Author for correspondence: Ovidiu Paun Tel:+44 2083325378 Email: firstname.lastname@example.org
• Hybridization and polyploidy are now hypothesized to have regularly stimulated speciation in angiosperms, but individual or combined involvement of these two processes seems to involve significant differences in pathways of formation, establishment and evolutionary consequences of resulting lineages. We evaluate here the classical cytological hypothesis that ploidy in hybrid speciation is governed by the extent of chromosomal rearrangements among parental species.
• Within a phylogenetic framework, we calculate genetic divergence indices for 50 parental species pairs and use these indices as surrogates for the overall degree of genomic divergence (that is, as proxy for assessments of dissimilarity of the parental chromosomes).
• The results confirm that genomic differentiation between progenitor taxa influences the likelihood of diploid (homoploid) versus polyploid hybrid speciation because genetic divergence between parents of polyploids is found to be significantly greater than in the case of homoploid hybrid species.
• We argue that this asymmetric relationship may be reinforced immediately after hybrid formation, during stabilization and establishment. Underlying mechanisms potentially producing this pattern are discussed.
Hybrid speciation refers to the mode of origin of a new species in which gene flow between species plays a major role. More than 25% of plant species seem to be involved in hybridization with other species (Mallet, 2005), but its frequency seems to vary considerably between groups, for example being more prevalent in rapidly radiating lineages (see Ellstrand et al., 1996). Closely related species are most likely to hybridize, but this phenomenon often persists for millions of years after initial diversification (Mallet, 2005). The rate of hybrid speciation is definitely much lower than the statistic of 25% owing to many disadvantages that early generation hybrids need to overcome to achieve successful establishment (e.g. reduced fertility and viability, lack of reproductive and ecological isolation from the parents, lack of mates of the same type, hybrid dysgenesis and necrosis, etc.). An estimate from five regional floras indicated that c. 11% of species are putative hybrids (Ellstrand et al., 1996).
Homoploid hybrid speciation (‘recombinational’, sensu Grant, 1981) appears to be facilitated by several factors, for example, availability of a suitable ecological niche or an available fitness peak, and rapid chromosomal evolution (Rieseberg, 1997; Mallet, 2007). To be evolutionarily successful, even fertile and ‘stable’ homoploid hybrids must be reproductively isolated from the parental species either by sorting genic or chromosomal sterility factors that already differentiate the parental species (Grant, 1981; Rieseberg, 2001; Wu, 2001) or by prezygotic barriers, such as spatial/temporal isolation and/or divergence into a new ecological niche (Rieseberg, 1997; Gross & Rieseberg, 2005). Indeed, hybrids may combine characteristics from both parents and/or exhibit transgressive traits that allow ecological distinctiveness (Andersson, 1949; Arnold, 1997; Rieseberg et al., 2003; Seehausen, 2004).
Homoploid hybrid speciation seems to proceed at a rapid tempo (Rieseberg et al., 1996; Rieseberg, 1997; Buerkle & Rieseberg, 2008), and diploid hybrid genomes are likely to be stabilized quickly, for example after 10–60 generations in the case of Helianthus anomalus (Ungerer et al., 1998). However, Buerkle & Rieseberg (2008) have recently shown that in Helianthus recombination continues to shape genomic composition of homoploid hybrid species for hundreds of generations. Even at this scale, diploid hybrid speciation can still be considered one of the fastest modes of speciation.
Hybrid speciation may, however, happen much more suddenly when combined with polyploidy, which immediately provides a hybrid with a high degree of post-zygotic reproductive isolation from its progenitors: backcrossing to either parent will produce nonviable or mostly sterile offspring of odd-numbered ploidy (triploids, pentaploids etc.: Stebbins, 1950, p. 308; Grant, 1981; Ramsey & Schemske, 1998). Allopolyploidy can be the product of gametic nonreduction (frequently via a ‘triploid bridge’; Ramsey & Schemske, 1998), and, more rarely, can also result from somatic chromosome doubling of a homoploid hybrid or polyspermy (Thompson & Lumaret, 1992; Ramsey & Schemske, 1998; Mallet, 2007). Of all these pathways, nonreduction during meiosis seems to be the most frequent route to polyploidy, as parents of spontaneous polyploids often produce a substantial number of unreduced gametes (see reviews by Thompson & Lumaret, 1992; Ramsey & Schemske, 1998).
Even if allopolyploidy can be viewed as abrupt or saltational speciation (Mallet, 2007), most neopolyploids will fail to become established because of meiotic abnormalities (Ramsey & Schemske, 2002) and/or their isolation, resulting in a frequency-dependent minority cytotype disadvantage (Husband, 2000). However, the latter may be overcome with the help of perenniality, asexual reproduction, assortative mating and loss of self-incompatibility barriers. Originating in sympatry (or parapatry) with progenitors, allopolyploids still require niche divergence to escape direct competition with parental taxa (Coyne & Orr, 2004). The co-joined genomes in polyploids usually have to face a complicated process of reorganization before full stabilization: chromosomal rearrangements within parental genomes, loss of low-copy DNA sequences, epigenetic effects on expression in duplicated genes and activation of transposable elements (for reviews see Comai, 2005; Chen, 2007; Paun et al., 2007). Such genomic responses also have the potential to induce novel expression patterns, which together with permanent heterozygosity (potentially resulting in hybrid vigour) and gene redundancy, might result in significant shifts in morphology, breeding system and ecological tolerances, and, finally, in elevated evolutionary flexibility and major ‘jumps’ in evolution (De Bodt et al., 2005; Comai, 2005; Otto, 2007; Paun et al., 2007).
Speciation via polyploidy is likely to be a major mode of sympatric speciation in plants. A model-based estimated frequency of polyploid (usually allopolyploid) speciation in angiosperms points to at least 2–4% of recent speciation events (Otto & Whitton, 2000). However, recent direct estimates indicate that 15–25% of angiosperm speciation events are accompanied by an increase in ploidy (Wood & Rieseberg, 2005). Moreover, up to 70% of extant flowering plant species are currently polyploids (Otto & Whitton, 2000); in fact all angiosperms have descended from polyploid ancestors and are paleopolyploids (N.B. except probably Amborella; De Bodt et al., 2005; Cui et al., 2006; Soltis et al., 2009). Meyers & Levin (2006) suggested that the abundance of polyploids may result from a simple ratcheting mechanism; they argued that in evolution chromosome number can double but not halve. However, genome size (as DNA amount and chromosome number) can decrease, for example as observed in Nicotiana sect. Suaveolentes where multiple chromosome fusions have resulted in chromosome number reduction (Chase et al., 2003).
Because polyploidy and hybridization have been so central to plant evolution, it is important to identify processes responsible for origins of hybrid species and those that promote shifts in ploidy, changing the possible outcomes of hybrid speciation. The interest here is not simply limited to predicting results of hybrid evolution and of polyploid dynamics, it is also of great importance for our understanding of evolutionary processes that result in isolation between species, including those that influence establishment of new taxa and maintain biodiversity.
A relevant hypothesis was proposed in the early twentieth century: the level of (structural) differentiation between ancestor genomes influences ploidy of successful hybrids (Winge, 1917; Darlington, 1937; Stebbins, 1950). Winge (1917, as cited by Darlington, 1937), for example, considered that polyploid formation after somatic doubling of homoploid hybrids would be stimulated by the need for a partner with which chromosomes could pair. Therefore, higher chromosomal differentiation between parents would increase the chance of shifts in ploidy. Decades later, Grant (1981, pp. 247–248 and 320) referring also to the initial formation of an allopolyploid, stated that pre-existing chromosomal rearrangements within parental genomes ‘upset the course of meiosis in the hybrid’, resulting in reduced pairing, and that the latter ‘sets (the stage) for (gamete) nonreduction and amphiploid formation’. Other authors extended this idea by referring more to the moment of polyploid establishment, rather than initiation. Darlington (1937, p.136), for example wrote: ‘the characteristic properties of hybrids depend not on the properties of the parents, but on the differences between these properties’. He considered that a ‘differential affinity’ between parental chromosomes governs long-term successful pairing in structural hybrids and polyploids (pp. 160, 172, 199): ‘The greater the (parental) dissimilarities, the more regularly do the identical chromosomes pair in the allotetraploid derived, and therefore the less frequent are the multivalents in the tetraploid’. In 1945 Clausen et al. (as cited by Buggs et al., 2008) reached the conclusion that the ‘success and constancy’ of allopolyploids must be linked with the ‘degree of relationships’ found between their parents. Even Stebbins (1950, p. 354) referred to the genetic relationships of the parental diploid species to each other as one of the factors promoting development of allopolyploidy in plants.
The potential cause–effect relationship between the level of chromosomal (genetic) divergence of the parents and ploidy of hybrids has recently been revisited by Chapman & Burke (2007) and Buggs et al. (2008), who, from different perspectives, reached partly contradictory conclusions. Based on 11 cases of homoploid hybrids (plus a misclassified polyploid Eupatorium) and 26 cases of allopolyploids, Chapman & Burke (2007) demonstrated that, in angiosperms, parental nuclear ribosomal internal transcribed spacer (ITS) divergence is significantly greater for allopolyploids than for homoploid hybrids. However, the method employed disregarded the variable substitution rates expected across such unrelated cases even in the same molecular marker (a caveat also discussed by the authors) and it included in the analysis hybrids formed by parents with different basic chromosome numbers (e.g. Arabidopsis suecica, Spiranthes diluvialis and Symphyotricum ascendens) or even different ploidies (Artemisia douglasiana, Primula scotica and Rubus maximus). Hybrid speciation starting from such parental pairs is particularly prone to result in allopolyploids and might follow special routes and rules (Ramsey & Schemske, 1998). By contrast, Buggs and collaborators (2008) took a molecular phylogenetic approach to the issue, but relied on subjectively defined clades as a measure of genetic divergence. Moreover, the latter study considered any naturally occurring hybrid individual reported in the literature within eight selected plant genera and, therefore, focused on polyploid formation not evolutionary success (effective speciation). Long-term success in meiosis is key to operation of the mechanism that governs ploidal shifts, which means that only taxa that appear to be valid species in their own right should be included in the calculations. Therefore, by including ephemeral homoploid hybrids, sterile triploids and neopolyploids, Buggs et al. (2008) did not directly evaluate the classical cytological hypothesis and failed to find convincing evidence showing that ploidal increase in established species is determined by the phylogenetic distance between progenitor species. However, their findings point to a restriction of homoploid formation to parental pairs less divergent than expectation if crossings were random between all species pairs in a genus.
In the light of this recent debate, we approach the potential relationship between ancestor divergence–descendant ploidy by uniting methodologically the two recent studies mentioned earlier. Like Chapman & Burke (2007) we use the extent of genetic divergence between parental pairs as a surrogate for chromosomal differentiation, but we attempt to standardize the method by taking into account the rate of evolution in the respective marker(s) and genus from a phylogenetic approach. We extend the sampling to more cases, but we include only diploid parental pairs with identical base chromosome numbers and only fertile, successful hybrids that have a long species history.
Materials and Methods
Selection of taxa
This analysis is based on 50 case studies (Table 1) chosen from the literature following several criteria: (1) the hybrid status for the respective species has been documented with some certainty by molecular means in addition to (at least) morphology; (2) an extensive and representative molecular phylogenetic analysis for the genus including the parental taxa was already available; (3) the parents are diploids and have the same chromosome number; and (4) the hybrids are natural and stable, with proven evolutionary success (neopolyploids and unnamed suspected hybrids were excluded). Owing to methodological constraints, we did not consider in our analysis intergeneric hybrids (e.g. allotetraploid Triticum turgidum), hybrids produced by more than two parents (e.g. homoploid hybrid Iris nelsonii) or hybrids from genera of uncertain delimitation (e.g. Tarasa and Brassica). The last exclusions were followed to try to minimize the influence of clearly artificial taxonomies.
Table 1. Details of hybrids included in this analysis (taxonomy used follows the original papers; P-GDI, uncorrected P-derived parental genetic divergence index)
We classified the data into three categories: (1) homoploid hybrids (n = 16); (2) allopolyploids (n = 32); and (3) two cases of both diploid and polyploid hybrids formed by the same parental species (Table 1). We counted the parental pairs, so we considered just once the instances where more than one homoploid or polyploid hybrid was formed by the same parental pair. In this way, we were able to identify half as many homoploid hybrids as allopolyploid species. This pattern may result from a biological rarity of homoploid hybrid speciation versus allopolyploid speciation, but it also mirrors a more general problem, namely that detecting and rigorously documenting homoploid hybrid species is much more difficult than detecting polyploid ones (Rieseberg, 1997).
Molecular data and statistical analyses
Some DNA sequence matrices were obtained directly from authors of published analyses (see the Acknowledgements); for the others, DNA sequences were collected from GenBank (http://www.ncbi.nlm.nih.gov) and realigned using clustal w (http://www.ebi.ac.uk/Tools/clustalw/; Chenna et al., 2003). Based only on ingroup taxa (species within genera), we calculated with paup* 4.0b10 (Swofford, 2003) all intrageneric pairwise genetic distances, using both uncorrected p-distances (P) and Kimura's (1980) two-parameter (K2P) distances. Uncorrected P is the observed number of changes between two sequences, with no correction for multiple changes. By contrast, the K2P model addresses this problem by considering equal base frequencies but different rates for transitions and transversions (Kimura, 1980).
Because species-level phylogenetic analyses use molecular markers exhibiting different substitution rates, we standardized our data among cases by calculating for each parental pair a genetic divergence index (GDI). For each instance, the genetic distance between parental pairs (Pd) was divided by the average genetic distance (Av) in the genus based on the same molecular markers. Under this definition, GDI is always positive; if GDI > 1.0, then Pd is higher than Av. When multiple sequences were available for a given taxon, an average of the genetic distance for all possible parental pairs was used in further analyses.
To check for potential bias in our analysis created by uneven sampling in phylogenetic trees, we performed a nonparametric, one-tailed Spearman rank order correlation of Av with the number of taxa included in each tree for both homoploid and allopolyploid species.
All statistical analyses were performed using SPSS 15.0 (SPSS, Chicago, IL, USA). As Pd, Av and GDI are not expected to be normally distributed, we treated our data as nonparametric.
The two genetic measures applied in this study, P and K2P, gave significantly congruent results (Spearman's correlation coefficient based upon ranks rho = 1, P < 0.0001, independently for Pd, Av and GDI). As expected, K2P values of Pd and Av were generally slightly higher then those calculated with P (see the Supporting Information, Table S1). However, GDI values based on the two genetic distances were identical up to the second decimal, confirming the value of our standardizing approach. In the following tests, we generally focused on the P-derived GDI because of the simpler assumptions.
Nonparametric comparisons of GDI values (calculated overall, exclusively on nuclear data, or just with nuclear ribosomal ITS data) for homoploid versus polyploid hybrid species using the Mann–Whitney test indicated statistically significant asymmetric relationships (P < 0.0001, Table 2). Parents of polyploids are generally more divergent than the average intrageneric distance (i.e. GDI > 1), whereas for most homoploid hybrids GDI is < 0.5 (Fig. 1). In addition, in all cases of direct comparisons between homoploid and polyploid hybrids in the same genus (i.e. Achillea, Actinidia, Gossypium, Helianthus and Lithophragma; Table 1) parents of polyploids are more divergent.
Table 2. Nonparametric comparisons of parental divergence indices for homoploid versus polyploid hybrid species using the Mann–Whitney test as calculated in SPSS
Type of data
The indices are calculated using either the uncorrected-p (P) or Kimura's (1980) two-parameter (K2P) distance.
A histogram (Fig. 2a) illustrating frequency distributions of classes of parental GDI for homoploid hybrid species and allopolyploids indicates that both categories have unimodal but distinct distributions. The relationships between frequency of occurrence and degree of chromosomal divergence of parental pairs for allopolyploids and homoploid hybrids (Fig. 2b) meet at a GDI ≈ 0.75, indicating an equal probability of a hybrid formation with and without a change in ploidy when Pd is c. three-quarters of Av.
The nonparametric, one-tailed Spearman rank order correlation of Av with the number of taxa included in each phylogenetic analysis was not significant (Spearman's rho = 0.069, P = 0.322).
By comparing frequency distributions of parental genetic distance (used here as a surrogate for chromosomal differentiation) for homoploid and allopolyploid hybrid species, we demonstrate the relevance of progenitor divergence as a determinant of ploidy in resulting hybrid species: although the range of genetic divergence between the parents of homoploid hybrids is similar to those of allopolyploids, the actual values of divergence are significantly higher in the latter (Fig. 1).
Our standardized approach, integrating each parental pair within its generic context, has several advantages: it makes our method independent of assumptions implied by specific genetic distances or defining clades (cf. Buggs et al., 2008), it allows us to include molecular markers and more cases (cf. Chapman & Burke, 2007) and gives our analysis greater predictive power. The last derives from our suggestion that species pairs with a divergence smaller than three quarters of the average divergence between species within the genus (i.e. GDI ≤ 0.75) have chromosomes mostly displaying colinearity of genes. Most homoploid hybrids (75%) included here were formed between such parental pairs, but this category included just 12.5% of allopolyploids (Fig. 2a). Furthermore, if a parental pair has a divergence greater than three-quarters of the average in a given genus, most of their corresponding chromosomes are likely to be sufficiently heterologous to act as homeologs, and hybridization is most likely to result in an increase in ploidy.
Two cases (in Paeonia and Stephanomeria; see Table 1) identified in the literature for which the same parental pair has successfully produced both homoploid and allopolyploid hybrid species substantiate our results. Their calculated divergence index (Table 1) is indeed close to the estimated value for which there should be an equal probability of hybrid formation with and without ploidy change (i.e. GDI ≈ 0.75, Fig. 2b). Our results parallel those of Chapman & Burke (2007): their analysis indicated that parents of allopolyploids are, on average, more than twice as divergent as parents of homoploid hybrids, a significant relationship that is also visible in GDI (Fig. 1).
Assumptions, limitations and alternatives
The general premise that genetic divergence provides the best available surrogate for differentiation of chromosome sets is often employed (Edmands, 2002). As early as 1937, Darlington (p. 197) hypothesized ‘a correlation between genetic differentiation of the chromosomes of the species and their structural differentiation’. Indeed, both genetic distance and magnitude of difference in genomic rearrangements between two species are expected to be proportional to the evolutionary time since common ancestry.
Calculating an average genetic distance within each genus adds a subjective component to our analyses. We cannot eliminate taxonomic inconsistencies created by differences in taxonomic practice among authors working on different taxa. We also start from the assumption that the modern taxa studied here are closely related to the actual progenitors of the hybrid species and that after hybridization genetic divergence between these species has remained largely unchanged. More appropriately, these taxa should be considered as the closest living descendants of the donor species. However, most of the cases included here, both homoploid and polyploid hybrids, are likely to be relatively recent, as with time such cases become increasingly difficult to detect (Chase et al., 2003; Clarkson et al., 2004).
Underlying processes: allopolyploids
A theoretical model for polyploid speciation along a continuous variation of genomic divergence between diploid progenitors was developed by Sang and collaborators (2004). They treated the origin of a successful polyploid lineage as a function of (1) polyploid formation (production of polyploid individuals from diploid populations), further broken up into probability of unreduced gamete production and frequency of hybridization, and (2) successful establishment of polyploid populations. Such a model would imply that frequency of polyploid formation would have a negative exponential distribution on parental genomic divergence. The overall probability of successful polyploid speciation, however, would have a unimodal distribution on parental divergence (Sang et al., 2004), with established autopolyploids being much less frequent than allopolyploids, despite the fact that autopolyploids occur spontaneously in nature at relatively high rates (see Ramsey & Schemske, 1998).
The main route to allopolyploid speciation is represented by the fusion of an unreduced gamete with a haploid gamete resulting in a ‘triploid bridge’ (Ramsey & Schemske, 1998; Husband, 2000), and after self-fertilization or backcrossing to diploids a new allotetraploid may originate. Alternatively, allopolyploid speciation can also result after fusion of two unreduced gametes, with better chances in dense hybrid zones, marginal or disturbed habitats and/or other limiting conditions (e.g. temperature variation; Thompson & Lumaret, 1992; Ramsey & Schemske, 1998). Such unreduced gametes are thought to be rare (Mallet, 2007) and will be unsuccessful and lost, especially if enough haploid gametes are produced. However, poor chromosome pairing in unbalanced diploid F1 hybrids leads to asynapsis at the first meiotic division, and such organisms form unreduced gametes with much greater frequency. Ramsey & Schemske (1998) reported that unreduced gametes are produced at rates c. 50 times higher in hybrids than in nonhybrid lineages. As predicted by Grant (1981, see the Introduction), it seems plausible that greater levels of parental divergence increase meiotic abnormalities at the homoploid level and, thus, the rate of nonreduction. For example, in Lilium, most gametes produced by intersectional hybrids are unreduced (van Tuyl et al., 1989). The same trend is expected in the case of allopolyploidy resulting from somatic chromosome doubling in meristematic tissues of a diploid hybrid or in a zygote/young embryo (e.g. Primula kewensis; Ramsey & Schemske, 1998). Such somatic doubling may be directly triggered by structural differences between parental homeologs (Winge, 1917; see the Introduction).
However, both somatic chromosome doubling and unreduced gametes require spontaneous occurrence of diploid hybrids that are at least partly fertile and self-compatible (Sang et al., 2004), thus limiting the possible maximal extent of parental divergence. Such a bounded distribution of parental divergence, between lower chances of gamete nonreduction or somatic doubling at minimal parental divergence and full diploid hybrid sterility at increased divergence, may result in apparently random occurrence of allopolyploidization events as suggested by Buggs et al. (2008). In addition, rare allopolyploid formation events via unreduced gametes produced directly in nonhybrid parents will not follow the rules presented above and may blur the trends concerning formation of allopolyploid individuals.
Independent of formation, a nascent allopolyploid must become established and expand its population/s in order to produce a new species. Establishment of the polyploid lineage will depend not only on stochastic events, such as the availability of appropriate environments, but also on its degree of viability, fertility, heterozygosity (hybrid vigour) and fitness. Darlington (1937, p. 196) hypothesized ‘a negative correlation between the fertility of diploids and that of the tetraploids to which they give rise.... The greater the dissimilarities in the diploid, the more regularly do the identical chromosomes pair in the allotetraploid derived from it, and therefore the less frequent are the multivalents in the tetraploid’. A diploid hybrid with reduced chromosome pairing will exhibit a high degree of sterility, which doubling overcomes. Fertility of allopolyploids might increase with parental divergence, because of fewer meiotic abnormalities. Ramsey & Schemske (2002) provided evidence that the degree of allopolyploid fertility is positively correlated with frequency of bivalents, but not with other configurations. They also concluded that allopolyploids generated by semisterile diploid hybrids are generally much more fertile than their progenitors, an attribute partly reflecting genic incompatibilities independent of and in addition to meiotic behaviour. A significant increase in effective population size seems to be the consequence of selection on fertility acting on neopolyploids, which rapidly increases pollen viability and seed set. The picture here is more complex than at homoploid level, and it is still unclear how parental incompatibilities will behave at the polyploid level, when they result in sterility/reduced fitness or breakdown. This is also the result of a lack of study; allopolyploids do not facilitate gene flow between diverging parental taxa and, hence, they are not considered in current research and debates regarding reproductive isolation and genetics of speciation (see for example Widmer et al., 2009).
It is expected that heterozygosity (resulting in heterosis) generally provides increased fitness and adaptive potential, through enhancing the potential for spatial, temporal and functional variation in gene expression (Flagel et al., 2008; Leitch & Leitch, 2008). The proportion of homeologous loci that are stably heterozygous should be positively correlated with genomic divergence between progenitors. In addition, alterations of gene expression in allopolyploid genomes have the potential to trigger fitness differences in the parental environment, which will be available to selection. A few case studies have indicated that the extent of genomic alterations and changes in gene expression may depend on the degree of divergence between the parental diploid genomes. For example, Song et al. (1995) observed fewer rearrangements in the allopolyploid genome formed from closely related Brassica rapa and Brassica oleracea, but many more in the allopolyploid combining more divergent B. rapa and Brassica nigra. Another such example is Nicotiana where allopolyploids Nicotiana arentsii and Nicotiana rustica show only minimal genetic changes, but Nicotiana tabacum (resulting from widely divergent Nicotiana sylvestris and Nicotiana tomentosiformis) exhibits intergenomic translocations (Lim et al., 2004). Rapid genomic repatterning will also increase genetic variability available to new polyploid populations. In addition, extensive changes in gene expression seem to be triggered by wide hybridization rather than polyploidy (Paun et al., 2007); recent studies show that genome duplication can, in fact, have widespread ameliorating effects on altered levels of gene expression arising from hybridization (as, for example in Senecio; Hegarty et al., 2006). In allopolyploid cotton, however, a significant proportion of expression novelty is likely triggered by polyploidy after long-term evolutionary processes on duplicated genes (Flagel et al., 2008). Finally, shifts in breeding system, such as the breakdown of self-incompatibility, seem to be initiated by polyploidization alone, whereas others, such as apomixis, usually seem to be triggered by the effects of combining hybridization and polyploidy (see Paun et al., 2006). In conclusion, the frequency distribution of allopolyploids along the continuum of ancestral genomic divergence (Fig. 2b) will have an optimum between low fertility, heterozygosity and fitness at minimal parental divergence (as compared to diploid progenitors and homoploid hybrids) and low probability of polyploid formation towards maximal divergence of progenitors (due to increased prezygotic and postzygotic barriers).
Autopolyploids and allopolyploids
Buggs et al. (2008) argued that the low frequency of allopolyploids in lower parental divergence classes in the study of Chapman & Burke (2007) was caused by exclusion of autopolyploids from analysis of the latter. Similarly, our study does not include any autopolyploids, but we definitely expect that parental divergence in this case should not span such a high interval (at least until GDI = 1, i.e. Pd = Av) to make the polyploid distribution fit a negative linear/exponential function (Fig. 2). The overall polyploid distribution would instead be bimodal, indicating the presence of different phenomena influencing the frequency of the two types of polyploids (see also Sang et al., 2004). We hypothesize that the adaptive valley between autopolyploid and allopolyploid frequency may have resulted from more or less equal combinations of bivalents and quadrivalents in meiosis. However, we regard inclusion of autopolyploids in analyses considering hybrid speciation as inappropriate because autopolyploid formation should correspond, at the diploid level, to frequent events of intraspecific processes, and these do not necessarily have a significant contribution to speciation.
Underlying processes: homoploid hybrids
The formation of homoploid hybrid individuals is partly shaped by stochastic events (e.g. shifts in distribution ranges) but is directly limited by the strength and nature of reproductive isolation between species pairs. Plant species are typically isolated by many prezygotic and postzygotic barriers and their complex interactions (Coyne & Orr, 2004). Recent studies have suggested that prezygotic isolation is usually much stronger than postzygotic isolation (Lowry et al., 2008; Widmer et al., 2009; but see Cozzolino et al., 2004), because of factors such as distribution, immigrant inviability, phenological differences, pollinator specificity, mating system and pollen competition. Significant trends obvious in hybrid formation without a change in ploidy (Buggs et al., 2008) represent, to a great degree, the entire speciation process in homoploid hybrids (this study). Studying hybrid formation, Buggs and colleagues (2008) reached the conclusion that parents of homoploid hybrids are less divergent than would be expected with random crossing. We found a similar, but more significant pattern (Fig. 2): the probability of production of a diploid hybrid is highest if the parental divergence is less than or equal to (more or less) half the average in the genus and decreases as reproductive barriers (both prezygotic and postzygotic) become stronger between diploid parents with increased genomic divergence. With greater parental divergence, genic and/or chromosomal incompatibility will occur with greater probability. Examples of characterized genic incompatibilities include: genes involved in hybrid necrosis or weakness (Bomblies & Weigel, 2007) and cytonuclear incompatibility (Chase, 2007). Models have suggested that the level of incompatibility between species should increase with evolutionary time at least as rapidly as the square of the divergence time between the two species (snowballing effect; Orr, 1995). Therefore, the degree of fertility of a diploid hybrid decreases rapidly with increased genomic divergence. Classical models of chromosomal speciation have stated that after a particular level, meiotic mismatches of parental chromosomes or karyosignificant reduction in fitness (White, 1978; but see Lowry et al., 2008). In addition, chromosomal divergence may also increase the strength of genic incompatibilities by suppressing recombination and therefore maintaining the effects of linked isolation genes (Rieseberg, 2001).
The diploid form of hybrid speciation occurs mainly in sympatry or parapatry, and hence the greatest challenge of the nascent taxon is to achieve reproductive isolation. Isolation from progenitors often occurs as a byproduct of the process that stabilizes the hybrid lineage and may be based on ecological factors (e.g. habitat divergence; Gross & Rieseberg, 2005), sorting pre-existing sterility factors and/or chromosomal rearrangements (Grant, 1981). All these pathways are potentially influenced by the extent of parental differentiation. Ecological divergence may be acquired by positive heterosis (hybrid vigour; Lippman & Zamir, 2007) and transgressive segregation (Gross & Rieseberg, 2005). Numerous experimental crosses have suggested that the optimal degree of genetic divergence for maximal expression of positive heterosis occurs within a range of divergence that is narrow enough for cytological irregularities not to be apparent (Moll et al., 1965; Cox & Murphy, 1990). Reproductive isolation can also be achieved by sorting pre-existing chromosomal rearrangements that differentiate the parental species, resulting in the formation of a novel recombinant genotype that is homozygous for these rearrangements (‘recombinational’ speciation, Grant, 1981, p. 250). Stronger genetic isolation from progenitors is most likely when a barrier differentiating the parents is genetically and/or chromosomally extensive and complex (Rieseberg, 2000) However, the parental species must have genomes similar enough for pairing and recombination to occur.
There are at least two important implications of our results. First, there may be a fitness and fertility valley between homoploid hybrids and allopolyploids with increasing genomic divergence of progenitors (see also Darlington, 1937). A second adaptive valley may be represented by intermediates between typical autopolyploids and allopolyploids, which will suffer the effects of mixed multivalent and bivalent formation in meiosis. Greater support for these hypotheses requires further experimental study.
We thank D. Albach, R. Bateman, E. Conti, A. Davis, Y. Guo, G. Mansion, A. Mast, E. E. Schilling and R. E. Timme for providing aligned DNA sequence matrices, M. Winkler for comments on appropriate statistical analysis, I. Leitch and A. Leitch for insightful discussions, and J. Wendel and one anonymous reviewer for useful criticisms. This research was funded by an Erwin Schrödinger fellowship (Austrian Science Fund, FWF; project J26406-B03) and an Intra-European Marie Curie fellowship (DactGene, MEIF-CT-2007-040494) to O.P.