Speciation genes in plants

Abstract

Background

Analyses of speciation genes – genes that contribute to the cessation of gene flow between populations – can offer clues regarding the ecological settings, evolutionary forces and molecular mechanisms that drive the divergence of populations and species. This review discusses the identities and attributes of genes that contribute to reproductive isolation (RI) in plants, compares them with animal speciation genes and investigates what these genes can tell us about speciation.

Scope

Forty-one candidate speciation genes were identified in the plant literature. Of these, seven contributed to pre-pollination RI, one to post-pollination, prezygotic RI, eight to hybrid inviability, and 25 to hybrid sterility. Genes, gene families and genetic pathways that were frequently found to underlie the evolution of RI in different plant groups include the anthocyanin pathway and its regulators (pollinator isolation), S RNase-SI genes (unilateral incompatibility), disease resistance genes (hybrid necrosis), chimeric mitochondrial genes (cytoplasmic male sterility), and pentatricopeptide repeat family genes (cytoplasmic male sterility).

Conclusions

The most surprising conclusion from this review is that identities of genes underlying both prezygotic and postzygotic RI are often predictable in a broad sense from the phenotype of the reproductive barrier. Regulatory changes (both cis and trans) dominate the evolution of pre-pollination RI in plants, whereas a mix of regulatory mutations and changes in protein-coding genes underlie intrinsic postzygotic barriers. Also, loss-of-function mutations and copy number variation frequently contribute to RI. Although direct evidence of positive selection on speciation genes is surprisingly scarce in plants, analyses of gene family evolution, along with theoretical considerations, imply an important role for diversifying selection and genetic conflict in the evolution of RI. Unlike in animals, however, most candidate speciation genes in plants exhibit intraspecific polymorphism, consistent with an important role for stochastic forces and/or balancing selection in development of RI in plants.

INTRODUCTION

No other recent advance in speciation studies has received as much attention as the cloning and characterization of genes that contribute to the cessation of gene flow or reproductive isolation (RI) between populations (Orr and Presgraves, 2000; Butlin and Ritchie, 2001; Noor, 2003; Orr et al., 2004; Rieseberg et al., 2004; Noor and Feder, 2006; Mallet, 2006; Bomblies and Weigel, 2007b; Rieseberg and Willis, 2007; Bomblies, 2010; Presgraves, 2010). These so-called ‘speciation genes’ are of interest because knowledge of their identities and attributes offers clues to the ecological settings, evolutionary forces and molecular mechanisms that drive the divergence of populations and species (Orr et al., 2004, 2007). For example, information about gene identity and normal function may suggest particular traits or functions that are prone to disruption by deleterious interactions among divergent alleles. Patterns of sequence evolution can be used to detect the action of positive selection during speciation, and functional information may provide insight into the roles of different kinds of mutations (coding, regulatory, copy number, micro-chromosomal rearrangements, etc.) in diversification.

Barriers to gene flow caused by speciation genes can arise at multiple prezygotic and postzygotic life-history stages (Fig. 1). Despite widespread interest in speciation genes, most discussions have focused on genes that contribute to intrinsic postzygotic reproductive barriers such as hybrid sterility or inviability (Ting et al., 1998; Coyne and Orr, 2004; Wu and Ting, 2004; Orr, 2005; Haerty and Singh, 2006; Mallet, 2006; Bomblies and Weigel, 2007b; Orr et al., 2007; Presgraves, 2010). This focus has some justification because until very recently there were few a priori expectations regarding the identities and normal functions of genes that cause hybrid incompatibilities. In contrast, genes underlying prezygotic reproductive barriers were expected to be associated with the genetic pathways or networks that underlie the barrier phenotype of interest. However, as discussed below, candidate genes (or gene families) for both prezygotic and postzygotic isolation are increasingly predictable in a broad sense from barrier phenotype. Moreover, given the principal role of prezygotic barriers in the origins of many plant and animal species, a more inclusive discussion of the genes and mutations that underlie species formation is warranted.

Fig. 1.

Reproductive isolating barriers occur at multiple prezygotic and postzygotic life-history stages.

Another characteristic of the literature on speciation genes has been an emphasis on animal systems, particularly Drosophila. Botanists are partially to blame for this bias because relevant plant genes have rarely been discussed in the context of RI and speciation, at least until recently (although see Rieseberg et al., 2004; Bomblies and Weigel, 2007a; Moyle, 2008; Widmer et al., 2008; Bomblies, 2010). Also, many potential speciation genes in plants have been identified in crop species or exhibit intraspecific polymorphism, so their role in speciation is unclear (Rieseberg and Willis, 2007).

In this review, we partially rectify these earlier omissions by compiling a list of genes that underlie both prezygotic and postzygotic reproductive barriers in plants (Table 1). We also discuss the normal functions of these genes and describe the mutations that are the cause of RI (Appendix 1). We use this information to address a number of fundamental questions about the genetics of speciation, including the following. (1) What kinds of genes and mutations underlie reproductive barriers in plants and are there differences between plants and animals, prezygotic and postzygotic barriers, and hybrid sterility and inviability? (2) What is the relative importance of regulatory versus coding mutations in the evolution of RI? (3) What role has genetic redundancy played in the evolution of hybrid incompatibilities? (4) What evolutionary forces (i.e. natural selection, sexual selection, drift) are responsible for the divergence of the genes underlying RI? And (5) what are the taxonomic and geographical distributions of RI inducing mutations and do these patterns differ between plants and animals and between prezygotic and postzygotic barriers?

Table 1.

Genes underlying reproductive barriers in plants, with complementary genes indicated by superscript numbers

Gene	Normal function	Organism	Level	Barrier phenotype	Likely genetic cause	Reference
Genes that underlie pre-pollination barriers
ANTHOCYANIN-2 (AN2) – a myb-type transcription factor	Regulation of anthocyanin production (floral colour)	Petunia axillaris/P. integrifolia	Inter-specific	Flower colour differences leading to pollinator shift	Loss of function mutations in coding sequence	(Quattrocchio et al., 1999; Hoballah et al., 2007)
ROSEA1, ROSEA2 and VENOSA – all are myb-type transcription factors	Regulation of anthocyanin production (floral colour)	Antirrhinum species	Inter-specific	Flower colour differences leading to shifts in pollinator fauna	?	(Schwinn et al., 2006)
FLAVONOID-3′-HYDROXYLASE (F3′H)	Key enzyme in anthocyanin pathway (floral colour)	Ipomea species	Inter-specific	Flower colour differences leading to pollinator shift	cis-regulatory mutations that downregulate F3′H	(Des Marais and Rausher, 2010)
FLOWERING LOCUS C (FLC) – a MADS-box transcription factor	Represses flowering	Allopolyploid A. suecica isolated from parental species, A. thaliana and A. arenosa	Inter-specific	Delayed flowering of allopolyploid, A. suecica	cis-regulatory mutations in both A. thaliana and A. arenosa FLC	(Wang et al., 2006a)
STYLE2·1 – encodes a helix-loop-helix (HLH)	Transcription factor	Solanum lycopersicon/S. pennellii	Inter-specific	Recessed style leading to mating system shift outcrossing to selfing	cis-regulatory mutation(s)	(Chen et al., 2007)
Genes that underlie post-pollination prezygotic barriers
S-RNase-SI	Rejection of self pollen	Nicotiana species	Inter-specific	Rejection of pollen from other species	?	(Murfett et al., 1996)
Genes that underlie intrinsic postzygotic barriers: hybrid inviability
1Cf2 – an extracellular leucine-rich repeat receptor-like gene	Resistance to fungal pathogen	Solanum lycopersicon/S. pimpinellifolium	Inter-specific	Hybrid necrosis	Gene copy number variation	(Dixon et al., 1996; Kruger et al., 2002)
1RC3 – encodes an extracellular cysteine protease	Perception of fungal Avr proteins	Solanum lycopersicon/S. pimpinellifolium	Inter-specific	Hybrid necrosis	Changes in protein sequence	(Kruger et al., 2002; Rooney, 2005)
DANGEROUS MIX 1 (DM1) – an NBS-LRR gene	Disease resistance	Arabidopsis thaliana	Intra-specific	Hybrid necrosis	Gene copy number variation	(Bomblies et al., 2007)
HISTIDINOL-PHOSPHATE AMINO-TRANSFERASE (HPA1 and HPA2)	Synthesis of essential amino acid, histidine	Arabidopsis thaliana	Intra-specific	Arrest of hybrid seed development	Reciprocal silencing of duplicate genes	(Bikard et al., 2009)
TRANSPARENT TESTA GLABRA2 (TTG2) – a WRKY transcription factor	Regulates epidermal cell fate	Arabidopsis thaliana	Intra-specific	Lethality of interploidal hybrids	cis-regulatory mutations	(Dilkes et al., 2008)
HBD2 – encodes a casein kinase	Root development and hormone sensitivity	Oryza sativa	Inter-sub-specific	Hybrid necrosis	Change in protein sequence	(Yamamoto et al., 2010)
HWH1 – encodes a GMC oxidoreductase	Catalyses oxidation–reduction reactions	Oryza sativa	Inter-sub-specific	Hybrid necrosis	?	(Jiang et al., 2008)
Genes that underlie intrinsic postzygotic barriers: hybrid sterility
S5 – encodes an aspartate protease	Disease resistance signalling and cell death	Oryza sativa	Inter-sub-specific	Hybrid female sterility (embryo sac sterility)	Changes in protein sequence	(Chen et al., 2008)
2SaM – a SUMO E3 ligase-like gene	Post-translational modification	Oryza sativa	Inter-sub-specific	Hybrid male sterility (pollen abortion)	Substitution in intron-splicing site, leading to truncated protein	(Long et al., 2008)
2SaF – encodes an F-box protein	Mediation of protein-protein interactions	Oryza sativa	Inter-sub-specific	Hybrid male sterility (pollen abortion)	Amino acid substitution	(Long et al., 2008)
mtRPL27 – Nuclear-encoded mitochondrial ribosomal protein L27	Translation of mitochondrial genes	Oryza sativa/O. glumaepatula	Inter-specific	Hybrid male sterility (pollen abortion)	Reciprocal silencing of duplicate genes	(Yamagata et al., 2010)

WHAT IS A SPECIATION GENE?

A speciation gene can be strictly defined as a gene that contributes to the splitting of two lineages by reducing the amount of gene flow between them. One concern with this definition is that if the cessation of gene flow is solely a consequence of geographical isolation, then one could argue that no genes contributed to speciation in these instances. Yet, we would not consider speciation to be complete until genetically based barriers to gene flow had evolved between the geographically isolated populations. Another concern is that speciation is an ongoing process and additional changes that result in further RI can accumulate after speciation is already complete. This creates two uncertainties. First, genes contributing to incipient speciation may be segregating within species. Second, for genes contributing to RI between species, it can be difficult to distinguish between those genetic changes that played causal roles in speciation and those that arose after speciation was complete. In considering these problems, we have chosen to treat genes contributing to prezygotic and postzygotic isolating RI differently. For postzygotic barriers, we have broadened the definition of speciation genes in this review to include any gene that contributes to reproductive isolation between populations, even if the contribution is likely to be small. This definition therefore accommodates genes that exhibit intraspecific polymorphism, as well as genes that diverged late in the speciation process and had no causative role. For prezygotic barriers, we retain the criteria that the genes contribute to barriers associated with a cladogenic event because prezygotic barriers may be more readily reversible. In an earlier discussion, Rieseberg et al. (2004) argued that the term ‘isolation genes’ would be a more accurate descriptor of genes that contribute to RI than ‘speciation genes’. Likewise, Noor and Feder (2006) employ the term ‘barrier genes’. However, these terms can refer to processes other than reproductive isolation (e.g. barriers to the absorption of nutrients in the small intestine); hence our use of ‘speciation genes’ herein, while recognizing that many of these genes are unlikely to be the cause of speciation. Also, although we applied strict selection criteria, many genes we have chosen to list have not been examined with a comprehensive battery of functional, organismal and population-level studies and thus should still be considered as candidate speciation genes.

CANDIDATE SPECIATION GENES AND THEIR ATTRIBUTES

For inclusion in Table 1, we required convincing genetic and/or functional validation of candidate speciation genes. As a consequence, many excellent candidates (e.g. Wang et al., 1997; Whittall et al., 2006; Escobar-Restrepo et al., 2007; Case and Willis, 2008; Scalliet et al., 2008) were not included on the list, although several of these are discussed briefly in the text. In some instances, changes in the expression of candidate genes were shown to be the cause of RI, but it was not clear whether the expression shift was controlled by cis- versus trans-acting factors (e.g. Wang et al., 1997; Whittall et al., 2006). In other cases, barrier phenotypes in interspecific crosses were similar to those of Arabidopsis mutants (Escobar-Restrepo et al., 2007) and/or rapid adaptive diversification was observed in proteins that seem likely to contribute to RI (Mayfield et al., 2001).

We also required evidence that the candidate speciation genes potentially contribute to RI between populations. For genes responsible for intra- or interspecific incompatibilities we considered this to be self-evident, although we recognize that incompatibility alleles at some of these genes might be too rare to have a significant effect on gene flow between populations. For pre-pollination barriers, we considered that floral changes associated with observed shifts in the pollinator community were likely to contribute to RI, even if this had not been proven by fieldwork in each instance. On the other hand, we found that genetic changes underlying adaptation to environmental differences were less clearly associated with RI. An example comes from Arabidopsis halleri, which is able to colonize heavy-metal-polluted soils due to enhanced expression of HEAVY METAL ATPASE 4 (HMA4) (Hanikenne et al., 2008). However, there is no evidence that Ah-HMA4 contributes to RI because, although zinc tolerance in A. halleri is constitutive, the species is found in both metalliferous and non-metalliferous sites and there is little evidence of population structure associated with soil type (Pauwels et al., 2006).

Genes and mutations that contribute to pre-pollination prezygotic isolation

Our review of the literature identified seven genes that have been shown to underlie pre-pollination postzygotic barriers (Table 1). Five of these genes were responsible for variation in floral pigmentation, one for differences in flowering time and one for a shift in mating system from outcrossing to selfing. Surprisingly, we were unable to find convincing examples of genes contributing to habitat isolation, even though ecogeographical isolation is viewed by many students of plant speciation as the most common and important type of reproductive barrier (Stebbins, 1950; Schemske, 2000; Rieseberg and Willis, 2007; Sobel et al., 2010). However, there are some promising candidates, such as Ha-CDPK3, which co-segregates with quantitative trait loci (QTLs) for the uptake of toxic mineral ions and immigrant inviability in a hybrid sunflower and its parental species (Lexer et al., 2004). Likewise, reciprocal transplant experiments have shown that the cytoplasm frequently contributes to habitat isolation (e.g. Wu and Campbell, 2007; Kimball et al., 2008; Sambatti et al., 2008), but the genes involved have not been isolated.

We were also surprised to find only a single case in which interspecific genetic differences in flowering times had been characterized (Chen et al., 2007), and even in this case the main reproductive barrier is a ploidy shift. However, genes have been connected with flowering time differences among accessions of Arabidopsis thaliana and among cultivars of many domesticated plant species (e.g. Yano et al., 2000; Caicedo et al., 2004; Yan et al., 2004, 2006; Turner et al., 2005; Balasubramanian et al., 2006; Takahashi et al., 2009; Schwartz et al., 2009; Blackman et al., 2010). Although we did not include these genes on our list, a reasonable argument could be made that they contribute more to reproductive isolation between populations than many of the genes underlying hybrid incompatibilities that are included in Table 1.

Despite the small number of pre-pollination RI genes on our list, several trends have emerged. First, it appears that the same genes and pathways may frequently contribute to pre-pollination RI. Currently, the main evidence for this trend comes from studies of variation in flower colour (Table 1; Appendix 1). Differences in the intensity and patterning of anthocyanin production in flowers appear to result mostly from mutations in the same small subfamily of MYB-related transcription factors that regulate the anthocyanin biosynthetic pathway (Quattrocchio et al., 1999; Schwinn et al., 2006). Another frequent change – the transition from bee-pollinated, blue/purple flowers to red, hummingbird-pollinated flowers – appears to have repeatedly involved mutations in a structural gene, FLAVONOID-3′-HYDROXYLASE (F3′H), that redirect flow down different branches of the anthocyanin pathway (Des Marais and Rausher, 2010). Although these inferences are based on a small number of studies, they are corroborated by broader surveys of spontaneous mutations, natural intraspecific polymorphisms and inferences from partially characterized candidate genes (Rausher, 2008; Streisfeld and Rausher, 2009a, b). In addition to the anthocyanin biosynthetic pathway, we suspect that genes in the photoperiod, vernalization and scent-biosynthetic pathways will often be associated with speciation as genes from these pathways frequently appear in studies of cultivated and model plants (Dudareva et al., 1996; Nam et al., 1999; Shindo et al., 2005; Jones et al., 2008; Scalliet et al., 2008; Takahashi et al., 2009; Koeduka et al., 2009; Schwartz et al., 2009).

Another interesting trend is that six of the seven pre-pollination speciation genes listed have regulatory functions, whereas only one is a structural gene (Table 1). This ratio may be slightly biased in favour of regulatory genes because repeated involvement in speciation is not accounted for and the structural gene, F3′H, may have been repeatedly involved in floral colour changes (Des Marais and Rausher, 2010). Nonetheless, it does imply that genes involved in transcriptional regulation play a disproportionately significant role in the evolution of pre-pollination RI.

Mutations contributing to pre-pollination RI appear to be mainly loss-of-function (LOF) mutations in coding sequence or cis-regulatory mutations (Table 1). The phenotypes associated with pre-pollination RI are often fixed within species (or nearly so), possibly suggesting that the causative mutations are widespread as well. This conjecture may not be warranted, however, as LOF mutations in Pa-ANTHOCYANIN2 (AN2), the only pre-pollination RI gene for which intraspecific variation has been examined, arose independently at least five times (Hoballah et al., 2007). These observations may be reconciled if flower colour is a late evolving character during pollinator shifts (Quattrocchio et al., 1999). Likewise, based on studies of these loci, little can be said about the role of selection versus drift in the evolution of pre-pollination speciation genes. Although the traits associated with pre-pollination RI seem likely to have experienced divergent natural selection because of their effects on pollinator behaviour, timing of reproduction or mating system, only for ROSEA1 is there direct evidence of selection on the underlying genes (Whibley et al., 2006).

Genes that contribute to post-pollination prezygotic barriers

As far as we are aware, the only genes contributing to post-pollination prezygotic barriers that have been functionally validated are S RNases, the functional products of the S-locus in the Solanaceae, which is responsible for self-incompatibility (SI) within species. S-RNase is secreted by stylar tissue and degrades the RNA of incompatible pollen tubes. The probable role of the S-locus in the evolution of interspecific or ‘unilateral’ incompatibility has long been suspected because of asymmetric patterns of pollen rejection in crosses between SI and self-compatible (SC) species in the Solanaceae, Plantaginaceae, Rosaceae and Brassicaceae (Lewis and Crowe, 1958; Hiscock and Dickinson, 1993; Hancock et al., 2003). Unilateral incompatibility has been mapped to the S-locus in tomato (Chetelat and De Verna, 1991; Bernacchi and Tanksley, 1997), but only in Nicotiana have functional studies been performed to show that that S RNase genes contribute to interspecific pollen rejection (Murfett et al., 1996). S RNase-SI genes are known to be under negative frequency-dependent selection, which results in extreme allelic polymorphism (Richman and Kohn, 2000).

In crucifers, the S-locus contains two tightly linked, highly polymorphic genes, S-receptor kinase (SRK) and its ligand S-locus cysteine rich protein (SCR), which determine SI in the stigma and pollen, respectively (Stein et al., 1991; Schopfer et al., 1999). Allele-specific binding of these proteins initiates a signalling cascade in the stigma that prevents self pollen from germinating. Transformation of SC A. thaliana with SRK and SCR alleles from its SI relative A. lyrata results in restoration of SI but has no demonstrated effect on unilateral incompatibility between these species (Nasrallah, 2002).

A promising candidate gene for post-pollination prezygotic RI in plants is FERONIA (FER), a receptor-like kinase involved in pollen tube reception. FER mutants in A. thaliana are characterized by a failure to arrest pollen tube growth in ovules. A similar phenotype is observed in interspecific crosses with pollen from A. lyrata or Cardamine flexuosa, which exhibit accelerated amino acid diversification in the extracellular domain of FER. Thus, coding changes in FER may contribute to the crossability barriers between these species, but this possibility has not yet been functionally verified (Escobar-Restrepo et al., 2007).

Genes and mutations that contribute to hybrid inviability

Empirical studies in plants and animals have shown that intrinsic postzygotic barriers to reproduction – hybrid inviability and hybrid sterility – frequently evolve through mechanisms consistent with the classic Dobzhansky–Muller (DM) model (Dobzhansky, 1937; Muller, 1942). As adaptive or nearly neutral substitutions accumulate in diverging lineages, substitutions may be fixed in one lineage that are incompatible with substitutions in the other lineage. Therefore, hybrid dysfunction results when these incompatible alleles are brought together in hybrid progeny. Although most often considered as a two-locus model with one substitution at each locus occurring in each diverging lineage, analyses of genes underlying DM incompatibilities in plants (Tables 1 and 2; Appendix 1) have shown that diverse evolutionary paths also yield similar outcomes (Fig. 2).

Table 2.

Genes involved in cytoplasmic male sterility, with complementary genes indicated by superscript numbers

Gene	Normal function	Organism	Level	Likely genetic cause	Reference
1Radish Ogura and Kosena cytoplasms (ORF138 and ORS125)	Disruption of pollen development	Raphanus sativus	Intra-specific; Inter-generic	Recombination created chimeric gene	(Bonhomme et al., 1991; Iwabuchi et al., 1999)
1Rfo/RFK1 – a mitochondria-targeting PPR gene	Regulation of organelle gene expression	Raphanus sativus	Intra-specific; Inter-generic	Amino-acid substitutions in PPR domains	(Koizuka et al., 2000; Brown et al., 2003; Desloire et al., 2003)
Brassica pol cytoplasm (ORF224)	Disruption of pollen development	Brassica napus	Intra-specific	Recombination created chimeric gene	(Singh and Brown, 1991)
Brassica nap cytoplasm (ORF222)	Disruption of pollen development	Brassica napus	Intra-specific	Recombination created chimeric gene	(L'homme et al., 1997)
Brassica tour cytoplasm (ORF263)	Disruption of pollen development	Brassica juncea/B. tournefortii	Inter-specific	Recombination created chimeric gene	(Landgren et al., 1996)
Brassica tour cytoplasm (ORF193)	Disruption of pollen development	Brassica napus/B. tournefortii	Inter-specific	Recombination created chimeric gene	(Dieterich et al., 2003)
Moricandia arvensis cytoplasm (ORF108)	Disruption of pollen development	Moricandia arvensis/Brassica napus	Inter-generic	Recombination created chimeric gene	(Ashutosh et al., 2008)
Sunflower PET1 cytoplasm (ORF522)	Disruption of pollen development	Helianthus annuus/H. petiolaris	Inter-specific	Recombination created chimeric gene	(Horn et al., 1991)
2Petunia pcf cytoplasm (ORF402)	Disruption of pollen development	Petunia hybrida	Inter-specific	Recombination created chimeric gene	(Young and Hanson, 1987)
2Rf-PPR592 – a mitochondria-targeting PPR gene	Regulation of organelle gene expression	Petunia hybrida	Inter-specific	Promoter deletion resulting in lack of expression	(Bentolila et al., 2002)
Maize cytoplasm S (ORF355/ORF77)	Disruption of pollen development	Zea mays	Intra-specific	Recombination created chimeric gene	(Zabala et al., 1997)
3Maize cytoplasm T (URF13)	Disruption of pollen development	Zea mays	Intra-specific	Recombination created chimeric gene	(Dewey et al., 1987)
3RF2 – encodes an aldehyde dehydrogenase	Oxidation of aldehydes	Zea mays	Intra-specific	Amino acid substitution leading to loss of Aldh activity	(Cui et al., 1996)
Sorghum A3 cytoplasm (ORF107)	Disruption of pollen development	Sorghum bicolor	Intra-specific	Recombination created chimeric gene	(Tang et al., 1996)
Sorghum RF1 – a mitochondria-targeting PPR gene	Regulation of organelle gene expression	Sorghum bicolor	Intra-specific	?	(Tang et al., 1996)
Wheat As or Tt cytoplasm (ORF256)	Disruption of pollen development	Triticum aestivum/T. timopheevi	Inter-specific	Recombination created chimeric gene	(Hedgcoth et al., 2002)
Common bean CMS (ORF239)	Disruption of pollen development	Phaseolus vulgaris	Intra-specific	Recombination created chimeric gene	(Abad et al., 1995)
4Rice Boro II cytoplasm (ORF79)	Chimeric orf	Oryza sativa	Inter-sub-specific	Recombination created chimeric gene	(Wang et al., 2006b)
4Rice RF1A – a mitochondria-targeting PPR gene	Regulation of organelle gene expression	Oryza sativa	Inter-sub-specific	Frameshift mutation in protein	(Wang et al., 2006b)
4Rice RF1B – a mitochondria-targeting PPR gene	Regulation of organelle gene expression	Oryza sativa	Inter-sub-specific	Nine amino acid substitutions, one of which likely causes loss of restoration function	(Wang et al., 2006b)
RETROGRADE-REGULATED MALE STERILITY (RMS) – encodes a 178-amino-acid protein of unknown function	unknown	Oryza rufipogon	Intra-specific	Upregulation of RMS due to cis-regulatory mutation(s)	(Fujii and Toriyama, 2009)

Fig. 2.

Diverse evolutionary paths lead to the origin of Dobzhansky–Muller incompatibilities in plants. (A) Two substitutions within a single lineage illustrated by Cf-2 and RCR3 loci in Solanum species. (B) Divergent resolution of an ancestral gene duplication as illustrated by the HPA1 and HPA2 loci in Arabidopsis thaliana populations. (C) Divergent resolution of an ancestral polymorphism at one locus and lineage-specific substitution at a second tightly linked locus as illustrated by the SaF and SaM loci in domesticated Oryza sativa subspecies.

In a recent review, Bomblies and Weigel (2007b) suggested that disease resistance genes might play an important role in the evolution of hybrid inviability. Their arguments were based on the following observations: (1) inter- and intra-specific hybrids often exhibit tumours, as well as hybrid necrosis or weakness; (2) symptoms of hybrid necrosis are similar to necrotic symptoms typically associated with environmental stress and pathogen attack; (3) hybrid necrosis usually results from the interactions of complementary genes, similar to classic DM incompatibilities; and (4) at least one example is known where hybrid necrosis impedes interspecific gene flow in the wild (McNaughton and Harper, 1960). Genetic characterization of hybrid necrosis in crosses between tomato species and between Arabidopsis ecotypes (Appendix 1) has revealed that incompatibilities among complementary disease resistance genes are indeed the cause of the necrosis (Kruger et al., 2002; Rooney, 2005; Bomblies et al., 2007). Disease resistance genes have also been implicated in mediating hybrid necrosis in crosses between cultivated rice varieties (Yamamoto et al., 2010) as well as between cultivated lettuce and an evolutionarily distant wild relative (Jeuken et al., 2009). Interestingly, in the tomato and Arabidopsis examples, one of the parental lines lacks a member of the interacting gene pair, a circumstance that might have facilitated the evolution of the incompatibility in the first place.

Copy number variation played a more obvious role in the evolution of a hybrid incompatibility involving HISTIDINOL-PHOSPHATE AMINO-TRANSFERASE (HPA) in Arabidopsis (Table 1; Appendix 1). Botanists have long speculated that the reciprocal silencing of duplicate genes might be a frequent cause of intrinsic postyzgotic barriers (Werth and Windham, 1991), but there has been little empirical evidence to support such a mechanism. The reciprocal silencing of HPA duplicates in natural populations of A. thaliana and mtRPL27 duplicates in AA genome species of rice represent the first clear examples of this process (Bikard et al., 2009; Yamagata et al., 2010). The transposition of JYALPHA, a hybrid sterility gene in Drosophila, was probably mediated by gene duplication as well, but this has not been proven (Masly et al., 2006).

Another potential trend is a role for maternally expressed regulatory genes in the development of dosage-sensitive incompatibilities. Several dosage-sensitive incompatibilities involving cis-regulatory changes (Dilkes et al., 2008) or epigenetic causes (loss of maternal imprinting) have been identified in interploidal crosses of Arabidopsis (Josefsson et al., 2006; Walia et al., 2009; Erilova et al., 2009).

Unlike the genetics of pre-pollination barriers, transcriptional regulators are not over-represented among genes that cause hybrid inviability. Nor is there an excess of LOF or cis-regulatory mutations. Alleles contributing to hybrid inviability vary widely in their geographical distributions. For example, DM1 is restricted to a small fraction of the range of A. thaliana, whereas reciprocal divergence at HWH1 and HWH2 between the indica and japonica subspecies of rice is substantially more complete (Jiang et al., 2008). The role of positive selection has not been fully resolved. Disease resistance loci in plants typically contain multiple, related, tightly linked genes that have arisen through evolutionary recent gene duplication events and often are the targets of diversifying selection (Mondragon-Palomino et al., 2002; Kuang et al., 2004). On the other hand, the haphazard distribution of LOF mutations in duplicate At-HPA genes is consistent with near neutral processes (Bikard et al., 2009).

Genes and mutations that contribute to hybrid sterility

Cytoplasmic male sterility

By far the largest numbers of sterility loci that have been characterized in plants (or animals) are found in the plant mitochondrial genome and cause cytoplasmic male sterility or CMS (Table 2; Appendix 1). Molecular genetic studies indicate that CMS typically results from rearrangements in the mitochondrial genome, most frequently via the formation and expression of chimeric open reading frames (ORFs; Hanson, 1991; Hanson and Bentolila, 2004; Chase, 2007; Carlsson et al., 2008). In most species, CMS is characterized by the absence of pollen. However, in a few instances (e.g. sunflower, petunia and maize), anthers are missing as well. The chimeric ORFs that cause CMS often include ATP synthase subunit gene promoter regions and/or coding regions or occur near ATP synthase genes (Appendix 1). Less frequently, the chimeric genes include subunits of cytochrome oxidase or NADH dehydrogenase. Essentially all CMS-associated loci also include unique unidentified sequences that show no similarity to chloroplast or nuclear sequences in plants and whose origins remain unknown (Hanson and Bentolila, 2004). Although these genes are included in our overall tally, we have catalogued CMS and restorer loci in a separate table (Table 2) to acknowledge that many of these alleles are specific to crop lines or are rare and thus may have had little impact on speciation.

Mitochondria are usually maternally inherited, so CMS is typically transmitted through ovules (but see McCauley et al., 2007). In contrast, nuclear genes are transmitted through both ovules and pollen. This difference in inheritance pattern creates a genetic conflict between nuclear and cytoplasmic genes. Theoretical studies indicate that CMS will spread in outcrossing populations if the CMS mutation provides even a slight fitness advantage in female function (Fig. 3; Lewis, 1941; Frank, 1989; Hodgins et al., 2009). Because CMS plants are freed from the cost of producing pollen, some CMS mutants have been shown to increase the fitness of females relative to hermaphrodites (Delph et al., 2007). That ‘selfish’ CMS elements can spread in spite of their effects on sterility casts doubt on the effectiveness of CMS as a reproductive barrier. In addition to genetic conflict, the successful origin of CMS loci represents a kind of diversifying selection, as it appears that novel proteins are required to evade regulation by restorer genes (see below).

Fig. 3.

Two evolutionary scenarios for the spread of CMS and RF alleles in hermaphroditic populations. (A) Mitochondrial CMS-causing allele arises first, followed by evolution of a new nuclear restorer allele. (B) Invasion of a CMS allele in a population segregating for RF and rf alleles. Females and hermaphrodites are represented by standard symbols. Frequencies of the female permissive alleles at the mitochondrial CMS locus and nuclear restorer locus are shown in pink.

Although all of the fully characterized CMS-associated loci are associated with crop plants, the CMS cytoplasms often derive from natural populations of a wild relative. For example, the PET CMS cytoplasm in cultivated sunflower (Helianthus annuus) was discovered in a naturally occurring population of the prairie sunflower, Helianthus petiolaris, from St Louis, USA (Leclercq, 1969). The only well-characterized CMS cytoplasm not associated with a crop is in Mimulus guttatus, where mtDNA transcripts containing NAD6 were perfectly correlated with CMS (Case and Willis, 2008). However, functional studies to confirm which of the ORFs present in these Mg-NAD6-containing transcripts cause sterility have not yet been performed.

We are aware of only three studies that have examined the distribution of the molecular variants underlying CMS in natural populations. In hermaphroditic species of Helianthus and Mimulus, the CMS loci are exceedingly rare, known only from the wild population in which they were originally found (Rieseberg et al., 2004; Case and Willis, 2008). In contrast, in gynodioecious populations of radish (Raphanus sativus) the frequency of the CMS locus ranged from 0 to 1 (Murayama et al., 2004). Case and Willis (2008) hypothesize that in the absence of gynodioecy, CMS loci will be rare and unlikely to contribute significantly to reproductive isolation and speciation.

Nuclear restorer of fertility (RF) genes

CMS-associated loci represent only one member of the pair of DM genes that cause cytoplasmic male sterility. The other member of these DM pairs is the non-restoring allele (rf) or homologue of nuclear restorer of fertility genes. RF genes restore fertility in CMS plants in several different ways (reviewed in Hanson and Bentolila, 2004). Most frequently, they regulate the transcript profile or protein accumulation of the CMS locus. Less frequently, they act to ameliorate the negative consequences of metabolic effects (Liu et al., 2001) or to reduce the abundance of mitochondrial segments carrying the CMS locus (Mackenzie and Chase, 1990).

We are aware of seven RF genes that have been cloned to date (Table 2; Appendix 1). Five of these belong to the pentatricopeptide repeat (PPR) family. The PPR family is one of the largest gene families in plants, with 441 and 655 PPR genes recognized in Arabidopsis and rice, respectively. The evolution of PPR genes resembles that of disease resistance genes in that they encode proteins with repeat domains, occur in small clusters of closely related genes that have arisen through evolutionarily recent gene duplication and transposition, and frequently exhibit an excess of non-synonymous substitutions indicative of diversifying selection (Geddy and Brown, 2007; Foxe and Wright, 2009). Notably, the nuclear restorer loci involved in CMS in Mimulus map to a region containing recent tandem expansions of PPR genes (Barr and Fishman, 2010).

The geographical distribution of RF genes and their non-restoring rf allele is only known from radish, where the Rf gene frequency varies from 0·41 to 1 (and the non-restoring allele from 0 to 0·59). There is indirect evidence that rf alleles are infrequent in hermaphrodites, as fertility is restored in most progeny from wild × CMS crosses. For example, all of the crosses that our group has made between natural populations of sunflower and cultivated lines carrying the PET CMS locus have generated fertile progeny (Rieseberg et al., 1995, 1996; Burke et al., 2002, 2004; Lai et al., 2005). These patterns imply either that rf alleles occur at low frequency or that there are multiple RF genes capable of restoring a given cytoplasm (e.g. Wang et al., 2006b).

The molecular changes responsible for the evolution of the rf sterility alleles have been characterized for five of the seven RF loci (Table 2; Appendix 1). Three of the mutations result in changes in protein sequence or structure, whereas two occur in cis-regulatory regions. Interestingly, most of these changes appear to represent LOF mutations, perhaps implying that the sequence of events responsible for the evolution of CMS may differ from that typically envisaged (Fig. 3). Both theoretical and empirical studies of CMS typically assume that a mitochondrial rearrangement that causes CMS arises first. The spread of the CMS locus generates strong selection for the evolution of a nuclear restorer allele. This sequence of events may be correct for gynodioecious species and for some cytotypes in hermaphrodites. However, in other instances (perhaps the majority), a newly arisen CMS locus may simply exploit a defect in the regulation of organellar genes that was caused by an LOF mutation in an RF gene. This might also account for the apparently limited geographical distribution of CMS-associated loci and rf alleles in hermaphroditic species.

Other hybrid sterility genes

A mystery associated with the genetics of plant speciation has been the frequent discovery of loci that reduce fitness when heterozygous (i.e. underdominant loci). Such loci should be rare because the establishment of alleles with negative fitness effects is unlikely except in small, inbred populations (Hedrick, 1981; Walsh, 1982; Lande, 1984). Some of the earliest examples of underdominant loci were found in crosses between the indica and japonica subspecies of rice (Oka, 1953). These examples were interpreted by Stebbins (1958) to most likely result from small structural changes involving as few as one to five genes. However, subsequent genetic analyses have largely failed to find evidence of structural differentiation in the vicinity of these loci.

The recent characterization of two underdominant loci in rice has provided two different solutions to this mystery (Chen et al., 2008; Long et al., 2008). At the S5 locus, a single underdominant gene was found to underlie hybrid female sterility; individuals heterozygous for alternative alleles from the indica and japonica are sterile. However, a third non-functional allele restores cross-compatibility between the subspecies. The existence of this compatible allele in both subspecies would make it feasible for the sterility-inducing mutations to arise without a loss of fitness. In contrast to S5, the Sa locus, which underlies male sterility, resolves into two adjacent DM genes. We suspect that other apparent examples of underdominance in plants will have similar explanations.

Although the evolutionary forces responsible for the establishment of the sterility-causing alleles in rice are unclear, the alleles are geographically widespread (Appendix 1). In contrast, of the incompatible alleles at two loci involved in a DM incompatibility between two Mimulus species, one allele is geographically widespread in one species whereas the other is extremely geographically restricted in the other species (Sweigart et al., 2007).

DISCUSSION

The identities of candidate speciation genes

An important conclusion from this review is that for many reproductive barriers, the identities of many candidate speciation genes in plants can be predicted in a broad sense from the barrier phenotype. This is not surprising for prezygotic barriers, such as changes in floral colour or flowering time. As expected, the genes underlying these barriers do indeed derive from the genetic pathways or networks known to be associated with these phenotypes. A similar generalization can be made in animals: genes causing prezygotic RI generally belong to the expected genetic pathway (Wheeler et al., 1991; Swanson and Vacquier, 1998; Palumbi, 1999; Fang et al., 2002; Barrett et al., 2009; Wittkopp et al., 2009). A caveat is that many studies of prezygotic RI employed a candidate gene approach, which undoubtedly biases findings toward well-circumscribed genes and pathways.

What is more surprising is that this conclusion also holds for multiple, common intrinsic postzygotic barriers in plants. CMS, for example, is typically caused by interactions between chimeric mitochondrial genes (particularly ATP synthase subunits) and members of the PPR gene family. Likewise, hybrid necrosis often results from changes in disease resistance genes. In contrast, few if any predictions can be made in animals, and the identity of each new speciation gene is a surprise (Ting et al., 1998; Froschauer et al., 2001; Barbash et al., 2003; Presgraves et al., 2003; Brideau et al., 2006; Harrison and Burton, 2006; Masly et al., 2006; Lee et al., 2008; Mihola et al., 2009; Phadnis and Orr, 2009; Tang and Presgraves, 2009). Unlike plants, there is little evidence that disease resistance genes contribute to RI in animals. Mitochondrial genes have been implicated in reduced hybrid viability in Tigriopus californicus copepods (Harrison and Burton, 2006) and in hybrid sterility in yeast (Lee et al., 2008), but mitochondrial-associated RI appears to be much rarer in animals than in plants. These trends aside, various other types of genes are represented among the few genes with known involvement in other hybrid sterility and inviability phenotypes. Only once additional genes involved in these forms of reproductive isolation are cloned will we know whether there is a broadly predictable relationship between these other barrier phenotypes and the normal functions of the underlying genes as well.

The nature of the genetic changes underlying the evolution of RI

A few general trends about the genetics of RI have emerged from this review. First, regulatory changes (both cis and trans) dominate the evolution of pre-pollination RI in plants. This contrasts with the genetics of intrinsic postzygotic barriers, where a mix of regulatory changes and changes in protein-coding genes are found. Second, LOF mutations frequently contribute to both prezygotic and postzygotic RI in plants. This trend is most apparent in genes underlying changes in floral colour and in the evolution of CMS, where LOF mutations in PPR genes appear to provide an opportunity for the invasion of a CMS-causing mitochondrial locus. A third emerging trend is the importance of copy number variation in the evolution of hybrid sterility and inviability. Both disease resistance genes and PPR genes belong to large gene families and high rates of gene turnover in these families appear to contribute to the evolution of RI. Even for low-copy-number genes, there is now evidence that gene duplication followed by reciprocal silencing of the duplicate copies can lead to RI (Bikard et al., 2009; Yamagata et al., 2010).

Few if any of these trends hold in animals based on the candidate speciation genes identified so far. Differences between plants and animals include the rarity of LOF mutations in animal speciation genes, as well as frequent reports of rapid protein evolution underlying animal RI (e.g. Ting et al., 1998; Presgraves et al., 2003; Tang and Presgraves, 2009). Also, copy number variation appears to be less important in the evolution of animal than plant RI (although see Masly et al., 2006).

The nature of the evolutionary forces underlying the evolution of RI

In contrast to animal speciation genes, which often exhibit the signature of positive selection, there is surprisingly little direct evidence that selection has been responsible for the divergence of candidate speciation genes in plants. The much higher levels of intraspecific polymorphism for RI alleles in plants versus animals (see below) is also consistent with a larger role for stochastic evolutionary forces in the evolution of plant RI. Nonetheless, there is indirect support for a key role for selection in the evolution of many of these genes, including: (1) disease resistance genes and PPR genes belong to large gene families and members of these families have been shown to be the targets of diversifying selection; (2) S RNase-SI genes are known to be under negative frequency-dependent selection; (3) CMS loci are selfish genetic elements that invade populations because they increase fitness through female function; and (4) studies in natural populations indicate that changes in floral pigmentation, floral scent and flowering time often are under divergent natural selection (Harder and Barrett, 2006).

Taxonomic and geographical distributions of RI-inducing mutations

Another apparent difference between plants and animals is the higher level of intraspecific polymorphism for RI alleles in the former (Rieseberg and Willis, 2007). This is especially true for CMS loci, rf alleles and R genes, which often have extremely restricted geographical distributions (Table 1). The seeming difference between plants and animals might be a consequence of ascertainment bias, as most of the plant genes in Table 1 were cloned for reasons other than their effects on RI. In contrast, the majority of speciation genes in animals were cloned because of their probable role in speciation, and intraspecific polymorphism may have been viewed as a liability. Alternatively, this is a real difference that stems from the greater importance of balancing selection and/or drift in the evolution of plant RI.

Future directions

We have reviewed what is known about candidate speciation genes in plants and identified possible patterns and trends that should be the subject of more rigorous testing in the future. However, there are notable gaps in our knowledge. In particular, with the possible exception of Pa-AN2 (Quattrocchio et al., 1999; Hoballah et al., 2007), none of the studies reviewed here provides a truly comprehensive analysis of a speciation gene. Such an analysis should ideally include functional characterization of the gene and mutation(s) that underlie RI, analyses of the phylogenetic and geographical distribution of the RI-inducing alleles, evolutionary analyses to test for selection and field studies that examine the effects of the allelic variants on RI in natural populations. In addition, more studies are needed of speciation genes in non-model and non-crop systems. This would reduce possible biases or artefacts due to artificial selection and/or a selfing mating system. Studies that identify genes underlying plant reproductive barriers for which no genes are currently known (e.g. ecogeographical isolation, mechanical isolation, conspecific pollen precedence, extrinsic postzygotic isolation) are also sorely needed. Lastly, it would be useful to conduct more rigorous evolutionary analyses of the genes listed in Table 1. Although we found few reports of selection on these genes, in most cases the appropriate tests were not conducted. If such data were collected for multiple genes underlying multiple barriers within a single system, then insights into speciation dynamics – the timing and order in which different types of reproductive isolating barriers arise – may be gained as well.

APPENDIX 1

Descriptions of genes that appear to contribute to reproductive isolation in plants

Genes that contribute to pre-pollination barriers.

A ANTHOCYANIN2 (AN2). Loss of function (LOF) mutations in Pa-AN2, a myb-type transcription factor, are responsible for flower colour differences between the white, hawk-moth-pollinated flowers of Petunia axillaris, and the violet-reddish flowers of P. integrifolia (Quattrocchio et al., 1999). Although all accessions of P. axillaris tested carry LOF mutations in Pa-AN2, the LOF mutations arose independently at least five times and show little evidence of having been favoured by positive selection (Hoballah et al., 2007). This observation is puzzling because pollinator preference experiments using genetic introgressions and transgenics show that the Pa-AN2 mutations have a major effect on pollinator attraction.

B ROSEA1, ROSEA2 and VENOSA. Wild snapdragon (Antirrhinum) species differ in flower colour and intensity – traits that appear to be associated with pollinator attraction and possibly speciation in the group (Jones and Reithel, 2001; Whibley et al., 2006). Three genes underlying differences between at least six different wild species have been shown to be allelic to loci identified from mutations within the garden snapdragon, A. majus (Schwinn et al., 2006). All are Myb transcription factors. ROS1 and ROS2 are tightly linked and are responsible for differences in pigment intensity among Antirrhinum species, whereas VENOSA underlies variation in pigmentation patterning. ROS1 was recently shown to be under divergent natural selection in a hybrid zone between Antirrhinum striatum and A. pseudomajus (Whibley et al., 2006).

C FLAVONOID-3′-HYDROXYLASE (F3′H). Like snapdragon, transitions in flower colour among morning glory (Ipomea) species often appear to be associated with shifts in the pollinator fauna and speciation (Wolfe and Sowell, 2006). Independent transitions from primarily bee-pollinated, blue/purple flowers to red, hummingbird-pollinated flowers in the Mina lineage (which includes I. quamoclit) and I. horsfalliae result from down-regulation in the same gene, Ipo-F3′H (Streisfeld and Rausher, 2009b; Des Marais and Rausher, 2010). In the former case, cis-regulatory mutations in Ipo-F3′H have been demonstrated to be at least partly responsible, but involvement of a trans-acting factor has not been ruled out in I. horsfalliae.

D FLOWERING LOCUS C (FLC). Natural and synthetic allotetraploids of Arabidopsis thaliana × A. arenosa flower later than either parental species due to up-regulation of At-FLC, which is trans-activated by Aa-FRI (Wang et al., 2006a). FLC is a MADS-box transcription factor that represses flowering, whereas FRI is a coiled-coil nuclear protein that positively regulates FLC expression (Michaels and Amasino, 1999). The non-additive effects on flowering in synthetic allopolyploid hybrids are due to divergence of both FRI and FLC during the 60Myr divergence of the parental species. While cis-regulatory mutations in At-FLC and Aa-FLC derived FLC copies appear to be involved in the flowering time divergence of the natural allopolyploid A. suecica, sequence evolution of As-FRI copies is unexamined.

E STYLE2·1. The cultivated tomato, Solanum lycopersicon, is partially reproductively isolated from related species due to a recessed style that results in predominantly self-pollination. Style length is controlled by a short 99-amino-acid protein containing a helix-loop-helix (HLH) motif (Chen et al., 2007). HLH proteins are thought to act as transcription factors. Functional studies indicate that reduced style length is caused by a mutation in the Sl-STYLE2·1 promoter, which downregulates expression during floral development.

Genes that contribute to post-pollination prezygotic barriers.

A S-RNase-SI. In plant groups that exhibit self-incompatibility (SI), the S locus sometimes also causes interspecific or ‘unilateral’ incompatibility. The situation in Nicotiana is typical: species with SI reject pollen from other species, whereas species that are self-compatible (SC) will accept pollen from both SI and SC species. Genetic studies have shown that unilateral incompatibility maps to the S locus and that different S-alleles can sometimes have different impacts on interspecific compatibility (Hancock et al., 2003). As in other plants from the Solanaceae, the functional products of the S locus are S-RNAses (McClure et al., 1989). To verify that S-RNAse genes were responsible for interspecific incompatibility, Murfett et al. (1996) transformed three SC species (N. glutinosa, N. tabacum and N. plumbaginifolia) with S-RNAse-SI genes from an SI species, N. alata. Na-S-RNase-SI contributed to rejection of pollen from all three species, but via different mechanisms. Na-S-RNase-SI alone was sufficient to reject N. glutinosa and N. tabacum pollen, but N. alata also rejected pollen from these species via an S-RNase-independent mechanism. Rejection of N. plumbaginifolia pollen required both S-RNase-SI and other genetic factors from N. alata, including HT-B [a pistil-expressed, extracellular, small asparagine/aspartate (N/D)-rich protein] from N. alata (Hancock et al., 2005). Np-HT-B (a non-S factor) does not appear to be expressed in styles of N. plumbaginifolia, but the genetic factor responsible for reduced expression is unknown.

Genes that contribute to intrinsic postzygotic barriers: hybrid inviability.

A Cf2 and RCR3. Domesticated lines of tomato (Solanum lycopersicon) that lack the fungal resistance gene, Cf2, exhibit weakness or necrosis when two tightly linked and essentially identical copies of the gene are introgressed from a wild species, S. pimpinellifolium (Dixon et al., 1996; Kruger et al., 2002). However, necrosis is not observed when a second gene, RCR3 (an extracellular cysteine protease), from S. pimpinellifolium is also introduced. Sl-Cf2 encodes an extracellular leucine-rich repeat receptor-like protein that activates a hypersensitive response when in the presence of AVR2 protein from the fungal pathogen, Cladosporium fulvum. Cf-AVR2 binds and inhibits Sl-RCR3, apparently leading to a conformation change in the latter that triggers the Sl-Cf2-mediated hypersensitive response (Rooney, 2005). The RCR3 protein from S. lycopersicon differs from Sp-RCR3 by one amino acid deletion and six amino acid substitutions. These changes are speculated to mimic the conformation imposed on Sl-RCR3 by Cf-AVR2 binding, thereby stimulating a hypersensitive response by Sl-Cf2 protein even in the absence of infection (Rooney, 2005).

B DANGEROUS MIX (DM1 and DM2?). Hybrids between different wild accessions of Arabidopsis thaliana sometimes exhibit necrosis due to deleterious epistatic interactions that induce autoimmune responses. Several of the genes underlying these interactions have been cloned and characterized (D. Weigel, Max Planck Institute for Developmental Biology, Germany and K. Bomblies, Harvard University, USA, pers. comm.). In the one published study (Bomblies et al., 2007), mapping and functional analyses show that an allele of an NB-LRR gene family (the largest family of resistance genes in plants) member triggers necrosis when combined with an allele from a second locus (DM2) from another accession. As in the Sl-CF2 example (above), a functional copy of DM1 is missing in the innocuous accession. DM2 has been mapped to an approx. 148-kb region that includes two NB-LRR genes, which can be viewed as strong candidates. Interestingly, the expression of this incompatibility is temperature sensitive, with necrosis observed in F₁ plants at 16 °C but not at 23 °C. However, F₂ plants that are partially or doubly homozygous at DM1 and DM2 suffer from necrosis, suggesting that increased dosage of incompatible alleles can overcome the ameliorating effects of temperature. A species-wide geographical survey revealed that DM1 was locally restricted to a region near Umkirch, Germany.

C HISTIDINOL-PHOSPHATE AMINO-TRANSFERASE (HPA1 and HPA2). HPA encodes a key enzyme in the synthesis of histidine, an essential amino acid. Columbia-0 (Col) and Cape Verde Island (Cvi) accessions of Arabidopsis thaliana contain different functional copies of the essential HPA gene (Bikard et al., 2009). Seed development is arrested in progeny that are homozygous for silenced copies at both genes. A survey of the 30 other accessions indicates that silencing of one or the other gene copy occurred in at least six different ways, including deletions, early stop codons and/or loss of expression. Silencing is widespread so that approx. 25 % of crosses in A. thaliana are likely to exhibit At-HPA incompatibility.

D TRANSPARENT TESTA GLABRA2 (TTG2). Polyploid species are frequently reproductively isolated from their diploid progenitors due to dosage-sensitive incompatibilities in progeny from inter-ploidal crosses. Landsberg erecta (Ler) and Columbia-0 (Col) accessions of Arabidopsis vary in their tolerance to inter-ploidal matings. Genetic analyses of the progeny from Ler × Col recombinant inbred lines crossed with Col indentified a major QTL for interploidal lethality (Dilkes et al., 2008). Fine-mapping and mutant analyses revealed that a maternally expressed WRKY transcription factor, At-TTG2, was mainly responsible for the QTL effects. Sequence and expression comparisons of the Ler and Col alleles at At-TTG2 further imply that cis-regulated differences in expression levels rather than changes in coding sequence are the cause of variability in inter-ploidal hybrid inviability.

E HBD2 and HBD3? Fine mapping studies in near isogenic lines derived from a cross between the indica variety, ‘Habataki’, and the japonica variety, ‘Koshihikari’, showed that hybrid necrosis was caused by an interaction between two unlinked loci (Yamamoto, 2010). The first, hbd2, is a 17-kb region on Lg2 that contains a single predicted gene encoding a casein kinase 1 (CKI1) homologue. Although the two varieties did not differ in HDB2 expression level, the coding sequences differed by a single charge-changing amino acid substitution, and overexpression of the ‘Habataki’ allele in the ‘Koshihikari’ background causes necrosis. The second locus, hbd3, maps to a region in Lg11 that contains an NBS-LRR gene cluster and has a great deal of sequence and structural variation. Although the causal mutations have not been identified, immune response genes are upregulated in the hbd2/hbd3 double mutant and HBD2 overexpression lines, consistent with a role for disease response genes. The causal HBD2 mutation is limited to only a few varieties of O. sativa indica derived from the Indonesian landrace ‘Peta’.

F HWH1 and HWH2? Segregation analyses of recombinant inbred lines of the indica and japonica subspecies of rice (Oryza sativa) showed that hybrid inviability was caused by recessive alleles at two unlinked loci, HWH1 and HWH2 (Jiang et al., 2008). HWH 1 mapped to an interval of 11·8 kb, which contains a single predicted gene, a putative glucose-methanol-choline (GMC) oxidoreductase family protein. Unfortunately, no functional studies have been performed, so it is not clear whether changes in coding versus regulatory regions are responsible for the hybrid incompatibility phenotype. The identity of HWH2 is less clear. Mapping studies place it in a 117-kb region on chromosome 11, which contains 12 predicted genes. One of these, a putative hexose transporter, was put forward by the authors as the best candidate, but no functional data are provided to support this claim. Cultivar surveys indicate that the HWH1 haplotype is fairly common in indica, whereas the HWH2 haplotype is almost fixed in japonica, so this incompatibility should be frequently observed in inter-subspecific hybrids.

Genes that contribute to intrinsic postzygotic barriers: hybrid sterility.

A S5. Hybrids between indica and japonica exhibit a reduction in female (embryo sac) fertility. Map-based cloning and transgenic complementation indicated that sterility is caused by interactions between alleles of an aspartate protease gene (Chen et al., 2008). Thus, it appears to be a rare example of true underdominance. Aspartate proteases are a large family of proteolytic enzymes that contribute to disease resistance signalling and cell death in reproductive tissues in Arabidopsis. Other than its effect on embryo sac fertility, the function of the S5 locus in rice is unknown. Interestingly, a third non-functional allele at this locus restores compatibility between the two subspecies, which might provide a means by which an underdominant mutation could become established. The functional alleles of S5 in indica and japonica differ by two amino acid substitutions in the central domain of the protein, but it is unclear how these might contribute to hybrid sterility.

B SaM and SaF. The Sa locus causes pollen to abort in hybrids between the indica and japonica varieties of rice. A combination of fine-mapping studies and functional analyses by transformation indicate that the Sa locus comprises two adjacent genes, SaM, which encodes a small ubiquitin-like modifier (SUMO) E3 ligase-like protein, and SaF, which encodes an F-box protein (Long et al., 2008). Pollen abortion requires the presence of three alleles: both the indica and japonica alleles at SaM (SaM + and SaM–, respectively), as well as the indica allele at SaF (SaF +). Abortion occurs in pollen grains carrying SaM– alleles, but not those carrying SaM +. The SaM– allele encodes a truncated protein due to a substitution in an intron-splicing site, resulting in negative interactions with SaF + . In contrast, the SaM+ protein contains a self-inhibitory domain that blocks interactions with both SaF alleles, thereby preventing sterility of SaM + pollen. The SaF– and SaF + alleles differ by a single amino acid substitution that does not alter physical interactions with SaM–, but apparently affects male sterility in another way. SUMO proteins are involved in the post-translational modification of other proteins, whereas F-box proteins contain a motif that mediates protein–protein interactions and are frequently involved in signal transduction and cell cycle regulation.

C Nuclear-encoded mitochondrial ribosomal protein L27 (mtRPL27). A chromosomal segment containing mtRPL27 appears to have been duplicated prior to the evolution of the AA genome species of rice (Yamagata et al., 2010). Subsequent to duplication, one copy (on chromosome 8) has been lost in a wild rice species, Oryza glumaepatula. The other copy (on chromosome 4) has lost functionality in domesticated rice, O. sativa, apparently due to a loss of promoter activity. Functional mtRPL27 protein appears to be required for pollen development. As a consequence, first-generation hybrids between O. glumaepatula and O. sativa exhibit complete pollen sterility. The loss of the chromosome 8 copy of mtRPL27 has occurred in all populations of O. glumaepatula and in some populations of O. barthii and O. longistaminata (Yamagata et al., 2010). The taxonomic and geographical distribution of LOF mutations in the chromosome 4 copy of mtRPL27 was not reported.

D Chimeric ORFs in mitochondrial DNA. Rearrangements in plant mitochondrial genomes frequently lead to the formation of chimeric genes, which are the cause of CMS. CMS loci typically include unique, unidentified sequences, as well as portions of one or several mitochondrial genes (see below). Most commonly, plants carrying the CMS locus fail to produce pollen, but sometimes they fail to produce anthers as well. We list 13 well-characterized CMS genes and the mitochondrial gene fragments they are associated with, but our list is by no means exhaustive. References can be found in Table 2.

1. Radish Ogura cytoplasm (ORF138) and Kosena cytoplasm (ORF125), two alleles of the same ORF, associated with ATP synthase subunits.

2. Brassica pol cytoplasm (ORF224), associated with ATP synthase subunits.

3. Brassica nap cytoplasm (ORF222), associated with ATP synthase and NADH dehydrogenase subunits.

4. Brassica tour cytoplasm (ORF263), associated with ATP synthase and NADH dehydrogenase subunits.

5. Moricandia arvensis cytoplasm (ORF108), associated with ATP synthase subunits.

6. Sunflower PET1 cytoplasm (ORF522), associated with ATP synthase subunits.

7. Petunia pcf cytoplasm (ORF402), associated with ATP synthase, NADH dehydrogenase, and cytochrome oxidase subunits, as well as ribosomal protein genes.

8. Maize cytoplasm S (ORF355/ORF77), associated with ATP synthase subunits.

9. Maize cytoplasm T (URF13), associated with ATP synthase subunits.

10. Sorghum A3 cytoplasm (ORF107), associated with ATP synthase subunits.

11. Wheat As or Tt cytoplasm (ORF256), associated cytochrome oxidase subunits.

12. Common bean CMS (ORF239), composed entirely of unique sequence.

13. Rice Boro II cytoplasm (ORF79), associated with ATP synthase and cytochrome oxidase subunits.

Although knowledge of the distribution of the CMS and their (restorers) within and among species is incomplete, Brassica pol, Brassica nap, Maize cytoplasm S and Maize cytoplasm T were identified in intraspecific crosses; Rice Boro II cytoplasm in crosses between subspecies; Brassica tour, Sunflower PET1, Petunia pcf, and Wheat As or Tt cytoplasms in interspecific crosses; and Radish Ogura cytoplasm in an intergeneric cross.

A Nuclear Restorers of CMS. Nuclear loci that restore male fertility in CMS plants are referred to as restorer of fertility (RF) genes. Restorers typically act by regulating the transcript profile and/or protein accumulation of the CMS locus (Hanson and Bentolila, 2004). Thus far, seven RF genes have been cloned and characterized:

(1) Maize RF2. The first restorer gene to be cloned was an aldehyde dehydrogenase that restores the fertility of the URF13 CMS locus in maize (Cui et al., 1996). However, restoration only occurs in the presence of a second restorer gene, Zm-RF1. Zm-RF1 is known to downregulate URF13 expression, whereas Zm-RF2 appears to biochemically compensate for detrimental consequences of residual URF13 expression, possibly by oxidizing toxic aldehyde. A non-restoring Zm-RF2 homologue is characterized by an amino acid substitution in the substrate binding pocket that appears to result in a loss of aldehyde dehydrogenase activity (Liu et al., 2001).

(2) RF-PPR592. The RF gene in Petunia restores fertility by reducing the expression of the PCF CMS locus, leading to a substantial reduction in the amount of PCF protein (Nivison and Hanson, 1989). Map-based cloning indicates that the Ph-RF locus comprises at least two duplicated genes containing pentatricopeptide repeats (Bentolila et al., 2002). The pentatricopeptide repeat (PPR) family is one of the largest gene families in plants and is known to be involved in regulating organelle expression (Hanson and Bentolila, 2004). Transgenic experiments indicate that one of the duplicated PPR genes, RF-PPR592, can restore fertility of plants carrying the PCF CMS locus. A non-restoring homologue of RF-PPR592 has a 530-bp deletion in the promotor region, which appears to account for its lack of expression in floral buds of CMS plants. Sequence analyses indicate that the non-restoring allele, rf-PPR592, probably arose via recombination between the duplicate PPR genes, similar to RF-PPR591 and RF-PPR592.

(3) RFK1/RFO. Two nuclear loci are known to be capable of restoring fertility of the ORF125 CMS locus (the so-called Kosena cytoplasm) in radish. One of these loci, RFK1, has been cloned and shown to be a member of the PPR gene family (Koizuka et al., 2000). There was no change in the expression of ORF125 in restored plants, but the amount of ORF125 protein was reduced. RFK1 differs from a non-restoring homologue by four amino acid substitutions in the region of PPR repeats. RFK1 is allelic to RFO, which restores the Ogura CMS, which is caused by the chimeric mitochondrial gene, ORF138 (Brown et al., 2003; Desloire et al., 2003). Like RFTK1, RFO appears to downregulate ORF138 at the translational or post-translational level (Uyttewaal et al., 2008). RFK1/RFO is flanked by two additional PPR genes, one of which appears to be a pseudogene, implying a role for gene duplication in the evolution of the RFO locus.

(4) Rice RF1A (see below).

(5) Rice RF1B. In rice, co-transcription of mitochondrial genes yields a cytotoxic protein (ORF79) that is responsible for the Boro II (BT) CMS. BT CMS is restored by the Os-Rf1 locus, which is composed of a cluster of PPR genes, two of which can independently restore male fertility: Os-RF1A and Os-RF1B (Wang et al., 2006b). Os-RF1A reduces levels of the ORF79 protein by endonucleolytic cleavage, whereas Os-RF1B degrades ORF79 transcripts. Sequence analyses of non-restorer Os-RF1B alleles revealed nine amino acid substitutions, one of which seems likely to cause a loss of restoration function. Likewise, Os-RF1A alleles were found to encode a truncated protein due to a frameshift mutation.

(6) RETROGRADE-REGULATED MALE STERILITY (RMS). The RMS gene was identified by positional cloning of Os-RF17, which restores the Chinese wild rice (CW)-type cytoplasmic male sterility (Fujii and Toriyama, 2009). The RMS gene encodes a 178-amino-acid protein of unknown function. Sterility is caused by upregulation of RMS, apparently due to cis-regulatory changes.

(7) Sorghum RF1. The Sh-RF1 gene restores fertility of the sorghum A1 cytoplasm, which is the primary cytoplasm used for hybrid seed production. Fine-mapping localized the RF1 gene to a 19-kb region containing three genes (Klein et al., 2005). Two of these were completely conserved between restored and non-restored plants, whereas 19 differences were observed in the region spanning the third gene, which encoded a PPR protein.