Introduction

The importance of hybridization as a trigger for evolutionary change and speciation has long been appreciated in plants (Stebbins 1959; Grant, 1981; Rieseberg, 1997; Rieseberg et al., 2003; Arnold, 2004) and also more recently in animals (Nolte et al., 2005; Gompert et al., 2006; Mavárez et al., 2006; Mallet, 2007; Jiggins et al., 2008). Traditionally, hybridization associated with a multiplication of chromosome number, that is, allopolyploid speciation, has been considered to be the major mode of speciation resulting from hybridization for two reasons. First, it is a conceptually simple means of conferring balanced chromosome numbers that restore meiosis and fertility in new hybrid taxa while creating a powerful reproductive barrier between hybrids and their progenitors. Second, allopolyploid species can be recognized relatively easily on the basis of altered chromosome numbers and associated hybrid phenotypes (Grant, 1981; Rieseberg, 1997). However, there is growing recognition that hybridization is also frequently associated with speciation without changes in chromosome number, known as homoploid hybrid speciation. This recognition has been facilitated by the increasing applicability of molecular genetic techniques to non-model organisms (Rieseberg, 1997; Gross and Rieseberg, 2005; Mallet, 2007; Abbott et al., 2010; Nolte and Tautz, 2010).

There are additional challenges associated with homoploid hybrid speciation compared to allopolyploid speciation. Incompatibilities between parental genomes after hybridization leading to hybrid unfitness due to increased mortality and/or reduced fertility might occur in the absence of polyploidy, thus tending to restrict successful homoploid hybrid speciation events to more closely related parents compared to allopolyploid speciation (Chapman and Burke, 2007; Paun et al., 2009). In addition, lack of reproductive, ecological and spatial isolation between new hybrids and their progenitors frequently leads to ongoing introgression and genetic swamping by more frequent parents, preventing hybrid establishment and subsequent independent evolution (Buerkle et al., 2000; Sobel et al., 2009; Nolte and Tautz, 2010). All known examples of homoploid hybrid speciation have overcome one or, more usually, several of these barriers to speciation.

Frequently, reproductive isolation seems to have been achieved by large-scale chromosomal rearrangements between parental genomes, termed recombinational speciation, that maximize hybrid fitness and fertility but also incidentally impose reproductive isolation with progenitors (Grant, 1981; Rieseberg, 1997). Recombinational speciation has been most comprehensively demonstrated and investigated in homoploid hybrid desert and salt marsh Helianthus species (Rieseberg, 1997). However, ecological differentiation also appears to be a highly important factor contributing to successful homoploid hybrid speciation (Rieseberg, 1997; Buerkle et al., 2000; Gross and Rieseberg, 2005; Abbott et al., 2010). Hybrid superiority in a distinctive niche from either parent is a selective force for continued hybrid persistence (Buerkle et al., 2000). Furthermore, niche differentiation is often implicitly associated with decreased gene flow between hybrids and parents for various reasons such as increased spatial separation, different phenology, resource use or mate recognition cues (Filchak et al., 2000; Gross and Rieseberg, 2005; Mavárez et al., 2006; Jiggins et al., 2008; Abbott et al., 2010; Nolte and Tautz, 2010). In general, these factors are rarely absolute barriers to gene flow, but in combination they may be sufficient to tip the balance in favour of hybrid persistence leading to eventual speciation.

Divergence towards speciation is frequently aided by population isolation (allopatry and parapatry) and/or founder effects associated with colonization (Turelli et al., 2001; Templeton, 2008; Sobel et al., 2009). While it remains controversial whether founder effects and isolation ever lead to speciation without additional divergent selection, both can contribute to kick-starting the process (Slatkin, 1996; Turelli et al., 2001; Templeton, 2008). Founder events during colonization involve population bottlenecks followed by population growth that cause considerable changes in genetic diversity relative to the source population as the rate of genetic drift changes from high to low, generating novel variance on which selection can act (Slatkin, 1996; Templeton, 2008). Geographic population isolation limits or prevents gene flow and so is in itself an important prezygotic reproductive barrier (Coyne and Orr, 2004; Lowry et al., 2008; Sobel et al., 2009). In addition, divergence in isolation of quantitative traits (QTs) not directly associated with reproductive isolation, such as differences in phenology (Hall and Willis, 2006; Lowry et al., 2008), or niche partitioning (Gompert et al., 2006; Hall and Willis, 2006) frequently contribute to prezygotic mating barriers that are effective at limiting gene flow. Finally, conditions of population isolation accelerate genetic drift and divergent selection contributing to reproductive isolation due to Dobzhanksy–Muller incompatible interactions between loci with alternately fixed alleles between populations (Turelli and Orr, 2000).

The subject of this investigation, Senecio squalidus, combines elements of both hybrid and founder speciation. The recent and well-documented origin of S. squalidus makes it a useful and fascinating case study of ecogeographic homoploid hybrid speciation ‘in action’ (Abbott et al., 2010). S. squalidus is a homoploid hybrid species that recently arose following human-mediated introduction of material from a hybrid zone between S. aethnensis and S. chrysanthemifolius on Mount Etna, Sicily, to the Oxford Botanic Garden in Britain (Harris, 2002; James and Abbott, 2005). S. aethnensis grows on the higher slopes of Mount Etna, above 2000 m on recent lava flows, whereas S. chrysanthemifolius grows below 1000 m on arable agricultural and waste land. All three species are short lived, outcrossing perennial herbs with generalist hoverfly pollinators and wind-dispersed fruits. Furthermore, all species are self-incompatible but highly interfertile without disruption of self-incompatible function in hybrids (Chapman et al., 2005; AC Brennan and SJ Hiscock, unpublished). Thus, on Mount Etna, gene flow between S. aethnensis and S. chrysanthemifolius has generated a stable hybrid zone that exists at intermediate altitudes around 1100–1900 m between the two species’ ranges (James and Abbott, 2005; Brennan et al., 2009). However, strong selection against hybrids prevents long-term persistence of individual hybrid lineages and prevents independent evolution and hybrid speciation within the hybrid zone itself (Brennan et al., 2009). In contrast, since its origin, S. squalidus has become a successful invasive species throughout large parts of the British Isles occupying disturbed urban habitats distinct from its parents, such as railway lines, motorway verges and waste land (James and Abbott, 2005). Here we present a comprehensive investigation measuring and comparing molecular genetic and QT differentiation between S. squalidus, S. aethnensis and S. chrysanthemifolius, and hybrid zone individuals from Sicily. In doing so, we identify the extent and nature of the differences that make S. squalidus distinct from Sicilian Senecio as a starting point for assessment of the relative contributions of hybridization, selection and population history to these differences.

Materials and methods

Plants

Plants of S. aethnensis, S. chrysanthemifolius and their hybrids investigated in the present study were the same as those described in Brennan et al. (2009). They were derived from seeds collected from separately sampled wild plants growing on Mount Etna, Sicily, typically within a 500 m2 area per sampled population. Sample individuals were classified as S. aethnensis, S. chrysanthemifolius or their hybrids on the basis of field observations, sample location and subsequent analysis of QTs as described in Brennan et al. (2009). This resulted in a sample size of 28 S. aethnensis, 56 S. chrysanthemifolius and 101 hybrid individuals. In addition, 23 S. squalidus individuals were sampled as seed from separate maternal plants from urban sites of 500 m2 area in Edinburgh and Cardiff in the United Kingdom, and were cultivated along with the wild-sampled Senecio from Sicily in a fully randomized design in a glasshouse. Most of the individuals cultivated (208) were surveyed for both molecular genetic and QT variation as detailed below.

Molecular genetic measures

Genomic DNA was extracted from plants as described in Brennan et al. (2009). Variation at 22 molecular genetic loci was surveyed across all 208 individuals. Results for 13 of these molecular genetic loci, including 6 loci encoding allozymes (AAT, ACP, GDH, IDH, PGI and PGM) and 7 anonymous simple sequence repeat loci (SSRs: S4, S10, S15, S20, S26, V44 and V45), were previously reported in Brennan et al. (2009) for the Sicilian Senecio. New molecular markers were developed by analyzing expressed sequence tag (EST) data for S. aethnensis, S. chrysathemifolius, S. squalidus and S. vulgaris from the online Senecio database (www.seneciodb.org/database.htm). Sequence analysis involved using CAP3 to produce contigs (Huang and Madan, 1999), EMBOSS etandem (Rice et al., 1999) to detect SSRs and Perl scripts to visualize ESTs aligned against contigs and highlight polymorphisms resulting in six novel SSR markers (EC1019, ES1, ES29, ES43, ES45 and ES87; Supplementary Table S1). A further three insertion–deletion (indel) polymorphic markers were developed from sequence data for single copy genes in S. aethnensis and S. chrysanthemifolius (A26, C19 and D11, Supplementary Table S1; Chapman et al., 2007; M Chapman, unpublished). All markers were fluorescently PCR amplified and capillary sequencer genotyped as described in Brennan et al. (2009).

Molecular genetic analysis

Rare alleles that were observed in a single individual were excluded from analysis. Significant deviations from Hardy–Weinberg equilibrium and the presence of possible null alleles (specific non-detected alleles) were tested across loci and populations using GenAlex v6.1 (Peakall and Smouse, 2006) and Microchecker v2.2.0.3 (Van Oosterhout et al., 2004). To compare molecular genetic diversity between species, allelic richness (Rs) and expected heterozygosity (He) were calculated for each marker class (SSRs, EST SSRs, indels and allozymes) and corrected for sample size differences by rarefaction and data bootstrapping, respectively, for unbiased molecular diversity comparisons using Microsatellite Analyser Software (MSA v4.05; Dieringer and Schlötterer, 2003). Molecular genetic differentiation was investigated by measuring patterns of allele sharing between species using GenAlex v6.1 (Peakall and Smouse, 2006) and by measuring paired-species differentiation values (Fst and Fst’ corrected for different marker diversities) and their 95% confidence intervals following bootstrapping using MSA v4.05 (Dieringer and Schlötterer, 2003). The influence of marker type (SSRs, EST SSRs, indels or allozymes) on patterns of differentiation was tested by comparing divergence measured for each marker type separately. Genotype data were used in a principal coordinate analysis and also a discriminant coordinate analysis of an absolute genetic distance matrix to discriminate between S. aethnensis, S. chrysanthemifolius, S. squalidus and Sicilian hybrid samples using the ade4 package in R v2.9 software (Dray and Dufour 2007; R Development Core Team 2009). Percentage variance explained was calculated from eigenvalues. The effectiveness of the first and second principal coordinates and discriminant scores at explaining paired-species differences was assessed by one-way analysis of variance after Bonferroni correction for multiple tests (three species pairs multiplied by two principal coordinates) using R v2.9 (R Development Core Team 2009). In addition, the influence of high, medium and low marker diversity on discriminant coordinate analysis results was investigated by analyzing different marker types separately divided into SSRs, EST SSRs, and combined indels and allozymes, respectively. Multivariate analysis of variance of the first two discriminant scores was used to assess the effectiveness of each marker class at discriminating between S. aethnensis, S. chrysanthemifolius, S. squalidus and Sicilian hybrid samples using R v2.9 (R Development Core Team 2009).

QT measures

Twenty QTs were measured on all sample plants raised from seed in the glasshouse and included: days to germinate, days to development of first true leaf, days to flowering, height, branch number, capitulum number, pedicel length, ray display, pollen number, pollen fertility, floret number, seed length, pappus length, photosynthetic rate at 20 °C, photosynthesis at 7 °C, leaf auricle width, leaf dissection, stomata number, leaf chlorophyll and leaf anthocyanin content as described in Brennan et al. (2009). Square root or natural logarithm transformations of the QT data were used when necessary to improve homogeneity of variance as described in Brennan et al. (2009). Missing data (9.2%) were conservatively replaced with mean QT values.

QT analysis

Differences in paired-species QT means were assessed by one-way analysis of variance after Bonferroni correction for multiple tests (four species pairs multiplied by 20 QTs). Principal component analysis (PCA) and linear discriminant analysis (LDA) were performed using the ade4 package in R v2.9 software (Dray and Dufour 2007; R Development Core Team 2009). Percentage variance explained was calculated from standard deviations of eigenvalues and important traits were identified from principal component and discriminant function axis rotations. The effectiveness of the first and second QT principal components and discriminant scores at explaining paired-species differences was assessed by one-way analysis of variance after Bonferroni correction for multiple tests using R v2.9 (R Development Core Team 2009).

Results

Molecular genetic analysis

Genotypes at 22 molecular genetic marker loci were successfully recorded for 203 of the 208 sample individuals of S. aethnensis, S. chrysanthemifolius, their Sicilian hybrids, and S. squalidus (Supplementary Table S2). Overall, 26/294 (8.8%) of Hardy–Weinberg equilibrium tests indicated non-equilibrium genotype frequencies and 27/294 (9.2%) of permutation tests indicated null allele presence at a 5% significance level (Supplementary Table S3). This is a small excess over the 5% significant results expected by chance alone and indicate that sample populations are generally at or close to Hardy–Weinberg equilibrium with loci exhibiting few null alleles, satisfying the assumptions of later analyses. A comparison of different classes of markers indicated a pattern of increasing molecular diversity in terms of allelic richness and heterozygosity (Rs and He) moving from allozymes to indels to EST SSRs to anonymous SSRs (Table 1). A comparison of the different species exhibited increasing molecular genetic diversity in the order S. squalidus, S. aethnensis, S. chrysanthemifolius and Sicilian hybrid Senecio (Table 1). Analyses of molecular genetic differentiation (Fst and Fst’ paired-species estimates) indicated that S. squalidus was more highly differentiated from S. aethnensis and S. chrysanthemifolius than from their Sicilian hybrids. This pattern held across all different marker types with different levels of diversity, but was strongest for SSR Fst’ measures due to relatively fewer shared SSR alleles between S. squalidus and Sicilian Senecio than for other less diverse marker types (Supplementary Figure S1). Patterns of between-species allele sharing indicated that, while S. squalidus was not distinctive in terms of many unique alleles, many more alleles were shared between Sicilian hybrids and both parents than between S. squalidus and both parents (Figure 1). Molecular genetic differentiation in paired-species allele frequencies as measured by Fst was significantly greater than zero and of a similar magnitude for comparisons between S. aethnensis and S. chrysanthemifolius and between each parent and S. squalidus meaning that S. squalidus is as distinct from each of its parent species as they are from each other (Fst (95% CI)=0.21 (0.14–0.31), 0.23 (0.15–0.32), 0.21 (0.15–0.28); aeth. vs chrys., aeth. vs squal., chrys. vs squal., respectively; Supplementary Figure S1). There was also significant molecular genetic differentiation between S. squalidus and the full range of Sicilian hybrids (Fst [95% CI]=0.16 [0.11–0.22]; hybrid v. squal; Supplementary Figure S1) that further emphasizes the genetic distinctiveness of S. squalidus relative to its Sicilian progenitors. The discriminant coordinate analysis was highly effective at distinguishing distinct species clusters with the first and second discriminant scores (DS1 and DS2) explaining 41.3 and 33.4% of between-group variance, respectively (Figure 2). The principal coordinate analysis was also highly effective at distinguishing distinct species clusters with the first and second principal coordinates (PC1 and PC2) explaining 13.9 and 7.0% of total genetic distance variation, respectively (Supplementary Table S4 and Supplementary Figure S2). S. squalidus was highly distinct from S. aethnensis, S. chrysanthemifolius and Sicilian hybrids based on DS1 values (Figure 2, Table 2), while the parental species were distinguished by their DS2 values with Sicilian hybrid Senecio and S. squalidus having intermediate DS2 values (Figure 2, Table 2). This pattern of species differentiation was similar for subsets of high, medium and low diversity markers indicating that the molecular genetic distinctiveness of S. squalidus is not simply a feature of more highly discriminating high diversity markers (Supplementary Figure S3).

Table 1 Molecular genetic diversity indices of SSRs, indels and allozymes for 203 wild-sampled Senecio aethnensis, S. chrysanthemifolius, S. squalidus and Sicilian hybrid Senecio individuals
Full size table
Figure 1
figure 1

Patterns of allele sharing (a) between Senecio aethnensis, S. chrysanthemifolius and their hybrids from Sicily, and (b) between S. aethnensis, S. chrysanthemifolius and the hybrid species S. squalidus from the United Kingdom. Values within labelled circles and their overlaps are the percentage of alleles observed in each sharing class derived from 180 individuals and 111 alleles and from 108 individuals and 102 alleles for comparisons involving Sicilian hybrids and UK S. squalidus, respectively.

Full size image
Figure 2
figure 2

Plot of first and second discriminant coordinate scores resulting from an analysis of a between-individual genetic distance matrix for 22 molecular genetic markers surveyed across 203 wild-sampled Senecio aethnensis, S. chrysanthemifolius, S. squalidus and Sicilian hybrid individuals. Axes are labelled with the percentage variance explained by the first and second discriminant functions, respectively. The same scale has been used for both axes.

Full size image
Table 2 Comparison of quantitative traits and discriminant scores between S. aethnensis, S. chrysanthemifolius, their Sicilian hybrids and S. squalidus
Full size table

QT analysis

Phenotype data for 20 QTs for 208 individuals of S. aethnensis, S. chrysanthemifolius, their Sicilian hybrids, and S. squalidus are presented in Supplementary Table S5. The LDA of QT variation distinguished species based on the first and second discriminant scores that explained 55.4 and 39.4% of the between-group variation, respectively (Figure 3). The corresponding PCA of QT variation also distinguished species with PC1 and PC2 explaining 23.5 and 9.3% of the total variation, respectively (Supplementary Table S4, Supplementary Figure S4). The Sicilian parent species were distinguished by DS1 while their Sicilian hybrids exhibited intermediate values for DS1 (Figure 3, Table 2). The three traits that contributed most to the DS1 axis were fruit length, pappus length and stomata number. The LDA DS2 values showed that S. squalidus individuals were distinct from Sicilian Senecio (Figure 3, Table 2) due largely to differences in plant height, leaf dissection and ray display. Comparisons of paired-species means for each of the twenty QTs permitted assessment of the S. squalidus phenotype relative to its parents (Table 2). This showed that although there was no significant variation between the three species for seven of twenty QTs, S. squalidus was intermediate to its parents for leaf dissection and ray display, more similar to S. chrysanthemifolius for seven QTs and more similar to S. aethnensis for just one QT, capitulum number. We defined trangressive traits in S. squalidus as not intermediate and significantly (or close to significantly) different from both parents. According to this definition, S. squalidus was transgressively shorter for plant height, had shorter pedicel length and contained more concentrated leaf chlorophyll than either parent. Thus, while for many QTs, S. squalidus is phenotypically similar to S. chrysanthemifolius, for one QT it is more similar to S. aethnensis, while for others it is transgressively distinct from both parents. Differences between S. squalidus and Sicilian hybrids were identified for fruit length, pappus length and pollen number, as well as for the three trangressive traits described above (Table 2).

Figure 3
figure 3

Plot of first and second discriminant component scores resulting from an analysis of 20 quantitative genetic traits measured on 208 wild-sampled Senecio aethnensis, S. chrysanthemifolius, S. squalidus and Sicilian hybrid individuals. Axes are labelled with the percentage variance explained by the first and second discriminant functions, respectively. The same scale has been used for both axes.

Full size image

Discussion

Our detailed comparisons of molecular and QT diversity between the recently originated homoploid hybrid species, S. squalidus, and its progenitor species, S. aethnensis and S. chrysanthemifolius, have emphasized this hybrid species’ distinctiveness from both of its parents and their Sicilian hybrids. Thus, S. squalidus supports the consensus that successful, established hybrid species are genetically and phenotypically distinct from progenitors (Rieseberg, 1997; Rieseberg et al., 2003; Arnold, 2004; Gross and Rieseberg, 2005; Nolte and Tautz, 2010). However, a particularly interesting finding to emerge from the present study was that molecular genetic divergence between S. squalidus and its parents is greater than QT divergence.

Molecular genetic divergence of S. squalidus from its parents

S. squalidus was as differentiated from both parents for molecular genetic diversity as the latter were from each other, in striking contrast with Sicilian hybrids that only exhibited intermediate discriminant scores between the parental species (Figure 2 and Supplementary Figure S2). Importantly, S. squalidus was also highly genetically differentiated from the full range of Sicilian hybrid genotypes and these patterns of differentiation were remarkably consistent across different marker types that exhibited different levels of allelic diversity (Figure 2, Supplementary Figures S1 and S2).

Investigation of the homoploid hybrid origin of S. squalidus is greatly facilitated by historical records documenting the timing and approximate population sizes of key events in its population history such as its introduction to the United Kingdom and its period of cultivation before exponential population growth (Harris, 2002; Abbott et al., 2009). Thus, it is unsurprising to find that unbiased comparisons of molecular genetic diversity indicate reduced diversity in S. squalidus relative to Sicilian Senecio in terms of allele richness and expected heterozygosity (Rs and He) across all investigated marker types that themselves exhibit a wide range of molecular genetic diversities (Table 1). However, the extensive molecular genetic differentiation observed between S. squalidus and its parental species and Sicilian hybrids (Figure 2) is not all simply due to reduced genetic diversity and therefore requires further explanation. One consequence of our background knowledge of the population history of S. squalidus is that we know that the extensive molecular genetic differentiation between S. squalidus relative to its parents and Sicilian hybrids must have accumulated in the 300 generations (assuming one generation per year) since introduction to the United Kingdom. Founder effects and genetic drift have probably had an important, if not predominant, role in generating and shaping genetic differentiation between S. squalidus and its progenitors during this period. The known population history has involved an extreme population bottleneck at the time of introduction of hybrid material from Mount Etna to the United Kingdom, followed by a long period of 90–150 years/generations of small cultivated population size prior to invasiveness. Such demographic history is likely to have contributed to rapid genetic change through loss of diversity and drift (Slatkin, 1996). Subsequently, as S. squalidus experienced rapid population growth during its invasive phase, the remaining genetic diversity would have been preserved and augmented by mutation while preserving altered allele frequencies from the introduction phase (Slatkin, 1996; Templeton, 2008). This study identified that at least three alleles present in UK S. squalidus were absent from both its parents and Sicilian hybrids (Figure 1). In theory, selection generating further molecular genetic differentiation is enhanced in growing populations because relaxed drift prevents more new favourable mutations or allele interactions from being lost when initially very rare soon after appearance (Slatkin, 1996).

A feature of the allele sharing analysis was that 10% of alleles were observed in Sicilian hybrid samples that were not shared with either parent species (Figure 1). It is possible that this is an effect of limited sample sizes failing to identify some of these alleles if they were present at low frequency in parent species samples. It is also possible that the hybrid zone could be acting as a reservoir of molecular genetic diversity retaining alleles introgressed from both parent species that have later been lost due to drift in the parents. The Senecio hybrid zone on Mount Etna is both persistent and large, having persisted for more than 300 generations and estimated to extend between 1.5–3.0 km in width (Brennan et al., 2009). The pattern of alleles from both parents in S. squalidus is suggestive of the importance of the hybrid zone as the potential original source of introduced S. squalidus. We are currently using individual-based, forward simulations, under a plausible range of demographic population histories for S. squalidus, to assess the relative contribution of natural selection and neutral processes to its divergence from its parental species (Barker et al., submitted).

Quantitative trait divergence of S. squalidus from its parents

LDA revealed a detailed and complex picture of QT differentiation between the species (Table 2). For some traits, S. squalidus exhibited an intermediate phenotype between its two differentiated parents, as expected in the case of QTs controlled by multiple quantitative trait loci (QTLs) of individually small additive effect acting mainly in the same trait direction within parental species and in mainly opposite directions between them (Rieseberg et al., 1999; Lexer et al., 2003). Frequently, however, QTs exhibited a parental-like phenotype in S. squalidus that instead indicates dominant expression of the QTL alleles of one parent over the other, or a biased contribution in S. squalidus of QTL alleles from one parent only. The majority of parental-like QTs in S. squalidus were in the direction of S. chrysanthemifolius (seven versus one S. aethnensis-like QT, Table 2) placing S. squalidus close to S. chrysanthemifolius in terms of the first LDA discriminant function (Figure 3).

The pattern of QT divergence between S. squalidus and its progenitors resembles in some respects that reported for certain other homoploid hybrid species and their progenitors. For example, a glasshouse-based QT investigation of the hybrid species Argyranthemum sundingii and its progenitors, A. broussonetii and A. frutescens, observed a similar pattern of hybrid QT variation partially overlapping parental and synthetic hybrid QT variation, whereas for certain QTs the hybrid species resembled the A. frutescens parent in particular (Brochmann et al., 2000).

Three of the QTs we examined exhibited transgressive or more extreme QT means relative to both parent species (Table 2). S. squalidus was also distinct from Sicilian hybrids on the basis of six QTs, including the three transgressive traits (Table 2). These findings are in accordance with observations of other homoploid hybrid species, such as H. anomalus and H. deserticola, which showed that the hybrid species were distinct from early-generation hybrids and exhibit particular QTs that are transgressive relative to their parents H. annuus and H. petiolaris (Schwarzbach et al., 2001; Rosenthal et al., 2002, 2005). Transgressive QT expression can be spontaneously generated in new hybrids when QTLs with fixed but opposing directions of effects within species segregate in heterozygous hybrids and also by interaction of divergent gene expression mechanisms in hybrids (Rieseberg et al., 1999; Hegarty and Hiscock, 2008). Furthermore, in the aforementioned Helianthus homoploid hybrids, selection among parental QTL alleles in novel hybrid habitats has been implicated as a major cause of adaptive transgressive character expression (Rieseberg et al., 1999, 2003; Lexer et al., 2003; Gross and Rieseberg, 2005).

Comparison of amounts of molecular genetic and QT divergence between S. squalidus and its parent species

Direct comparisons between levels of QT differentiation (Qst) and molecular genetic differentiation (Fst or Gst) are a complementary means of investigating the extent to which adaptation has contributed to quantitative differentiation (Merilä and Crnokrak, 2001; McKay and Latta, 2002). However, the population sampling approach of our study does not permit adequate partitioning of genetic and environmental contributions to quantitative variance, preventing Qst estimates that, validly, can be directly compared with molecular genetic estimates. Therefore, we chose instead to focus on multivariate analysis of QTs that have been successfully applied in other hybrid studies (for example, Brochmann et al., 2000; Rosenthal et al., 2002, 2005).

Our analyses indicated that there was less QT divergence between S. squalidus and its progenitors relative to molecular genetic divergence. Thus, whereas S. squalidus showed some overlap with S. chrysanthemifolius and the Sicilian hybrids for QTs, there was no such overlap in terms of molecular genetic divergence (Table 2 and Figures 2 and 3). In contrast to randomly sampled molecular genetic markers representing neutral variation, QTs represent phenotypic variation that is potentially exposed to selection. Accordingly, greater QT differentiation than neutral molecular genetic differentiation is typically observed in studies investigating both forms of variation (Merilä and Crnokrak, 2001; McKay and Latta, 2002). Furthermore, divergent ecological selection is increasingly seen as a key feature required for hybrid species establishment and speciation more generally (Rieseberg, 1997; Buerkle et al., 2000; Sobel et al., 2009; Abbott et al., 2010; Nolte and Tautz, 2010), and invasion of novel urban habitats by S. squalidus appears to conform to this expectation. Indeed, some local adaptation has been identified across the UK range of S. squalidus for flowering time and drought and temperature stress tolerance traits indicating that rapid QT evolution is possible for this species (Allan and Pannell, 2009). In theory, for QTs controlled by multiple QTLs of individual small effect, interactions between alleles at different loci permit evolution of a considerable range of QT variation with only small changes in allele frequency at individual loci (Latta, 1998; Le Corre and Kremer, 2003). However, our observation of greater molecular genetic differentiation than quantitative genetic differentiation contradicts these expectations and requires explanation.

Some of the reasons for this might be due to biases associated with estimating QT differentiation using the PCA approach. The PCA might be prone to underestimating overall QT differentiation due to variance of QTs that are not divergent between species. Thus the resulting principal components will consist of eigenvectors that seek to explain total variation that might not necessarily correspond to maximum species divergence. However, we found that the corresponding LDA analysis that determines the optimal discriminant function of QT loadings to distinguish defined species groups, and that is not susceptible to the potential weakness of the PCA, provided a similar picture of species differentiation to the PCA. A different possible explanation might therefore be that complex interactions between QTs and their underlying genetics, such as pleiotropic constraints, imposed by interactions between multiple QTLs or QTs could generate evolutionary stasis despite selection due to the complex multivariate nature of adaptation (Merilä et al., 2001; Walsh and Blows, 2009). Alternatively, divergent selection pressures might have been weak in this particular case of hybrid speciation because complete geographic isolation between the introduced hybrid and Sicilian progenitors would have removed selection pressures for ecological differentiation as a means of spatial and reproductive isolation that seems to be such an important feature of in situ homoploid hybrid speciation (Buerkle et al., 2000; Lexer et al., 2003; Gross and Rieseberg, 2005). That said, it is unlikely that no adaptation to ecological differences between the intermediate altitude slopes of Mount Etna and urban British habitats was required in the transition from Sicilian hybrid Senecio to S. squalidus, given that the QT analysis presented here finds them to be distinct (Figure 3 and Table 2) and that local adaptation may have subsequently occurred within the UK range of S. squalidus (Allan and Pannell, 2009). Finally, it is worth emphasizing that the observed QT differentiation between S. squalidus and its progenitors is very recent and is likely to represent an example of speciation-in-progress having been underway for only ∼300 generations, in comparison with other new homoploid species examined where speciation has been underway for far longer (Gross and Rieseberg, 2005). Further studies of adaptation, particularly reciprocal transplant experiments, are required to better understand selection for QT differentiation between this new hybrid species and its parents.