SEARCH

SEARCH BY CITATION

Keywords:

  • adaptation;
  • barley;
  • crop wild relatives;
  • domestication;
  • population differentiation;
  • QSTFST comparison

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Phenotypic variation in natural populations is the outcome of the joint effects of environmentally induced adaptations and neutral processes on the genetic architecture of quantitative traits. In this study, we examined the role of adaptation in shaping wild barley phenotypic variation along different environmental gradients. Detailed phenotyping of 164 wild barley (Hordeum spontaneum) accessions from Israel (of the Barley1K collection) and 18 cultivated barley (H. vulgare) varieties was conducted in common garden field trials. Cluster analysis based on phenotypic data indicated that wild barley in this region can be differentiated into three ecotypes in accordance with their ecogeographical distribution: north, coast and desert. Population differentiation (QST) for each trait was estimated using a hierarchical Bayesian model and compared to neutral differentiation (FST) based on 42 microsatellite markers. This analysis indicated that the three clusters diverged in morphological but not in reproductive characteristics. To address the issue of phenotypic variation along environmental gradients, climatic and soil gradients were compared with each of the measured traits given the geographical distance between sampling sites using a partial Mantel test. Flowering time and plant growth were found to be differentially correlated with climatic and soil characteristic gradients, respectively. The H. vulgare varieties were superior to the H. spontaneum accessions in yield components, yet resembled the Mediterranean types in vegetative characteristics and flowering time, which may indicate the geographical origin of domesticated barley.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

…Man selects only for his own good; Nature only for that of the being which she tends…

(Darwin, 1859)

Local adaptation may lead to differentiation between wild populations over time if migration is low and selection is sufficiently strong (Endler, 1986; Charlesworth et al., 1997). Genetic drift may also result in genetic differentiation, especially in plants reproducing mainly by self-fertilization (Wright,1943; Lande, 1976; Ellstrand & Elam, 1993). Distinguishing between these two processes as the cause for differentiation in quantitative traits is problematic and considered as a major challenge in the study of adaptation (Turelli, 1988). Nevertheless, a comparison of genetic differentiation calculated for neutral markers (FST; Weir & Cockerham, 1984) and quantitative traits (QST; Spitze, 1993) enables distinguishing between the effects of natural selection and genetic drift on the differentiation of natural populations in various species (Spitze, 1993; Palo et al., 2003; Volis, 2011). Higher differentiation of a quantitative trait as compared to neutral loci (QST > FST) implies adaptive differentiation in the respective trait between populations. On the other hand, a lower differentiation of a quantitative trait compared with neutral loci (QST < FST) suggests convergent evolution, implying that similar phenotypic values are favoured in different environments. When the distributions of genetic and phenotypic variation are not statistically different, the effect of genetic drift and natural selection cannot be distinguished for the respective quantitative trait (Merilä & Crnokrak, 2001). The estimation of QST distributions is affected by the precision of method and the level of introduced bias (Whitlock & Guillaume, 2009). A Bayesian framework has an advantage over the traditional analysis of variance because QST estimates are obtained directly from the posterior distribution, which is the product of a likelihood function and a prior distribution. This approach allows estimating the effect of genetic variation on phenotypic distributions at higher accuracy, especially when the data are limited due to a small number of populations and replicates or extended missing data (Palo et al., 2003; O'Hara & Merilä, 2005).

Environmental characteristics affect population genetic variation along spatial gradients. This may result in gradual differentiation that varies between traits (Volis et al., 2002; Hübner et al., 2009). These environmental factors may differentially affect migration rates, the strength of selection and genetic drift in each part of the distribution range in accordance with natural barriers or environmental extremes. However, the interplay between environmental and demographic factors influences the population genetic structure and our ability to detect it (Schmidt et al., 2008; Hughes, 2011). Therefore, clinal or clustered patterns of variation along environmental gradients should be considered when mapping adaptive trait loci in wild populations.

Barley (Hordeum vulgare) is one of the most important crops cultivated worldwide, especially in poor regions where it is used as a staple food for both animals and humans (Ullrich, 2011). Nevertheless, the artificial selection it experienced since domestication approximately 12 000 years ago (Salamini et al., 2002) has narrowed the genetic variation available for breeding robust varieties with stable yields under changing environmental conditions (Nevo, 1992; Caldwell et al., 2006). This bottleneck effect on the genetic variation in the cultivated barley gene pool has led to proposals to use its wild relative (H. spontaneum) as a valuable source of germplasm for breeding purposes. Although the wide genetic variation in natural populations holds great promise for plant breeding (Ellis et al., 2000; Zamir, 2001), linkage drag caused by the effect of undesired alleles in the wild gene pool challenges the introgression of novel alleles into breeding material. Moreover, it was previously proposed (Waugh et al., 2009) that natural populations can be used for fine mapping of beneficial alleles. However, one limitation in implicating such genetic material for association studies is the possible confounding effect caused by the interplay of demographic and adaptive forces. While adaptation may be the true force underlying causative polymorphism, demographic processes may add to the association analysis a statistical noise of genetic background. Wild populations in their natural habitats are subject to natural selection and demographic effects such as migration, genetic drift and population bottleneck. Wild ancestor populations of modern crops are affected, in addition to other evolutionary forces, by introgressions from the cultivated gene pool (Ellstrand, 2003). Their study usually involves technical adjustments of the genotyping platform based on different genetic markers, which are powerful but subjected to different mutation models, and a phenotyping procedure that introduces bias and limits the interpretation of the results (Jana & Pietrzak, 1988; Hübner et al., 2012). Nevertheless, an advanced and detailed sampling scheme combined with a well-established genotyping platform can improve the reliability of the genetic associations and enable to draw credible conclusions about the genetic basis of adaptation in plants.

In this study, we focus on differentiation for both vegetative and reproductive quantitative traits along environmental gradients. In particular, we were interested in studying (i) the phenotypic distribution of wild barley along environmental gradients in Israel, (ii) the genetic and phenotypic divergence between ecotypes, and (iii) the phenotypic differentiation between wild and cultivated barley. Based on the correspondence between quantitative traits and the ecogeographical regions, we found that three major ecotypes can be distinguished among wild barley populations. Although the combination of all quantitative traits together differentiated the two Mediterranean ecotypes from the desert ecotype, analysis of each trait separately revealed different and sometime opposite mode of divergence. Reproductive traits were less differentiated among the three wild ecotypes than vegetative traits, and the opposite pattern was found in comparisons of wild and cultivated barley. Environmental gradients were also differentially correlated with reproductive and vegetative variation. This study is of interest in the context of wild plant adaptation and the use of exotic germplasm for deciphering the genetic and mechanistic basis of complex traits relevant to modern crop productivity under threat of climatic changes.

Materials and methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Plant material

Wild barley seeds were obtained from a single seed descent propagation of the Barley1K collection, which was sampled during the spring of 2007 along the distribution range of wild barley in Israel in a hierarchical sampling mode (Hübner et al., 2009; https://sites.google.com/site/barley1k/). This collection consists of more than 1000 accessions sampled at five microsites in each of 51 sampling sites. To obtain a manageable and reliable representation of this wide collection, representative accessions were randomly chosen from each sampling site. In total, 164 accessions from 47 sites were phenotyped under common garden conditions in four replicates (656 plants) to study the phenotypic diversity of wild barley from Israel. The cultivated collection consisted of 18 accessions divided into 10 two-rowed and eight six-rowed cultivars of the worldwide barley core collection (Mano et al., 2001; Haseneyer et al., 2010). A list of wild and cultivated barley accessions used in this study is shown in Table S1.

Quantitative genetic data

A field trial was conducted in a nethouse at the experimental station in Rehovot. The seed germination and growing procedure in all experiments included sowing in trays covered with tin foil and saran wrap (to maintain darkness and humidity) and kept at 4 °C for 21 days to break dormancy and to induce tillering. The plants were transplanted as single plants to an irrigated field in a randomized plot design with 0.3 m distance between plants and 1 m between rows. Plants were characterized on a daily basis throughout the growing season (November to May) from planting to harvesting and included the following measurements: flowering time, plant height, number of spikes, spike length, spike weight, vegetative dry weight, grains per spike, grain size and grain dry weight. All traits measured are characterized by a high reproducibility due to a high heritability ranging between 0.49 (GrainPerSpike) to 0.9 (Flowering, GrainSize) calculated from experiments conducted in different regions (Talamè et al., 2004; Volis et al., 2005). This enabled obtaining reliable results from a one-year common garden experiment conducted under optimal conditions for screening the optimal performance of accessions as measurement for the additive genetic potential (Atlin et al., 2000; Ghalambor et al., 2007).

Flowering time (Flowering) was measured as the number of days from planting to the emergence of the first awns from the main tiller; plant height (Height) was measured from the plant base to the end of the tallest spike excluding awns. To optimize grain filling and to avoid spontaneous seed shattering, the timing for harvest of each accession was determined as the day that the first 10 spikes were dried but had not shattered seeds. These criteria resulted with full maturation for at least the five spikes used for grain and spike phenotype without losing seeds. At the day of harvest, number of spikes per plant (NumSpikes) was scored, and five spikes of each plant were separated to determine average spike length. Next, harvested plants and spikes were oven dried for 48 h at 37 and 70 °C, respectively, followed by measurements of the spike dry weight (SpikeWeight) and vegetative dry weight (VegDryWeight). The five spikes were manually threshed (to avoid breakage of seeds), and grains per spike (GrainsPerSpike) and grain surface (GrainSize) were measured by image analysis using the imagej 1.45 (http://imagej.nih.gov). Raw data were entered into a database and inspected for typing errors, data authenticity and outlier removal prior to analysis. Data deposited in the Dryad repository: doi:10.5061/dryad.ph35f

Statistical analyses

Statistical analyses were performed in r v.2.10 (http://www.r-project.org/). If more than one replicate was missing, all data from the corresponding accession were excluded. The missing replicate out of four (< 5% in all data set) was imputed by the mean value of the remaining three replicates of the same accession. The data were normalized by scaling trait scores to zero mean and unit variance to obtain a balanced effect of phenotype profiles for principal components analysis. Bayesian model–based classification of the wild collection normalized scores was conducted with the mclust package (Fraley & Raftery, 2007) in r to define phenotypic clusters across the Barley1K collection sites (Hübner et al., 2009). Model-based clustering approaches are preferred over heuristic or interpreting methods such as PCA, because they enable to statistically test the level of classification in multivariate data. Clusters of accessions are detected in mclust using EM (expectation-maximization) algorithm which fits the data to different spatial models. Then, a maximum likelihood approach is used to estimate the best fit, and a Bayesian information criterion (BIC) is applied for model selection. The number of clusters tested ranged from one to nine. A pairwise correlation matrix between all nine traits was calculated with the ltm package (Rizopoulos, 2006) and visualized with the correlogram r packages (Friendly, 2002). Descriptive statistics and the Mann–Whitney test for difference between each wild cluster and the cultivated barley were conducted for each of the nine traits. A partial Mantel test (Clarke, 1993) was conducted using the vegan package (Oksanen et al., 2008) in r to test the correlation (rM) between each trait (t) and each of the four ecogeographical (e) parameters [bulk density (BD), midday temperature in January (MDT1), mean annual rainfall (Rain) and Elevation] given the geographical distance (g) calculated between sampling sites based on X–Y coordinates [rM (t,e|g)]. A spatial presentation of the trait scores and ecogeographical parameters was performed in quantumgis 1.7.1 (http://www.qgis.org) by producing an inverse distance weighting interpolated layer (Burrough & Mcdonnell, 1998) with 300 columns and rows based on the average trait score in each sampling site.

Quantitative and population genetic analyses

The genotype for the whole Barley1K collection using 42 expressed-sequence-tag-derived microsatellite markers (EST-SSR) was described previously (Thiel et al. 2003; Hübner et al., 2009). Pairwise quantitative genetic divergence between wild clusters was estimated for each trait using a hierarchical Bayesian model. This approach fits the hierarchical nature of the Barley1K collection and enabled to estimate QST directly from the posterior distribution of phenotypes scored for natural populations grown under common garden conditions (Palo et al., 2003). The effect of genetic variation is articulated with high accuracy even when the data are limited due to small number of populations and/or replicates, or considerable missing data, because the posterior distribution is estimated using a likelihood function that fits the data to a prior independent distribution. Two steps were included in the hierarchical Bayesian model; hence, each of the nine traits (y) was assumed to follow a normal distribution with mean μ for each individual i within cluster c with mean φ for the respective cluster:

  • display math
  • display math

Variance estimates were considered to represent a solely additive genetic effect although multiplicative effects of traits, scaled variance effect of QTL and dominance effect may all influence the estimated variance distributions (Korol et al., 1996). Nevertheless, even when dominance effects are present, QST > FST would still represent good evidence for adaptive differentiation, because dominance effects tend to decrease the QST estimates. On the other hand, QST < FST may not indicate adaptive convergence and the null hypothesis of neutrality could not be easily rejected for the same reason (Goudet & Büchi, 2006). Therefore, the test is conservative with respect to adaptive phenotypic divergence. Random effects were modelled as the quantitative genetic variances:

  • display math

and

  • display math

where VGBt is the cluster effect and VGWt is the additive individual effect within the cluster. Variance components were estimated using a Bayesian model and a Gibbs sampler implemented in winbugs (Lunn et al., 2000). Flat prior distributions were used in the estimation due to little knowledge about the true population distributions:

  • display math

and variation was expected to follow a gamma distribution with little prior knowledge:

  • display math

Markov chain Monte Carlo algorithm was used for iterative data sampling and was updated every 10 cycles from two chains iterated 50 000 times after a burn-in period of 10 000 iterations giving a total of 5000 draws of each chain. Components of variance drawn from the posterior distribution were used to calculate QST for each trait (Spitze, 1993; Merilä & Crnokrak, 2001):

  • display math

Pairwise genetic differentiation for molecular markers was computed as FST and GST statistics using microsatellite analyzer 4.05 (Dieringer & Schlötterer, 2003) for each locus and globally (over all marker loci) with 10 000 permutations. Both indices gave similar results, and therefore, only FST was used in the following analyses. Normal distributions were constructed for QST and FST based on the calculated means and standard deviations. Divergence was tested by comparing QST and FST distributions using 10 000 permutations of each distribution.

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

Phenotypic variation within the Barley1K collection

A representative sample of the Barley1K collection (Hübner et al., 2009) was investigated under field conditions to obtain a detailed morphological and reproductive phenotype. Overall, high level of morphological and phenological variation was observed in this sample for the nine traits scored in the field experiment (Table 1, Table S2). The average coefficient of trait variance among accessions replicates was 26% and ranged from 6% (flowering time) to 53% (grain dry weight). The average coefficient of trait variance between accessions was 60% and ranged from 11% (flowering time) to 73% (grain dry weight). The low experimental variance effect compared with the additive genetic effect indicates the reliability and reproducibility of the experimental design. A pairwise correlation test between traits was significant (= 0.15–0.69, < 0.05) for most comparisons (78%) and indicated that vegetative and reproductive plant traits work under trade-off constrains (flowering time and number of spikes; = 0.24 < 0.05) or have a pleiotropic effect on different morphological types (vegetative dry weight with spike length, number of spikes and plant height; = 0.61–0.66, P < 0.05, Fig. S1).

Table 1. Means and standard deviations (in parentheses) of each trait in cultivated barley (Hordeum vulgare) and the three wild (H. spontaneum) clusters
 CultivatedDesertCoastNorth
  1. Significant divergence between cultivated barley and each of the three wild clusters obtained from Mann–Whitney test for each trait is indicated (*0.05; **0.01; ***0.0001).

Sample size18455762
Flowering (days)66.1 (6.46)57.5 (4.47)***64.3 (4.21)63.5 (5.06)
Height (cm)123 (11.8)107 (14.5)***125 (11.8)122.6 (11.7)
Spike length (cm)11.2 (1.09)9.51 (1.16)***11.1 (1.30)11.02 (1.06)
No. spikes47.5 (19.3)44.2 (13.5)43.1 (11.3)38.4 (10.3)
Spike weight (g)2.91 (1.16)1.02 (0.43)***1.55 (0.36)***1.63 (0.41)***
Veg dry weight (g)82.3 (29)39.5 (15.9)***61.8 (21.4)*58.1 (18.8)**
Grains per spike42.9 (21.2)15.8 (4.07)***16.6 (2.94)***15.01 (3.11)***
Grain size (cm2)0.29 (0.03)0.21 (0.04)***0.26 (0.03)**0.27 (0.04)*
Grains dry weight (g)105 (39.2)17.9 (13.9)***24.6 (9.79)***20.02 (8.25)***

The number of phenotypic clusters found in the Barley1K collection was inferred from a model-based analysis using an EM algorithm and maximum likelihood estimates to fit the best model to the data. All multivariate ellipsoid models indicated that three clusters best represent the Barley1K collection based on the phenotypic data (Fig. 1). Assignment of accessions to clusters was performed using the k-means analysis for = 3 as was inferred from the model-based analysis. Most accessions (137) were assigned to clusters in accordance with their geographical location. When contradictive assignments of accessions from the same sampling site to different clusters were noticed, a majority rule was used for correction based on the assignment of all accessions from the corresponding sampling site. This clustering and assignment procedure enabled obtaining a reliable classification mode of individuals because clustering was based on both geographical and morphological information. Each accession was marked according to its assignment to cluster, and principal components analysis was conducted (Fig. 2) using all nine traits normalized to equalize the contribution of each trait (see Methods). The first three principal components explained 76% of the variation. A pairwise manova test was conducted between each pair of clusters using the three PCs. This analysis revealed a significant differentiation of the desert cluster from the northern (F1 = 48.2, = 2.49 × 10−7) and coastal clusters (F1 = 37.7, = 3.29 × 10−7), and to a lesser extent between the northern and coastal clusters (F1 = 4.41, = 0.005). These three ecotypes correspond to the northern Galilee area and Golan Heights, the coast Mediterranean regions and the desert along the Jordan valley and the southern Negev.

image

Figure 1. Clustering mode of the wild barley collection based on their phenotypic data using an expectation-maximization (EM) algorithm. Model-based approach using multivariate ellipsoidal models (EEV – equal volume, equal shape; VEV – equal volume, variable shape; VVV – variable volume, variable shape) to define the number of clusters best representing the data based on the Bayesian information criterion (BIC).

Download figure to PowerPoint

image

Figure 2. Principal components analysis of the normalized phenotypic scores for wild barley accessions based on common garden experiment phenotypes. Ecogeographical origin for each accession based on assignment analysis is indicated by different characters (circle – desert; triangle – coast; diamond – north). Plots designate PC1 against PC2 (a), PC1 against PC3 (b), and PC2 against PC3 (c).

Download figure to PowerPoint

Levels of phenotypic divergence

Representative sampling sites of each cluster were selected for further analysis, and all accessions from these sites were used for the QST/FST tests. This reduction step was based solely on the geographical location of the selected sampling sites. Although it may introduce some bias to the analysis, selecting a balanced representation of the three clusters and reducing the confounding effect introduced by the mixture of phenotypes in hybrid zones are expected to yield a reliable and robust indication of adaptive differentiation between ecogeographical regions. Altogether, 83 accessions (26 – desert; 29 – coast; 28 – north) were used for the pairwise QST and FST estimates. Pairwise FST varied much between loci (Table 2) and indicated a stronger genetic differentiation between the coastal and desert clusters than between the northern cluster and the desert or coastal clusters, respectively. QST estimates for most traits were characterized by a large standard deviation as obtained from the MCMC sampling procedure of the prior distribution due to little prior knowledge (Table 3). In the QSTFST comparison, the three clusters differed significantly from each other for grain size and spike weight, which are highly correlated (= 0.67, P = 0.0001), implying differentiation in grain characteristics. Although the mean grain dry weight was different between the three clusters, the null hypothesis of neutrality could not be rejected (= 0.3). Flowering time, plant height and vegetative dry weight were more divergent between the desert and each of the north and coastal clusters than between the north and the coastal clusters based on the QST analysis (Table 3), although the null hypothesis of neutrality could not be rejected (= 0.22, = 24, = 0.27). Reproductive traits such as number of spikes and number of grains per spike were similar in different clusters (QST < FST), and the null hypothesis for neutrality could not be rejected (Table 3). Overall, this analysis shows the differential contribution of the analysed traits to the divergence to three clusters, with spike weight and grain size being the sole traits for which the null hypothesis for neutrality was rejected.

Table 2. Global pairwise FST between clusters and standard deviation (in parentheses) over all loci for all accessions (upper) and for the clusters' representatives (lower)
 DesertCoastNorth
Desert 0.08 (0.08)0.05 (0.04)
Coast0.16 (0.13) 0.04 (0.03)
North0.10 (0.08)0.09 (0.07) 
Table 3. Permutation test of QSTFST for divergence in each trait. Among- and within (additive genetic)-population variance components, and QST estimates (mean and SD) between clusters from the hierarchical Bayesian analysis for each trait and the corresponding FST value
TraitCoast–North (FST = 0.09)Desert–Coast (FST = 0.16)Desert–North (FST = 0.10)
σ2 Amongσ2 Within Q ST P-valueσ2 Amongσ2 Within Q ST P-valueσ2 Amongσ2 Within Q ST P-value
  1. Significantly differentiated traits are marked with three asterisks.

Flowering (days)1.0222.10.04 (0.06)0.2713.317.90.31 (0.20)0.2411.0523.00.25 (0.19)0.22
Grain dry weight (g)14.141040.15 (0.08)0.3013.382060.13 (0.05)0.1911.281750.11 (0.02)0.32
Grain size (cm2)***0.850.030.90 (0.08)0.0010.840.040.89 (0.09)0.0010.830.030.90 (0.09)0.001
Grains per spike2.608.010.09 (0.06)0.382.2711.80.13 (0.13)0.440.9510.70.08 (0.09)0.39
Height (cm)2.031360.04 (0.13) 0.3411151560.55 (0.35) 0.1076.281230.40 (0.30) 0.15
No. spikes4.021140.05 (0.08) 0.311.201290.01 (0.02)0.075.071200.05 (0.09) 0.33
Spike length (cm)0.861.380.29 (0.18) 0.151.761.410.42 (0.20)0.131.661.110.45 (0.20)0.04
Spike weight (g)***0.840.210.66 (0.18) 0.0010.930.170.71 (0.15)0.0010.960.20.70 (0.16)0.001
Veg dry weight (g)1.533060.01 (0.01)0.091702630.30 (0.20) 0.2792.52140.24 (0.18)0.23

Spatial distribution of phenotypes

To test for a gradual differentiation of each trait along ecogeographical gradients, a partial Mantel test was performed. A pairwise dissimilarity matrix between accessions for each trait was tested against each environmental factor (‘BD’, ‘Rain’, ‘MDT1’, ‘Elevation’) given the geographical distance between sampling sites calculated from GPS coordinates (Fig. 3). Flowering time gradient was significantly correlated with climatic parameters, that is, temperature [midday temperature in January (MDT1), rM = 0.21; P = 0.001] and precipitation (Rain, rM = 0.33; P = 0.001), and the above sea level elevation (rM = 0.19; = 0.001), which is highly correlated with temperature (= 0.78; = 0.0001) and to a lower extent with precipitation (= 0.21; P = 0.006). Vegetative traits, that is, plant height and spike length, were significantly correlated with bulk density (rM = 0.26; < 0.001 and rM = 0.24; P = 0.001, respectively), suggesting an effect of soil compactness on plant morphology. Total biomass (‘VegDryWeight’) was not significantly correlated with any environmental gradient; however, grain characteristics such as grain size and spike weight were significantly correlated with the precipitation gradient (rM = 0.21; = 0.001 and rM = 0.17; = 0.001, respectively).

image

Figure 3. Partial mantel test for each of the nine quantitative traits with each of four environmental gradients (Rain; MDT1 – midday temperature in January; Elevation; and BD – bulk density of the soil) given the geographical distance. Significant correlations are indicated with asterisk.

Download figure to PowerPoint

A graphic presentation was performed using interpolated scores of the phenotypic average in each sampling site to visually examine the distribution of phenotypic landscapes along environmental gradients (Fig. 4). We observed a similar pattern of variation for flowering time, plant height, spike length, and weight and vegetative dry weight. These vegetative traits fit the ecogeographical landscape; that is, they are differentiated between the desert and the Mediterranean ecotypes. On the other hand, reproductive traits, that is, number of spikes and grains, did not correlate with the ecogeographical landscape or any spatial pattern (Fig. 4).

image

Figure 4. Geographical presentation of interpolated scores for each of the nine traits. Sampling sites are coloured by their ecogeographical assignment following the k-means analysis (desert – yellow/light grey; coast – blue/dark grey; north – green/mid-grey) and greyscale varies according to the sampling site average for each trait in the common garden experiment.

Download figure to PowerPoint

Phenotypic differentiation between wild and cultivated barley

Pairwise comparisons of each wild cluster with the cultivated barley collection used in this study were conducted for each trait using nonparametric Mann–Whitney test (Table 1). Cultivated barley was significantly different from the desert type in all traits except for the number of spikes, which was similar between wild and cultivated barley and within the wild clusters. Mediterranean ecotypes were significantly different from the cultivated barley in both vegetative and reproductive traits but were similar in flowering time, plant height and spike length. Although vegetative characteristics, that is, height and spike length, were similar for Mediterranean wild types and cultivated barley, grain characteristics (e.g. grains dry weight, spike weight) were significantly higher in cultivated than in wild barley.

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

In this study, wild barley phenotypic variation was investigated along the south-western part of the Fertile Crescent, which harbours a high proportion of the total genetic variation of wild barley (Nevo, 1998). The wide phenotypic variation found in this region can be grouped into three major ecotypes in accordance with the main ecogeographical regions: north and coastal Mediterranean and the desert. Different levels of divergence between clusters and correlation with environmental gradients were found for the nine traits analysed. Comparison of each cluster with domesticated barley indicated different levels of differentiation for each trait; it was also found that Mediterranean types resemble the domesticated barley more than the desert type.

Differential adaptation of quantitative traits in wild barley

Wild barley is a predominantly self-fertilizing species encountered in different habitats along the Fertile Crescent (Harlan & Zohary, 1966). Here, we investigated the phenotypic variation of wild barley in the south-western part of its distribution range, which is characterized by a wide array of environmental variation along short geographical distances. Two major ecogeographical regions can be distinguished: Mediterranean and the desert. Although several studies were conducted on the adaptive differentiation between these regions (Nevo et al., 1984; Volis et al., 2002), we were able to further split the Mediterranean region into a coastal and northern part based on the phenotypic analysis. The two major regions of desert and Mediterranean climate were well differentiated according to phenotypic data based on both PCA- and model-based approaches, suggesting strong adaptive and geographical effects on phenotypic divergence (Fig. 4). Within the Mediterranean ecotypes, accessions from the coastal and northern regions were less differentiated, possibly because of higher levels of gene flow or a relatively recent differentiation between the two regions. Moreover, the mixture of phenotypes in the shared hybridization zone is a major obstacle in discriminating between the two regions. Therefore, excluding accessions from hybridization regions is a preliminary essential step for adaptive differentiation analysis. Notably, this segregation pattern of the Barley1K collection to the three ecotypes is in accordance with previous analyses of genetic variation with a high resolution platform, including the levels of gene flow between clusters (Hübner et al., 2012).

Phenotypic differentiation between the three clusters indicated that both adaptive and demographic processes, such as genetic drift and migration, affect wild barley phenotypes. Here, we used a hierarchical Bayesian model to estimate QST for different traits, which, in contrast to previous studies (Volis et al., 2005), suggests a complex interplay of the two evolutionary forces for different traits. Genetic differentiation (FST) was calculated using 42 microsatellites, which revealed that the pairwise differentiation between the desert and coast clusters was relatively high compared with the differentiation between other clusters. This indicates the lower gene flow between these clusters presumably due to both geographical and environmental barriers caused by the Judah and Samaria mountain ridge. The wide variation in FST estimates found between loci is probably due to the hypervariable nature of microsatellites, which extended the 95% confidence interval (Table 2). Such high variation hinders the ability to reject a null hypothesis of neutrality in quantitative traits and to detect divergence of phenotypes between clusters (Ritland, 2000). Although QST/FST studies can provide evidence to local adaptation through trait differentiation, reciprocal transplant field experiments are required to fully reject neutrality. Nevertheless, our results showed a stronger phenotypic divergence between the desert and the Mediterranean ecotypes than between the two Mediterranean ecotypes (Figs 2 and 4), possibly as a consequence of diversifying selection between the two regions. Synchrony in flowering time, which is essential for successful mating and the introgression of new alleles into populations, was found to differentiate the desert type from the Mediterranean type in accordance with previous studies (Verhoeven et al., 2008; Volis, 2011). Grain characteristics (weight and size), which are mostly determined by the investment of the maternal plant and its susceptibility to environmental stress, were found to differentiate between the three clusters while reproductive characteristics (number of grains, number of spikes) were similar for all three ecotypes.

Gradual variation along environmental clines

Quantitative traits change gradually over environmental clines when they are sensitive to the respected environmental effect (Endler, 1986); however, when geographical barriers (e.g. rivers or mountain ridges) impede genetic exchange, phenotypic changes may amend these clines. In this study, four environmental gradients were tested for correlation with each of the nine quantitative traits given the geographical distance between sampling sites. As previously described (Craufurd & Wheeler, 2009), flowering time is significantly correlated with temperature and rainfall gradients, which are the main contributors to spatial differentiation between habitats in this region (Hübner et al., 2009). Plant elongation (i.e. plant height and spike length) was significantly correlated with the soil characteristics affecting the availability of nutrients and water for the plant growth (Fig. 3). Grain characteristics that were mostly differentiated between the three clusters were significantly correlated with the rainfall gradient, which is highest in the north Mediterranean, lower along the coast and the lowest in the desert. This demonstrates the fundamental importance of rainfall in plant adaptation and population differentiation in this region. A visual presentation of the phenotypic scores interpolated over the spatial distribution of the study region using a GIS platform distinguished between traits that fit the ecogeographical distribution and traits that did not (Fig. 4). Similar to the trait differentiation and the partial Mantel test, the visual demonstration emphasizes that reproductive traits follow less the ecogeographical distribution as compared to morphological and phenological characteristics. The size of the grains, which was found to be the main diversifying traits between the three clusters, corresponded to the three ecogeographical regions.

The spatial distribution of traits and their clustering mode are important in the context of genetic mapping of complex adaptive traits. Traits that are spatially distributed in accordance with the genetic population structure (e.g. flowering time, height etc.) will be less amenable to genetic dissection by means of association mapping without controlling for population structure due to the masking effect of genetic drift (Nordborg et al., 2005). On the other hand, for traits whose geographical distribution is different from the genetic population structure (e.g. grains per spike or number of spikes), it should be easier to genetically map them at higher resolution in natural populations even without stringent control of the population structure (Bergelson & Roux, 2010). This may be of great interest to association studies aimed at zooming in on suspected regions associated with grain yield characteristics, based on the notion that linkage disequilibrium is gradually reduced from cultivated barley to landraces and to wild barley (Morrell et al., 2005).

Divergence under domestication

Since its domestication, cultivated barley was subjected to intense directional selection for traits that are important in the cultivation process and for higher yielding varieties. In this study, we support previous studies claiming that barley was domesticated from a Mediterranean ecotype distributed in Israel and northward (Badr et al., 2000; Kilian et al., 2006). This ecotype was found to resemble the cultivated barley in more traits than the desert type, especially in flowering time and plant size (Table 1). Nevertheless, cultivated barley was found to diverge significantly from wild barley in yield components, that is, number of grains and their weight. This suggests that the primary trait our ancestors selected for during domestication was grain yield. Although the cultivated barley is superior to the wild barley in all yield components, the rich repertoire found in wild barley may still be beneficial for understanding some unknown mechanisms of adaptation to changing environments. These may be proven as useful as increasing crop resilience under the threat of global climate change.

Acknowledgments

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information

We thank Prof. Andreas Graner, Dr. Nils Stein (IPK) for hosting S.H. during the genotyping and for providing the Hordeum vulgare seeds. We also thank Elad Oren, Lital Nakar, Anat Garzon, laboratory members of the Fridman laboratory and the Lehava project for their assistance in the plant phenotyping, as well as Naama Rona for her technical assistance. This work was supported by the German Academic Exchange Service grant (DAAD; S.H.), the Young GIF grant no. 2181-1802.12/2007 (E.F.).

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information
  • Atlin, G.N., McRae, K.B. & Lu, X. 2000. Genotype X region interaction for two-row barley yield in Canada. Crop Sci. 40: 16.
  • Badr, A., Müller, K., Schäfer-Pregl, R., El Rabey, H., Effgen, S., Ibrahim, H.H. et al. 2000. On the origin and domestication history of Barley (Hordeum vulgare). Mol. Biol. Evol. 17: 499510.
  • Bergelson, J. & Roux, F. 2010. Towards identifying genes underlying ecologically relevant traits in Arabidopsis thaliana. Nat. Rev. Genet. 11: 867879.
  • Burrough, P.A. & Mcdonnell, R.A. 1998. Principles of Geographical Information Systems. Oxford University Press, Oxford, UK.
  • Caldwell, K.S., Russell, J., Langridge, P. & Powell, W. 2006. Extreme population-dependent linkage disequilibrium detected in an inbreeding plant species, Hordeum vulgare. Genetics 172: 557567.
  • Charlesworth, B., Nordborg, M. & Charlesworth, D. 1997. The effects of local selection, balanced polymorphism and background selection on equilibrium patterns of genetic diversity in subdivided populations. Genet. Res. 70: 155174.
  • Clarke, K.R. 1993. Non-parametric multivariate analyses of changes in community structure. Aust. J. Ecol. 59: 117143.
  • Craufurd, P.Q. & Wheeler, T.R. 2009. Climate change and the flowering time of annual crops. J. Exp. Bot. 60: 25292539.
  • Dieringer, D. & Schlötterer, C. 2003. Microsatellite analyser (MSA): a platform independent analysis tool for large microsatellite data sets. Mol. Ecol. Notes 3: 167169.
  • Ellis, R.P., Forster, B.P., Robinson, D., Handley, L.L., Gordon, D.C., Russell, J.R. et al. 2000. Wild barley: a source of genes for crop improvement in the 21st century? Plant Mol. Biol. 51: 917.
  • Ellstrand, N.C. 2003. Current knowledge of gene flow in plants: implications for transgene flow. Phil. Trans. R Soc. Lond. 24: 217242.
  • Ellstrand, N.C. & Elam, D.R. 1993. Population genetics consequences of small population size: implications for plant conservation. Ecology 24: 217242.
  • Endler, J.A. 1986. Natural Selection in the Wild. Princeton University Press, Princeton, NJ.
  • Fraley, C. & Raftery, A.E. 2007. Model-based methods of classification: using the mclust software in chemometrics. J. Stat. Softw. 18: 113.
  • Friendly, M. 2002. Corrgrams: exploratory displays for correlation matrices. Am. Stat. 56: 316324.
  • Ghalambor, C.K., McKay, J.K., Carroll, S.P. & Reznick, D.N. 2007. Adaptive versus non-adaptive phenotypic plasticity and the potential for contemporary adaptation in new environments. Funct. Ecol. 21: 394407.
  • Goudet, J. & Büchi, L. 2006. The effects of dominance, regular inbreeding and sampling design on QST, an estimator of population differentiation for quantitative traits. Genetics 172: 13371347.
  • Harlan, R. & Zohary, D. 1966. Distribution of wild wheats and barley. Science 153: 10741080.
  • Haseneyer, G., Stracke, S., Piepho, H.P., Sauer, S., Geiger, H.H. & Graner, A. 2010. DNA polymorphisms and haplotype patterns of transcription factors involved in barley endosperm development are associated with key agronomic traits. BMC Plant Biol. 10: 5.
  • Hübner, S., Höffken, M., Oren, E., Haseneyer, G., Stein, N., Graner, A. et al. 2009. Strong correlation of wild barley (Hordeum spontaneum) population structure with temperature and precipitation variation. Mol. Ecol. 18: 15231536.
  • Hübner, S., Günther, T., Flavell, A., Fridman, E., Graner, A., Korol, A. et al. 2012. Islands and streams: clusters and gene flow in wild barley populations from the Levant. Mol. Ecol. 21: 1151129.
  • Hughes, A.L. 2011. Evolution of adaptive phenotypic traits without positive Darwinian selection. Heredity 108: 347353.
  • Jana, S. & Pietrzak, L.N. 1988. Comparative assessment of genetic diversity in wild and primitive cultivated barley in a center of diversity. Genetics 119: 981990.
  • Kilian, B., Ozkan, H., Koh., J., von Haeseler, A., Barale, F., Deusch, O. et al. 2006. Haplotype structure at seven barley genes: relevance to gene pool bottlenecks, phylogeny of ear type and site of barley domestication. Mol. Genet. Genomics 276: 230241.
  • Korol, A.B., Ronin, Y.I., Tadmor, Y., Bar-Zur, A., Kirzhner, V.M. & Nevo, E. 1996. Estimating variance effect of QTL: an important prospect to increase the resolution power of interval mapping. Genet. Res. 67: 187194.
  • Lande, R. 1976. Natural selection and random genetic drift in phenotypic evolution. Evolution 30: 314334.
  • Lunn, D.J., Thomas, A., Best, N. & Spiegelhalter, D. 2000. winbugs – A Bayesian modeling framework: concepts, structure, and extensibility. Stat. Comput. 4: 325337.
  • Mano, Y., Kawasaki, S., Takaiwa, F. & Komatsuda, T. 2001. Construction of a genetic map of barley (Hordeum vulgare L.) cross “Azumamugi” × “Kanto Nakate Gold” using a simple and efficient amplified fragment-length polymorphism system. Genome 44: 284292.
  • Merilä, J. & Crnokrak, P. 2001. Comparison of genetic differentiation at marker loci and quantitative traits. J. Evol. Biol. 14: 892903.
  • Morrell, P.L., Toleno, D.M., Lundy, K.E. & Clegg, M.T. 2005. Low levels of linkage disequilibrium in wild barley (Hordeum vulgare ssp. spontaneum) despite high rates of self-fertilization. Proc. Natl Acad. Sci. USA 102: 24422447.
  • Nevo, E. 1992. Origin, evolution, population genetics and resources for breeding of wild barley, Hordeum spontaneum, in the Fertile Crescent. In: Barley: Genetics, Biochemistry, Molecular Biology and Biotechnology (P.R. Shewry, ed.), pp. 1943. C.A.B. International, Alden Press, Oxford.
  • Nevo, E. 1998. Genetic diversity in wild cereals: regional and local studies and their bearing on conservation ex situ and in situ. Genet. Resour. Crop Evol. 45: 355370.
  • Nevo, E., Beiles, A., Yitzchak, G., Storch, N. & Kaplan, D. 1984. Genetic resources of wild cereals in Israel and vicinity. II. Phenotypic variation within and between populations of wild barley, Hordeum spontaneum. Euphytica 33: 737756.
  • Nordborg, M., Hu, T.T., Ishino, Y., Jhaveri, J., Toomajian, C., Zheng, H. et al. 2005. The pattern of polymorphism in Arabidopsis thaliana. PLoS Biol. 3: 12891299.
  • O'Hara, R.B. & Merilä, J. 2005. Bias and precision in QST estimates: problems and some solutions. Genetics 171: 13311339.
  • Oksanen, A.J., Kindt, R., Legendre, P., Hara, B.O., Simpson, G.L., Stevens, M.H.H. et al. 2008. The vegan Package. http://cran.r-project.org/
  • Palo, J.U., O'Hara, R.B., Laugen, A.T., Laurila, A., Primmer, C.R. & Merilä, J. 2003. Latitudinal divergence of common frog (Rana temporaria) life history traits by natural selection: evidence from a comparison of molecular and quantitative genetic data. Mol. Ecol. 12: 19631978.
  • Ritland, K. 2000. Marker-inferred relatedness as a tool for detecting heritability in nature. Mol. Ecol. 9: 11951204.
  • Rizopoulos, D. 2006. ltm: an r package for latent variable modeling and item response theory analyses. J. Stat. Softw. 17: 125.
  • Salamini, F., Ozkan, H., Brandolini, A., Schäfer-Pregl, R. & Martin, W. 2002. Genetics and geography of wild cereal domestication in the Near East. Nat. Rev. Genet. 3: 429441.
  • Schmidt, P.S., Serräo, E.A., Pearson, G.A., Riginos, C., Rawson, P.D., Hilbish, T.J. et al. 2008. Ecological genetics in the North Atlantic: environmental gradients and adaptation at specific loci. Ecology 89: 91107.
  • Spitze, K. 1993. Population structure in Daphnia obtusa: quantitative genetic and allozymic variation. Genetics 135: 367374.
  • Talamè, V., Sanguineti, M.C., Chiapparino, E., Bahri, H., Ben Salem, M., Forster, B.P. et al. 2004. Identification of Hordeum spontaneum QTL alleles improving field performance of barley grown under rainfed conditions. Ann. Appl. Biol. 144: 309319.
  • Thiel, T., Michalek, W., Varshney, R.K. & Graner, A. 2003. Exploiting EST databases for the development and characterization of gene-derived SSR-markers in barley (Hordeum vulgare L.). Theor. Appl. Genet. 106: 411422.
  • Turelli, M. 1988. Phenotypic evolution, constant covariance, and the maintenance of additive variance. Evolution 42: 13421347.
  • Ullrich, S.E. 2011. Barley: Production, Improvement, and Uses. Wiley-Blackwell, Oxford, UK.
  • Verhoeven, K.J.F., Poorter, H., Nevo, E. & Biere, A. 2008. Habitat-specific natural selection at a flowering-time QTL is a main driver of local adaptation in two wild barley populations. Mol. Ecol. 17: 34163424.
  • Volis, S. 2011. Adaptive genetic differentiation in a predominantly self-pollinating species analyzed by transplanting into natural environment, crossbreeding and Q(ST)-F(ST) test. New Phytol. 192: 237248.
  • Volis, S., Mendlinger, S. & Ward, D. 2002. Differentiation in populations of Hordeum spontaneum along a gradient of environmental productivity and predictability: life history and local adaptation. Biol. J. Linn. Soc. 77: 479490.
  • Volis, S., Yakubov, B., Shulgina, I., Ward, D. & Mendlinger, S. 2005. Distinguishing adaptive from nonadaptive genetic differentiation: comparison of QST and FST at two spatial scales. Heredity 95: 466475.
  • Waugh, A., Jannink, J.L., Muehlbauer, G.J. & Ramsay, L. 2009. The emergence of whole genome association scans in barley. Curr. Opin. Plant Biol. 12: 218222.
  • Weir, B.S. & Cockerham, C.C. 1984. Estimating F-statistics for the analysis of population structure. Evolution 38: 13581370.
  • Whitlock, M.C. & Guillaume, F. 2009. Testing for spatially divergent selection: comparing QST to FST. Genetics 183: 10551063.
  • Wright, S. 1943. Isolation by distance. Genetics 28: 114138.
  • Zamir, D. 2001. Improving plant breeding with exotic genetic libraries. Nat. Rev. Genet. 2: 983989.

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgments
  8. References
  9. Supporting Information
FilenameFormatSizeDescription
jeb12043-sup-0001-FigS1.docxWord document346KFigure S1 Pairwise correlation matrix between traits.
jeb12043-sup-0002-TableS1.xlsapplication/msexcel45KTable S1 List of wild and cultivated barley used in this study.
jeb12043-sup-0003-TableS2.xlsapplication/msexcel116KTable S2 Phenotypic data of the wild accessions including all nine traits and ecogeographical region.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.