence of the first spikelet to determine days to first flower.

At the end of May 2002, the experiment was harvested. The height, number of reproductive tillers and total spikelet number of each plant were recorded, then the tillers were separated from the leaves and each was bagged and dried at 55 °C for 2 weeks. Leaves and flowering tillers were weighed separately to assess allocation to vegetative growth and reproduction, respectively, and summed to give total mass.

Reciprocal transplant field experiment

In the fall of 2002, two fenced common gardens were established in California. The first site, Hopland Research and Extension Centre in the Coast mountains, is located within the mesic part of A. barbata’s range, and the second, at Sierra Foothills Research and Extension Centre, lies within the monomorphic xeric part of A. barbata’s range (cf. Clegg & Allard, 1972; Hutchinson, 1982). Rainfall is approximately 30% greater at Hopland than at Sierra Foothills, and we situated the garden plots to accentuate the difference in available moisture. The Hopland garden was situated in a shallow valley, whereas that at Sierra Foothills was situated on the crest of a hill to mimic micro-environmental differences between the ecotype habitats (cf. Hamrick & Holden, 1979). The A. barbata population at Hopland exhibited the morphological traits of the mesic ecotype (e.g. pubescent stems), and the population at Sierra Foothills exhibited the traits of the xeric ecotype (glabrous). These traits are genetically determined.

At each site, three randomized complete blocks were established. Nine individuals from each of six families of each parental ecotype and three seeds from each of 188 RILs were randomly assigned to the three blocks with each generation/line represented in each block. Seeds were prepared for germination on filter paper following Latta et al. (2004). Seedlings were sown in individual plastic cones (Stuewe and Sons Inc., Corvallis, OR, USA) filled with moistened germination soil mix (Sunshine Mix, SunGro, Vancouver, Canada). The plastic cones had been previously cut in half such that the bottoms could be removed upon planting, so as to allow the seedling’s roots to grow in the native soil during the experiment. The seedlings were transported to the field sites and planted in the blocks, each of which was 1.5 by 25 m in size. The surrounding vegetation was undisturbed, and seedlings were planted at 30-cm intervals. Thus, we maximized the exposure of the plants to the native environment both above and below ground. The experiment was left unattended for 7 months to allow plants to flower and reach senescence.

Once the growing season was complete and plants were senescent in early June 2003, each site was harvested. Maximum height, number of tillers and number of spikelets produced were recorded. The above ground portion of each plant was then bagged, dried at 55 °C and weighed. The field trial was repeated in the 2003–2004 growing season using the same methods.

Statistical analysis

For the univariate analysis of genetic variation of each trait, we analysed each garden trial separately. In essence, this treats measures of the same trait in different environments as two separate, but potentially correlated traits (cf. Falconer, 1952; Via & Lande, 1985), and allows us to address univariate heritability separately from genetic correlation across environments. To assess variation within and between the parental ecotypes, ecotype was treated as a fixed effect, and block and family within ecotype were treated as random factors.


A simpler model was used to measure heritability in the F6 recombinant liness.


Interactions were not tested because we used only one individual per line per block. Analyses were conducted using the Univariate procedure of spss 11.5. Variance components were estimated using the restricted maximum likelihood estimation function of spss with line (random) and block (random) as main effects. Broadsense heritabilities were estimated as the variance due to line/family differences as a proportion of the total.

We calculated two estimates of the genetic correlation for all pairwise combinations of traits among the F6 lines. We used correlations of line means for all pairs of traits. This method works well when separate individuals are measured for separate traits, as is the case for cross environment correlations. However, within each common garden trial, all traits were measured on each individual. This can inflate the genetic correlation, because line means correlations will be influenced by the environmental correlation (Lynch & Walsh, 1998). Therefore, within environments, we also estimated genetic correlations by partitioning the sums of squares and cross products attributed to among line differences, dividing by the degrees of freedom, and equating these values to their expected mean square and mean cross product respectively (Lynch & Walsh, 1998). The estimates of among line variances and covariances for the traits were then incorporated into the standard equation for determining correlation. We applied a sequential Bonferoni correction to estimates of statistical significance.

Factor analysis was carried out on line means. Factor analysis seeks to identify underlying (‘latent’) variables that are observed through their effect on a correlated suite of observed variables (as for example, when the length, width and height of a structure are indicators of an underlying latent factor ‘size’–Crespi, 1990). Eigenvectors were extracted using the Factor function in spss, and factors having eigenvalues above 1 were retained. We applied an oblique Promax rotation that allowed some correlation between the final factors. There is no a priori reason to assume that the underlying factors are orthogonal, and the use of oblique rotation facilitated interpretation of the component loading matrix in terms of the underlying factors.


Variation within and between the ecotypes

In the greenhouse, the parental ecotypes showed genetic differentiation for the majority of traits, including spikelet production, flowering time and mass allocation (Table 1). On average, the mesic ecotype flowered earlier and allocated more biomass to reproductive structures (i.e. tillers), producing nearly double the number of spikelets than the xeric ecotype. In contrast, the xeric plants were on average larger, devoting more resources to vegetative growth as opposed to reproductive structures, and flowered over 1 month later than the mesics. As well, the xerics produced fewer, albeit taller, tillers. There were negligible differences between ecotypes for early growth traits such as germination time and dry mass at 60 days, although the mesic seedlings were taller after 20 days growth.

Table 1.   Means and heritabilities of traits in Avena barbata.
 XericMesicDiff F(1,19)F6 linesCV linesCV environmentRelative meanRelative variance
  1. Mean and heritability are given separately for xeric and mesic ecotypes and for F6 recombinant inbred lines. Heritabilities that are significantly different from zero are indicated in bold. ‘Diff’ is the F-ratio for the difference between the mesic and xeric means (full anovas are given in the Appendix S1, see Supplementary material). Genetic and environmental coefficients of variation are given for the F6. The mean and variance of the F6 are given relative to the average of the parents.

  2. †Because of extremely high mortality at Hopland in 2004, few of the F6 recombinant lines contained more than one individual that survived to reproduce. As a result, it was not possible to estimate heritabilities among the F6 at Hopland in 2004.

 Germination time5.190.0145.12011.65.270.1510.0410.0961.02221.571
 Height day 2010.770.0813.650.246314.212.220.3900.0690.0861.0012.393
 Mass day 600.520.0130.501.220.390.0860.1390.4530.76513.231
 Final height149.460.168135.320.05416.51137.60.5790.1110.0950.9665.216
 Days to flower155.70.279120.10.258472.7135.810.5850.0670.0560.9852.179
 Tiller number5.320.09315.260.132436.011.830.5460.3360.3071.1504.853
 Tiller mass6.160.0349.15069.717.790.5670.2390.2091.01833.353
 Leaf mass10.410.2045.010.06233.07.380.5810.2270.1920.9574.402
 Total mass16.360.05914.19010.915.170.3320.0940.1330.99311.254
 No. spikelets275.120.049504.430.107159.0425.080.5900.2930.2441.0917.564
Sierra 2003
 Tiller number1.9400.0002.2780.0007.481.7960.2030.1920.3810.852
 No. spikelets 23.360.00025.550.0005.3216.4430.1580.3170.7320.713
Hopland 2003
 Tiller number0.9800.0851.2450.0343.261.0210.0010.0120.4920.9180.009
 No. spikelets5.8370.0119.1840.1293.095.7250.0410.1790.8630.7620.590
Sierra 2004
 Tiller number0.3330.0110.8420.08611.750.5010.0370.2061.0630.8530.744
 No. spikelets2.8440.0306.5250.0091.433.0830.1150.6711.8610.6585.842
Hopland 2004
 Tiller number0.0430.0000.2330.0503.650.045
 No. spikelets0.0850.0001.2000.0780.760.270

There was limited heritable variation among families within the two parental ecotypes (Table 1). Among xeric families, there was significant genetic variation for flowering time (H2 = 0.279) and mass of leaves (H2 = 0.204) with the remaining traits, including fitness, generally having heritabilities less than 10–15% and lower confidence limits including zero. Similarly, there was significant genetic variation for flowering time among mesic families (H2 = 0.258), as well as height at 20 days (H2 = 0.246). No heritable variation was detected for any of the remaining traits measured among mesic families (Table 1).

In the field, plants were much smaller and produced far fewer spikelets than in the greenhouse. For example, mean spikelet numbers for the F6 ranged from 0.193 (Hopland 2004) to 16.4 (Sierra 2003) in the field, as compared with over 400 in the greenhouse (Table 1). In 2003, height and total mass differed between mesic and xeric ecotypes at both field sites (Table 1). Overall, the mesic ecotype grew taller, produced slightly more tillers and accumulated more biomass than the xeric ecotype at both field sites. However, the difference between the ecotypes for spikelet production was statistically significant only at the Sierra Foothills field site (Table 1). In addition, no significant genetic variation was detected among any of the families within either parental ecotype. Estimates of the broadsense heritability within ecotypes (H2) for height, tiller number, mass and spikelet number at each field site were close to zero with confidence limits including zero (Table 1).

In 2004, mortality was substantial in the field. Dead individuals were assigned a fitness of zero, because we could be certain that no spikelets had been produced, and thus mean spikelet number includes a large proportion of individuals with zero spikelets. Mesic plants produced markedly more spikelets than xeric individuals at both Sierra and Hopland (Table 1), and much of this can be attributed to higher survival of the mesic genotype (see Johansen-Morris & Latta, 2006 for a detailed analysis of mortality).

Variation in the recombinants

After accounting for significant block effects, highly significant genetic variation was evident among the F6 lines in the greenhouse, (Table 1). For most adult traits, the among line component of variation (i.e. broadsense heritability) accounted for over 50% of the total trait variation (Table 1). The notable exception is total (adult) mass, which has a hertitability of 0.33 despite the fact that it’s component parts (tiller mass and leaf mass) have heritabilities greater than 0.55. Seedling traits had markedly lower heritabilities ranging from 0.08 for mass at 60 days to 0.39 for height at 20 days.

Heritabilities were much lower in the field than in the greenhouse and were significantly above zero only at Sierra Foothills. Because of the extremely high mortality at Hopland in 2004, most lines had zero mean fitness. As a result, estimates of heritability were strongly influenced by the few lines with more than one survivor and were statistically unreliable. We therefore omitted Hopland 2004 from the analysis. The reduction of heritability in the field derives more strongly from an increase in the environmental coefficient of variation relative to the greenhouse (0.2–0.8 vs. 0.05–0.4) than from a decrease in the genetic coefficient of variation that exhibited a similar range of values in both environments (Table 1). Heritability of field performance ranged from 4% to 25%, which represents a substantial release of variation over that within either of the parent ecotypes. In all traits, in the greenhouse, heritability in the F6 was greater than the average heritability among parental accessions by several fold (Table 1), and substantial transgressive segregation was apparent (Fig. 1). In the field, this pattern was most clearly seen at Sierra Foothills.

Figure 1.

 Boxplots depicting the release of genetic variation in the recombinant progeny of the cross. Family mean scores on each of seven factors (Table 2) describing different axes of variation are shown. X, xeric ecotype; M, mesic ecotype; F, F6 recombinant inbred lines.

Genetic correlations and underlying factors

The full correlation matrix (including both correlation estimates) is given in Appendix S1 (see Supplementary material). Genetic correlations estimated by the method of Lynch & Walsh (1998) did not differ substantially in either sign or magnitude from those estimated using family means. Thus, it appears that family means correlations are not inflated by environmental variation, and we restrict ourselves here to family means correlations.

In the greenhouse, there was a strong negative relationship (r187 = −0.55, P < 0.001) between leaf mass and tiller mass indicating a trade-off in resource allocation, and this trade-off was associated with flowering time (Fig. 2). Lines that flowered earlier tended to allocate more biomass to reproductive structures (i.e. tillers) and produced more spikelets. In contrast, the later flowering lines tended to produce more vegetative biomass and fewer tillers resulting in fewer spikelets. This is the strongest suite of associated traits, each loading on the first Factor, which accounts for 26% of the total variation among F6 family means (Table 2).

Figure 2.

 Graphical depiction of the pattern of statistically significant (following sequential Bonferroni correction) positive or negative genetic correlations among traits in the F6 recombinant inbred lines derived from the cross between mesic and xeric ecotypes. Particularly strong correlations (r > 0.7) are indicated in bold. The full correlation matrix is given in Appendix S1 (see Supplementary material).

Table 2.   Factor analysis of genetic variation in Avena barbata.
  1. Variance explained by each factor is shown along with factor loadings of each trait and the correlation among the rotated factors. The highest loading of each trait is given in bold. Other loadings > 0.25 are shown in italics. For ease of interpretation, we give an approximate name for the underlying variation captured by each factor. Note that all traits are included in one analysis, treating the same trait measured in separate environments as two (possibly correlated) traits (cf. Falconer, 1952).

(a) Eigenvectors and factor loadings
 Variance explained (%)25.80216.21411.89410.0317.2525.4635.224
 Germination time0.106−0.1750.053−0.109−0.032−0.1450.805
 Height day 20−0.1420.160−0.0400.1010.0690.1550.717
 Mass day 60−0.0030.066−0.0980.1110.6680.1450.060
 Final height−0.026−0.0420.034−0.0100.1070.962−0.016
 Days to flower0.911−0.0290.035−0.011−0.012−0.084−0.040
 Tiller number0.9610.042−0.0210.0680.0540.3930.019
 Tiller mass0.805−0.0120.033−0.0080.2740.304−0.006
 Leaf mass0.659−0.0230.024−0.0360.5670.286−0.025
 Total mass0.197−0.0480.071−0.0550.9040.035−0.033
 No. spikelets0.940−0.0320.017−0.0670.055−0.062−0.060
Sierra 2003
 Tiller number0.1140.1130.0660.7590.019−0.1650.098
 No. spikelets0.003−0.0650.0510.970−0.057−0.017−0.057
Hopland 2003
 Tiller number0.052−0.0500.8240.0040.031−0.1100.091
 No. spikelets−0.0210.0060.9180.0340.008−0.013−0.131
Sierra 2004
 Tiller number0.0950.889−0.024−0.0450.025−0.022−0.047
 No. spikelets−0.0210.9430.039−0.0230.001−0.028−0.018
 Factor ‘name’PhenologySierra 2004Hopland 2003Sierra 2003SizeHeightSeedling
(b) Correlations among Factors
 Factor 1 0.0690.1360.1490.0180.256−0.013
 Factor 2  0.2650.2490.0790.2280.085
 Factor 3   0.3030.1980.1190.130
 Factor 4    0.1060.1250.076
 Factor 5     −0.0230.111
 Factor 6      0.052

Three other factors describe variation in the greenhouse, but these account for much smaller amounts of the total genetic variation (5–7% each –Table 2). Factor 5 accounts for variation in the size of the plants, with strong loadings of total mass and seedling mass (at 60 days). Tiller and leaf mass, which are components of total mass, both loaded positively on this axis of variation, although not as strongly as on Factor 1 (and the total mass did not correlate with Factor 1). This was the only case of an association between an early growth trait (mass at 60 days) and adult traits. Germination time and height at 20 days both load on Factor 7, which seems to describe the variation in seedling traits. Factor 6 is most strongly associated with adult height, but also correlates with increased reproductive mass and decreased tiller number. It thus seems to capture variation between lines that produce fewer taller tillers, vs. more and smaller tillers.

Within each field environment in 2003 and 2004, there was a highly significant positive genetic correlation between dry mass and spikelet production (Fig. 2). The correlations were strongest at Sierra Foothills, with the correlation between total mass and spikelets estimated as approximately 0.90 in each year, and only slightly smaller at Hopland in 2003. The lack of variation among lines for fitness at Hopland in 2004 precluded calculating genetic correlations for this trial.

Between greenhouse and field environments, correlations among traits were generally small (r < 0.25) and few were statistically significant, especially after bonferoni correction. However, significant positive genetic correlations were identified among traits across the two field sites in 2003 (Fig. 2). The most notable of these traits was spikelet production, which had a cross-environment correlation of r187 = 0.325 (P < 0.001). In addition, total mass at Sierra was positively correlated with both mass and spikelet number at Hopland. Cross-year genetic correlations are only possible at the Sierra site because of mortality at Hopland in 2004, but significant correlations are still seen. Some correlation is also seen between traits expressed in both different site and year (i.e. Hopland 2003 is weakly correlated with Sierra 2004). Within each field trial, all three size measures and spikelet number loaded heavily on the same Factor, with Factors 2–4 describing performance variation in Sierra 2004, Hopland 2003 and Sierra 2003 respectively. These axes of variation account for 10–16% of the total variation each (Table 2). However, these three factors were also somewhat correlated (Table 2), reflecting the positive correlation of performance across environments. Few of the other factors were correlated with each other.


In this study, we have described the release of genetic variation in RILs made between naturally occurring ecotypes. The results document both the release of considerable variation in the recombinants, as well as several pronounced genetic correlations. Thus, hybrid recombination greatly increases the potential for adaptive evolution, through increases in both univariate and multivariate genetic variation. Some constraint is evident, but as many of the correlations are positive, our findings suggest that broadly adapted genotypes may be possible.

Release of heritable variation

Our first finding is the near absence of heritable variation within both the mesic and xeric ecotypes. These ecotypes were first recognized based upon fixed differences at allozyme loci, such that there was no polymorphism within the ‘pure’ ecotypes, and variation was observed only at zones of contact (Clegg & Allard, 1972). The families used here were examples of such pure ecotypes, collected from locations in Northern California between 1984 and 1989. In our mapping studies (Gardner & Latta, 2006), we found complete monomorphism at all but one of over 200 AFLP markers surveyed prior to constructing the genetic map of A. barbata. This monomorphism extends to most of the quantitative traits (Table 1), indicating that the mesic and xeric ecotypes can be thought of as true-breeding lines in the tradition of line cross analyses (cf. Mather & Jinks, 1982). The ecotypes are diverged for a wide variety of growth and phenological traits, but few of these traits show any variation within ecotypes, and thus the potential for response to selection within the ecotypes is quite limited.

The pattern of divergence (Table 1) mirrors that seen in early quantitative genetic studies of A. barbata. In a greenhouse study, Hamrick & Allard (1975) reported earlier flowering and greater tiller and seed production by the mesic genotype, a divergence mirrored here. Hutchinson (1982) also found earlier flowering of the mesic along with greater field performance than the xeric ecotype. Latta et al. (2004) extended these findings by showing that the earlier germination and height of the mesic seedlings were accompanied by greater seed size, and resulted in a considerable competitive advantage to the mesic ecotype under crowded conditions.

The recombinants exhibit greatly increased genetic variation and heritabilities for all of the traits studied here. Only two traits – seedling mass at 60 days in the greenhouse and tiller number at Hopland in 2003 – lacked significant heritable variation. Thus, the most obvious consequence of the cross is the greatly increased evolutionary potential among the recombinants, which are therefore much more capable of responding to natural selection. Line crosses between true-breeding lines always have released variation and have been a staple of textbooks on genetic variation in domestic or laboratory species (Mather and Jinks, 1992; Falconer & MacKay, 1996; Kearsey & Pooni, 1996; Lynch & Walsh, 1998). Of particular recent interest is the occurrence of transgressive segregation in recombinant progeny (Rieseberg et al., 1999; Seehausen, 2004). This is the occurrence of recombinant genotypes with trait values that fall outside the range of values exhibited by the parental genotypes. Each of the major axes of variation in A. barbata exhibits pronounced transgressive segregation (Fig. 1). This increased variation thus permits a response to selection not just between the trait values of the parental ecotypes, but further allows adaptation to transcend these boundaries into novel phenotypes.

Heritabilities were substantially greater in the greenhouse than in the field. There are several hypotheses regarding the effect on heritability of novel (Holloway et al., 1990), stressful (Hoffmann & Merila, 1999) or heterogeneous environments (Weigensberg & Roff, 1996). Several such hypotheses assume that novel environments are more stressful (Hoffmann & Merila, 1999), which is not the case here – the greenhouse environment is both novel, but also considerably less stressful than the field. Reduced heritability in the field can theoretically be due either to a greater environmental variance in the field, or greater genetic variance in the greenhouse (Roff, 1997; Hoffmann & Merila, 1999). Simons & Roff (1994) reported that both phenomena were responsible for the increased heritability in Gryllus firmus under laboratory conditions. Increased genetic variance in the greenhouse can come about if there are genes that become expressed only when moved to the novel greenhouse environment (Holloway et al., 1990; Kawecki, 1995). However, in A. barbata, the genetic coefficients of variation are similar in both greenhouse and field, whereas the field environment exhibits markedly greater environmental variance. Thus, trait expression in the field (at least for the performance traits measured here) is strongly buffeted by the more heterogeneous field environment. This increased environmental variance will mask heritable variation and thus reduce the capacity to respond to natural selection.

Genetic correlations

Genetic correlations give insight into both the constraints upon the range of multi-trait phenotypes that can exist, as well as the major underlying axes of variation along which variation occurs. In the greenhouse, the major axis of genetic variation shows a strong trade-off between vegetative and reproductive biomass allocation, which is mediated by date of first flowering. The strong negative correlation between vegetative and reproductive allocation resulted in substantially less genetic variation for total size than for the components of mass individually. That is to say, genetic variation is greater for the allocation of mass than it is for the accumulation of mass. Indeed, total mass is more strongly correlated with juvenile mass (at 60 days) than it is with the principle component that captures the relationship between flowering time and fitness.

In the field, the correlations among traits are quite different from those in the greenhouse. A single underlying factor seems to capture most of the variation within each field trial. Height, tiller number, mass and spikelet number are all strongly correlated in the field (Fig. 2), and all load strongly on the same factor within each trial (Table 2). This is in contrast to the greenhouse, where only tiller and spikelet numbers are correlated and the four traits load on three separate factors: phenology (Factor 1), size (Factor 5) and height (Factor 6). Thus, the contrasting conditions experienced in the greenhouse vs. the field produce a change in not only the heritability, but also the pattern of association among traits.

This is in contrast to correlations among floral measurements in Raphanus sativa reported by Conner et al. (2003). In that study, common principle components were found among floral traits in both greenhouse and field, such that the change in heritability was largely a matter of the amount, rather than the axis of variation. The difference between the two results likely stems from two sources. First, floral traits are likely more functionally and developmentally integrated than are general performance measures such as height and mass. Second, the field environment induced a much greater variance in total size than was evident in the greenhouse. Both the genetic and environmental coefficients of variance for mass (0.094 and 0.133, respectively) were substantially lower in the greenhouse than the field (0.159–0.518 and 0.771–1.87 respectively). This implies that much of the variation in the more challenging field environment reflected differences among the genotypes in their ability to acquire resources, and this appears to have created a positive correlation among mass, height and tiller number (van Noordwijk & de Jong, 1986; Houle, 1991; de Jong, 1993). In the more benign greenhouse environment (as judged by the vastly greater mass and spikelet numbers –Table 1), it seems that a reduced variation of resource acquisition has allowed a greater variation in allocation patterns to dominate the genetic correlations.

The pattern of correlations between traits does to some degree mimic the divergence between the ecotypes. The xeric ecotype flowered later than the mesic, and consequently had higher vegetative mass, lower reproductive allocation and fewer seeds in the greenhouse. These results are consistent with earlier observations by Hamrick & Allard (1975) and Hutchinson (1982). Moreover, the major axes of variation among the recombinants (as assessed by the eigenvalues of the factors) summarize those traits that also exhibit the largest difference between the parental ecotypes. For example, days to flowering, tiller and leaf mass, tiller number and spikelet number in the greenhouse all show large differences (relative to the mid-parent) between the mesic and xeric ecotypes (Table 1), and these are the traits that load primarily on the first factor (Table 2). The second factor (performance at Sierra Foothills in 2004) also captures variation in traits for which the parents are strongly diverged. In contrast, Factors 5–7 capture less of the total variation in the progeny and summarize traits for which there is less divergence between the ecotypes. This pattern is suggestive of the mesic and xeric ecotypes having diverged along the axis of variation for which there is the greatest genetic variance (cf. Schluter, 1996).

The cross released considerable genetic variation, not only in the individual traits (Table 1) but also in the combinations of traits. As two ecotypes can only diverge in one multivariate dimension (by definition), the observation of many uncorrelated axes of variation in the RILs indicates the release of multivariate variation – that is, novel variation in the combinations of trait values. Such recombination of traits has been documented earlier in ecological traits (Latta et al., 2004), most notably root allocation and competitive ability. Indeed, aside from the trade-off between vegetative and reproductive allocation, there are relatively few strong genetic correlations that would impose a marked constraint on the range of multivariate phenotypes that can be realized by A. barbata. This release of variation is a critical component of the adaptive potential of hybridization in creating novel genotypes for novel ecological situations, whether during speciation (Grant, 1981; Arnold, 1997; Rieseberg et al., 1999) or colonization (Allard, 1965; Ellstrand & Schierenbeck, 2000). In A. barbata, we have shown that recombinants can have greater fitness than the parental ecotypes in the novel greenhouse environments (i.e. there is transgressive segregation for fitness –Johansen-Morris & Latta, 2006, 2008). The advantage of recombinants over the parents is less pronounced in the field consistent with the hypothesis that recombination is a key factor in the adaptation to novel conditions (Burger, 1999; LeNormand & Otto, 2000).

Targets of selection and niche breadth

Formal analysis of natural selection on these traits using multiple regression (Lande & Arnold, 1983;Philippi, 1993), path analysis (Mitchell, 1993) or structural equation modelling (Crespi, 1990) is beyond the scope of the present study that focuses on the release of variation following hybrid recombination. Nevertheless, spikelet number is a strong indicator of fitness, because each spikelet contains two selfed seeds. As the plants are selfing annuals, spikelet number accounts for lifetime reproductive success through both male and female functions. Thus, the association of spikelet production with other traits gives a reasonable first indication of the targets of natural selection.

Under benign greenhouse conditions, it appears that those genotypes making an early switch to reproductive mode are able to produce more spikelets before senescing than other genotypes that flower later. We did not impose an artificial termination of the experiment, but rather plants had all senesced by the time of harvest. In the greenhouse, size had relatively little influence on spikelet number. In contrast, spikelet number in the field is strongly correlated with plant size, especially mass. Although we could not monitor our field sites for date of first flowering, and therefore we cannot know what relationship this trait has with fitness in the field, the change in selection intensity on mass indicates a substantial shift of selection pressure in the field compared with the greenhouse. The field environment is considerably more stressful than the greenhouse, showing considerable mortality in 2004 (Johansen-Morris & Latta, 2006). Thus, we might hypothesize that selection in the field favours those plants that allocate biomass into resource acquisition (vegetative growth) rather than to early flowering (as in the greenhouse), and this would favour those plants that become larger before flowering, rather than those that simply flower soonest. Our genetic correlations between traits and spikelet production are concordant with our earlier results mapping quantitative trait loci (QTL) for fitness (Gardner & Latta, 2006), which revealed that QTL for plant size were also under selection in the field, but not in the greenhouse, where QTL for flowering time are under selection.

Because of the separate targets of selection in the greenhouse and field, fitness is poorly correlated between greenhouse and field environments. However, although each field trial tended to load on a separate factor (Table 2a), there were weakly positive correlations of fitness between field environments (Table 2b, Fig. 2). Indeed, given the extensive rather than intensive sampling scheme in this study, much of the scatter in the correlations among environments may be because of the error in estimating each line mean in each environment, and thus it seems possible that the cross-environment correlations are in fact underestimated. Moreover, the mesic ecotype appears to consistently outperform the xeric across all environments. A full discussion of issues related to G × E interactions in these genotypes is presented elsewhere: Gardner & Latta (2006) found that the selection tended to favour the same alleles in each field environment, whereas Johansen-Morris & Latta (2008) found no genotype-by-environment interaction between field trials. Particularly relevant in the current results is the absence of any negative correlations between performance across environments (Table 2), which would be indicative of a trade-off between performance in one environment and that in another. Such trade-offs are an important hypothesis for the limitation to niche breadth of genotypes, and their absence suggests that a single broadly adapted genotype can have high fitness in both field environments. It remains to be determined whether such a generalist genotype is indeed spreading in the field. However, anecdotal comparisons of visible genetic markers in extant populations with historical records (e.g. Clegg & Allard, 1972) and herbarium specimens suggests that some form of genetic change has occurred within these populations and may be continuing in the present day (Latta et al., 2007).


We are very grateful to Dr. Pedro Garcia who provided the original seed stocks from which these lines were developed. We are also grateful to Hopland Research and Extension Center and Sierra Foothills Research and Extension Center for supporting our field trials. Mark Johnston and Christophe Herbinger provided helpful comments on the work and Carman Mills and Joanna MacKenzie provided valuable greenhouse assistance. This work was funded by a Post Graduate Scholarship to KMG and a Discovery Grant to RGL.