Adaptive genetic differentiation in a predominantly self-pollinating species analyzed by transplanting into natural environment, crossbreeding and QSTFST test


Author for correspondence:
S. Volis
Tel: +972 8 6477197


  • Both genetic drift and natural selection result in genetic/phenotypic differentiation over space. I analyzed the role of local adaptation in the genetic differentiation of populations of the annual grass Hordeum spontaneum sampled along an aridity gradient.
  • The study included the introduction of plants having desert vs nondesert origin into natural (desert) environment, analysis of population differentiation in allozymes and random amplified polymorphic DNA (RAPD) markers vs phenotypic traits (QSTFST comparison), and planting interpopulation hybrids under simulated desert conditions in a glasshouse.
  • The results of the home advantage test, QSTFST comparison and crossbreeding were consistent with local adaptation; that is, that differentiation of the desert plants from plants of nondesert origin in phenotypic traits was adaptive, giving them home advantage. Each method used provided additional, otherwise unavailable, information, meaning that they should be viewed as complementary rather than alternative approaches.
  • Gene flow from adjacent populations (i.e. populations experiencing the desert environment) via seeds (but not pollen) had a positive effect on fitness by enhancing natural selection and counteracting drift. At the same time, the effect of genes from the species distributional core (nondesert plants) by either seed or pollen had a negative fitness effect despite its enriching effect on neutral diversity. The pattern of outbreeding depression observed in interpopulation hybrids (F1) and their segregating progeny (F2) was inconsistent with underdominance, but indicated the presence of additive, dominance and epistatic effects.


Plant species often show genetic/phenotypic differentiation over space which may or may not result from natural selection. Recognition of the adaptive significance of such differentiation has been a challenging task in evolutionary biology (Endler, 1977; Bradshaw, 1984; Conover & Schultz, 1995; Linhart & Grant, 1996; Lenormand, 2002; Hedrick, 2006; Rasanen & Hendry, 2008; Volis, 2008).

Both nonselective (drift, gene flow) and selective (natural selection) evolutionary forces determine population subdivision and the extent of genetic variation in a species. However, nonselective forces influence all loci equally, while selective forces affect loci differentially. Cavalli-Sforza (1966) introduced the idea that heterogeneity of population differentiation (FST) values across loci reflects differential loci evolvement in an adaptation process. This led to the idea of comparing population differentiation in adaptive quantitative traits (QST) with population differentiation obtained from neutral molecular markers (FST). Since the introduction of the formal comparison of phenotypic and neutral molecular variation (Prout & Barker, 1989, 1993; Spitze, 1993), the QSTFST comparison has become a routine test for adaptive population divergence (Merila & Crnokrak, 2001; McKay & Latta, 2002; Leinonen et al., 2008), despite a lack of rigorous tests of its validity (but see Porcher et al., 2004) and recognized limitations of the QSTFST comparison (Leinonen et al., 2008; Pujol et al., 2008; Whitlock, 2008; Whitlock & Guillaume, 2009; Volis & Zhang, 2010).

One of the potential problems with the QSTFST comparison is the underlying assumption that a neutral trait has an additive genetic basis with linkage equilibrium among loci (Lynch et al., 1999). Diversifying selection is expected to relax the linkage equilibrium assumption, creating positive associations among loci across populations and leading to high QST-values (Latta, 1998). However, several other factors can also violate the neutral assumptions, for example, demographic history (Miller et al., 2008) and presence of dominance and epistasis (Whitlock, 1999; Lopez-Fanjul et al., 2003; Goudet & Buchi, 2006). Genetic variability in quantitative traits is known to vary depending on environmental conditions (Mitchell-Olds & Rutledge, 1986; Bennington & McGraw, 1996; Jenkins et al., 1997) and estimates of population differentiation are sensitive to the conditions under which they are measured (Merila & Crnokrak, 2001; Palo et al., 2003; Gomez-Mestre & Tejedo, 2004). Genotype × environment interactions can affect QST values even in favorable common garden environments if the latter do not correspond to the conditions of the natural environment (Hoffmann & Merila, 1999).

How reliable is the QSTFST test in detecting the effect of natural selection and what is its relative value in comparison with other techniques for studying diversifying selection? With a notable exception (Porcher et al., 2004, 2006), I am unaware of any study that addresses this question, and which, at the same time, analyzes QST vs FST differentiation and demonstrates an empirically local selection effect.

There are several experimental approaches to testing for local selection. Local adaptation can be demonstrated and measured in common garden experiments where a single environmental factor is tested for its selective effect on individuals of different origin (Kawecki & Ebert, 2004). Although this method has limited use as a result of its assumption that the agent of selection is known and is not confounded by the other factors across tested environments, it can be of high value for analysis of certain clines, for example, salinity or aridity (Weider & Hebert, 1987; Volis & Zhang, 2010).

Reciprocal-transplant experiments employing cross-relocation of individuals originating in different habitats are the classical approach to test for spatially heterogeneous selection (Turesson, 1922; Clausen et al., 1940). In these experiments, higher fitness of native vs alien genotypes in their respective environments is evidence that populations are locally adapted. The reciprocal-transplant designs can be informative about the pattern of differentiation among populations, and – when accompanied by environmental manipulations – about the environmental factors inducing selective responses (Santamaria et al., 2003; Byars et al., 2007; Petru & Tielborger, 2008; Liancourt & Tielborger, 2009).

Although the reciprocal transplanting is a powerful tool for detecting adaptive differentiation, it cannot evaluate the contribution of dominance, genetic linkage and pleiotropic effects of genes under selection to the phenotype. Experimental hybridization between ecotypically differentiated populations and planting of the hybrids together with parents in the common garden or reciprocal-transplant field experiments is a way to elucidate the genetic architecture of local adaptation (i.e. genetic effects responsible for fitness differences) (Fenster & Galloway, 2000; Erickson & Fenster, 2006; Johansen-Morris & Latta, 2006). This is also the method of choice in testing for a scale of local adaptation because it allows testing across large numbers of populations, habitats or distance classes (Waser & Price, 1994; Galloway & Fenster, 2000; Montalvo & Ellstrand, 2001; Heiser & Shaw, 2006; Wright & Stanton, 2007). Finally, crossbreeding can provide valuable information on effects of gene flow on a locally adapted population. These effects can be negative (introduction of maladaptive genes) or positive (restoration of depleted heritable variation and adaptive potential) (Holt & Gomulkiewicz, 1997; Kirkpatrick & Barton, 1997; Boulding & Hay, 2001; Lenormand, 2002; Garant et al., 2007; Rasanen & Hendry, 2008).

I set out to test the role of local adaptation in intraspecific variation along an aridity gradient in wild barley, Hordeum vulgare ssp. spontaneum Koch (hereafter H. spontaneum), a progenitor of cultivated barley. This species has wide geographic distribution and high genetic differentiation despite considerable gene flow (Morrell et al., 2003). A high degree of structuring of genetic variation is expected in this species because of its mating system (predominant self-fertilization). Self-pollination is a life history character limiting gene flow, heterozygosity and recombination, and enhancing creation of multilocus associations (Brown, 1979; Golding & Strobeck, 1979; Hastings, 1989; Charlesworth, 2003). At the same time, ecotypic variation is a distinct feature of this species (Snow & Brody, 1984; van Rijn et al., 2000). Because of the importance of wild barley as a source for cultivated barley improvement, evolutionary processes that lead to adaptation in this species were a subject of intensive research over the last several decades. However, a genetic architecture of local adaptation, as well as roles of gene flow and genetic drift in this species, is not well understood yet.

In wild barley distributional range, desert populations represent the species periphery and, characteristic for marginal populations, have small size and a high degree of isolation. On the other hand, desert plants possess morphological, phenological and life history traits that distinguish them from plants originating from other climatic zones of Israel (Snow & Brody, 1984; Volis et al., 2002a, 2004). Evidence of local adaptation was found when plants were reciprocally transplanted between one desert and one Mediterranean location (Volis et al., 2002b). Early reproduction in the desert ecotype was clearly advantageous in the desert but not the Mediterranean location.

In the present study, I analyzed the role of local adaptation in the genetic differentiation of populations on a much larger scale using populations sampled along an aridity gradient from desert (with c. 100 mm annual precipitation) to mountain (with > 1000 mm annual precipitation). In addition to direct transplantation, I performed targeted inter-population crossing to analyze the genetic architecture of adaptation to the desert environment and an effect of gene flow from populations of desert and nondesert origin, and I also performed a QSTFST test to compare genetic differentiation in selectively neutral vs adaptive loci. In the latter technique, genetic differentiation was analyzed in a pairwise fashion, using either one desert and one nondesert population or a desert vs nondesert regional gene pool. Joint application of these approaches to detect local adaptation allowed assessment of their relative merit and better understanding of the effects of genetic architecture and gene flow on adaptive population differentiation.

Materials and Methods

Study species and sampling

Hordeum spontaneum Koch, an annual, diploid grass with predominant self-pollination (Brown et al., 1978), is widely distributed in Israel across different environments, including deserts (< 200 mm annual rainfall) (Harlan & Zohary, 1966).

The plant material used in this study consisted of four groups of five populations sampled in Israel in 1996 employing a nested sampling design (Supporting Information, Table S1). The groups represented four climatic zones (ordered by increasing mean annual rainfall and predictability): desert, semi-arid, Mediterranean, and mountain. Relief, slope exposition, vegetation, and soil types were kept constant in five sampling localities within a group that were ≤ 5 km (mountain) or ≤ 20 km (other three groups) from each other. All 20 populations were used in a study of population genetic structure, 12 populations (three populations per group) were used in a crossbreeding experiment, and one population per group was used for transplanting into an intact natural desert location. In each population, seeds were collected from at least 30 plants separated by > 2 m, stored under 5°C and had 100% germination upon usage.

Introduction into natural environment

The transplantation experiment was conducted in 1997 in the wadi (Arabic for ‘ephemeral river valley’) near kibbutz Sede Boqer in the Negev Desert. A 100 m2 plot was fenced and cleared of vegetation. Each of four seed pools that represented climatic zones (desert, semi-arid, Mediterranean, and mountain) had a single population origin (Table S1) and consisted of selfed progeny of 10 mother plants grown under uniform conditions within a plastic growth facility (hereafter referred to as a ‘glasshouse’). Two-week-old plants were transplanted within 2 wk after the first effective rain (> 10 mm rainfall). A randomized block design was established, with each of eight blocks containing plants from all four populations arranged as four × four plants 10 cm apart, that is, four sets of 16 plants. At seed maturation, the number of spikes and the number of spikelets in a spike were counted for each plant. In addition, 100–200 seeds from each population/block were collected. After obtaining average spikelet weights of seeds harvested over 16 plants/population/block, the mean plant reproductive biomass per population/block was calculated.

The relative performance (RP) of plants of nondesert origin was calculated from the averages of 16 plants for each block separately as RP = (wawl)/max(wa,wl), where wa and wl designate the fitness of plants of alien and local origin, respectively. The ‘local vs foreign’ criterion (Kawecki & Ebert, 2004) was used to infer local adaptation from plant differences in fitness-related traits, such as reproductive traits. Local adaptation was defined as a superiority of plants of local origin over foreign plants in their home site.

QST vs FST experiment

Two concurrent common garden experiments were performed to separately estimate trait heritability and population differentiation. Plants were grown under the same uniform conditions in a glasshouse at the Bergman Campus, Beer Sheva, and the same set of quantitative traits was measured. The traits included tiller height (TH), flag and penultimate leaf length and width (FLL, PLL, FLW, and PLW, respectively), spike length (SPL), awn length (AWL), and the number of days to awning (DAW). At senescence, mean spikelet weight (SWT) was obtained from the three first by appearance spikes.

In the first common garden experiment, I used four populations (one population per climatic zone) with eight randomly chosen accessions per population for assessment of trait broad-sense heritability, H2. Before the experiment, the mother plants representing different accessions were planted under uniform conditions in a glasshouse and four offspring per mother plant were used in the experiment. As wild barley is predominantly autogamous, the offspring of each mother plant can be considered as a genetically identical single genotype. Each of four offspring was randomly assigned to one of four blocks containing 32 plants.

In the second common garden experiment, I used nine randomly chosen accessions from each of 20 populations (five populations per climatic zone) to partition the quantitative genetic variation into between- and within-population components needed for estimation of QST. Before the experiment, the accessions were planted under uniform conditions in a glasshouse to minimize maternal effects.

From the same 20 populations, 25 accessions per population were examined for nine enzyme systems encoding 15 loci by starch gel electrophoresis, and 14 accessions per population were analyzed with the random amplified polymorphic DNA (RAPD) markers. The enzymes examined were acid phosphatase (Acph), catalase (Cat), general protein (Gp), glutamate dehydrogenase (Gdh), esterase (Est), malate dehydrogenase (Mdh), phosphoglucomutase (Pgm), phosphoglucose isomerase (Pgi), and 6-phosphogluconate dehydrogenase, (6-Pgd). To evaluate differentiation based on RAPD markers, nine primers of 10 bp in length were used. The detailed protocols for the two procedures are given in Volis et al. (2001, 2003).

Crossbreeding experiment

Crosses were performed in 2005 in a glasshouse at the Bergman Campus, Beer Sheva, by artificial pollination of a mother plant of desert origin with pollen of plants of either desert or nondesert origin (P plants) (Fig. 1). Nine accessions separated by at least 100 m in the wadi near kibbutz Sede Boqer in the Negev Desert (SB population) acted as pollen recipients, and three father plants per climatic zone (one randomly chosen accession per population) were used as pollen donors. Crossing was done using a protocol developed for H. vulgare ssp. vulgare as described in Volis et al. (2011). Using this method, I crossed nine mother plants with plants of different origin and analyzed the produced hybrids (F1) and their self-pollinated offspring (F2) for their relative performance as compared with the self-pollinated mother plant in subsequent experiments (Fig. 1).

Figure 1.

Diagrammatic presentation of the crossbreeding experiment in which the progeny of the mother plant (P) derived through self-pollination was compared with: crosses (F1) derived through pollination of the mother plant with pollen of different geographic origin; progeny of self-pollinated F1 (F2); or progeny of the father plant (P) derived through self-pollination.

The hybrids (F1), regarded as families of full-sibs nested within half-sib families, and their self-pollinated offspring (F2) were grown together with plants derived from self-pollinated parents in two consecutive seasons, 2008–09 and 2009–10. The experiments had identical experimental design and differed only in the amount of water supplied (adjusted based on ambient temperature, see later in this paragraph). Owing to the limited number of F1 seeds, because of the relatively low success of artificial pollination, it was not possible to plant F1 and F2 generations simultaneously. Seeds were simultaneously germinated in an incubator at 24°C and transferred into 3 l pots arranged in a glasshouse using block design. There were nine blocks representing families that differed in mother origin. Within a block, the half-sibs (either F1 or F2) differed only in father origin. The half-sibs were grown in each block with the corresponding mother and father plants, and the block comprised 12 hybrids, one P and 12 P. In total, 225 plants were grown in each of the two experiments. The pots were filled with the sieved loess soil from the original desert population location. During these experiments, conducted in a glasshouse at the Bergman Campus, Beer Sheva, plants received an amount of water close to the multiyear average rainfall amount recorded for years 1961–2010 at Sede Boqer location (95.2 mm). The amount of water applied was adjusted to compensate for the higher evaporation rate in the glasshouse resulting from different soil and ambient temperatures. The winter season of 2008–09 was warmer than the following season and this is reflected in the amount of water supplied to the plants (221 and 150 mm, respectively). Watering was done twice weekly using a drip-irrigation system. The estimates of plant fitness included the total number of seeds and total mass of seeds produced by a plant.

To estimate outbreeding depression or heterosis effect in hybrids, as well as performance of plants that acted as parents, RP was calculated for each generation/family following Agren & Schemske (1993) as RP = (wiw)/max(wi, w), where w was the fitness of progeny of the mothers (P) derived through selfing, and wi was the fitness of either the F1 or F2 plants or selfed progeny of plants that acted as pollen donors (P). Usage of the RP values that are bounded by −1 and 1, allowed standardization of the data obtained in two seasons.

Degree of dominance (d/a) was estimated for the two performance traits, seed number and total seed mass as a deviation of F1 from the midparent value. A positive d/a indicated that alleles of the genes contributing to the trait and conferring adaptation to the desert environment were (mostly) dominant, while a negative d/a indicated that alleles maladaptive in the desert environment were (mostly) dominant.

Statistical analyses

In the introduction into the natural environment experiment, RP values of plants of each of three nondesert origins were analyzed over eight plots by one-sample t-test, and three origins were compared with each other by pairwise Mann–Whitney test.

In the QST vs FST experiment, the structure of variation in phenotypic traits was analyzed after running nested ANOVA by partitioning the total variance into several components. Two random effects included populations and accessions nested within populations. The analysis was done in a pairwise fashion, comparing each desert population with one nondesert population at a time. The REML procedure of Statistica (StatSoft Inc, 2004) was used to calculate variance components.

Data obtained in common garden 1 (four populations) were used to estimate broad-sense heritability, H 2 = VG/(VG + VE). The within-accession variance, σ2E, estimated the environmental variance, VE, and the among-accession variance provided an estimate of the genetic variance, VG. The data from common garden 2 (20 populations) allowed the determination of the among-population (σ2Pop) and within-population (σ2Acc) genetic variances.

Population pairwise estimates of QST were obtained using its surrogate PST (a phenotypic alternative for QST, a measure of genetic differentiation in quantitative traits) (Merila et al., 1997; Merila & Crnokrak, 2001; Storz, 2002; Saint-Laurent et al., 2003; Leinonen et al., 2006). PST was calculated as VB/(VB + H2VW), where VB is a proportion of genetic variance distributed among populations (σ2Pop), VW is a proportion of genetic variance distributed within populations (σ2Acc), and H2 is trait heritability.

For both RAPD markers and allozymes, the distribution of genetic variability within and among populations and regions was investigated by an analysis of molecular variance (AMOVA; Excoffier et al., 1992). The number of permutations for significance testing was set at 1000. I also estimated regional differentiation by the DEST measure that is based on the effective number of alleles (Jost, 2008). DEST values were averaged over polymorphic loci and provided with 95% confidence intervals obtained by bootstrapping.

In the crossbreeding experiment, the general linear model of Statistica (StatSoft Inc, 2004) was used to analyze the two performance traits (seed number and total seed mass). Before analysis, the original values were converted into RP values. The model included as terms generation (either F1 or F2), origin (desert, semi-arid, Mediterranean, or mountain), pedigree (hybrid or parent) and maternal effect (different mother plants arranged in blocks including all combinations of the other three effects). Maternal effect was treated as a random effect and the other three factors as fixed effects.


Introduction into the intact environment

The average RP of plants of nondesert origin analyzed over eight plots by one-sided t-test was significantly different from zero for seed number for semi-desert (−0.44 ± 0.10, df = 7, t = 4.5, P < 0.01), Mediterranean (−0.65 ± 0.13, df = 7, = 5.0, < 0.01), and mountain origin (−0.85 ± 0.04, df = 7, = 20.0, < 0.001). For total seed mass, the average RP values were negative, but significantly different from zero only for mountain origin (semi-desert, −0.15 ± 0.13, df = 7, = 1.2, > 0.05; Mediterranean, −0.36 ± 0.18, df = 7, = 2.0, > 0.05; and mountain, −0.77 ± 0.07, = 10.4, < 0.001; one-sided t-test). There was substantial variation in RP of plants across transplanting plots and this variation decreased from semi-arid to Mediterranean and then to mountain origin of transplanted plants (Fig. 2). There were significant differences among RPs of semi-arid, Mediterranean and mountain plants for seed number (Z = 1.7, 2.9 and 1.7, < 0.05, 0.01 and 0.05, df = 7 and 7, pairwise Mann–Whitney tests). For total seed weight, Mediterranean and mountain plants (Z = 2.1, < 0.05, df = 7 and 7, Mann–Whitney test) and semi-arid and mountain plants differed in RP (Z = 2.8, < 0.01, df = 7 and 7, Mann–Whitney test), but not semi-arid and Mediterranean plants (Z = 1.5, > 0.05, df = 7 and 7, Mann–Whitney test).

Figure 2.

Relative performance of the Hordeum spontaneum plants of nondesert origin (aliens) as compared with plants of desert origin (natives) introduced into a natural desert location using block design. Bars designate blocks. Positive and negative values denote better and poorer performance, respectively, of the aliens (plants of semi-arid, Mediterranean, and Mountain origin) as compared with the natives in eight blocks. Black bars, fitness of the aliens relative to the desert natives. PST, phenotypic differentiation; DEST, neutral genetic differentiation.

Genetic variation and QST vs FST

All traits exhibited high heritabilities, ranging from 0.69 to 0.95, and these values were used for the calculation of PST (as a surrogate for QST, see Materials and Methods section for details) values in pairwise comparisons of two regional pools comprising five populations and in population pairwise comparisons. The results of the two comparative approaches were in agreement.

Partitioning of quantitative trait and marker variation in regional pool comparison revealed a very high regional component in quantitative trait variation and PST values greatly exceeding both types of FST estimates (ΘST and DEST) obtained from allozymes and RAPD markers (Table 1). There was a negligible overlap of frequency distributions of molecular marker and quantitative trait estimates of regional differentiation and a clear trend for the increase in frequency of high PST values from desert–semi-arid comparison to desert–Mediterranean comparison and then to desert–mountain comparison (Fig. 3).

Table 1.   Pairwise population differentiation in quantitative traits and two marker classes, allozymes and random amplified polymorphic DNA (RAPD) markers
Genetic variation typePairwise population differentiation
EstimatePopulation pair
  1. In pairwise comparisons a regional pool comprising five populations was treated as a single population.

QuantitativePST (95% CI)0.603 (0.432–0.697)0.859 (0.824–0.881)0.866 (0.835–0.891)
RAPD markersΘST0.1030.2320.270
Figure 3.

Frequency distribution of locus/trait estimates of population differentiation in pairwise regional comparisons of desert Hordeum spontaneum plants with plants of semi-arid, Mediterranean, and mountain origin. Numbers of loci/traits for quantitative, random amplified polymorphic DNA (RAPD) marker and allozyme data are in parentheses: PST, white bars (9); DEST RAPD (59), black bars; DEST, gray bars, allozymes (17). PST, phenotypic differentiation; DEST, neutral genetic differentiation.

In pairwise comparisons of five desert populations with populations of nondesert origin, the average ΘST values were 0.07, 0.13, and 0.35 for allozymes and 0.24, 0.43, and 0.54 for RAPD markers, and the average PST values were 0.63, 0.88, and 0.88 (desert–semi-arid, desert–Mediterranean and desert–mountain comparisons, respectively). The range of values and differences between ΘST and PST were highly consistent across populations of the same origin (Fig. 4).

Figure 4.

Average (± SE) QST (open bars) and FST estimates (closed bars) over five pairwise comparisons of each of the five desert populations with populations of semi-arid, Mediterranean, and Mountain origin. The FST estimates were obtained from the allozyme and random amplified polymorphic DNA (RAPD) marker data separately.

The desert population that supplied mother plants in the crossbreeding experiment did not differ from the other three desert populations in the extent of variation in either quantitative traits or molecular markers (Fig. S1). Similarly, the desert region did not differ from the other three regions in the extent of variation in the analyzed traits/markers (Fig. S1). The population differentiation within the desert region was weak for allozymes (ΘST = 0.041), but similarly high for RAPD markers and quantitative traits (ΘST = 0.219 and QST = 0.169).

Analysis of quantitative trait variation revealed a significant difference between the desert region and all the other regions in eight traits out of nine. In most of the traits, a clinal variation with an increase in aridity (from Mediterranean to semi-arid and to desert) was observed, that is, a decrease in days until flowering, height of tillers, spikelet weight, size of spikes and their awns, and an increase in size of the leaves (Table 2).

Table 2.   Quantitative trait regional mean (±SE) and heritability
TraitMean regional value ± SEH2
  1. AWL, awn length; DAW, number of days to awning; FLL, flag leaf length; FLW, flag leaf width; PLL, penultimate leaf length; PLW, penultimate leaf width; SPL, spike length; SWT, mean spikelet weight TH, tiller height.

  2. Letters denote the results of Tukey-Kramer test with different letters indicating significant differences at P < 0.05.

DAW120.0 ± 0.8a131.4 ± 0.8b140.3 ± 0.8c145.3 ± 0.6d0.86
SWT (mg)48.6 ± 1.2a70.3 ± 1.9b79.4 ± 1.6c70.4 ± 1.2b0.76
SPL (cm)20.3 ± 2.7a24.3 ± 2.6b28.6 ± 5.0c28.9 ± 5.2c0.94
AWL (cm)6.0 ± 1.7a10.1 ± 2.6b15.2 ± 5.1c14.0 ± 4.5c0.95
FLL (cm)13.4 ± 5.5c9.2 ± 3.7b6.0 ± 2.0a7.2 ± 3.0a0.83
FLW (mm)10.9 ± 2.8c9.0 ± 2.7b5.3 ± 1.7a5.1 ± 1.7a0.75
PLL (cm)26.8 ± 0.8c17.8 ± 0.5b13.2 ± 0.3a14.8 ± 0.4a0.69
PLW (mm)15.1 ± 0.3c14.1 ± 0.4c10.1 ± 0.2b8.9 ± 0.2a0.76
TH (cm)63.9 ± 1.4a77.6 ± 1.3b85.7 ± 1.6c87.4 ± 1.3c0.82


There was a significant difference in seed number and a marginally significant difference in total seed mass of F1 and F2 plants relative to the P (−0.37 ± 0.03 and −0.51 ± 0.03, respectively) (Table 3). There were no significant maternal effects, that is, genetic and nongenetic differences between the mother plants representing the same population, while the factors paternal effect (pollen origin) and pedigree (hybrid vs parent) were highly significant (Table 3).

Table 3.   Analysis of variance on Hordeum spontaneum seed number and total seed weight used to estimate performance of F1, F2 and P plants relative to the corresponding P plants
SourcedfF (seed number)F (total seed weight)
  1. †, < 0.10; *, < 0.05; **, < 0.01; ***, < 0.01.

Generation (G)17.5*5.0
Maternal effect (M)80.70.8
Pollen geographic origin (O)3156.4***92.9***
Pedigree (hybrid vs P) (P)142.6***55.3***
G × O31.52.6
O × P351.0***43.7***
G × O × P30.90.7
G × O × P × M240.90.9

In the majority of the crosses, F1 plants had reduced performance (seed number and total seed mass) as compared with their corresponding half-sibs derived through self-pollination of the mother plants. However, fitness of the F1 plants was almost ubiquitously higher than the fitness of progeny of self-pollinated father plants (Fig. 5). Heterosis was observed only in crosses involving plants of desert origin (Fig. 5). Although performance of F2 was, on average, lower than that of F1, a range of RP values in F2 was wider than in F1 as a result of recombination and segregation (Fig. 5).

Figure 5.

(a) Relative performance (RP) estimated as differences in Hordeum spontaneum seed number and total seed mass in F1 and F2 plants, as well as in P plants (± SE). Positive and negative values denote better and poorer performance, respectively, of the hybrids and P as compared with the self-pollinated P plants. Open bars, fitness of F1 and F2 relative to P; closed bars, fitness of P relative to P. (b) Degree of dominance (d/a) estimated from F1 and RP of F1 and F2 in seed number. Dots denote hybrids of a particular population and climatic zone origin.

The generation effect, that is, a general decrease in performance of plants from F1 to F2, was independent of pollen origin (Table 3, Fig. 5), while the interaction between pollen origin and pedigree (hybrid vs parent) was significant (Table 1). When pollen was of nondesert origin, seed number and total seed mass were higher in the progeny of hybrids as compared with the self-pollinated father plants. The reverse was true when pollen donors were plants of desert origin (Fig. 6).

Figure 6.

The effect of interaction between pollen origin and pedigree on performance of Hordeum spontaneum plants (relative to the P).

The performance of plants grown in a simulated desert environment progressively decreased with increase in the difference in aridity between the original locations of the mother and the father plants in the following order: desert, semi-arid, Mediterranean, mountain (Table S2). This effect was consistent over generations (F1 and F2), and in the two types of pedigree (P and crosses).

The degree of dominance (d/a) varied highly between the half-sibs, but positive values predominated for both seed number and total seed weight (Fig. 5, only seed number is shown), indicating that alleles conferring adaptation to the desert environment in these traits are mostly (but not exclusively) dominant.


One goal of this study was to supplement a home advantage transplant experiment directly testing for local adaptation with the QSTFST approach to see whether the latter test can reliably detect local adaptation and reveal otherwise unavailable information. Transplanting into a desert environment provided clear evidence of local adaptation. Performance of plants of nondesert origin was lower than that of natives, progressively decreasing from semi-arid to Mediterranean to mountain origin.

The results of the QST –FST test were also in agreement with adaptive selection along the aridity gradient. Population differentiation in molecular markers was much lower than in phenotypic traits despite self-pollination that is predicted to reduce the neutral genetic polymorphism within populations and to promote differentiation at all hierarchical levels as a result of genetic hitchhiking, lower effective population size, and limited gene flow (Charlesworth, 2003), thus making the QSTFST test more conservative for a self-pollinating than an outcrossing species. Early flowering is a known adaptation of annual plants to aridity (Aronson et al., 1992; Bennington & McGraw, 1995; Franke et al., 2006) and, indeed, desert plants had a shorter time period to flowering than nondesert plants. In addition, desert plants had larger flag and penultimate leaves and smaller spikes that contained smaller spikelets. Trait heritability was high, and this was because of high within-population genetic variation (Volis et al., 2005, 2010), which, in turn, is a result of fine-scale spatial mosaic of various locally abundant genotypes (Volis et al., 2010). Mosaic-like spatial genetic variation is created in wild barley by combined effects of selection, limited seed and pollen dispersal, and infrequent recombination (Volis et al., 2010, 2011).

Whitlock (2008) argued that although the strength of the QSTFST comparison is that it can rule out drift as an explanation for population divergence, the same information can be gained from a correlation of the traits with environmental gradient. However, the additional value of the QSTFST comparison is that it can reveal the intensity of gene flow along an environmental gradient by quantifying the neutral population differentiation.

In this study, desert populations were found to be locally adapted and strongly differentiated from nondesert populations in phenological and reproductive traits. This differentiation increased with environmental difference in aridity. At the same time, neutral divergence also increased with environmental difference in aridity because the latter was positively correlated with geographic distance and therefore with decrease in gene flow. However, while increase in neutral divergence was approximately linear, increase in selective divergence was not. Therefore the QSTFST approach provided not only a null expectation, but also a quantitative description of the amount of gene flow that selection has opposed.

The second goal of this study was to understand the genetic architecture of local adaptation (i.e. genetic effects responsible for fitness differences) in wild barley by analysis of the performance of interpopulation hybrids (F1), their recombined progeny (F2) and the parents (P) in a simulated desert environment. The results of the crossbreeding experiment also strongly supported adaptation of plants of desert origin to their local desert environment. Lower performance of nondesert parents, F1 and F2 hybrids than parents of desert origin was detected in a simulated desert environment. However, the crossbreeding experiment allowed additional insights into the genetic architecture of spatially structured phenotypic variation in wild barley.

Hybrid breakdown was observed in the F1 plants that resulted from pollination of a desert mother plant with pollen of nondesert origin. However, when the desert mother plant was pollinated with pollen of plants from other desert populations, a moderate heterosis was observed, potentially indicating the presence of deleterious alleles as a result of drift, despite strong environment-specific selection. This is not surprising as, in addition to being a predominant selfer, wild barley in deserts occupies only the most favorable water-accumulating microhabitats, such as ephemeral river beds (wadis), and therefore the desert populations are small (from hundreds to a few thousands of plants). Surprisingly, there was no reduction in either neutral or quantitative genetic variation of the desert populations as compared with the populations from the other three regions.

The Negev Desert is the area where wild barley reaches southern geographic limits of its distribution in Israel. Such a peripheral part of the species range can either be a result of range expansion with no current gene flow between the species core and periphery or undergo a repeated colonization. Under the first scenario, a colonization bottleneck is expected with depleted genetic variation (Pujol & Pannell, 2008), while under the second scenario no depletion and high adaptive potential as a result of recombination are expected (Pérez de la Vega et al., 1991; Lavergne & Molofsky, 2007). The heterosis observed in crosses between different desert populations coupled with no reduction in within-population genetic variation indicates effect of drift as a result of small population size despite the presence of some gene flow from the species core. In other words, gene flow from the core maintains high neutral genetic variation in peripheral populations but does not improve their adaptive potential.

In the F2 generation (progeny of selfed F1 plants), a general increase in outbreeding depression was observed, including selfed progeny of F1 plants of pure desert origin. These results are not fully consistent with either underdominance or break-up of co-adapted gene complexes as possible mechanisms of observed hybrid breakdown. Underdominance is expected to lead to a rapid restoration of fitness with an increase in homozygosity from the F1 generation on, while the break-up of epistatic interactions among loci as a result of recombination is expected not to be expressed until the F2 generation (Hufford & Mazer, 2003). An increase in outbreeding depression from F1 to F2 rules out underdominance, while the fact that the hybrid breakdown was already evident in the F1 plants indicates the presence of several mechanisms that are not mutually exclusive and may operate simultaneously. It could be that decreased performance of the F1 relative to the mother is caused by a loss of local adaptation while the decrease in the F2 relative to the F1 is a result of genetic effects not related to extrinsic selection, such as Dobzhansky–Muller incompatibilities.

Comparison of the performance of the F2 relative to F1 and P generations allows distinguishing additive effects from effects of dominance and gene interactions (epistasis) (Fenster & Galloway, 2000). If the basis of genetic differentiation between the two parent populations is additive, in a test of plant performance relative fitness of F1 should be intermediate between that of the parental plants, while performance of the F2 should not differ from the F1. If dominance is present, the performance of F1 will depend on whether the alleles associated with adaptation are dominant or recessive and the performance of the F2 should be intermediate to the F1 and midparent value because heterozygosity in the F2 will be half that of the F1 as a result of segregation. And if the two parental populations differ in gene interactions (i.e. epistasis is present), the fitness of the segregating and recombining F2 generation will be less than (F1 + midparent value)/2 (Lynch, 1991; Lynch & Walsh, 1998).

The observed relative fitness of the F1 varied with the origin of P, being close to 0.5 for pollen of mountain origin (the most extreme contrast to the desert environment), with higher values for pollen from Mediterranean and semi-arid environments (more similar to the desert). On the other hand, the relative fitness of the F2 was lower than the F1 in all the crosses. These results are consistent with a contribution of additive and epistatic but not dominance effects to the adaptive genetic differentiation of desert populations. However, a degree of dominance in the two performance traits, estimated from F1 plants, was nonzero and mostly positive; that is, alleles associated with adaptation are mostly dominant. This means that not only additive-by-additive epistasis (expected for highly homozygous species) but also dominance-by-additive epistasis (expected for outcrossing species with high heterozygosity) (Rousselle et al., 2010) operates in wild barley.

Thus, joint usage of the direct test of home advantage, QSTFST comparison and crossbreeding experiment provided evidence of simultaneous operation of three genetic factors in the desert part of wild barley distribution in Israel. The desert environment imposed a strong selection that caused differentiation of the desert plants from plants of nondesert origin in phenotypic traits, giving them home advantage. Gene exchange between adjacent populations (i.e. populations experiencing desert environment) via seeds (but not pollen) had a positive effect on the average population fitness by bringing in already selected (but new for this population) genotypes, thus enhancing natural selection and counteracting drift. Such an effect was theoretically predicted by Alleaume-Benharira et al. (2006), who showed that immigrants from the larger populations can be closer to the local optima than residents in smaller populations experiencing stronger drift. On the other hand, the positive effect of gene exchange via pollen was limited to F1 only, and disappeared after one generation of segregation (i.e. in F2). As for the effect of genes from the species’ distributional core (nondesert plants), it was negative for either seeds or pollen already in F1 despite its enriching effect on neutral diversity. These findings suggest that a short-term effect of gene flow by pollen in wild barley in general is negative because the majority of the new combinations of alleles created by recombination are maladaptive.

However, the long-term effect of occasional pollen flow can be positive in this species because recombination will create new combinations of alleles with enhanced fitness (Fig. 6). Similar effects of pollen flow were detected in wild barley at the fine within-population scale (Volis et al., 2011). These results agree well with known properties of this species, such as predominant but not complete selfing (98%), very limited major seed dispersal (95% of seeds are dispersed within 1.2 m) but with seed adaptations for entrapping in the fur of animals allowing long-distance dispersal. These results have important implications for considering bottleneck effects associated with population expansion (Pujol & Pannell, 2008; Pujol et al., 2009) or invasion (Facon et al., 2011). The new findings suggest that gene flow can restore not only neutral genetic variation, but also adaptive potential of a species depleted by the bottlenecks.

In addition, this study shows that a proper QSTFST study design can accomplish the goal of quantifying the intensity of selection in a similar manner to how it is done through reciprocal transplanting or crossbreeding. In this study, I used hierarchical sampling design with multiple populations per environmental type along an environmental gradient of aridity for both QSTFST comparison and crossbreeding. The results were in agreement, showing strong diversifying selection and increase in its intensity along an aridity gradient.

To conclude, the three approaches – the home advantage transplant experiment, crossbreeding with analysis of progeny in a common garden experiment, and the QSTFST test – should be viewed not as alternatives but rather as complementary methods. The first two approaches can be combined into one in which parental genotypes are transplanted together with several generations of their crossed/selfed progeny, allowing inferences to be drawn about the genetic architecture of adaptive population differentiation. Also, proper design and use of the same accessions for both crossbreeding with transplantation and a QSTFST study could allow disentangling and quantification of the effects of natural selection and drift.


I am grateful to Moshe Feldman, Marc Stift, Frank Sorensen, and three anonymous reviewers for their constructive and detailed comments on an earlier version of the manuscript. A grant from the Israel Academy of Sciences (86293101) supported this study.