Phenotypic consequences of polyploidy and genome size at the microevolutionary scale: a multivariate morphological approach


Author for correspondence:
Francisco Balao
Tel: +34954559887


  • Chromosomal duplications and increases in DNA amount have the potential to alter quantitative plant traits like flower number, plant stature or stomata size. This has been documented often across species, but information on whether such effects also occur within species (i.e. at the microevolutionary or population scale) is scarce.
  • We studied trait covariation associated with polyploidy and genome size (both monoploid and total) in 22 populations of Dianthus broteri s.l., a perennial herb with several cytotypes (2x, 4x, 6x and 12x) that do not coexist spatially. Principal component scores of organ size/number variations were assessed as correlates of polyploidy, and phylogenetic relatedness among populations was controlled using phylogenetic generalized least squares.
  • Polyploidy covaried with organ dimensions, causing multivariate characters to increase, remain unchanged, or decrease with DNA amount. Variations in monoploid DNA amount had detectable consequences on some phenotypic traits. According to the analyses, some traits would experience phenotypic selection, while others would not.
  • We show that polyploidy contributes to decouple variation among traits in D. broteri, and hypothesize that polyploids may experience an evolutionary advantage in this plant lineage, for example, if it helps to overcome the constraints imposed by trait integration.


Phenotypic shifts induced by chromosome duplications have been reported since the early 20th century (see Gates, 1909; Stebbins, 1947). A well-known (although not general) effect of polyploidy in plants and animals is cell enlargement (the ‘gigas’ effect; Stebbins, 1971; Levin, 2002; Knight & Beaulieu, 2008), but less evident effects can also occur. In plants, for example, polyploidy often modifies physiological traits like transpiration, and photosynthetic or growth rates (Otto & Whitton, 2000; Levin, 2002; Maherali et al., 2009). Following such changes in physiology, shifts in ecological tolerance have been demonstrated for some taxa (Levin, 2002). Polyploidy can also induce phenotypic modifications in reproductive traits but, surprisingly, these effects have received less attention. Sometimes polyploids have reproductive organs that are larger than their diploid counterparts, more flowers per inflorescence, and (as incompatibility may be lost) increased selfing (Robertson et al., 2010), but data on the reproductive consequences of polyploidy are relatively scarce (see Vamosi et al., 2007 for a review). After more than a decade of molecular studies, theory predicts ‘evolutionary advantages’ of polyploids over their diploid ancestors (Otto & Whitton, 2000; Comai, 2005; Parisod et al., 2010). These advantages may issue from increased heterozygosity, gene redundancy, genomic rearrangements, epigenetic reprogramming, or variations in gene expression. According to a recent study by Wood et al. (2009), however, the rate at which a plant lineage diversifies does not necessarily increase as a result of polyploidy.

Following instantaneous multiplication in DNA amount, polyploids can experience processes that either expand (e.g. retroelement insertion) or shrink their genomes (e.g. deletions; Leitch & Leitch, 2008). This has a great potential to induce phenotypic variation (Bennett & Leitch, 2005; Chen, 2007). The relationships between genome size and phenotypic traits have been discussed in comparative studies at broad phylogenetic levels (Beaulieu et al., 2007, 2008; Knight & Beaulieu, 2008; Münzbergová, 2009), but few studies have analysed how and whether genome size or polyploidy can modify phenotypic traits at the microevolutionary (i.e. population) scale (but see Sugiyama et al., 2002; Lavergne et al., 2010).

Understanding how polyploidy modifies phenotypic traits is a major goal in evolutionary biology, and although the genetic and genomic attributes of polyploids have received much attention, other crucial aspects of polyploid evolution are relatively less well known (e.g. ecology, pollination biology, and physiology; Soltis et al., 2010). This type of study has a number of pitfalls. For example, the covariation among traits need to be determined in order to get an insight into the magnitude of genetic constraints (Cheverud, 1984; Arnold, 1992). In addition, ‘phylogenetic constraints’ (i.e. a tendency of closely related organisms to be similar because of their shared history) can be operating on the traits (McKitrick, 1993; Miles & Dunham, 1993). Lastly, phenotypic characters may experience environmentally induced responses (i.e. phenotypic plasticity; Bradshaw, 1965; Pigliucci, 2001) to an unknown extent. Our aim here is to take a comparative approach to polyploidy in relation to phenotypic trait variation at the population level.

The genus Dianthus has c. 200 Eurasian species which evolved during the last 0.9–2.1 million yr (Valente et al., 2010) and shows great morphological diversity. Pollinator shifts and polyploidy are considered two major contributors to the striking radiation of this group (Carolin, 1957; Weiss et al., 2002; Balao et al., 2010). Dianthus broteri s.l. (Bernal et al., 1990) is an Iberian endemic that presents the most extensive polyploid series for the genus, with 2x, 3x, 4x, 6x and 12x cytotypes that rarely, if ever, coexist on the same population (Balao et al., 2009). Furthermore, significant variations in monoploid genome size (1Cx DNA) have been documented, both within and among populations/cytotypes (Balao et al., 2009). Lastly, some D. broteri cytotypes show distinct morphological traits, and have been considered separate species (Willkomm, 1859; Gallego, 1986). Therefore, D. broteri s.l. is a good system to study (over a microevolutionary scale) phenotypic evolution associated with polyploidy and genome size dynamics.

To date, no comparative study has analysed the connections of polyploidization and genome size dynamics with phenotypic evolution at a microevolutionary scale. In the present study we use phylogenetic comparative methods to test the prediction that polyploidy drives the evolution of phenotypic traits (both reproductive and vegetative) across 22 populations of D. broteri. We address the following specific questions: do the dimensions of the organs covary with chromosome number or DNA amount; is the effect of polyploidy consistent across vegetative and reproductive attributes; and does genetic relatedness (among populations) limit phenotypic evolution – or, in other words, is there significant phylogenetic constraint?

Materials and Methods

Organ measurements

During the summers of 2006 and 2007, vouchers (four to 11 per population) were collected in the field (see Supporting Information, Table S1 for details) to obtain linear measurements of flower parts, bracts and stomata. Studied characters were selected either because they were known correlates of ploidy in other taxa (e.g. stomata, see Levin, 2002) or because they have a diagnostic value for the identification of Dianthus taxa (Williams, 1893; Bernal et al., 1990), which in all likelihood would indicate these traits have little phenotypic plasticity and are genetically based. The organs and quantitative variables studied are depicted in Fig. 1. Floral parts provided 11 quantitative variables: petal length (PETL) and width (PETW); length of the longest petal lacinia (LACL); anther length (ANTL) and width (ANTW); calyx length (CALL) and diameter (maximum, MAXCAD; minimum, MINCAD); length of calyx teeth (CALTL) and width (CALTW); and length of the anthophore (i.e. the elongated stalk between the sepals and the petals, ANTPHL). In addition, seven nonfloral traits were studied, including the number of bracts subtending the flower (BRAN); the length (BRAL) and width (BRAW) of the longest bract; the length (LEAL) and width (LEAW) of the longest leaf on the first node below the flower; stoma size, estimated as the longest dimension of the stomatal guard cells (STOD); and mean stomatal density (STODEN). In all, we studied 130 individuals from 22 populations encompassing the whole range of the species (for detailed information on its distribution, see Balao et al., 2009) and representing all major cytotypes (i.e. 2x, 4x, 6x and 12x).

Figure 1.

A summary of the quantitative traits analysed. ANTL, anther length; ANTPHL, length of the anthophore; ANTW, anther width; BRAL, length of the longest bract; BRAW, width of the longest bract; CALL, calyx length; CALTL, length of calyx teeth; CALTW, width of calyx teeth; LACL, length of the longest petal lacinia; LEAL, length of the longest leaf; LEAW, width of the longest leaf on the first node below the flower; MAXCAD, maximum calyx diameter; MINCAD, minimum calyx diameter; PETL, petal length; PETW, petal width; STOD, longest dimension of the stomatal guard cells.

Most organs were measured with digital calipers with 0.01 mm precision, while traits related to stomata were quantified with a microscope. Stoma size (in μm) and density (stomata mm−2) were estimated on one randomly chosen basal leaf per plant which was boiled and cleared in a 30% NaClO solution for 10 min. Strips of the epidermis were peeled from the adaxial surface, mounted in glycerol, and photographed with a Carl-Zeiss Axiophot photomicroscope (Carl-Zeiss, Inc.) equipped with a Leica DFC290 (Leica Microsystems GmbH, 35578 Wetzlar, Germany). Photographs of the preparations were then imported into image analysing software (UTHSCSA Image Tool 3.0, University of Texas Health Sciences Center, San Antonio, TX, USA), the number of stomata in five randomly chosen fields counted at 10× magnification, and the length of one (randomly chosen) guard cell in the stoma closest to the centre of the preparation measured at 40×. The five measurements/counts performed on each leaf were averaged and used as plant estimates for, respectively, stoma size and density. Lastly, the mean values of the traits were calculated on a population basis and analysed.

Identifying patterns of multivariate phenotypic evolution

We performed a principal component (PC) analysis of the (log-transformed) morphological variables aimed at describing the pattern of covariation among floral and nonfloral quantitative traits. We used varimax rotation to facilitate the interpretation of the multivariate pattern described by the PC analysis, which in addition helped to maintain orthogonality in the data set (Rencher, 2002). The five morphological PC components (i.e. combinations of morphological variables) which had eigenvalue variances greater than one were identified and investigated for covariation with ploidy levels and amounts of cell DNA (discussed later). Computations were performed with R statistical software version 2.10.1 (R Development Core Team, 2009).

Determining phylogenetic relatedness among populations

The D. broteri complex has, in all likelihood, a very recent origin (Valente et al., 2010), and therefore the phylogenetic relatedness among populations cannot be determined from the usual sequence markers that are often employed in phylogenetic studies (e.g. ITS, trnK-matK, psbA-trnK, trnH-psbA). In the present study, relatedness was estimated from a bootstraped consensus phylogram (Fig. 2) based on amplified fragment length polymorphism (AFLP) data (Balao et al., 2010). The AFLP tree had a number of polytomies attributable to rapid divergence, so we collapsed branches with bootstrap values < 50, and the six distinct, independent clades that resulted were investigated for associations between morphological PC components and chromosome numbers/DNA amount.

Figure 2.

Phylogenetic relationships among the 22 Dianthus broteri populations of the study. Polyploidization events: 4x, white circles; 6x, grey circles; 12x, black circles.

Polyploidy vs morphology

To rule out the possibility that any phenotype–polyploidy correlations, if detected, were the result of underlying spatial/ecological similarity among the populations (rather than being genetically based) we performed a Mantel’s test between morphological (Euclidean) and spatial (from geographical coordinates) distances. The test also incorporated the phylogenetic distance among populations (package vegan in R software; Oksanen et al., 2010) so that it should be best described as a partial Mantel’s test (Legendre & Legendre, 1998).

To investigate the relationship between morphology and ploidy level/genome size we used data from a previous study by Balao et al. (2009) in which genome size and ploidy level had been determined from flow cytometry. Our approach was a multivariate analysis of variance (MANOVA) which had the five main morphological PC scores as response variables, and ploidy level as the main factor. The association between morphology and degree of ploidy could, in part, be the result of phylogenetic constraint, so the ‘phylogenetic signals’ that affected each morphological PC score were assessed with independent Abouheif’s tests (adephylo package in R software; Abouheif, 1999; Jombart et al., 2010). The Abouheif’s C statistic tested the null hypothesis that traits did not experience phylogenetic autocorrelation (based on the topology of the tree) and phylogenetic relatedness among population was therefore uncorrelated with morphology.

To evaluate more exactly the joint effect of ploidy level and phylogenetic relatedness on trait variation, we used two different approaches. The first was an ANOVA of morphological PC scores that had ploidy level (nested under clade) as the main, fixed factor. This method only takes into account the main phylogenetic clades (i.e. it does not consider the actual relatedness of populations within clades), so this approach can be considered rough and conservative. The second, more exact approach was a phylogenetic generalized least squares ANOVA (PGLS, Martins & Hansen, 1997; Garland & Ives, 2000; Rohlf, 2001). In this analysis, the matrix of phylogenetic variance-covariance (i.e. a summary of tree topology and branch lengths) is used to rectify the correlation observed among morphological variates, so that any phylogenetic effects are accounted for. The evolutionary assumptions under which phylogenetic effects were examined were as follows: an Ornstein–Uhlenbeck model (OU), in which the evolutionary constraint of the trait under stabilizing selection depends on α, the rate of selection (in our analyses this was either α = 1 or α = 10); a Brownian model (BM), in which α = 0 (i.e. changes in morphology are not constrained by selection, and are therefore assumed to occur stochastically); and a nonphylogenetic model (TIPS), in which α approaches infinity (i.e. morphological variations occur without any phylogenetic constraint). As advised by Posada & Buckley (2004), whenever two models had identical numbers of degrees of freedom (e.g. the OU model with α = 1 vs α = 10) we used the Akaike information criterion (AIC) to decide which one was more appropriate. If the numbers of degrees of freedom were different (e.g. OU vs BM), a likelihood-ratio test (assuming a χ2 distribution) was employed instead. All analyses were performed with R software using ape (Paradis et al., 2004), geiger (Harmon et al., 2008) and nlme packages (Pinheiro et al., 2010).

The OU model assumed that there existed only one morphological optimum, which might not be the case. We also analysed two complex scenarios in which morphological evolution presented several optima: one had one optimum per-ploidy level (i.e. four optima in all; OU4), and another with two optima for the tetraploids and one for the remaining levels of ploidy (i.e. five optima in all; OU5, see Fig. 2). Computations were performed with the ouch package in R (Butler & King, 2004). Models were compared by their AICs, and using a likelihood-ratio test against the BM model.

Additional PGLS analyses were run on morphological PC scores as predicted by both overall (2C) and monoploid (1Cx) DNA amounts (Balao et al., 2009). These analyses were based upon the same evolutionary models described previously, but since confidence intervals for intercepts often included zero, we had to constrain the models accordingly.


Morphological covariation of vegetative and reproductive traits

The PC analysis identified five axes with eigenvalues > 1 (Table 1), which altogether accounted for 65% of the observed morphological variation (for population-specific means and standard errors, see Tables S2, S3). Factor loadings showed a pattern of covariation with floral and nonfloral traits intermingled. For example, the main component (PC 1, accounting for 27% of variance) basically combined the dimensions of nonfloral organs (bract and leaf), but also the length of the calyx. Component PC 2 was again a compound of floral (petals) and nonfloral characteristics (stoma size and density), and the same applied to PC 4 (petal, lacinia, anthophore, number of bracts). Only components PC 3 and PC 5 (which summarized calyx and anther attributes, respectively) could be considered purely reproductive.

Table 1.   Principal components (PCs, eigenvalues > 1) of morphological attributes in Dianthus broteri
 PC 1PC 2PC 3PC 4PC 5
  1. *Eigen scores > 50. Factor loadings for floral and nonfloral traits are reported separately.

  2. ANTL, anther length; ANTPHL, length of the anthophore; ANTW, anther width; BRAL, length of the longest bract; BRAN, number of bracts subtending the flower; BRAW, width of the longest bract; CALL, calyx length; CALTL, length of calyx teeth; CALTW, width of calyx teeth; LACL, length of the longest petal lacinia; LEAL, length of the longest leaf; LEAW, width of the longest leaf on the first node below the flower; MAXCAD, maximum; MINCAD, minimum calyx diameter; PETL, petal length; PETW, petal width; STOD, longest dimension of the stomatal guard cells; STODEN, mean stomatal density.

% variance explained26.8711.509.759.016.66
Floral traits
Nonfloral traits

The net effect of polyploidy on organ sizes

A partial Mantel’s test between morphological and spatial distances failed to detect any significant autocorrelations in our data set (= 0.02, = 0.38). Therefore, phenotypic correlates of polyploidy (if any) could not be attributed to spatial/ecological similarity between the studied populations.

Multivariate scores of organ linear dimensions were significantly dependent of ploidy level (MANOVA analysis: Wilks’s λ = 0.067 27, F4, 18 = 4.3171, < 0.001). However, and as can be seen in Table 2, three of the morphological PC scores (PC 1, PC 2 and PC 4) also had significant phylogenetic signals, so we could not rule out the possibility that nonindependence was actually the result of the underlying phylogenetic structure of the studied populations (i.e. phylogenetic constraint), rather than of ploidy level by itself. To get an insight into the importance of phylogeny, we incorporated clade membership (nested under ploidy) into more complex ANOVA models also reported in Table 2. Clade ownership was a constraining condition only in the case of PC 1 (for which it explained 38% of morphological variance; Fig. 3), whereas it did not affect morphological scores PC 2, PC 3, PC 4 and PC5 in any significant ways. By contrast, ploidy level had significant effects on PC 1, PC 2 and PC 4 (Table 2), for which it explained 26, 65 and 53% of variance, respectively (Fig. 3). The amount of variance that remained unexplained (i.e. among-population variation) could be considerable on occasions (e.g. up to 80% for PC3; Fig. 3).

Table 2.   Quantitative trait variation (principal component (PC) scores) in Dianthus broteri as predicted by ploidy level and phylogeny
 Phylogenetic signalANOVAPGLS
Ploidy (F3,16)Clade (F2,16)F3,18ModelaAICblog-Lc
  1. The ANOVA section reports the effects of both ploidy level (2x, 4x, 6x and 12x; see Fig. 2) and phylogeny on morphological scores, with clade ownership nested under ploidy level. The PGLS (phylogenetic generalized least squares) section includes results of regressions on ploidy level under best-fitting evolutionary models. The phylogenetic signal for each PC score is represented by their C-statistic values in Abouheif’s tests.

  2. aBM, Brownian; TIP, nonphylogenetic model.

  3. bAkaike information criterion.

  4. cLog-likelihood value; ***, P < 0.001; **, P < 0.01; *, P < 0.05.

PC 10.34**3.948*8.487**0.64BM61.62−25.81
PC 20.49***13.139***2.5042.278*BM39.30−14.65
PC 3−0.050.8741.0950.707TIP48.71−19.36
PC 40.37**7.032**1.3425.542**TIP45.31−17.65
PC 50.162.5322.1151.84TIP46.32−10.16
Figure 3.

Variance components for each of the ANOVA analyses that used the five morphological principal component scores as response variables (Table 2). For each score, the relative contribution of ploidy level is represented by the black bars, and that of clade membership by grey bars. White bars depict among Dianthus broteri populations variances (calculated from the error term).

Phylogenetic generalized least squares analyses were for the most part consistent with ANOVAs (Table 2). Once phylogenetic relatedness among populations was taken into account, PC 1 was no longer sensitive to variations in chromosome numbers, whereas PC 2 and PC 4 varied significantly with ploidy level. Furthermore, variations in morphological score PC2 fitted the OU5 model (i.e. the one considering a scenario with five morphological optima; AIC = 38.7) rather than the BM (AIC = 45.8), the single-optimum OU model (AIC = 51.9), or the four-optima OU4 model (AIC = 42.31). In turn, morphological score PC 4 matched the OU4 model better (AIC = 47.3) than either OU5 (AIC = 48.8), single-optimum OU (AIC = 54.97), or BM (AIC = 50.2; for more details see Tables S4, S5).

In summary, the relative importance of polyploidization and phylogenetic relatedness varied among traits, and the ways in which ploidy levels modified organ size (when and if they did) were far from straightforward. As exemplified in Fig. 4, some morphological PC scores increased with increasing chromosome numbers (e.g. PC 2), while others (e.g. PC 4) decreased. In addition, our analyses supported the hypothesis that some combinations of traits (PC 2 and PC 4) had experienced a natural selection regime which had several optima.

Figure 4.

Box-plot representation of morphological variations with ploidy levels in Dianthus broteri. Principal component (PC) 2 (upper panel) summarizes information on petal and stoma dimensions, and PC 4 (lower panel) on petal, lacinia and anthophore size, as well as the number of bracts. Horizontal lines represent the median, and boxes and whiskers represent the interquartile range and the nonoutlier ranges, respectively.

Genome size vs morphology

Models for the evolution of the studied traits as explained by cell DNA content are reported in Table 3. Absolute DNA amount (2C) successfully predicted PC 2 (r2 = 0.27) and PC 4 (r2 = 0.17). The evolution of these PC scores was best described by an OU model, meaning that the involved organs (i.e. stoma size and density, petal dimensions, anthophore length and number of bracts) would have experienced phylogenetic and selective constraints. As for monoploid DNA content, the OU model predicted adequately only the morphological scores on PC 3 (r2 = 0.28), and did so under a TIP model (i.e. calyx shape variability would have evolved without phylogenetic constraint). It should also be noted that the trend with genome size depended on the traits considered: as shown in Fig. 5, PC 2 increased with overall DNA cell content; PC 3 varied in direct proportion with monoploid genome size; and PC 4 decreased with absolute DNA amount. This, along with the fact that principal components did exhibit orthogonal patterns of variation, seem to indicate that the effect of polyploidization varied depending on the subset of traits under consideration.

Table 3.   Phylogenetic generalized least squares (PGLS) models of morphological principal component (PC) scores as predicted by overall (2C) and monoploid (1Cx) cell DNA amount
DNA contentMorphological responseInterceptSlopePModelaAICblog-Lcr2
  1. aBM, Brownian; OU1, Ornstein–Uhlenbeck model with α = 1; Ornstein–Uhlenbeck model with α = 10; TIP, nonphylogenetic model.

  2. bAkaike information criterion.

  3. cLog-likelihood value; ns, nonsignificant.

2CPC 10.03nsBM59.56−27.780.00
PC 20.810.017OU144.53−20.260.27
PC 3−0.05nsBM51.49−23.750.02
PC 4−0.660.088OU151.12−23.560.17
PC 50.10nsOU1048.38−22.190.01
1CxPC 1−0.29nsBM55.12−25.560.00
PC 21.44nsBM44.61−20.310.02
PC 30.886.920.011TIP39.27−16.640.28
PC 4−0.83nsBM49.49−22.740.06
PC 50.34nsOU1044.30−20.150.06
Figure 5.

The relationships between morphology and genome size in Dianthus broteri. Variations in principal component (PC) 2 (a) and PC 4 (b) refer to overall DNA content, and in PC 3 (c) they refer to monoploid DNA. Statistically significant regression lines (with phylogeny incorporated) are also shown.


Sources of variation in phenotypic traits

Our study shows that the evolution of some phenotypic characters in D. broteri is under strong phylogenetic constraint (Table 1), so that the phenotypes of plants of related populations were more similar than by chance (note that these relationships were necessarily genetic, as the Mantel test ruled out any spatial/ecological autocorrelations). This could be attributed to founder effects, i.e. nearby, related populations retained significant imprints from their source. Our data also lead to the conclusion that, for any relationship between polyploidy/DNA content and phenotype to be properly ascertained, population relatedness must be controlled. Once population relatedness was taken into account, our PGLS analyses indicated that increasing polyploidy levels and genome sizes had a significant net effect on the evolution of D. broteri phenotype.

Our data set showed a great amount of unexplained phenotypic variation in some traits (Fig. 3), which could be reasonably attributed to phenotypic plasticity. Trait plasticity has not been specifically addressed in the present study, but since we used population averages, any environmentally induced variation would increase the error term in the analyses, thus making it harder to detect statistically significant effects. Therefore, our conclusions on the effect of ploidy level/DNA amount on phenotypic variation will likely fall on the conservative side.

Phenotypic consequences of polyploidy and genomic dynamics

As in many other plant taxa (Otto & Whitton, 2000), the sizes of organs in general increased with both chromosome number and absolute cell DNA amount in D. broteri. Bennett (1971) proposed that allometric increases of organ size (e.g. stoma diameter, petal length) following genome duplications can be expected as a result of ‘nucleotype’ effects (i.e. the organs would be larger simply because the cells forming the organ had become larger). It must be pointed out, however, that some attributes actually decreased with increasing chromosome numbers/DNA amount (e.g. PC 4, relating to anthophore and laciniae length; Fig. 5b), which indicates that the relationship between ploidy level and organ size need not be always direct and allometric (Otto & Whitton, 2000). Unexpected phenotypic responses can arise from developmental tradeoffs (Maynard Smith et al., 1985; Prusinkiewicz et al., 2007), changes in the pattern of gene expression (Broz et al., 2009), or even epigenetic effects (Madlung et al., 2002; Wang et al., 2004).

The amount of monoploid DNA in D. broteri does vary among and within cytotypes (Balao et al., 2009). A reason often invoked to explain this sort of variation is that the polyploid genome is actually dynamic and undergoes rapid structural alterations following chromosome duplication (Chen, 2007; Leitch et al., 2008; Parisod et al., 2010). We hypothesize that part of the phenotypic variation observed in the present study (e.g. calyx shape, Fig. 5c) could be the result, at least in part, of within-cytotype genomic readjustments. Rapid alterations in monoploid DNA amount have been shown to trigger morphological phenotypic changes in other plant species (Meagher & Costich, 1996; Lavergne et al., 2010), but our study demonstrates that the sign of covariation with ploidy level can change from one plant organ to another.

Trait uncoupling and polyploidization

Berg’s (1960) idea that functionally related traits must show correlated phenotypic variation opened the way to studies of ‘phenotypic integration’ in plants. Vegetative and reproductive attributes represent examples of sets of largely covarying traits that share a common function (‘correlation pleiades’; Conner & Via, 1993). In D. broteri, covariation among traits often occurred across floral and vegetative sets (e.g. PC 1, PC 2, PC 4), which seems to indicate that the phenotype is to some extent ‘disintegrated’. Previous studies have atributed this phenomenon to either low selective pressures from pollinators (Armbruster et al., 1999; Herrera, 2002), pleiotropy, genetic linkage, and/or developmental effects (Murren, 2002). In D. broteri, any of these factors can be invoked to explain the observed decoupling of continuous trait variation, plus the additional factor of polyploidy. Oswald & Nuismer (2011) have shown in Heuchera grossularifolia that experimentally induced autopolyploidy stimulates trait variation and also causes shifts in the structure of the phenotypic covariance matrix (i.e. trait decoupling).

Dianthus broteri plants can be either 2x, 4x, 6x or 12x, and since chromosome number variation can itself be a source of phenotypic instability (Levin, 1983; Osborn et al., 2003; Chen, 2007), a unified floral phenotype would seem difficult to attain. Two nonmutually exclusive reasons have been proposed to explain trait decoupling in polyploids. One is the weakening, or ‘dilution’, of the phenotypic effects of polyploidy, which can be relatively straightforward (genetic) in an individual cell, but relatively more complex (developmental) in many-celled plant parts (Knight & Beaulieu, 2008). Another way in which traits can become uncoupled is through nonrandom genetic readjustments (such as fragment loss, methylation or duplications; Adams et al., 2003; Adams & Wendel, 2005; Gaeta et al., 2007), so that, following polyploidy, some traits become more prone to vary than others. Lastly, a third factor that may have contributed to reinforce trait uncoupling is divergent selection. Phenotypic divergence among cytotypes could be driven by natural selection acting on specific characters (Nuismer & Cunningham, 2005). This is supported in our study species, since the model that relied on selective assumptions (i.e. the OU model) was only significant for PC 2 and PC 4 but not for other combinations of traits (with the corollary that some characteristics would experience phenotypic selection while others would not). We further hypothesize that in D. inoxianus there has been divergent selection on some traits following polyploidization, which ultimately may have led to decreased pheno-typic integration.

Recently formed polyploids are assumed to experience decreased fitness relative to their parentals, mostly because they appear in low frequencies and tend to mate with the latter (the ‘minority cytotype disadvantage’ of Levin, 1975). On the other hand, neopolyploids can also compensate for their initial unfavourable status via improved ecological tolerance (Levin, 2002; Sobel et al., 2010) or pollinator shifts (Segraves & Thompson, 1999; Husband, 2000). Whatever its causes, trait uncoupling brought about by polyploidy may allow natural selection to operate differentially on attributes otherwise constrained in the phenotype, which, on occasion, may help polyploid plants adapt to new suitable ecological niches (Otto & Whitton, 2000). Pollination-wise, D. broteri polyploids seem to benefit from larger corollas and a longer distance from the anthophore to the tip of calyx (i.e. the effective length of the flower tube), two attributes which may contribute to modify their fitness. On the other hand, they present fewer calyx bracts, which may represent a disadvantage when dealing with florivores. Stomatal size and density are other examples of traits that can contribute (through improved ecological tolerance) to the establishment and/or divergence of polyploids from their progenitors. Interestingly, all these effects are evidenced at the local, population level, suggesting that polyploidy itself can facilitate population differentiation.

Highest-order neopolyploids (12x, 6x) of D. broteri have larger flowers and stomata, occupy very specific habitats (e.g. sand palaeodunes, gypsum-rich soils; F. Balao, pers. obs.), and are serviced by an extremely narrow pollinator fauna (in the case of the dodecaploid cytotype, a single migrant hawkmoth species; Balao et al., 2011). Further-more, populations are almost invariably composed of a single cytotype (Balao et al., 2009, 2010). We hypothesize that uncoupling of morphofunctional traits brought about by polyploidy may represent, in itself, an evolutionary advantage. Dianthus is a large genus with over 200 species, a high incidence of polyploidy (Carolin, 1957; Weiss et al., 2002), and the fastest speciation rate in the flowering plants (Valente et al., 2010). Results from the present study suggest that phenotypic evolution through polyploidy may have promoted speciation in this genus.


We thank Dr M. Escudero, Prof. P. E. Gibbs and Prof. P. Jordano for helpful comments on a previous version of the manuscript; M. Lorenzo for assistance with organ measurements; and Dr Gonzalez-Voyer and Dr E. Rezende for input and discussions about the comparative methods. This study was supported by a predoctoral grant to F.B. from the Spanish Ministerio de Educación y Ciencia (AP2005-4314); Junta de Andalucía, Proyecto de Excelencia (2005/RNM848) and Ministerio de Educación y Ciencia, Flora Iberica 8 (CGL2009-08178). The authorities of Doñana National Park (project 34/2004) granted permission to access several sampling sites.