R.J.A.B. uses molecular genetic and bioinformatics approaches to study duplicate gene evolution in recent Tragopogon allopolyploids. P.S.S.’ research interests include: plant phylogenetics, polyploidy, gene family evolution, phylogeography and conservation genetics. D.E.S. is interested in angiosperm phylogeny, genome doubling, floral developmental genetics, phylogeography and molecular cytogenetics.
Dr Richard J. A. Buggs, Fax: +1 352 846 2154; E-mail: firstname.lastname@example.org
Hybridization and whole-genome duplication are both potential mechanisms of rapid speciation which sometimes act in concert. Recent surveys, showing that homoploid hybrid species tend to be derived from parents that are less evolutionarily divergent than parents of polyploid hybrid species (allopolyploids), have been interpreted as supporting a hypothesis that high divergence between hybridizing species drives whole-genome duplication. Here, we argue that such conclusions stem from problems in sampling (especially the omission of autopolyploids) and null model selection, and underestimate the importance of selection. The data simply demonstrate that hybridization between divergent parents has a higher probability of successfully producing a species if followed by polyploidization.
Hybrid speciation occurs when two existing species cross and produce phenotypically distinct offspring (or later-generation progeny) which then become reproductively isolated from their parents. This mode of speciation was once regarded as unimportant (Darwin 1859; Mayr 1963) but is now seen as a significant evolutionary process in both plants (Rieseberg & Willis 2007; Soltis & Soltis 2009) and animals (Mallet 2007; Jesús Mavárez & Linares 2008). This new view of hybrid speciation has developed through both the discovery of numerous hybrid species, and an increased understanding of processes which allow the obstacles to hybrid speciation to be overcome (Rieseberg 1997; Mallet 2007).
Several obstacles stand in the way of successful hybrid speciation (reviewed in Grant 1981; Mallet 2007; Rieseberg 1997; Soltis & Soltis 2009). Geographic separation or reproductive differences may prevent crossing between two potential parent species. If hybridization does occur, the progeny may abort or have low fitness due either to genetic incompatibilities between the parental genomes, or the generation of maladaptive phenotypes. Even if the hybrid has high fitness, it may never achieve reproductive isolation and will backcross with its parents, leading to introgression. Successful hybrid speciation is therefore a comparatively rare event that occurs under a limited range of conditions.
One process involved in the success of hybrid species is transgressive segregation: the expression of trait values that exceed the range between the parental means (McDade 1990; Rieseberg et al. 2003). Transgressive segregation raises the likelihood of speciation as the hybrids have new phenotypes not found in the parental species and may therefore occupy an underused fitness peak on the local adaptive landscape (Mallet 2007). Under a scenario involving transgressive segregation, ecological differentiation will occur rapidly following hybridization, conferring a degree of reproductive isolation which may then be reinforced with prezygotic isolation.
Another process frequently allowing success of hybrids is whole-genome duplication, which is commonly found in plants and more rarely in animals (Otto & Whitton 2000; Mable 2004; Rieseberg & Willis 2007). After genome doubling, backcrosses with parental diploid species will generate progeny with an odd number of genome sets and typically low fitness, providing postzygotic reproductive isolation (Ramsey & Schemske 1998). Parental genomic incompatibilities may be overcome because genome doubling causes every chromosome to have an identical homologue—thus pairing between divergent parental chromosomes at meiosis is not necessary for successful gamete formation. Whole-genome duplication can occasionally confer phenotypic changes such as increased size or a change in sexual system, which may also contribute to speciation (Otto & Whitton 2000).
The role of genetic divergence
The extent to which genetic divergence between parental species affects hybrid speciation processes is currently under debate. The probability of ecological differentiation following hybridization would seem to increase with the level of divergence between the hybridizing parental species, as increased divergence would probably expand the range of possible intermediate phenotypes. Recent evidence also suggests that transgressive segregation in hybrids increases with genetic distance between parental species (Stelkens & Seehausen 2009). Wide hybrids may therefore have more evolutionary potential than hybrids between closely related species.
However, genetic differentiation between parental species is likely to be positively correlated with some of the obstacles to hybridization. Under a general model of allopatric speciation, geographic separation is likely to be more pronounced between distantly related species, reducing the incidence of hybridization between distant relatives. In addition, genetic incompatibilities due to, for example, chromosomal rearrangements, Dobzhansky–Muller incompatibilities, or differential loss of duplicated genes, are also likely to increase with genetic divergence, leading to hybrid inviability (Lynch 1991; Orr 1995). Thus, wide hybrids are less likely to form than those between close relatives, and may have lower fitness.
As some of the negative effects of genetic divergence between parental species can be overcome by whole-genome duplication, we would expect formation of hybrid polyploid species to be successful over a greater range of parental divergences than homoploid hybrid species. At high parental divergences, allopolyploids may benefit from the effects of transgressive phenotypes, while homoploid hybrids between highly divergent parents, which are likely to be sterile, may not similarly benefit. Thus, we might expect the average divergence between parents of allopolyploids to be higher than that of homoploid hybrid species, simply because polyploidy enables wide hybrids to be viable.
Several researchers have made the intriguing suggestion that rather than being a random mutation that fortuitously confers viability to a wide hybrid, successful polyploidy is determined by wide hybridization: genomic differentiation between parental species in fact drives successful whole-genome duplication in their hybrids. Winge (1917) first suggested that ‘occasional hybridization might be the cause’ (p. 13) of polyploidy. He developed ‘a scheme showing the different degrees of physiogenetic likeness between gametes—and thus also between their chromosomes—endeavouring at the same time to ascertain what results we can expect in each case from the fusion of gametes’ (p. 196). In this scheme, a hybrid in which the chromosomes from parental species were unable to pair would have to undergo chromosome doubling in the zygote to have ‘any possibility at all of propagating’ (p. 199). Subsequent workers proposed that this doubling would occur through unreduced gamete formation. Darlington (1937) suggested that gametic doubling occurs in hybrids because meiosis has proved unworkable. Similarly, Grant (1981) considered that reduced chromosome pairing in hybrids between parents whose chromosomes had different structures would ‘set the stage’ for unreduced gamete formation. Ramsey & Schemske (1998) found unreduced gametes to occur at a frequency of 28% in hybrids but only 0.6% in nonhybrids. This could suggest that unreduced gamete formation is actually triggered in hybrids.
Darlington (1937) also suggested another level at which this drive could occur. He proposed an inverse relationship between the fertility of a diploid hybrid and that of a tetraploid to which it gives rise. He reasoned that at low parental divergences, homoploid hybrids will be fertile because chromosomes will be able to pair at meiosis, but allopolyploids will be of low fertility because pairing will occur between both duplicated chromosomes and homologous chromosomes from each parent, causing uneven segregation. In contrast, at high parental divergences, homoploid hybrids will be sterile due to failure of chromosome pairing, but allopolyploids will be fertile due to consistent bivalent formation at meiosis. This has sometimes been called ‘Darlington’s rule’ (not to be confused with Darlington’s rule in biogeography which states that with every ten-fold increase in area, the number of species doubles). A literature survey by Clausen et al. (1945) seemed to support this rule by showing that the success and constancy of allopolyploids is ‘linked with the degree of relationship found between their parents’ (p. 2). They argued that the ‘parent species…should be closely enough related to produce a vigorous F1 hybrid, but remotely enough so that the balance between their combined genomes can be perpetuated’ (p. 68–69). Stebbins (1950) agreed with this conclusion. However, a recent survey of neopolyploids (Ramsey & Schemske 2002) did not show significantly lower fertility in autopolyploids than allopolyploids, and we now know that many allopolyploids do not show consistent bivalent formation and that nonhomologous transposition can occur between parental genomes (Leitch & Leitch 2008). These findings suggest that although Darlington’s rule may describe the average fertility of hybrids and allopolyploids, selection may subsequently play a large role in preserving fertile autopolyploids, as well as allopolyploids that formed from closely related parents.
Three recent studies have re-visited the hypothesis that high divergence between hybridizing species drives whole-genome duplication, using molecular methods to assess divergence between the progenitor species of natural polyploid species, assuming as Darlington (1937) did that genetic differentiation will correlate with structural differentiation of chromosomes: Chapman & Burke (2007), Buggs et al. (2008) and Paun et al. (2009). Chapman & Burke (2007) provided the first study that directly compares the genetic distance between the parental species of homoploid and polyploid hybrid species. They calculated Kimura’s 2-parameter (K2P) genetic distance between DNA sequences from the internal transcribed spacer region of nuclear ribosomal RNA genes of 12 species pairs that have given rise to homoploid hybrid species and 26 species pairs that have given rise to allopolyploid species. They compared all hybrid vs. allopolyploid parental pairs and found a significantly larger divergence between the parents of allopolyploids. They concluded that ‘the extent of evolutionary divergence between hybridizing taxa plays an important role in determining the outcome of hybrid speciation’ (p. 1778).
Buggs et al. (2008) tested the hypothesis that closely related parents are less likely to form a successful polyploid than more divergent parents. They examined molecular phylogenies of eight genera that contain polyploids, using node-based and clade-based methods of calculating the phylogenetic distance between parental pairs. They compared these with expected divergences based on the null hypothesis that hybridization would occur successfully at random between all species of a genus. They found that the phylogenetic divergence between parents of polyploids was not significantly different from the divergence expected under the null hypothesis. The same analysis on homoploid hybrids in the same genera found a lower divergence between the parents of homoploid hybrids than the null expectation, even when unstable hybrids were included. They concluded that ‘contrasting patterns of divergence between the parents of polyploids and homoploid hybrids are…determined by the restriction of homoploid hybrid formation to low parental divergence, rather than the restriction of polyploid formation to high parental divergence’ (p. 87).
Paun et al. (2009) conducted an additional analysis that combined and improved some of the approaches of Chapman & Burke (2007) and Buggs et al. (2008). For 16 homoploid hybrids and 32 allopolyploids, they calculated uncorrected p-distances and K2P distances between parental pairs using nuclear and/or chloroplast sequences. They converted each of these distances to a genetic divergence index (GDI) by dividing parental divergence by the average genetic distance between all pairs in each genus based on the same molecular markers. The GDI gave very similar results for both distance measures, and parents of polyploids were found to be significantly more divergent than parents of hybrids (Fig. 1). Fitting a heuristic model to their data, Paun et al. (2009) suggested that at a GDI of around 0.75, there is an equal probability of a hybrid being homoploid or allopolyploid, but above this point, allopolyploidy is more likely, and below this, homoploidy is more likely. They concluded that ‘parental divergence drives ploidy’.
Although Paun et al. (2009) calculated the average divergence between all species pairs in each genus, they did not use this as a null hypothesis for the expected divergence between parents of allopolyploids as in Buggs et al. (2008). If we carry out a two-tailed paired t-test on the genetic distances between parental pairs and the average genetic distance between all species pairs in their respective genera, using the data from Table S1 of Paun et al. (2009), we find a significant difference between these values for homoploid hybrids (t =3.427, d.f. = 15, P <0.01), but no significant difference for allopolyploids (t =1.533, d.f. = 31, P >0.1). A Wilcoxon matched-pairs signed-ranks test gives similar P-values. Significantly, this result agrees with that of Buggs et al. (2008): homoploid hybrid formation occurs at low parental divergence, but polyploid formation fits a model of random hybridization.
Null models and sampling issues
The three studies summarized above therefore provide the same general pattern of results, despite differences in methodology and sampling. They allow us to predict that wide hybridization has a higher probability of producing a successful species if followed by genome doubling. The authors of the three studies disagree over whether or not the results constitute good evidence for high parental divergence driving ploidy. This difference is partly due to the use of different null hypotheses. By basing their conclusions on comparisons between homoploid hybrids and allopolyploids, Chapman & Burke (2007) and Paun et al. (2009) seem to view the parental divergence of homoploid hybrids as a null hypothesis of the distribution expected of allopolyploids without the action of drive. This is problematic for two reasons. First, homoploid hybrids are likely to be restricted in the parental divergences under which they can form. Second, polyploid and hybrid species may not be comparable: while polyploid species typically have postzygotic reproductive isolation from their parents, homoploid hybrids are likely to be more interfertile with their parents than the parents are with each other. Most homoploid hybrid species, such as Helianthus anomalus, Helianthus deserticola and Helianthus paradoxus (Rieseberg 2003), are introgressed ecological forms that survive in habitats unused by the parental species and are not clearly able to coexist with their parents due to a lack of postzygotic isolation (Rieseberg 1997).
Buggs et al. (2008) view random hybridization as a better null model than homoploid hybrid species formation as the former assumes that parental divergence has no a priori influence on the probability of allopolyploidization. This random hybrid formation model is likely to be overly simplistic, as acknowledged by Buggs et al. (2008), but fits the data well. Sang et al. (2004) suggested a model for the origin of tetraploids as a function of genomic divergence between diploid progenitors, but as they suggest, this needs to be made more mathematically rigorous, with better natural population estimates of the rates of hybridization, unreduced gamete formation, and establishment of the resulting lineages. It seems, for example, that wide hybridization may increase rates of unreduced gamete formation (see above) but, as Ramsey & Schemske (1998) argue, this rise in mutation rate is likely to be at least cancelled out by low hybridization rates and may therefore not increase the frequency of allopolyploids at high parental divergences.
Our ability to draw firm conclusions is also restricted by sampling limitations. Chapman & Burke (2007), Buggs et al. (2008) and Paun et al. (2009) agree that improved sampling is needed of homoploid hybrids, which are very difficult to detect and may in fact be quite rare in nature. We suggest that the lack of allopolyploids between closely related species (Fig. 1) may also be due to sampling bias as these may also be hard to detect (Rieseberg & Willis 2007); for example, a recently discovered allotetraploid formed by interspecific hybridization between Mimulus nasutus and Mimulus guttatus was identified as the former species until it gave anomalous results in a crossing experiment (Sweigart et al. 2008). There is an urgent need for thorough molecular analysis of many additional hybridizing plant groups, particularly those genera containing hybrids as well as polyploids, such as the genera Crepis, Clarkia, Betula and Gilia.
The authors of the three studies disagree about the relevance of autopolyploids (defined here as polyploids formed within a species) to the issue of parental divergence and whole-genome duplication. We have long recognized that there is a continuum from true autopolyploid to allopolyploid, with ‘hybrid autopolyploid’ and ‘segmental allopolyploid’ as intermediate points in this continuum (Stebbins 1950). In a restriction that appears to stem from the use of homoploid hybrids as a null model, Paun et al. (2009) argue that autopolyploids should not be included in the analysis as they do not directly correspond to hybrid speciation processes at the diploid level. In contrast, Buggs et al. (2008) argue that the hypothesis that parental divergence drives polyploidy cannot be tested fully without including divergences at or close to zero (i.e. autopolyploids).
These issues regarding the comparability of autopolyploids, allopolyploids and homoploid hybrids are part of the broader problem of defining species. Classing two groups as ‘species’ seems to imply that they are comparable evolutionary units, but due to the use of different species concepts and types of information, the classification of certain groups as species is not standardized and somewhat arbitrary. Autopolyploids seem to occur very frequently in nature (e.g. Soltis & Soltis 1993; Ramsey & Schemske 1998) but are rarely classified as separate species (Soltis et al. 2007), despite sometimes strong reproductive isolation from their parents. Reliance upon named species as a unit of comparison in surveys therefore introduces a strong bias against successful autopolyploidization events, whose frequent occurrence in nature certainly contradicts the idea that whole-genome duplication is less likely at low parental divergence. Likewise, with homoploid hybrids, the division between a recurrent unstable hybrid and a hybrid species is not straightforward.
If in our sampling of polyploids we were able to count all polyploidization events that have led to an established population (whether classified by taxonomists as a species or not), it is possible that the number of events would be highest for closely related parents and actually decline with parental divergence. This distribution would directly contradict the idea that parental divergence drives polyploidy, and would also contrast with the random pattern of allopolyploid species formation in relation to parental divergence noted by Buggs et al. (2008) in their survey and shown here in that of Paun et al. (2009).
Paun et al. (2009) make the interesting suggestion that inclusion of autopolyploids would cause a bimodal distribution of polyploid frequency in relation to parental divergence, indicating the presence of different phenomena. Whilst such a distribution could be an artefact of the lack of detection of polyploids of intermediate parental divergence (i.e. allopolyploids with closely related parents; see above), it might also be explained by selection. For an unoccupied fitness peak that is close to an occupied peak to be filled by a new variant, that variant must be reproductively isolated. Polyploidy typically confers reproductive isolation from its parents at both low and high parental divergences; in contrast, a homoploid hybrid is likely to occur and be reproductively isolated from its parents only at intermediate parental divergences (i.e. low enough for the hybrids to be viable but high enough for the hybrid to be isolated from both parents). Because homoploid hybrids often come into existence before an allopolyploid forms (following the Class 2 mode of allopolyploidization; Harlan & De Wet 1975), there will not be strong selection for polyploidy at intermediate divergences, unless the new polyploid would occupy a different adaptive peak from that of the extant homoploid hybrid.
In our view, the idea that parental divergence drives polyploidy is based on two factors. The first is an over-emphasis on potential bias in mutational mechanisms (e.g. unreduced gamete formation) without sufficient consideration of subsequent selection on the newly formed hybrid or polyploid. The second is misinterpretation of survey data due to problems in sampling (particularly the omission of autopolyploids) and null model selection. The three recent studies reviewed here of the relationship between parental divergence and hybrid speciation (Chapman & Burke 2007; Buggs et al. 2008; Paun et al. 2009) do not provide convincing evidence that polyploid species are less likely to form successfully at lower parental divergences and therefore do not demonstrate that parental divergence drives ecologically successful whole-genome duplication. Instead, they simply allow us to predict that wide hybridization has a higher probability of producing a successful species if followed by polyploidization.
The authors thank two anonymous referees for helpful comments on an earlier version of this article. This work was supported in part by NSF grant DEB-0614421 and the University of Florida.