The long wait for hybrid sterility in flowering plants


Related allopatric or parapatric species may interbreed when habitat disturbance, long-distance dispersal, and environment-mediated range shifts allow species contact. Correlatively, ecologically isolated sympatric species may hybridize following environmental/habitat alteration. If hybrids are fertile, such contact could lead to extensive introgression or to the production of homoploid hybrid derivatives (Hewitt, 2004; Arnold, 2006; Jaramillo-Correa et al., 2009; Soltis & Soltis, 2009). Even if hybrids are largely sterile, cross-breeding species may still generate allopolyploids. The lower the degree of fertility, the greater is the rate of unreduced gamete formation (Ramsey & Schemske, 1998).

The window of opportunity for hybridization is finite. As lineages diverge, the potential for gene exchange or genetic union declines, because genomes become increasingly incompatible, that is, they accrue Dobzhansky–Muller (DM) incompatibilities (Coyne & Orr, 2004). As a consequence, the fertility of hybrids in plants typically declines as we go from intrataxon crosses, to crosses between races or subspecies, to crosses between closely related species, and to crosses between more distantly related ones (Levin, 1978; Grant, 1981; Rieseberg et al., 2006). Correlatively, the ability to make fertile hybrids also declines with increasing genetic distance in plants (Moyle et al., 2004; Scopece et al., 2007, 2008). What is causal in this relationship remains to be determined.

Allozyme distance has been widely used as a surrogate for evolutionary time, as have DNA-based measures (Edmands, 2002; Coyne & Orr, 2004), but they are not substitutes for real time. The waiting time for hybrid sterility is a little-discussed metric in the plant literature, because a method for its measurement has not been forthcoming. We would like to know the approximate number of yr that two lineages can potentially influence each other's gene pool or form allopolyploids. Moreover, we would like to know how long it takes to achieve a major milestone of allopatric speciation. Whereas we appreciate that ecological speciation may be relatively rapid, how much slower is the tempo of allopatric speciation?

I propose that insights into this waiting time can be gained by relating the fertility of first-generation hybrids to estimates of the divergence times of species. The species considered will be the same ploidal level. Before applying this approach, it is important to recognize some of its limitations. First, data on hybrid fertility are available for few species employed in molecular phylogenetic studies. The small number of genes used in estimating divergence time may not be representative of the genome as a whole, in some instances residing in the nucleus and others in the cytoplasm, and may deviate from molecular rate constancy across lineages (Gaut et al., 2011). If species have exchanged genes, divergence time estimates may be substantially smaller than the actual time, especially if gene flow has been recent; and gene exchange may occur even if hybrid fertility is low (Strasburg et al., 2009; Sambatti et al., 2012). If the fertility relationships of two species vary among their populations (Levin, 1978; Grant, 1981), and if there are barriers to gene flow among conspecific populations (Scopece et al., 2010), then the few populations chosen for hybridization and divergence time studies may not be representative of the species as a whole. To the extent that hybrid fertility decline is the result of chromosomal rearrangement, lineages may differ in their penchant for rearrangement, which may occur relatively quickly (Levin, 2002). Finally, the approach used in this paper is restricted by the considerable error terms in phylogenetic estimates of divergence times (Ho & Phillips, 2009; Schwartz & Mueller, 2010).

In spite of the imperfection of relating hybrid fertility to divergence time estimates, we can gain a rough approximation of the waiting times for partial hybrid sterility in several plant lineages, recognizing that time has probably been underestimated, perhaps substantially. The greater the amount of interspecific gene exchange that has taken place, the more the time to sterility will be underestimated. The approximate waiting time has value because it provides temporal insight into an important evolutionary process associated with speciation; and we did not have this before. As noted by Rieseberg & Willis (2007), ‘We know surprisingly little about the speed of plant speciation.’ The approach taken here complements mathematical models that focus on how long it takes for lineages to develop postzygotic isolation as a result of DM incompatibilities. This time varies over orders of magnitude depending on model parameters (Gavrilets, 2000).

I will show that, even erring on the conservative side, the waiting time for partial hybrid sterility is considerable. The onset of substantial fertility loss seems to be fairly consistent across several lineages, thus providing a measure of confidence to the conclusions. The genera discussed in the following were all I could find after an exhaustive literature search. No genus or species combination was omitted because the time of fertility decline was inconsistent with others.

Waiting for hybrid sterility

One million years of divergence seems insufficient for a significant amount of hybrid sterility. Hybrids are largely fertile in Hawaiian Tetramolopium, which began to radiate ecologically < 1 million yr ago (mya; Lowrey et al., 1995). Argyranthemum species in the Canary Islands had a common ancestor roughly 1.5 mya (Francisco-Ortega et al., 1996); and hybrids between Argyranthemum broussonetti and Argyranthemum frutescens are largely fertile (Brochmann et al., 2000). Silene dioeca and Silene latifolia diverged c. 1 mya (Frajman et al., 2009); and they produce fertile hybrids (Kruckeberg, 1963). Hawaiian Bidens is < 3 million yr old (Knope et al., 2012), and its species are interfertile (Ganders et al., 1984). Fertile hybrids are produced by Cyrtandra grandiflora and Cyrtandra sandwicensis (Smith et al., 1996), which diverged c. 3.5 mya (Clark et al., 2009). Ipomopsis aggregata and Ipomopsis tenuituba split c. 5 mya (Porter et al., 2010) and their hybrids are fertile (Alarcón & Campbell, 2000). Finally, Senecio vernalis diverged from Senecio leucanthemifolius and Senecio squalidus between 2.4 and 4.8 mya (Comes & Abbott, 2001), but produces highly fertile hybrids with both species (Crisp, 1972; Alexander, 1979)

A pronounced reduction in hybrid fertility tends to appear after lineages have been separate for > 4 million yr. Reduced fertility is evident in hybrids between Aquilegia flabellata and Aquilegia viridiflora (45%), and between Aquilegia ecalcarata and Aquilegia sibirica (64%, Taylor, 1967), both of which diverged c. 5 mya (Bastida et al., 2010). Divergence times of approximately this magnitude (Xie et al., 2009) are associated with complete or nearly complete sterility in hybrids between Circaea lutetiana and Circaea alpina, and between Circaea cordata and Circaea erubescens (Boufford, 1990). Similarly, hybrids between Arabidopsis thaliana and diploid Arabidopsis arenosa, which split 5–6 mya (Koch et al., 2000), are largely sterile (Bomblies & Weigel, 2010). Fertile hybrids are obtained from crosses between Silene douglasii and Silene virginica (Kruckeberg, 1963), which split c. 3 mya (Frajman et al., 2009), but hybrids between Silene and Lychnis, which separated > 7 mya (Frajman et al., 2009), are sterile (Kruckeberg, 1962).

The Hawaiian silversword alliance provides valuable insights into the evolution of hybrid sterility. The maximum age of the lineage is only c. 5 million yr, cytogenetic studies have been conducted on interspecific hybrids of several different species combinations, and divergence times vary substantially among species pairs. The species with the most recent common ancestry have the highest fertility rates. For example, Dubautia ciliolata and Dubautia arborea split c. 1 mya (Baldwin & Sanderson, 1998), and have hybrids with 96% pollen fertility (Carr & Kyhos, 1986). Hybrids between D. ciliolata and Dubautia sherffiana, which diverged c. 2.5 mya, have 99% fertility. Proceeding to c. 4.5 mya since divergence, hybrids between D. sherffiana and Dubautia laevigata have 68% fertility. With roughly 5 mya since common ancestry, hybrids between Dubautia scabra and Dubautia paleata are 27% fertile, hybrids between Dubautia knudsenii and D. scabra are 39% fertile, and hybrids between Dubautia pauciflorula and Dubautia latifolia are 26% fertile.

Consider next the fertility relationships of Wilkesia and Dubautia species which split from a common ancestor c. 5 mya (Baldwin & Sanderson, 1998). Hybrids between various Dubautia species and Wilkesia gymnoxiphium have fertilities ranging from 28 to 44% (Carr & Kyhos, 1986). A greater fertility reduction among taxa that diverged c. 5 mya is realized only in hybrids between Argyroxiphium sandwicense and Dubautia menziesii, which are 11% fertile. Given that the hybrids are heterozygous for three translocations, this fertility value closely approximates the expected value of 12.5%, were there equal frequencies of alternate and adjacent anaphase I disjunction. DM incompatibilities seem not to be operative in the first-generation or in later-generation hybrids between these species. In general, fertility reduction in the silversword alliance is largely a result of a failure of normal chromosome pairing (Carr & Kyhos, 1986; Carr, 1995).

The genus Draba provides an exception to the slow pace of fertility decline (Grundt et al., 2006). Low degrees of genetic differentiation within and among Draba nivalis, Draba subcapitata, and Draba fladnizensis suggest that each species probably originated within the past 1 million yr. Yet progeny of crosses within regions among populations of Dfladnizensis were semisterile, and crosses among proximal D. nivalis populations were mostly sterile. Crosses among regions within the three species were largely sterile. The observed correlation between hybrid sterility and genetic distance implies that this postzygotic barrier accrues in a gradual (stepwise) manner. In D. nivalis, reduced interpopulation fertility is attributable in part to underdominant chromosomal rearrangements and in part to DM incompatibilities (Skrede et al., 2008). Genetic drift in these predominantly self-fertilizing species is the most likely explanation for the relatively rapid bloom of hybrid sterility (Grundt et al., 2006; Skrede et al., 2008). The establishment of underdominant mutations by genetic drift, as promoted by small effective population size, and by cycles of localized extinction and recolonization (Lande, 1985; Gavrilets, 2004), fosters the formation of cryptic species as seen in Draba and elsewhere (Grant, 1981).

As noted, hybrid sterility may have multiple causes. Genetic incompatibility contributes to the sterility of hybrids in many genera (e.g. Mimulus, Fishman & Willis, 2001; Solanum, Moyle & Nakazato, 2008; Oryza, Li et al., 2008). Indeed, an association between the levels of pairing anomalies and sterility in hybrids does not necessarily indicate that the anomalies are the prime cause of sterility (Lowry et al., 2008). Conversely, pairing anomalies do seem to explain sterility in the silversword alliance (Carr & Kyhos, 1986; Carr, 1995), and the recovery of fertility in polyploid derivatives of sterile hybrids indicates that DM incompatibilities were not responsible for those instances of hybrid sterility (Rieseberg & Willis, 2007). It remains to be determined whether a decline in hybrid fertility as a result of the accumulation of chromosomal rearrangements in the parental species proceeds at the same pace as a decline resulting from the emergence of DM incompatibilities. Although hardly enough to be conclusive, the information presented here does not suggest a major disparity.

In trees, partial fertility may persist for a very long time. The North American/European Platanus occidentalis and Platanus orientalis separated c. 50 mya (Feng et al., 2005), yet their hybrid is partially fertile (Panetsos et al., 1994). The North American/Asian Liriodendron tulipera and Liriodendron chinense are between 10 and16 million yr old, and form partially fertile hybrids (Parks & Wendel, 1990). Fertile hybrids are obtained between the North American disjuncts Acer rubrum and Acer saccharinum, which separated c. 4 mya (Santamour, 1965; Renner et al., 2008). The longer waiting time for sterility in trees than in herbs is not surprising, since the generation time in trees is much longer than that in herbs. Notably, the rate of molecular evolution in herbaceous plants is roughly 2.5 times faster than in woody plants based on a global phylogenetic analysis of angiosperms (Smith & Donoghue, 2008). Annuals have faster substitution rates than perennials (Yue et al., 2010).

Waiting for sterility is not equivalent to waiting for speciation

If we employ the biological species concept (Mayr, 1942, 1963), the tempo at which hybrid sterility builds is not indicative of the tempo of speciation. Prepollination isolation, as mandated by divergent flowering time, habitat preference, and floral attractants and architecture, is likely to appear sooner and develop at a faster rate than postpollination isolation. This is because the aforementioned attributes can be acted upon directly and forcefully by natural selection (Levin, 2003; Givnish, 2010), whereas hybrid sterility is thought to build through the gradual stochastic accumulation of many DM incompatibilities with small effects (Orr & Turelli, 2001; Coyne & Orr, 2004), and/or through chromosomal rearrangement.

Across organismic assemblages, the time required for the formation of postzygotic barriers may be several orders of magnitude longer than the time required for ecological divergence/speciation (Schluter, 2000; Seehausen, 2002; Mendelson, 2003; Fitzpatrick, 2004; Malone & Fontenot, 2008). A considerable time differential between the emergence of ecological barriers and hybrid sterility is well illustrated in many Hawaiian genera, wherein substantial ecological radiation within the past 3–4 million yr has not been accompanied by the emergence of strong postpollination barriers (Baldwin et al., 1998; Price & Wagner, 2004; Keeley et al., 2011). In general, chromosomal evolution in the Hawaiian flora has been rather muted (Stuessy et al., 1998). It is likely that many lineages evolve prezygotic barriers, but expire before the appearance of postzygotic barriers (Rosenblum et al., 2012).

In contrast to hybrid sterility, the degree of prepollination isolation does not progressively increase over time, because the niches of related lineages do not progressively diverge over time (Prinzing et al., 2001; Wiens & Graham, 2005; Couvreur et al., 2011; Peterson, 2011). Once established, the niches of related species tend to be relatively constant, being conserved by habitat selection, pleiotropy, and interpopulation gene flow, and the lack of genetic variation that might allow a niche shift (Wiens, 2004). Accordingly, when prezygotic barriers are breached or when species migrate long distances naturally or with the aid of humans, hybridization may occur between taxa long separated in the same or distant regions (Abbott et al., 2003).

The fact that herbaceous species that diverged 3–4 mya (or much longer in the case of trees) may retain the potential to exchange genetic material has important ramifications. First, the long time to hybrid sterility means that the patterns of genetic variation within contemporary species may have been shaped by numerous episodes of gene exchange with sister taxa at many times in the near and deep past. Whereas discussions of ancient hybridizations often focus on Holocene expansions from glacial refugia (e.g. Hewitt, 2004; Petit et al., 2004; Jaramillo-Correa et al., 2009), the gene pools of contemporary species (especially trees) may reflect much earlier and potentially repeated exchanges of genetic material. Secondly, there is a very long period during which substantive gene exchange may have led to the fusion of incipient species or to the retardation of species divergence (Seehausen et al., 2008; Gilman & Behm, 2011). Correlatively, there is a very large window of time during which a given taxon pair may have spawned hybrid species.

The accrual of chromosome pairing abnormalities (and attendant fertility reduction) in hybrids increases the production of unreduced gametes, and thus the likelihood of allopolyploid formation (Ramsey & Schemske, 1998). The notion of a very slow development of hybrid sterility is consistent with the observation that millions of yr have transpired between the split of two lineages and the origin of their allopolyploids. Consider that the B and C genomes of the Oryza officinalis complex split c. 4 mya, but tetraploids formed only between 0.3 and 0.6 mya (Wang et al., 2009). The lineages of Brassica oleracea and Brassica rapa diverged roughly 3.7 mya, whereas its allotetraploid derivative Brassica napus arose < 10 000 yr ago (Cheung et al., 2009). The progenitors of Nicotiana tabacum diverged c. 4.5 mya, whereas the latter arose only 0.2 mya (Clarkson et al., 2005). In these and other cases, the long time to allopolyploid formation may reflect a long wait for progenitor sympatry, as well as for reduced hybrid fertility.

We would hope to gain information that would provide a better understanding of the hybrid sterility dynamic. Divergence time estimates that were based on multiple markers, and on multiple populations of two species, would be most informative, as would hybrid fertility rates from the same multiple populations. If divergence times could be estimated for some populations that were allopatric and others that were sympatric with the related species, we might observe the effect of recent interspecific gene exchange (divergence time estimates would be reduced).

This paper is the first attempt at bringing a temporal perspective to the development of a postpollination barrier between species sharing the same ploidal level. In doing so, it brings us closer to understanding the tempo of allopatric speciation. In addition to better documenting fertility descent in time, it behoves us to consider the pace at which species crossability and hybrid viability decline, as they too are products of genomic dissonance. With the speed of barrier building in hand, we will finally appreciate the time required for two gene pools to become incompatible in all respects.


The author is grateful to Richard Abbott, Sally Otto, and three anonymous reviewers for their thoughtful critiques of the paper.