SEARCH

SEARCH BY CITATION

Keywords:

  • adaptive radiation;
  • convergent selection;
  • ecotypic differentiation;
  • elevational gradient;
  • inflorescence height;
  • microevolution;
  • rockiness;
  • speciation

Abstract

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

Divergent selection is a key in the ecological theory of adaptive radiation. Most evidence on its causes and consequences relies on studies of pairs of populations or closely related taxa. However, adaptive radiation involves multiple taxa adapted to different environmental factors. We propose an operational definition of divergent selection to explore the continuum between divergent and convergent selection in multiple populations and taxa, and its links with environmental variation and phenotypic and taxonomic differentiation. We apply this approach to explore phenotypic differentiation of vegetative traits between 15 populations of four taxa of Iberian columbines (Gen. Aquilegia). Differences in soil rockiness impose divergent selection on inflorescence height and the number of flowers per inflorescence, likely affecting the processes of phenotypic and, in the case of inflorescence height, taxonomic diversification between taxa. Elevational variation imposes divergent selection on the number of leaves; however, the current pattern of divergent selection on this trait seems related to ecotypic differentiation within taxa but not to their taxonomic diversification.


Introduction

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

The ecological theory of adaptive radiation places divergent selection, arising from differences in the environment, as the primary driver of phenotypic differentiation among populations and closely related taxa (Schluter, 2000). In a review of studies comparing the variability among populations for neutral molecular markers against variation in quantitative traits, Leinonen et al. (2007) concluded that divergence because of natural selection is the norm rather than the exception in these studies. This conclusion is also supported by studies directly analysing the association between traits, fitness and the environment across populations or closely related taxa (Linhart & Grant, 1996; Kawecki & Ebert, 2004; Mullen & Hoekstra, 2008). The vast majority of these studies focused on single pairs of taxa (or on intra-specific comparisons) and analysed the role of a single environmental factor as the causal agent of divergent selection. To understand adaptive radiation, we need a broader view because the process involves the proliferation of multiple taxa adapted to a variety of habitats which may differ in an array of environmental factors.

Experimental approaches have important limitations in their applicability to studies requiring the use of a wide array of populations and taxa under field conditions. For example, field experiments of genetic differentiation and local adaptation in plants usually include a few populations but require plantations of more than a thousand plants in each locality where the experiment is conducted (see for example Galloway & Fenster, 2000; Kittelson & Maron, 2001). Besides the logistical limitations, implementation of very large experimental designs in the field might distort the natural ecological scenario where the experiment is conducted, possibly biasing the results. The approach proposed by Lande & Arnold (1983) for measuring natural selection in the field is widely used (see Kingsolver et al., 2001 for a review), and it offers a more tractable way to study divergent selection and phenotypic differentiation in a wide array of populations and taxa (see for example Møller et al., 2006). The coefficients of selection obtained through the Lande–Arnold approach can be used to estimate the magnitude of divergent selection between pairs of populations. For example, compelling evidence for the action of divergent selection is provided by the correlation between the difference in selection coefficients (a measure of the strength of divergent selection) and the difference in phenotype between populations (Cruz et al., 2004; Nosil & Crespi, 2006; Carlson et al., 2009). The general hypothesis of this approach is that if current selection drives differentiation between populations, traits with large differences in selection coefficients between populations should show very different means in each population, whereas traits with similar selection coefficients should be more similar between populations. Moreover, if the differences in selection coefficients could be clearly linked to differences in some environmental factors, it could be concluded that the process of phenotypic differentiation has been driven by adaptation to the environmental factors in question. In this study, we use a similar rationale to (1) quantify the strength of divergent selection among multiple populations, (2) to relate the patterns of divergent selection among populations to variation in environmental gradients and (3) to test whether the current patterns of divergent selection may have been involved in the process of phenotypic diversification within and between closely related taxa.

We address these topics in a group of European columbines (genus Aquilegia; Ranunculaceae) to begin to understand how adaptation to different environments may have shaped the taxonomic diversity of the genus in this continent. Columbines are a classic example of adaptive radiation in relation to pollinator specialization (Hodges & Arnold, 1994). However, the diversity of this genus involves not only floral traits and the use of different types of pollinators, but also a wide variation in vegetative traits and the use of a wide variety of habitats, like springs, mesic forest understories, meadows or rocky outcrops, subarctic areas, temperate forests, Mediterranean climates and deserts in North America and Eurasia (Munz, 1946; Nold, 2003). Thus, the diversification of the genus may have been related also to reproductive isolation derived from the use of different habitats (Grant, 1952, 1976; Chase & Raven, 1975; but see Fulton & Hodges, 1999). We have shown (Bastida et al., 2010) that habitat differentiation may have been particularly important in the case of the European columbines, whose taxonomic diversity is similar to that of North American ones, even though they rely on a lower diversity of pollinators. Accordingly, Medrano et al. (2006) found that, even though floral and vegetative morphological differentiation occur between populations of the widespread Aquilegia vulgaris subsp. vulgaris and the narrow endemic Aquilegia pyrenaica subsp. cazorlensis in Southern Spain, only vegetative characters contribute significantly to discrimination between taxa.

The relevance of vegetative traits in processes of taxonomic diversification can be grasped from studies on habitat–trait associations between closely related species. Compared to congeneric widespread taxa, endemic plant taxa often occur at higher elevation (McDonald & Cowling, 1995, Kessler 2002), or in slopping and rocky habitats with sparse vegetation (Baskin & Baskin, 1988; Matthews et al., 1993, Lavergne et al., 2004). Accordingly, phenotypic differentiation in vegetative traits is common between closely related taxa occupying contrasting parts of these environmental gradients. Conspecific populations growing in higher elevation or in dryer environments usually have smaller stature and shorter leaves (e.g. Clausen et al., 1948; Jonas & Geber, 1999; Monty & Mahy, 2009). Lavergne et al. (2004) found that in Western Mediterranean plants, endemic species occur in habitats with shallow-rocky soils with low aboveground competition, and that these endemics have shorter inflorescences with fewer flowers than their widespread conspecifics. Physiological or functional vegetative traits [such as leaf size, specific leaf area (SLA), and photosynthetic rates and efficiencies] have been found to vary among closely related taxa occurring in habitats with different light regimes (Givnish et al., 2004; Santiago & Kim, 2009). These examples suggest that taxonomic diversification might be related to processes of adaptive differentiation of vegetative traits in response to divergent selection among environments.

Although many environmental factors have been identified as potentially influencing adaptive responses within and between species, directly linking natural environmental variation with adaptive phenotypic responses remains challenging (Nakazato et al., 2008). Such a link is provided by divergent selection. To further understand the process of adaptive diversification, it is critical to assess specifically which environmental factors impose the patterns of divergent selection that may account for the observed patterns of phenotypic differentiation within and between closely related taxa. To this end, we propose an operational definition of divergent selection: a trait experiences divergent selection between two populations if their observed differences in trait means could potentially increase as a consequence of the current pattern of selection (see Fig. 1). This definition differs from the standard definition (e.g. Schluter, 2000), but it has the advantage of incorporating the possibility of convergent and divergent selection in a single estimate, in a context of multiple populations. Using this approach, we evaluate the strength of divergent selection among 15 populations from four columbines from the Iberian Peninsula (A. vulgaris subsp. vulgaris, Aquilegia vulgaris subsp. nevadensis, A. pyrenaica subsp. pyrenaica and Aquilegia pyrenaica subsp. cazorlensis). Our results clearly show that occupancy of habitats differing in soil rockiness imposes divergent selection on inflorescence height and number of flowers per inflorescence, and that this divergent selection is likely involved in the processes of phenotypic differentiation and, in the case of inflorescence height, taxonomic diversification among the studied columbines. On the other hand, we have found that the number of leaves is subject to divergent selection in habitats at contrasting elevation, and that this pattern of divergent selection is likely involved in the process of differentiation among populations within taxa, but not in the process of taxonomic diversification.

image

Figure 1.  Idealized example of the patterns of divergence between populations and estimates of the coefficient of Divergent Selection (D). The observed means for trait z in seven populations (a–g; the vertical order of the populations is arbitrary) are indicated with black points. The estimated selection differential for the trait in each population (sz) is indicated with an arrow. The maximum population mean after selection (z*) in each population is indicated by a white point. D is estimated as the predicted distance between populations after selection (d*) minus the current (observed) distance of population means (d), as shown for populations f and g. For example, using this measure, the pairs of populations (a, b) and (a, d) would diverge if they respond to the current patterns of selection; the pairs (a, c) and (d, e) would converge, and the pairs (a, e) and (b, c) would not converge nor diverge as their differences would remain of the same magnitude after selection.

Download figure to PowerPoint

Methods

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

Study taxa and populations

The study populations belong to two pairs of conspecific subspecies: subspecies vulgaris and nevadensis of A. vulgaris, and subspecies pyrenaica and cazorlensis of A. pyrenaica. Columbines in general are perennial herbs with a slender rhizomatous stem. Their basal rosettes are formed by a few pubescent compound leaves. Mature plants produce a variable, although small, number of paniculate inflorescences bearing also a small number of flowers (mean values for these traits are given for each study population in Supporting Information). Aquilegia vulgaris subsp. vulgaris (A. v. vulgaris hereafter) is widespread from the Iberian Peninsula to Eastern Europe. It grows in permanently wet places, near streams and springs, in the forest understory and clearings, from the sea level to 2000 m elevation. In contrast, A. vulgaris subsp. nevadensis (A. v. nevadensis hereafter) is endemic from southeast Iberian Peninsula, with few known populations. It also grows in permanently moist soils in the forest but, in addition, it occurs also in wet alpine meadows and scrublands. According to Díaz González (1986), it occurs from 1100 to 2500 m elevation, although we found no populations below 1500 m elevation. Both subspecies grow in calcareous and siliceous soils. However, both are more common on developed soils than on rocky substrates. Aquilegia pyrenaica subsp. pyrenaica (A. p. pyrenaica hereafter) is distributed through the Pyrenees and Cantabrian Mountains (Northern Spain), occupying alpine meadows, rocky outcrops and calcareous rocky grasslands from 1200 to 2250 m elevation. Aquilegia pyrenaica subsp. cazorlensis (A. p. cazorlensis hereafter) is endemic from the southeast of the Iberian Peninsula, with few populations known in Sierra de Cazorla and Sierra de Castril. Like A. p. pyrenaica, it inhabits rocky outcrops and cliff bases, but always in shaded places and never in open meadows or grasslands. Its altitudinal distribution spans from 1600 to 2000 m elevation. Both subspecies of A. pyrenaica are restricted to shallow calcareous soils.

Operational definition of divergent selection

Estimates of selection coefficients

Between May and August 2007, we collected information on fruit production and phenotypic traits in 3–4 populations of each taxon. Sample sizes for each population are given in Supporting Information, and ranged from 21 to 52 plants (mean 45), depending on the number of reproductive individuals available or accessible during the season. The traits measured on each plant were: total number of ripe fruits, mean number of healthy ripe carpels (carpels with evidence of infection or any other damage were not counted), number of inflorescences, height of the tallest inflorescence, the number of leaves, length of the longest leaf, number of flowers per inflorescence, SLA and density of glandular and nonglandular pubescence on the leaves. SLA was determined in the laboratory from a sample of the longest leaf. The density of pubescence in the leaves was estimated under dissecting microscope as the mean of five samples (three along the petiole plus one sample in each face of the leaf blade), each sample from a strip (10 mm long by 1 mm wide) of fresh epidermal tissue.

To obtain estimates of phenotypic selection in each population, we followed the procedure developed by Lande & Arnold (1983). We note that our primary goal was parameter estimation rather than significance testing of selection coefficients, which is reasonable as estimates of selection gradients and differentials are not affected by departures from normality, but their tests of significance rely on assumptions about the error distribution which can be difficult to fulfil (Lande & Arnold, 1983; Nosil & Crespi, 2006). In any case, to improve normality of the traits and minimize possible correlations between trait means and variances, we log-transformed the traits before analyses. Focus on estimated parameters rather than in their statistical significance is common in studies estimating selection or genetic parameters in wild populations (see for example Andersson, 1996; Fedorka et al., 2007; Frentiu et al., 2007). From a statistical point of view, our sample sizes are small for detecting significant selection coefficients of moderate to small magnitude. However, note that our sample sizes can hardly be increased because the total number of reproductive individuals in our study populations is small (range 27–350, median = 120; see Supporting Information). Even sampling all the plants in most of our study populations, the power to detect significant selection differentials of small magnitude would still be low: for a population of average size in our study (mean N = 138; median = 115), the probability that one could reject the null hypothesis of no correlation between fitness and a trait (which is equivalent to the covariance between the variables, i.e. the selection differential, standardized to unit variance) would be lower than 95% for any correlation coefficient r < 0.3. This problem of small population size is inherent to many plant species, and it seems common in Aquilegia. For example, in a study of 40 populations of Aquilegia caerulea (Mavraganis & Eckert 2001), population size ranged from 4 to 768 individuals, with a mean of 140.9 and median value of 62 plants. In any case, analysis of selection in small populations is clearly relevant on biological grounds as small population size has many key implications in processes of phenotypic and genetic differentiation, and speciation.

We excluded the number of inflorescences from the analyses because it was largely constant among individuals in some populations. Similarly, we excluded the density of glandular pubescence as some populations lacked it. Thus, six phenotypic traits were eventually included in the analyses: inflorescence height, number of leaves, length of the longest leaf, mean number of flowers per inflorescence, SLA and density of nonglandular pubescence on the leaves. We used the number of healthy ripe carpels per plant as estimate of fitness.

Directional selection gradients (βzj) for each population (j) and trait (z) were estimated through multiple regression analysis of relative fitness on plant traits standardized to zero mean and unit variance. These gradients measure the strength of direct selection on a given trait, independent of selection on the other traits included in the analysis. They are suitable for identifying the traits directly targeted by selection within each population. Directional selection differentials (szj) were estimated as the covariance between relative fitness and unstandardized plant traits. Selection differentials estimate the strength of total selection on a trait, including direct selection and indirect selection through its correlation with other traits under selection. As differentials do not estimate the direct selection on a trait, they cannot be used to identify the traits targeted by selection. However, unlike selection gradients, estimates of selection differentials do not require all traits under selection to be included in the analysis.

Estimates of divergent selection

Coupled with information on trait heritability (inline image), selection differentials can be used to predict the change in population means (zj) from one generation to the next (Roff, 1997):

  • image

It is essential to note that szj represents the expected amount of change in the population mean if the trait was genetically free to respond to selection. In this case, zj + szj indicates the population mean that would be favoured by selection. The actual magnitude of change will depend on the heritability of the trait in the population. We estimated the magnitude of divergent selection on a trait between each pair of populations [Dz(i,j′)] as the difference between the Euclidean distance among their maximum population mean after selection (dz*) and the Euclidean distance among their observed trait means (dz):

  • image

This formulation can be considered an operational definition of divergent selection: a trait experiences divergent selection between two populations if their observed differences in trait means could potentially increase as a consequence of the current pattern of selection [Dz(j,j′) > 0; see Fig. 1]. Contrarily, the trait would be under convergent selection if observed differences in trait means could potentially decrease as a consequence of the current pattern of selection [Dz(j,j′) < 0; Fig. 1]. We can describe this formulation as the difference in selection differentials after controlling for the relative value of the phenotypic means. Although Dz(j,j′) measures both divergent and convergent selection, we will refer to it as a measure of divergent selection for simplicity.

Before estimating s, it is necessary to minimize the possible correlations between trait means and variances across populations. This is necessary because s is the covariance between relative fitness and traits, so populations with higher variance for a trait would show larger s even if direct selection on the trait was of the same magnitude in the populations. Thus, if trait means and variances were correlated, trait means and estimates of s might also be correlated across populations, even if they experienced the same strength of selection. This would induce a positive bias in estimates of Dz(j,j′) because populations with more different means would tend to show larger differences in s. There is no clear strategy for avoiding this problem in an observational study, as standardization of traits to unit variance within populations (as is common practice in analyses of selection) would make comparisons of Euclidean distances between different pairs of populations unreliable (because each population would be standardized to a different scale because of their different variances). An alternative to solve this problem is to apply some transformation to the whole data set that removes the correlation between mean and variance across populations. As mentioned before, we transformed traits to log-scale before estimating selection coefficients. Compared with results obtained with untransformed trait values (data not shown), the log-transformation substantially reduced the correlations between means and s in all traits, except in the number of leaves (see Results). In any case, it must be noted that observed population means and means predicted after selection must be back-transformed from log-scale to their original scale before computing the Euclidean distances, otherwise estimates would indicate ratios of trait means rather than linear distances between population means.

Estimation of Dz(j,j′) for a given trait between all the possible pairs of populations renders a matrix (Dz) of divergent selection on the trait. Although our formulation of Dz(j,j′) is easily adapted to multivariate phenotypic distances, multivariate directions of divergence may be difficult to interpret because different pairs of populations may differ in different traits and still render the same estimate for Dz(j,j′). Moreover, if the traits included in the multivariate estimate have different metrics, as in this study, the interpretation of the multivariate distance might be misleading. Thus, we built a univariate Dz matrix for each trait.

Measures of environmental differentiation

The environment in each population was characterized in terms of elevation above the sea level, rockiness (percentage of the surface occupied by rocks) and above ground cover (estimated as the inverse of mean Global Site Factor Index (GSF) obtained from hemispherical photographs; Clark et al., 1996; Ramírez et al., 2006). From these data, we estimated the difference between populations for each environmental factor as the Euclidean distance between their respective values. A large distance indicates that the populations occur in very different environments, whereas a zero distance indicates that the two populations occur in the same environment. As above, we did not use multivariate distances because the three environmental variables are measured in different scales. Distances between all possible pairs of populations were combined in a matrix (Ei) for each environmental variable (i). In addition to these matrices of environmental differentiation, we also obtained a matrix of geographic distances (K) estimated as the Euclidean distance between each pair of populations based on their UTM coordinates.

Patterns of phenotypic differentiation

As a first step in the analyses, we explored the variation of plant traits between species, subspecies and populations. For each trait, we conducted a nested anova testing for differences between species, subspecies nested in species, and population nested within subspecies and species. The population effect was considered random. Next, we explored the patterns of covariation between plant traits and the environment across populations. To this end, we conducted a multiple regression evaluating the relationship between each plant trait (dependent variable) and the environmental factors (including the latitude of the population) as independent variables. These analyses were conducted in Statistica (StatSoft, Inc, 2004).

Relationships between environmental variation, divergent selection and phenotypic differentiation

To assess whether populations occurring in different environments experience divergent selection, we regressed each Dz on the three Ei matrices and K. A significant positive effect of Ei on Dz indicates that populations occupying more contrasting environments (or, in the case of K, located further away from each other) experience more divergent selection on the focal trait. We used Multiple Mantel tests to perform multiple matrix regression, evaluating the significance of effects through 10 000 permutations of the dependent matrix.

To test whether the observed patterns of phenotypic differentiation could have been caused by the observed patterns of divergent selection, we computed the correlations between each Dz and dz (the matrix of observed phenotypic distances among populations for trait z) using Mantel correlation tests (significance tested through 10 000 randomizations of the matrices). A significant correlation would indicate that the patterns of current divergent selection acting on the trait might have caused the observed pattern of phenotypic differentiation. However, a relationship between divergent selection and phenotypic differentiation does not necessarily imply that divergent selection on a given trait has contributed to the process of taxonomic diversification. It would be possible that the patterns of divergent selection were only related with a process of local adaptation and ecotypic differentiation between populations. To ascertain whether the current patterns of divergent selection on a given trait might have been involved in the process of taxonomic diversification, we conducted a multiple matrix regression of T (a matrix of taxonomic distances between populations) on those Dz for which we detected a significant correlation with dz. The ideal information for building T would be a fully resolved phylogeny, including branch lengths, of the populations. As we lack such information, we assumed a simple phylogeny based on taxonomic information of the Iberian columbines (Díaz González, 1986). A population-level Amplified Fragment Length Polymorphism (AFLP) phylogeny of the Iberian columbines including most of our study populations supports the validity of this assumption (J. L. Garrido, personal communication). Populations of the same subspecies formed a politomy, and subspecies were nested within species. Branches in the phylogeny were assumed equal and of unit length: pairs of populations of the same subspecies were assigned distance (or branch length) 1, those of different subspecies within the same species were assigned distance 2 and those of different species were assigned distance 3. A positive relationship between Dz and T would indicate that the current pattern of divergent selection agrees with the pattern which led to the current taxonomic structure of the populations. All the analyses described in this section were implemented using the software R, using the ecodist package (Goslee & Urban, 2007).

As Dz measures current patterns of divergent selection, the lack of relationships between Dz and the other matrices does not imply that such relationships did not occur in the past. For example, if a trait evolved under divergent selection but is already at equilibrium in most populations we would find small or zero values of s, and no relationships between Dz and dz (or E, K or T). In such cases, one cannot conclude that divergent selection was not involved in the process of phenotypic differentiation. However, if the relationship were significant, it would strongly suggest that patterns of divergent selection similar to those we see nowadays have been involved in the process of phenotypic differentiation.

Results

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

Patterns of phenotypic differentiation

All the traits showed significant differences between populations, and all of them varied also between subspecies or species (Table 1; see Supporting Information for trait means). Inflorescence height, the number of leaves, SLA and the density of nonglandular pubescence showed significant variation along the environmental gradients under study (Table 2). Variation among populations in inflorescence height and the density of nonglandular pubescence were negatively related to variation in soil rockiness (Fig. 2a,d). These correlations were related to differences between species, as A. vulgaris occupies more developed soils and has taller inflorescences and denser pubescence on the leaves than A. pyrenaica. On the other hand, the number of leaves increased with elevation (Fig. 2b). The opposite extremes of the gradient are mostly occupied by populations of the two subspecies of A. vulgaris. One extreme of the gradient is occupied by populations of A. v. vulgaris occurring in low elevation and having fewer leaves. The other extreme is occupied by populations of A. v. nevadensis which occur at high elevation and have more leaves. Populations of both subspecies of A. pyrenaica occupy intermediate positions, with no clear segregation along the gradient. SLA decreased with latitude, but this relationship is strongly marked by the lower SLA of the northern subspecies A. p. pyrenaica. Excluding this subspecies from the analysis, there was no relationship between latitude and SLA.

Table 1.   Summary of univariate nested ANOVAs comparing plant traits (log-transformed) between species, subspecies and populations. Population effect was considered random. Error degrees of freedom for species and subspecies effects were adjusted using Satterthwaite’s method. Effects significant after Bonferroni correction are indicated in bold-type.
 SpeciesSubsp. (Sp.)Population [Subsp. (Sp.)]
Fd.f.PFd.f.PFd.f.P
  1. SLA, specific leaf area.

Inflorescence height75.051, 11.02< 0.00014.252, 11.02< 0.042729.7411, 649< 0.0001
Number of leaves2.461, 11.08< 0.03919.402, 11.07< 0.00418.1811, 646< 0.0001
Leaf length23.221, 11.03< 0.00067.422, 11.03< 0.009118.5111, 645< 0.0001
Number of flowers per inflorescence16.701, 11.02< 0.00183.232, 11.020.07926.1311, 649< 0.0001
SLA0.881, 11.040.36918.212, 11.02< 0.000339.1611, 613< 0.0001
Nonglandular pubescence90.791, 11.15< 0.00011.782, 11.150.21317.7311, 516< 0.0001
Table 2.   Summary of multivariate regression analyses describing environmental gradients of phenotypic variation in plant traits (log-transformed population means). Independent regression models were fit for each plant trait. The top three lines indicate model fit parameters. The rest of lines show standardized regression coefficients (± SE). Coefficients significant after table-wide Bonferroni correction are indicated in bold-type.
 Inflorescence heightLeaf lengthNumber of leavesNumber of flowers per inflorescenceSLANonglandular pubescence
  1. SLA, specific leaf area.

R20.780.600.610.590.720.72
F4,1013.596.166.536.129.8710.09
Model P <0.0010.010.0080.010.0020.002
Soil rockiness0.60 ± 0.15−0.30 ± 0.200.59 ± 0.20−0.33 ± 0.200.32 ± 0.170.82 ± 0.17
Elevation−0.04 ± 0.130.14 ± 0.180.80 ± 0.180.34 ± 0.18−0.23 ± 0.15−0.01 ± 0.15
GSF0.05 ± 0.160.03 ± 0.210.40 ± 0.210.25 ± 0.21−0.41 ± 0.18−0.28 ± 0.18
Latitude−0.50 ± 0.17−0.64 ± 0.23−0.43 ± 0.23−0.65 ± 0.240.68 ± 0.200.02 ± 0.19
image

Figure 2.  Relationships between population means for phenotypic traits and environment. Only significant relationships detected through multiple regression analyses (Table 2) are shown. Trait means are represented on natural log-scale. Populations from each subspecies are indicated through different symbols: (bsl00001) Aquilegia vulgaris subsp. nevadensis; (□) Aquilegia vulgaris subsp. vulgaris; (•) Aquilegia pyrenaica subsp. cazorlensis; (○) Aquilegia pyrenaica subsp. pyrenaica.

Download figure to PowerPoint

General patterns of selection

The mean of absolute values of our estimates of standardized selection gradients was 0.217, very similar to the mean values of 0.22 reported by Kingsolver et al. (2001) across a large sample of studies, or 0.20 reported by Geber & Griffen (2003) specifically for plants. We found significant selection gradients on at least one population of each subspecies (Table 3a). On the other hand, we could not detect significant direct selection on the density of nonglandular pubescence in any population. Leaf length and SLA had significant selection gradients only in one population each. Inflorescence height was under direct selection in four populations. The number of leaves and number of flowers per inflorescence had significant selection gradients in five populations. Most significant selection gradients were positive, with the only exception of selection on leaf length, which experienced significantly negative selection in one population of A. v. nevadensis.

Table 3.   Selection gradients (a) and differentials (b) estimated for each (log-transformed) trait and population. Traits were standardized to zero mean and unit variance within each population to estimate selection gradients. Unstandardized trait values were used to obtain selection differentials. Significant coefficients are indicated in bold-type. Traits are indicated as Inf. Height: height of the tallest inflorescence; N. leaves: number of leaves per plant; leaf length: length of the longest leaf; N. flowers: number of flowers per inflorescence; SLA: specific leaf area and NG. Pub.: density of nonglandular pubescence on the leaves. Taxa are indicated as A.v.v.: Aquilegia vulgaris subsp. vulgaris; A.v.n.: Aquilegia vulgaris subsp. nevadensis; A.p.c.: Aquilegia pyrenaica subsp. cazorlensis; A.p.p.: Aquilegia pyrenaica subsp. pyrenaica.
TaxonPopulationInf. heightN. leavesLeaf lengthN. flowersSLANG. Pub.
(a)
 A.v.v.F. Reina0.14−0.020.020.40−0.05−0.08
 A.v.v.S. Cabrilla0.770.33−0.02−0.34−0.06−0.31
 A.v.v.Garrotegordo0.72−0.470.241.210.290.38
 A.v.v.B. Jabalises0.460.19−0.13−0.48−0.09−0.02
 A.v.n.Pradollano0.050.49−0.010.27−0.01−0.13
 A.v.n.S. Maroma0.070.180.000.42−0.10−0.07
 A.v.n.Cortijuela0.300.460.210.13−0.09−0.02
 A.v.n.F. Fria0.190.160.150.320.150.01
 A.p.c.Cabañas0.580.16−0.540.240.09−0.02
 A.p.c.B. Canal0.500.46−0.240.480.300.30
 A.p.c.B. Charca−0.100.110.130.31−0.110.17
 A.p.c.C. del Aire0.080.050.050.320.06−0.12
 A.p.p.Tobazo−0.010.340.330.04−0.02−0.17
 A.p.p.Tortiellas−0.240.130.200.020.020.15
 A.p.p.Larra0.200.61−0.200.09−0.050.02
(b)
 A.v.v.F. Reina0.050.050.020.17−0.030.01
 A.v.v.S. Cabrilla0.150.140.160.100.01−0.08
 A.v.v.Garrotegordo0.01−0.050.070.210.010.05
 A.v.v.B. Jabalises0.06−0.01−0.01−0.060.03−0.02
 A.v.n.Pradollano0.090.330.140.10−0.040.15
 A.v.n.S. Maroma0.110.140.070.20−0.030.01
 A.v.n.Cortijuela0.060.250.040.10−0.03−0.01
 A.v.n.F. Fria0.110.260.110.150.040.01
 A.p.c.Cabañas0.150.100.070.100.02−0.09
 A.p.c.B. Canal0.190.310.190.210.090.05
 A.p.c.B. Charca0.040.050.040.09−0.010.04
 A.p.c.C. del Aire0.060.030.070.110.02−0.05
 A.p.p.Tobazo0.120.210.140.120.01−0.05
 A.p.p.Tortiellas−0.010.080.010.00−0.010.05
 A.p.p.Larra0.070.300.040.10−0.060.04

Similarly, we found significant selection differentials on at least one population of each subspecies (Table 3b). We detected significant total selection on SLA and nonglandular pubescence in only one population, whereas for the rest of traits, there were significant selection differentials in at least one population of each subspecies. All the significant selection differentials were positive except for selection on nonglandular pubescence in one population. Selection differentials were not significantly correlated with phenotypic variance of the traits (r: −0.07, 0.0016, 0.29, 0.19 and −0.33 for inflorescence height, leaf length, flowers per inflorescence, SLA and nonglandular pubescence, respectively; > 0.05 in all cases), except in the case of the number of leaves (= 0.75, < 0.05). As SLA and nonglandular pubescence were not related to any of the environmental gradients considered, and we detected so few significant selection differentials acting on them, we did not consider these traits in subsequent analyses.

Across populations, selection differentials were significantly correlated with selection gradients in the case of inflorescence height, number of leaves and number of flowers per inflorescence (r: 0.66, 0.72, 0.76, respectively; = 15, < 0.05 in all cases). This correlation was not significant in the case of leaf length (= 0.02, > 0.05). Thus, with the exception of leaf length, total selection on the traits is largely based on direct selection. In the case of leaf length, indirect selection seems to act in opposition to direct selection, because total selection on this trait was significant and positive in many populations, whereas direct selection was not significant and negative in many populations.

Analysis of divergent selection

Matrix regressions of Dz on Ei (Table 4) suggest that GSF is not significantly related to current divergent selection on the studied traits. The matrices of divergent selection on inflorescence height and number of flowers per inflorescence were significantly related to the matrix of differences in soil rockiness between populations (Table 4). Figure 3a shows that most differentiation in soil rockiness occurs in inter-specific comparisons, which also show the stronger divergent selection on inflorescence height. Intra-specific comparisons (between subspecies of the same species or between populations of the same subspecies) show lower differentiation in soil rockiness and also less divergent selection, or even convergent selection in comparisons between populations of A. p. cazorlensis. In the case of the number of flowers per inflorescence (Fig. 3b), the strength of divergent selection increases also with the magnitude of differentiation in soil rockiness, and this pattern depends on the level of comparison. Thus, with the exception of comparisons of populations within A. v. vulgaris, populations of the same subspecies experience less divergent selection than populations of different species or subspecies. Divergent selection on leaf length and the number of leaves was significantly related to differences between populations in elevation (Table 4). As Figs 3c,d show, inter- and intra-specific comparisons did not differ clearly in the magnitude of divergent selection on leaf number and length.

Table 4.   Summary of multivariate regression (Mantel) tests of trait divergence/convergence matrices on dissimilarity matrices of environmental factors. The top three lines indicate model fit parameters. The rest of lines show raw regression coefficients and their significance level (within brackets) according to permutation tests. Coefficients significant after Bonferroni correction are indicated in bold-type. Specific leaf area and nonglandular pubescence were not considered (see Results).
 Inflorescence heightLeaf lengthNumber of leavesNumber of flowers per inflorescence
R20.190.130.200.13
F6.063.786.063.63
Model P <0.00010.0070.00010.009
Diff. soil rockiness2.93 (0.0007)0.40 (0.283)−0.24 (0.398)0.34 (0.006)
Diff. elevation0.002 (0.090)0.002 (0.003)0.002 (0.0001)0.0001 (0.63)
Diff. GSF0.55 (0.758)−0.17 (0.841)−1.28 (0.053)−0.31 (0.22)
Geographic distance0.004 (0.003)0.001 (0.016)0.0002 (0.70)0.0004 (0.038)
image

Figure 3.  Relationship between environmental differentiation and the strength of divergent selection. Only significant relationships according to Mantel’s test are shown (see Table 4): (a) inflorescence height vs. soil rockiness, (b) number of flowers per inflorescence vs. soil rockiness, (c) leaf length vs. elevation and (d) number of leaves vs. elevation. Each point represents the mean environmental distance and divergence coefficient for pairs of populations grouped, for clarity, as follows: within subspecies (pairs where both populations belong to the same subspecies), between subspecies within species (pairs where each population belongs to a different subspecies of the same species) and between species (pairs where each population belongs to a different species).

Download figure to PowerPoint

The current patterns of divergent selection acting on inflorescence height, number of leaves and number of flowers per inflorescence were significantly correlated with the respective patterns of phenotypic differentiation (Mantel correlations between Dz and dz: 0.31, 0.62 and 0.44, respectively; < 0.008 in all cases). However, such correlation was not significant in the case of leaf length (Mantel correlation: 0.01; > 0.44). Multiple matrix regression of T on Dz (Table 5) indicates that the taxonomic structure of the populations agrees with the current patterns of divergent selection on inflorescence height: populations of different subspecies or species experience stronger divergent selection on inflorescence height. Thus, the current patterns of divergent selection on inflorescence height would act to reinforce the patterns of taxonomic diversification, favouring more similar phenotypes between populations of the same taxa, and increased phenotypic differentiation between populations of different taxa. Figure 4 represents the structure [as assessed by unweighted pair-group average (UPGA) clustering] of the Dz matrices for inflorescence height, number of leaves and number of flowers per inflorescence. As indicated by the matrix regression of T on Dz, the structure of divergent selection on inflorescence height resembles, to some extent, the taxonomic structure of the populations. At the species level, the strongest divergent selection on inflorescence height largely, although not perfectly, segregates populations of A. vulgaris from A. pyrenaica. Within A. pyrenaica, convergent selection predominates within A. p. cazorlensis. In the case of A. vulgaris, divergent selection would tend to segregate two populations of A. v. vulgaris which experience convergent selection with some populations of A. p. cazorlensis. Selection was largely convergent between the other populations of A. vulgaris, preventing a clear differentiation between its subspecies for this trait. In the case of number of leaves and flowers per inflorescence (Fig. 4), divergent selection segregates populations of the same species and subspecies, whereas convergent selection would tend to increase the similarity of populations of different taxa, all of which suggest that current patterns of divergent selection may lead to differentiation within taxa, but would prevent differentiation between taxa.

Table 5.   Summary of multivariate regression (Mantel) test of taxonomic distances on trait divergence/convergence matrices. Geographic distance was included in the analysis to control for the spatial distribution of the populations. The top three lines indicate model fit parameters. The rest of lines show raw regression coefficients and their significance within brackets (according to permutation tests). Significant coefficients are indicated in bold-type.
 Full model
R20.20
F4.81
Model P <0.0005
 Regression coefficient (P-level)
Inflorescence height0.065 (0.008)
Number of leaves0.050 (0.36)
Leaf length−0.083 (0.15)
Number of flowers per inflorescence0.129 (0.42)
Geographic distance0.001 (0.002)
image

Figure 4.  Results of unweighted pair-group average (UPGA) clustering of populations based on the matrices of divergent selection (D) for inflorescence height, number of flowers per inflorescence and number of leaves. Positive values on the x-axis indicate divergent selection, and negative values indicate convergent selection between groups.

Download figure to PowerPoint

Discussion

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

Covariance across populations between mean phenotype and environment suggests the existence of divergent selection between environments (Endler, 1986), and there are abundant examples in the literature of such covariation (Endler, 1977). Accordingly, many studies have detected divergent selection among populations (reviewed in Schluter, 2000). Most of these studies have analysed the patterns of divergent selection and phenotypic differentiation imposed by contrasting environments, like serpentine vs. nonserpentine soils (Wright & Stanton, 2007), lake vs. stream (Berner et al., 2008) or high vs. low elevation (Byars et al., 2007). Moreover, these studies focus on single pairs of taxa or on intra-specific comparisons. We present a new analytical approach to the study of divergent selection among multiple populations and taxa. In analyses of differentiation involving multiple populations or taxa, it is necessary to consider that the process may have been related to several environmental factors, each possibly affecting a different set of traits and possibly contributing to differentiation at different levels (within and between taxa). This approach is particularly useful for the study of processes of recent adaptive radiations, as it allows the evaluation of links between environmental variation, divergent selection and phenotypic and taxonomic differentiation. Moreover, as Dz reflects current patterns of divergent selection, its relationship with the observed phenotypic differences may fade away with increased phylogenetic distance between populations or taxa, because the patterns of selection that led to the current phenotypic differences may have changed or populations may be already at equilibrium. Thus, this approach is more appropriate for the study of diversification within taxa or in the study of recent or undergoing processes of diversification.

Previous considerations

Significant positive correlations between Dz and dz or between Dz and T clearly indicate that current patterns of selection can partly explain the process of phenotypic or taxonomic differentiation between our study taxa. However, the absence of such correlations may have many causes, both ecological (e.g. inter-annual variation in the strength of selection, selection acting on the traits through unmeasured fitness components or biased estimates of selection coefficients because of covariation between traits, environment and fitness), genetic (e.g. large differences in trait heritability between populations, strong genetic correlations between traits) and historical (the environmental factors causing the pattern of divergent selection that led to the current pattern of differentiation may not be operating nowadays). Our approach is specifically tailored to assess the ecological factors behind divergent selection and to which extent current divergent selection can explain the observed patterns of phenotypic and taxonomic differentiation independently from genetic and historical factors. Thus, discussion on the role of genetic and historical effects is beyond the scope of this study.

Selection coefficients estimated using the procedure of Lande & Arnold (1983) can be biased by simultaneous covariance between environmental factors varying within populations and both traits and fitness (Mitchell-Olds & Shaw, 1987). Several approaches have been proposed to control for this bias. Following Rausher (1992), some studies have used different breeding designs (half-sibs, full-sibs, recombinant inbred lines) in field experiments, using the obtained breeding values instead of raw phenotypic values to estimate selection coefficients (Tiffin & Rausher, 1999; Stinchcombe & Rausher, 2001; Agrawal et al., 2004; Etterson, 2004; Hall & Willis, 2006). Estimates of breeding values for even a single population require large sample sizes and field settings, posing strong logistic limitations to their use in many habitats (e.g. rocky places, scrublands or forest understory), for many types of species (e.g. large plants and animals, or species typically growing at low density or with small population size) and particularly in studies of a wide sample of populations. Alternatively, Scheiner et al. (2002) proposed inclusion of a measure of a relevant environmental factor or a measure of individual condition (like total body weight) in a path analysis to reduce environmental biases in the estimated selection coefficients. The environmental factors inducing the bias are likely to vary between populations (e.g. it may be soil moisture in one population and nutrient availability in another), so it would be necessary to include multiple environmental variables in the analyses, which might result in selection coefficients difficult to interpret and to compare between populations. Including a measure of individual condition could be an easier approach, but it should be a trait whose variation in the population was largely determined by the environment and directly linked to fitness, which would again require an experimental (ideally breeding) design to be demonstrated. A third approach could be the use of molecular pedigrees of wild plants to estimate heritability and breeding values (van Kleunen & Ritland, 2004; Hodgins & Barrett, 2008), but the procedures available to date have limitations for their use in many instances (Reed & Frankham, 2001; Garant & Kruuk, 2005; Thomas, 2005; Frentiu et al., 2008).

Thus, although these approaches may help to understand and ameliorate the problem, their implementation for analyses of selection in a wide sample of populations is limited. Whenever multiple populations or taxa are compared, it is likely that different environmental factors bias the estimates of selection in different populations, because these may occupy localities differing in several environmental factors that vary at small scale. Therefore, it seems likely that this type of bias may affect disparately the estimates of selection coefficients on different traits and populations. Rather than inducing spurious correlations between Dz and Ei or dz, this bias may have added noise in our estimates of divergent selection, reducing our power to detect true correlations but strengthening our conclusions when a significant correlation was found.

Another problem common in most studies measuring natural selection, and ours is not an exception, is the use of a single fitness component instead of several or a measure of lifetime fitness (Kingsolver et al., 2001). This problem will be of most concern in the case of traits participating in several functions linked to total plant fitness, because selection acting through one fitness component might be modified by selection acting through other fitness components (Schluter et al., 1991; Alcántara & Rey, 2003; Gómez, 2008). The traits included in our study can affect other plant functions besides fruit production (the fitness component we measured). For example, in a study of selection in Impatiens pallida, Gross et al. (1998) found that the number of leaves was under selection through both survival and fecundity. Similarly, the number of flowers per inflorescence can be subject to selection through male and female fitness components (Conner, 1997; Benitez-Vieyra et al., 2006). Like many studies of selection in wild plant populations, we assume that selection acting through fruit production in the set of traits analysed is not significantly modified by selection acting through other fitness components or on subsequent years. In any case, it seems likely that departures from this assumption would reduce our ability to detect true correlations between divergent selection and phenotypic or environmental differences between populations, but significant correlations found in spite of possible departures from this assumption could be considered as very well supported.

Patterns of phenotypic differentiation

All the traits studied varied among populations within taxa and between species or subspecies. Similar results were obtained comparing a wider set of vegetative and floral traits between populations of A. v. vulgaris and A. p. cazorlensis (Medrano et al., 2006). This variation can be partly attributable to adaptive differentiation in response to environmental variation, to phenotypic plasticity or to genetic drift. Vegetative plant traits frequently show plasticity in response to diverse environmental factors (e.g. Valladares et al., 2007 and references therein). We have conducted a common garden experiment with the same four taxa used in this study (J. Bastida, P. J. Rey and J. M. Alcántara, unpublished) analysing the plasticity of vegetative traits (number of leaves, leaf size, inflorescence height and SLA) in response to soil depth (deep vs. shallow-rocky soil) and type (calcareous vs. siliceous). All these traits varied significantly between species and subspecies in the common garden and did not show significant plasticity in response to soil depth. These results suggest that the patterns of differentiation between taxa found in this study probably reflect more genetic differentiation than phenotypic plasticity. However, we cannot rule out the possibility that differences between populations within taxa were affected by phenotypic plasticity.

On the other hand, species of the genus Aquilegia can be prone to diverge through genetic drift, as they typically occur as small populations (rarely exceeding a few hundreds of reproductive individuals; personal observations, Mavraganis & Eckert, 2001) and lack mechanisms for long distance seed dispersal, so gene flow between populations is expected to be very scarce, as it has been shown in Aquilegia chrysantha and Aquilegia longissima in North America (Strand et al., 1996). In our study, genetic drift can be ruled out in the case of traits whose variation was related to the patterns of divergent selection, like inflorescence height, number of leaves and number of flowers per inflorescence. However, variation in leaf length, SLA and the density of nonglandular pubescence was not related to the current patterns of divergent selection. Moreover, we detected very few significant selection differentials on nonglandular pubescence and SLA, so we cannot reject the possibility of differentiation through genetic drift in these traits. Alternatively, it is also possible that these last traits might be subject to selection through a different fitness component, or that they diversified under selection pressures not present nowadays, or that populations are already at equilibrium with the current patterns of selection.

Our results strongly suggest that differentiation in inflorescence height, number of leaves and the number of flowers per inflorescence among the studied taxa is the result of divergent selection promoting adaptation to habitats differing in soil rockiness and elevation. Soil rockiness and altitudinal gradients are commonly related to variation in plant traits and the abundance of endemic plants (Baskin & Baskin, 1988; Matthews et al., 1993; McDonald & Cowling, 1995; Kessler, 2002; Lavergne et al., 2004), what suggests that adaptation to environmental factors associated to these gradients may frequently contribute to the processes of ecotypic differentiation and taxonomic diversification in plants.

Lavergne et al. (2004) found that endemic plants from the Western Mediterranean occupy habitats with steeper slopes and higher rock cover, and that insect-pollinated endemics have lower stature and fewer flowers than their widespread congeners. Our results indicate a significant decrease in inflorescence height (and a nonsignificant decrease in the number of flowers per inflorescence) with increasing soil rockiness across populations. This relationship holds between subspecies of the same species and is particularly marked at the species level because the more narrowly distributed A. pyrenaica grows on shallow-rocky soils and has smaller inflorescences with fewer flowers than the more widely distributed A. vulgaris which grows on deeper soils. We found that divergent selection on inflorescence height and the number of flowers per inflorescence was stronger between populations occurring on soils of contrasting rockiness: selection differentials on both traits tended to be smaller in populations growing on rocky soils than in those growing on deeper soils. Our results support the hypothesis that divergent selection stemming from differences in soil depth promoted phenotypic differentiation between species in inflorescence height and the number of flowers per inflorescence. Moreover, the pattern of divergent selection on inflorescence height agrees with the pattern of taxonomic proximity between the studied populations, which clearly points to divergent selection between habitats differing in soil rockiness as an important driver in the process of taxonomic diversification of the studied columbines. Several reasons may explain variation in the strength of selection on inflorescence height and the number of flowers per inflorescence along a soil rockiness gradient. (1) The activity of vertebrate herbivores may severely reduce fruit production in perennial herbs, thus reducing the opportunity for selection through this fitness component (Herrera, 2000; Herrera et al., 2002; Gómez, 2003). For example, Gómez (2003) found that selection on inflorescence height in Erysimum mediohispanicum was positive in the absence of ungulates but much lower and not significant in their presence. This effect might explain our findings if vertebrate herbivores were more active in rocky habitats. (2) A second possibility is that taller inflorescences with more flowers attracted more pollinators in deep soils with a dense community of herbs, whereas this advantage would not exist in rocky soils where the density of herbs is much lower and even small inflorescences could be easily located by pollinators. A similar argument has been used to explain the maintenance of scape length dimorphism in Primula farinosa (Toräng et al., 2006). (3) Another possibility is the existence of higher physiological costs of maintaining a tall inflorescence with many flowers in shallow-rocky soils, because these soils are more prone to drought. Such costs might counteract the possible benefits of a tall inflorescence, lowering the strength of selection on inflorescence height and flower number.

Besides the effect of soil rockiness, the abundance of endemic taxa in many floras also increases with elevation, at least within the range of elevation occupied by columbines (Kessler, 2002; Zhang et al., 2009), and there seems to be a common trend for populations at higher elevation to be smaller in most vegetative traits (Clausen et al., 1948; Neuffer & Bartelheim, 1989, Oyama, 1994; Willis & Hulme, 2002; Kofidis et al., 2007; Alexander et al., 2009). However, our results suggest the opposite pattern of phenotypic differentiation with elevation. In our study area, populations of the narrow endemic A. v. nevadensis occurring at the highest elevations have more leaves than the populations of the widely distributed A. v. vulgaris which occur at the lowest elevation. The decrease in general plant size with elevation found by most studies has been interpreted as an adaptation to cope with lower temperature. It is possible that the major limitation for Mediterranean columbines occurs at lower elevation, where summer drought and higher temperatures would favour plants with fewer leaves to reduce water loss. This stress is less severe at higher elevation in Mediterranean mountains of southern Spain (lower maximum and minimum temperatures, and shorter periods with low relative humidity; Ramírez et al., 2006), what would allow the production of more leaves. In fact, this pattern of differentiation has been found in comparisons between plants in Mediterranean habitats of contrasting water availability (Aronson et al., 1990; Arafeh et al., 2002; Petru et al., 2006). In agreement with this pattern of phenotypic differentiation, we found that the magnitude of selection differentials on the number of leaves increased with elevation. Accordingly, divergent selection on this trait was stronger between populations occurring in contrasting elevation and was positively related to the pattern of phenotypic differentiation between populations. Thus, our results strongly suggest that divergent selection related to differences in elevation is responsible for the differentiation in the number of leaves per plant in the studied columbines. However, contrarily to what we found for inflorescence height, the pattern of divergent selection on the number of leaves does not match the pattern of taxonomic proximity. If our study populations were free to respond to selection on this trait, the pattern of divergent selection would favour the evolution of similar phenotypes in populations of different taxa (i.e. ecotypic differentiation). For example, Fig. 4 shows that some populations of A. v. nevadensis would become more similar to populations of A. p. pyrenaica or A. p. cazorlensis than to other populations of its subspecies. This suggests that this process of phenotypic differentiation for leaf number in response to divergent selection between contrasting elevations may be responsible for ecotypic differentiation within taxa, but it was not related to the process of taxonomic diversification of the studied columbines.

Concluding remarks

This study offers a new methodological approach to assess the links between environmental variation, divergent selection and phenotypic and taxonomic differentiation in the context of multiple taxa and populations. This makes it is particularly useful for the study of recent adaptive radiations, because the forces promoting divergent selection are likely to be operating nowadays (Schluter, 2000). More specifically, as our results show, by simultaneously considering populations and taxa this approach allows discerning which traits and environmental factors are involved in processes of ecotypic differentiation (i.e. among populations within taxa) and which ones are related to the process of taxonomic diversification. Future improvements of this approach would benefit from incorporating the phylogenetic relationships or genetic distances among the studied populations and taxa.

Acknowledgments

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

We thank Rafael Jaime Bueno for his help with field and laboratory work, and José L. Garrido for sharing with us his unpublished results on the AFLP phylogeny of Iberian columbines. We also thank the Consejería de Medio Ambiente (Junta de Andalucía), for permission to work in the protected areas of Sierras de Cazorla-Segura-Las Villas and Sierra Nevada. This work was co-financed by the Spanish Ministerio de Educación y Ciencia (projects BOS2003-03979-C02/01-02 and CGL2006-02848) and FEDER Funds from the EU. During this work, JMB was granted with a BES-2004-3387 of Spanish MEC.

References

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

Supporting Information

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

Appendix S1 Summary of environmental parameters and plant trait means (± SD) for the study populations.

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer-reviewed and may be re-organized for online delivery, but are not copy-edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

FilenameFormatSizeDescription
JEB_1981_sm_Appendix-S1.doc75KSupporting info item

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.