On the relative abundance of autopolyploids and allopolyploids

The prevalence of autopolyploids in angiosperms has long been a subject of debate. Müntzing (1936) and Darlington (1937) concluded that autopolyploids were common and important evolutionary entities. However, Clausen et al. (1945) and Stebbins (1947) subsequently considered them rare, in part because the criteria upon which interpretations of autopolyploidy were rendered were not rigorous. This position was reiterated by Grant (1981) decades later, although evidence was mounting that autopolyploid taxa might be important in natural populations (Lewis, 1980). As cytological and genetic data have accumulated, it has become increasingly apparent that the latter view is likely to be correct (Soltis et al., 2004b, 2007, 2010). However, it still appears that the majority of polyploids are allopolyploids (Parisod et al., 2010; Soltis et al., 2010), even though Ramsey & Schemske (1998, p. 467) conclude that ‘the rate of autopolyploid formation may often be higher than the rate of allopolyploid formation.’

In this letter we survey the literature to assess whether allopolyploids are indeed the prevailing cytotype in nature. Using our new estimates for the incidence of autopolyploidy and allopolyploidy, we discuss some of the evolutionary dynamics that may be driving their frequencies in nature. Finally, we suggest avenues for future research on polyploidy that build on our results and other recent progress in the field.

Prevalence of autopolyploids and allopolyploids within plant genera

Polyploidy has enjoyed a renaissance over the past decade (Soltis et al., 2014b). Advances in plant genomics have revealed a history of polyploidy throughout plant evolution (Cui et al., 2006; Barker et al., 2008, 2009; Soltis et al., 2009; Shi et al., 2010; Jiao et al., 2011, 2012, 2014; Arrigo & Barker, 2012; Vekemans et al., 2012; Kagale et al., 2014; Cannon et al., 2015; Edger et al., 2015). Among contemporary species, the wealth of plant phylogenetic and cytological data has been analyzed with new tools (Mayrose et al., 2010, 2015) to estimate the frequency, incidence, and diversification of polyploid species (Otto & Whitton, 2000; Meyers & Levin, 2006; Wood et al., 2009; Mayrose et al., 2011, 2015; Scarpino et al., 2014; Soltis et al., 2014a). Despite recent progress, these analyses have not addressed the relative contributions of autopolylpoidy and allopolyploidy to plant evolution. Although past researchers concluded that autopolyploid speciation is likely rare (Clausen et al., 1945; Stebbins, 1947), the current consensus is that autopolyploidy is probably a more common and important element of plant diversity than historic views suggest. Autopolyploid prevalence is likely underestimated, perhaps greatly, because taxonomists often do not recognize autopolyploids as species (Soltis et al., 2007). These taxonomically cryptic autopolyploid cytotypes may comprise a substantial fraction of plant diversity. A more precise census of the frequency and importance of autopolyploid speciation relative to allopolyploidy remains unavailable.

We surveyed the plant systematic literature to evaluate the prevalence of autopolyploids and allopolyploids within plant genera. Phylogenetic, cytological, and genetic data were integrated to develop an empirical estimate of autopolyploid and allopolyploid taxa. We analyzed data from 47 vascular plant genera representing 4003 species (Supporting Information Table S1, Methods S1). These data were assembled by an initial survey of the literature for phylogenies of genera with at least 30% species represented. Following this survey, we cross-referenced those taxa with chromosome counts (Wood et al., 2009; Rice et al., 2015) and additional literature on the nature of polyploids in the genus (e.g. allozyme, microsatellite, or amplified fragment length polymorphism (AFLP) studies). The combination of these data sets yielded 43 genera with > 30% of taxa represented and four genera with > 50 species with available data. Information collected from nearly 300 systematic publications (Methods S1) was used to assign ploidy level (diploid vs polyploid) and nature (autopolyploid vs allopolyploid). We inferred polyploids as taxa with a multiple of the base count for each genus as in Wood et al. (2009). Polyploids were also frequently identified in the literature by other methods, such as isozyme or microsatellite evidence, and these were included in our data set (Table S1). Note that our analyses do not estimate or account for multiple origins of a polyploid species. Thus, we estimated the incidence of polyploid taxa rather than the number of origins of polyploids in these genera. We also did not assess whether unnamed polyploids meet the criteria of various species concepts. Given that polyploidy itself is a substantial reproductive barrier, most polyploids would fulfill the requirements of at least the biological species concept. Finally, we divided polyploids into autopolyploids and allopolyploids based on reports by researchers in our survey. Although a continuum of parental divergences yields a corresponding distribution of polyploid natures from true autopolyploids, ‘hybrid autopolyploids’, ‘segmental allopolyploids’, and allopolyploids (Stebbins, 1950), most literature reports do not distinguish these gradations.

Across 47 vascular plant genera, we found that 76% of plant species were diploids and 24% were polyploids (Fig. 1a). The incidence of diploids and polyploids is similar to our previous estimates (Wood et al., 2009; Mayrose et al., 2011) with a largely different data set. Our survey confirmed recent views (Soltis et al., 2007; Parisod et al., 2010) that autopolyploids are more prevalent than indicated by taxonomy alone. Across all plant species in the survey, 13% were inferred as autopolyploids whereas 11% were allopolyploids. This near parity of autopolyploids and allopolyploids contrasts with a long history of expectation that allopolyploids are predominant (Clausen et al., 1945; Stebbins, 1947; Grant, 1981), but is consistent with recent views that many autopolyploids may have been overlooked (Soltis et al., 2007; Parisod et al., 2010; Husband et al., 2013). Notably, our estimate that autopolyploids, named and unnamed, represent c. 13% of plant taxa is very similar to an unpublished estimate from the Flora of California reported in Soltis et al. (2007). In that estimate, J. Ramsey and B. C. Husband (unpublished) reported that 334 of 2647 species contained unnamed cytotypes (Husband et al., 2013). Assuming that each of these represent a single autopolyploid taxon, 11.2% of the taxa in the California Flora data would be autopolyploids. Given that these estimates are based on different data and taxa, their consistency suggests that these values of autopolyploid incidence may be robust.

Differences in taxonomic practice appear to be the primary source of the perception that allopolyploids are predominant (Fig. 1b). More than 87% of the allopolyploids were recognized as named species, whereas only c. 12% of the autopolyploids were named. Much of this difference is likely attributable to the fact that allopolyploids often have more distinct morphological features relative to their parental taxa than autopolyploids (Stebbins, 1947). Despite the fact that many autopolyploids fulfill the requirements of different species concepts (Soltis et al., 2007), they are frequently relegated as cytotypes. Our survey finds that autopolyploids comprise a substantial fraction of plant diversity. Following the recommendation of Soltis et al. (2007), autopolyploids that meet the criteria of species should be named to provide an accurate accounting of plant species and further elucidate the processes of plant evolution.

The proportion of taxa that are autopolyploid or allopolyploid constitutes a nearly continuous distribution across our sampled genera (Fig. 1c). Although there were genera whose polyploids were either all autopolyploid or allopolyploid, most genera in our survey contained a mixture of autopolyploids and allopolyploids. On average, we found that 50% of the polyploids were autos and 50% were allos. The mean autopolyploid frequency within genera had a 95% confidence interval of 43% to 56%, and a standard deviation of 21%. Given that allopolyploidy requires opportunities for hybridization, and genera vary widely in their degree of interspecific hybridization (Whitney et al., 2010), the differential production and success of these ploidal types likely varies widely across the phylogeny. Whether differences in the proportions of ploidal types reflects differences in the rate of production, persistence, or diversification is not clear from our present study, but is an important avenue of future research.

How many unnamed polyploid flowering plant species exist in nature? Current estimates suggest there are c. 350 000 named flowering plant species with another 10–20% to be named in general (Joppa et al., 2011). Our survey found that 53.2% of polyploid taxa were unnamed and that c. 24% of all plant taxa were polyploids (Fig. 1b; Table 1). Assuming these numbers are representative for angiosperms, we estimate that there may be an additional c. 51 000–61 000 cryptic polyploid species. Given that our estimate of unnamed polyploids relied upon some recognition of these taxa in the literature, there may be more cryptic polyploid species of an unknown magnitude. Our results suggest that the majority of these cryptic polyploids are likely autopolyploids that will probably be challenging to recognize by morphological traits alone, but may fulfill the requirements of other species concepts.

Notably, naming of polyploid taxa does not vary with prevalence of the ploidal types (Table S1). Thus, even in genera where all or most of the polyploids are autos, these taxa are still often unnamed. Although many of these autopolyploid cytotypes may be relatively restricted geographically or ecologically, the lack of taxonomic recognition confounds research on the evolution of plant diversity. As biogeographic and genomic data sets are compiled and analyzed (Goff et al., 2011; Boyle et al., 2013; Matasci et al., 2014; Violle et al., 2014; Wickett et al., 2014; Rice et al., 2015), it will be important for individuals of different ploidal levels to be properly recognized. For some species, it is likely that their biogeographic or other data represents a mixture of cytotypes, often unknown to users of these databases. Genome duplication is associated with physiological shifts (Chao et al., 2013) that could distort biogeographic, functional genomic, and ecological analyses. Applying taxonomic names to unnamed cytotypes that meet the criteria of species would improve current databases.

The differential formation of autopolyploids and allopolyploids in space

Although autopolyploids and allopolyploids were near parity in our surveyed genera, the available evidence suggests a large disparity in formation rates. Ramsey & Schemske (1998) estimated the rates of polyploid formation per generation on the basis of unreduced gamete production in a range of species. When unreduced gametes come from fertile diploids, the rate of autotetraploid formation is 2.16 × 10⁻⁵ vs h (1.22 × 10⁻³) for allotetraploids, where h refers to the hybridization rate, assuming the plants are outcrossing. They (Ramsey & Schemske, 1998) estimated that a hybridization frequency of 0.0272 or 2.7% would be required for the rates of autotetraploid and allotetraploid formation to be equal. This low level of hybridization is possible for cytotype parity because the rate of unreduced gametes in diploid hybrids is almost 50 times greater than that in nonhybrids (27.5% vs 0.6%). In the case of self-fertilization, the rate of autopolyploid formation is 7.14 × 10⁻⁵ vs h (4.05 × 10⁻²) for allopolyploids, where h is the rate of hybridization. A hybridization rate of 0.17% would yield equal rates of autopolyploid and allopolyploid formation.

Is hybridization frequent enough in nature to yield equal rates of autopolyploid and allopolyploid formation? Very little is known about hybridization levels across the ranges of wild species (Ramsey & Schemske, 1998). If their ranges were only partially sympatric and/or local conditions allowing contact were rarely met, then most populations of species A would not be mixed with that of species B, and vice versa. For example, if only 1% of these species' populations were in contact and the hybridization level within was 2.7%, then the rate of autopolyploid formation would be c. 100 times greater than that of allopolyploid formation. If the hybridization level were 1.35%, then the rate of autopolyploid formation would be c. 200 times greater. Of course, the level of hybridization across species' ranges may be much < 1%. Accordingly, it seems that the rate of autopolyploid formation could be much greater than allopolyploid formation.

It is tempting to conclude that if two species were in contact over a good portion of their ranges, then cytotype parity may indeed be achieved. However, other factors, such as mating system, likely play a large role in the differential production of autopolyploidy and allopolyploidy. For example, as selfing rates increase, the incidence of hybridization (or outcrossing of any type) must go down. Accordingly, a transpecific value of 0.17% hybridization may not often be achieved, especially if species are only partially sympatric and have somewhat different ecological preferences. For autopolyploids, models indicate that higher selfing is a boon to their establishment and persistence (Rausch & Morgan, 2005). Modest spatial differentiation and limited dispersal, combined with increased rates of self-fertilization, will further enhance autopolyploid persistence (Baack, 2005). Thus it seems very likely that autopolyploid formation will usually far exceed allopolyploid formation with selfing.

The discussion thus far has assumed that hybridization occurs uniformly across the plant phylogeny. Although common in some genera, hybridization is apparently absent or rare in others. Ellstrand et al. (1996) reviewed two regional biosystematic floras in Europe, another two regional floras in North America, and the Hawaiian islands for reports of interspecific hybrids. The percentages of genera with hybrids ranged from a high of 16% (British Isles) to a low of 6% (Hawaii). Whitney et al. (2010) expanded this study to include two additional biosystematic floras in the United States and one in Australia. Globally, 16% of genera had interspecific hybrids, whereas 84% did not. These results suggest that for many genera coexisting congeners often are unable to foster viable hybrids or that sympatric species that have the potential to hybridize typically have divergent niche requirements which prevents their co-occurrence. Most relevant for our discussion, the absence or rarity of hybrids in the vast majority of the 3200 genera surveyed by Whitney et al. (2010) indicates that the potential for autopolyploid production far exceeds that of allopolyploid production.

The differential formation of autopolyploids and allopolyploids in time

The relative genesis of autopolyploids and allopolyploids would be an unchanging variable, if the fertility relationship of species A and B were constant in time. However, this is not likely to be the case. The closer we get to the time of the most recent common ancestor, the higher hybrid fertility would be. Hybrid fertility declines over time because sterility is a by-product of stochastic, nonadaptive genetic and chromosomal changes (Levin, 2000; Coyne et al., 2004; Lowry et al., 2008; Fierst & Hansen, 2010; Matute et al., 2010; Moyle & Nakazato, 2010; Presgraves, 2010; Sherman et al., 2014; Larcombe et al., 2015). This is important because the higher the fertility of hybrids the less prone they are to producing unreduced gametes, and thus polyploid progeny (Ramsey & Schemske, 1998). There has been somewhat conflicting evidence for this contention in the literature. Consistent with this expectation, two recent studies (Chapman & Burke, 2007; Paun et al., 2009) found evidence that allopolyploids are more likely to be produced from divergent diploids, whereas homoploid hybrid species are more likely to result from crosses between closely related species. However, Buggs et al. (2008) found that allopolyploids are actually produced from a random draw of diploid divergences within genera, and that only homoploid hybrids are restricted to crosses of low divergence. They (Buggs et al., 2009, 2011) came to a similar conclusion after re-assessing the Chapman & Burke (2007) and Paun et al. (2009) analyses. However, for the purposes of our discussion it remains unclear if the success rate of allopolyploid production varies with parental divergence even if allopolyploids may be formed from any congeneric species cross. Given that hybrids between diploids of varying divergence will have different levels of incompatibilities to resolve and potential differences in genetic novelty, allopolyploids from different parental combinations likely face distinct challenges to establishment and persistence. At the very least, allopolyploidy requires speciation among diploids and subsequent hybridization (Levin, 2013). This restricts allopolyploid speciation to a narrower window of time than autopolyploids, which may be produced at any time.

The arguments made in this letter assume that allopolyploids are produced via the formation of unreduced gametes in diploid hybrids. However, autopolyploids and allopolyploids also may be formed via a triploid bridge, a process which involves the production of unreduced gametes in two successive generations (Harlan & DeWet, 1975; Ramsey & Schemske, 1998). Allopolyploid production via a triploid bridge still requires interspecific hybridization, a process that is not ubiquitous across the ranges of species or in time, as argued earlier. Accordingly, far fewer allopolyploids are expected to be formed across the ranges of two species than are autopolyploids. According to Ramsey & Schemske (1998), the triploid bridge accounts for c. 1% of allopolyploid production in selfing species and 30% in outcrossers.

Do allopolyploids have an evolutionary advantage?

Although we found that autopolyploids and allopolyploids occurred near parity in nature, allopolyploids likely have a significant evolutionary advantage. Autopolyploid plants were probably produced in higher numbers than allopolyploid plants throughout contemporary species' ranges and over the course of deep time. The differential might have been five-fold or 20-fold. Current data and approaches do not provide any further insight into the production advantage of autopolyploids. If there were a substantial differential, autopolyploid lineages should be very much more frequent than allopolyploid lineages, all else being equal.

An evolutionary advantage may accrue to allopolyploids from higher rates of population establishment and persistence. This may occur because genome doubling, especially when two divergent genomes are present, provides an immediate vehicle for transcending parental niches. Greater niche differentiation between progenitor and neopolyploid increases the proximity and opportunity to mate with other neopolyploids, a key factor in polyploid establishment (Fowler & Levin, 1984; Rodriguez, 1996; Levin, 2002; Oswald & Nuismer, 2011). In addition, niche divergence may lead to less competition with progenitors or to a competitive advantage, and correlatively higher levels of population growth, which is essential if new populations are to survive in the face of demographic and environmental stochasticity. The ability to exploit novel habitats is manifested, for example, in the relative abundance of allopolyploids in recently deglaciated areas (Brochmann et al., 2004). Polyploids with two or more genomes seem especially adept at colonizing habitats which are beyond the tolerances of their progenitors (te Beest et al., 2012). An establishment advantage might also accrue for allopolyploids when neoautoploids have lower fertilities than neoalloploids (Ramsey & Schemske, 2002).

Allopolyploids may also endure environmental change better than autopolyploids, because the former may marshal more genetic resources through genome reorganization and greater flexibility in gene expression to cope with immediate stresses (Doyle et al., 2008; Leitch & Leitch, 2008; Jackson & Chen, 2010; Parisod et al., 2010). Allopolyploid lineages may also have a survival advantage because the union of the same two species in different populations may yield products with rather divergent ecological tolerances (Paun et al., 2011). Accordingly, if one nascent allopolyploid population cannot cope with a particular type of environmental change, others may survive, and thus the combination will persist. The multiple origins of allopolyploid species in Aegilops (Meimberg et al., 2009), Asplenium (Werth et al., 1985), Mimulus (Vallejo-Marín et al., 2015), Senecio (Abbott et al., 2007) and Tragopogon (Soltis et al., 2004a), for example, bode well for their longevity. [Correction added after online publication 6 October 2015: in the preceding sentence text ‘Galax (Servick et al., 2015)’ has been deleted.]

Whereas we cannot directly measure autopolyploid failure relative to allopolyploids, we can consider a contemporary variable which is associated with taxon vulnerability, namely geographical range size. Taxa most vulnerable to extinction have narrow geographical distributions (Gaston & Fuller, 2009; Birand et al., 2012), and relatively small numbers of individuals across populations (Dynesius & Jansson, 2014). Small population sizes and numbers are associated with increased sensitivity to environmental and demographic stochasticity, heightened inbreeding, and reduced genetic diversity (Mayr, 1963; Anacker & Strauss, 2014). Taxon vulnerability is also associated with narrow niche breadths (Morin & Lechowicz, 2013; Slatyer et al., 2013). Are autopolyploid taxa more prone to extirpation based on the aforementioned criteria, and are they apt to have a shorter duration time than allopolyploid taxa? There are relatively few data on the subject and this will be an important direction for future research on the evolutionary success of polyploids.

Given that allopolyploids may have an advantage in their production and generation of new species, we may briefly ponder the magnitude of this difference. Let us assume that each of two diverging, diploid lineages lives for 20 million years, and that the window for allopolyploidy is 5 million years (Levin, 2012, 2013). Let us also assume that over the course of time 10 times as many autopolyploid populations are produced as allopolyploid populations per million years. Since there are two diverging lineages in which autopolyploids may be formed over a time-frame four times that of allopolyploids, we would expect autopolyploid production to be 80 times greater than that of allopolyploids. Even if autopolyploid genesis per million years was only five times greater, the production of this cytotype would be 40 times greater. Yet, on average, there is parity in cytotype representation. To achieve parity, there would have to be a 40-fold allopolyploid advantage in lineage survival and diversification. Although this ‘fitness’ disparity may be quite surprising, we believe that it is quite conservative. The production differential of autopolyploid to allopolyploid in a pair of diverging lineages may well exceed 50:1, which would mean that the allopolyploid advantage could easily exceed 100-fold. Better models and data are needed to assess how these rates vary across the phylogeny and contribute to the success of autopolyploids and allopolyploids.

Conclusions

We have shown that, on average, autopolyploid and allopolyploid taxa are at approximate parity in terms of their numbers. We now have a mechanistic explanation why autopolyploids might be more common than we appreciate, as argued by Soltis et al. (2007, 2010). Using a holistic approach, we have shown how autopolyploid populations may have been formed in far greater numbers than allopolyploid populations, and thus may have had far greater opportunities to ultimately form novel, long-lived evolutionary entities. This would be true even if the hybridization rate in outcrossers exceeded 2.7% across contemporary species ranges because species A and B did not always have the level of sympatry they do today (assuming allopatric or parapatric speciation), and their hybrids probably produced fewer unreduced gametes as we move towards the time of lineage splitting.

Logic is not a substitute for data; and substantial data are lacking for many aspects of polyploid evolution. Although our estimate of autopolyploid and allopolyploid frequencies appears to be consistent with contemporary expectations and other previous estimates of polyploidy (Soltis et al., 2007; Wood et al., 2009; Mayrose et al., 2011), it reflects the biases of the studies available in the literature. A more inclusive survey with more tropical and woody taxa represented would be useful. Thus, it is important that cytological and genome size data continue to be collected as systematists study the phylogenetic relationships of new groups. We agree with a recent call for a global census of angiosperm genome sizes (Galbraith et al., 2011) to address many gaps in the study of plant genome evolution. There are also many aspects of unreduced gamete production that are in need of further research (De Storme & Mason, 2014). A comprehensive assessment of the number of unreduced gametes produced by environmental vs internal conditions is needed. It would be good to establish hybridization rates throughout areas of diploid sympatry, and to determine the proportions of species' ranges where contact is made with a cross-compatible congener. New insight into unreduced gamete production may be gained by extending to allopolyploids a recent model (Suda & Herben, 2013) that reasonably estimated proportions of unreduced gametes for autopolyploids. It would also be fruitful to extend recent work on meiosis in autopolyploids (Hollister et al., 2012; Wright et al., 2015) to allopolyploids to assess if potential differences in selection on the meiotic machinery differentially influence the establishment of autopolyploids and allopolyploids. Research on the genetics of unreduced gamete production (De Storme & Geelen, 2011, 2013a,b) show much promise for advancements in this area. Mining molecular phylogenies for insights on the relative persistence of autopolyploid and allopolyploid lineages and their penchant for branching would also be most useful. Hopefully, this letter will spur efforts to investigate the process and product of polyploidy from new vantage points and with renewed vigor.

Author contributions

M.S.B., N.A. and D.A.L. planned and designed the research. M.S.B., N.A. A.E.B. and Z.L. collected data and conducted analyses. M.S.B. N.A., and D.A.L. wrote the manuscript.