The selfing syndrome: a model for studying the genetic and evolutionary basis of morphological adaptation in plants

Abstract

Background

In angiosperm evolution, autogamously selfing lineages have been derived from outbreeding ancestors multiple times, and this transition is regarded as one of the most common evolutionary tendencies in flowering plants. In most cases, it is accompanied by a characteristic set of morphological and functional changes to the flowers, together termed the selfing syndrome. Two major areas that have changed during evolution of the selfing syndrome are sex allocation to male vs. female function and flower morphology, in particular flower (mainly petal) size and the distance between anthers and stigma.

Scope

A rich body of theoretical, taxonomic, ecological and genetic studies have addressed the evolutionary modification of these two trait complexes during or after the transition to selfing. Here, we review our current knowledge about the genetics and evolution of the selfing syndrome.

Conclusions

We argue that because of its frequent parallel evolution, the selfing syndrome represents an ideal model for addressing basic questions about morphological evolution and adaptation in flowering plants, but that realizing this potential will require the molecular identification of more of the causal genes underlying relevant trait variation.

INTRODUCTION

The change from reproduction by outbreeding to selfing is one of the most frequent evolutionary transitions in angiosperms (Stebbins, 1950, 1957, 1974; Barrett, 2002). Because of its frequency and its far-reaching implications for the genetic structure of individuals and populations, this transition has attracted considerable interest amongst taxonomists, ecologists and evolutionary biologists. Important theoretical and empirical work has addressed, for example, the conditions under which predominant selfing can evolve despite the deleterious effects of inbreeding and the genomic and population-genetic consequences of selfing (Charlesworth and Wright, 2001; Barrett, 2010). The most widely accepted ecological scenario favouring the evolution of selfing is a condition of limited pollinator and/or mate availability, for example under low population densities (Darwin, 1876; Stebbins, 1950; Baker, 1955; Lloyd, 1992). Alternative, not mutually exclusive views hold that selection either for rapid maturation in marginal habitats or for reduced predation by herbivores leads to smaller flowers with reduced spatial and temporal separation between dehiscing anthers and receptive stigma. These changes in turn would increase the rate of selfing as an incidental by-product, which could counterbalance the reduced attractiveness of the smaller flowers to pollinators (Guerrant, 1989; Snell and Aarssen, 2005; Eckert et al., 2006). In otherwise self-incompatible taxa, the transition to selfing is frequently associated with the breakdown of the genetic self-incompatibility system that prevents self-fertilization (Busch and Schoen, 2008). Different forms of self-incompatibility systems have evolved in different lineages, and for three main cases the molecular basis has been elucidated using model species (Takayama and Isogai, 2005).

The change from outbreeding (also called xenogamy or allogamy) to autogamous selfing, i.e. selfing within a hermaphroditic flower, is generally associated with a characteristic set of changes to the morphology and function of flowers, termed the selfing syndrome (Darwin, 1876; Ornduff, 1969; Richards, 1986). Predominantly selfing species often have much smaller flowers than their outbreeding relatives that also tend to open less (Fig. 1); they show a reduced pollen-to-ovule ratio, usually due to producing less pollen, and also produce less nectar and scent. The association of these floral traits with a high selfing rate has been demonstrated in a large number of genera, including Arenaria (Wyatt, 1984), Laevenworthia (Lloyd, 1965), Leptosiphon (Goodwillie, 1999) and Mimulus (Ritland and Ritland, 1989). Also in the widely used genetic model species Arabidopsis thaliana, a high rate of selfing is associated with typical selfing syndrome traits, such as a strongly reduced flower size compared with its outbreeding sister species A. lyrata. More recently, a negative correlation between flower size and the rate of selfing was demonstrated in a large taxonomic survey of angiosperms (Goodwillie et al., 2010); a similar negative correlation was observed between the number of open flowers at any time and the rate of selfing, indicating that both components of floral display (number of open flowers and flower size) are reduced in selfing species. However, in contrast to our increasingly detailed knowledge about the genetics and molecular biology of self-incompatibility vs. self-compatibility (Takayama and Isogai, 2005), much less is known about the genetic and molecular basis of the selfing-syndrome traits, despite their pervasive occurrence in angiosperm evolution. The aim of this review is to summarize our current knowledge about the evolution and the genetic basis of key selfing syndrome traits, i.e. the reduced pollen-to-ovule ratio, as well as the reduced flower size and separation between anthers and stigma (herkogamy).

Fig. 1.

Comparison of flowers from outbreeding and selfing species in the genus Capsella. Lateral views of flowers from the outbreeding C. grandiflora (left) and the autogamously selfing C. rubella (right). Note the reduced size of the petals, the reduced opening of the flower and the resulting smaller distance between stigma (arrow) and anthers in the flower of the selfer. See text for more details. Scale bar is 3 mm.

EVOLUTION OF SEX ALLOCATION DURING AND AFTER THE TRANSITION TO SELFING

Evolutionary biologists have long been fascinated by the problem of sex allocation in hermaphrodites. The theory of sex allocation is based on the assumptions of limited reproductive resources and an intrinsic trade-off between the investment in male and female function (Charlesworth and Charlesworth, 1981; Charnov, 1982; Charlesworth and Morgan, 1991; West, 2009). Even though empirical support for such a trade-off in hermaphroditic plants is limited (Campbell, 2000), clear experimental evidence in favour comes from artificial selection experiments, where selection for an increase in pollen or ovule numbers was correlated with reduced allocation to the respective other sex function (Mazer et al., 1999).

Why does sex allocation change during and after the transition to selfing?

In his table of characteristics differentiating outbreeding and selfing species, Ornduff (1969) lists a reduced number of pollen grains in selfers, as well as possibly a reduced number of ovules per flower or inflorescence. The first large-scale empirical study of how the amount of pollen and the number of ovules differ with the mating system was reported by Cruden (1977). When comparing 86 species that were classified in five groups according to their mode of reproduction, ranging from obligate xenogamy (outbreeding) to cleistogamy (self-pollination in closed flowers), the ratio of pollen grains to ovules per flower decreased sharply with an increasing tendency to self-fertilize. Thus, Cruden proposed using the P/O ratio as a more suitable indicator of the breeding system than other floral traits. The interpretation of the decreasing P/O ratio with increased selfing was based on the differential pollination efficiency in outbreeding vs. autogamously selfing species; in self-compatible plants adapted for self-pollination the amount of pollen that reaches a stigma where it is able to effect fertilization is much higher than in obligate or facultative outcrossing species, where fewer pollen grains will reach a compatible stigma. Thus, the high investment in male function found in outcrossing species would become unnecessary in autogamous species, where maximum seed set can be achieved with a much lower number of pollen grains, and resources could be reallocated to female reproductive function.

In contrast, Charnov (1982) interpreted the documented changes in P/O ratio in light of the local mate competition theory. According to this view, how the reproductive return depends on the investment in male or female function, the so-called gain curve, is the critical parameter for the evolution of the P/O ratio. The P/O ratio will change, for example, when the fitness gain from allocating more resources to female function outweighs the fitness loss due to reduced allocation to male function. Subsequent theoretical studies have discussed how the shapes of female and male gain curves are modified by the mating system, and how this will influence sex allocation (for reviews, see Brunet, 1992; West, 2009). Two main predictions from these analyses are that (1) the fitness gain from the allocation to male function will decrease as the selfing rate increases; and (2) in autogamous selfing, allocation to attractive structures will decrease with increased selfing (cf. below the section on changes to flower morphology).

A major issue in studying sex allocation in plants is quantifying the relative investment in male and female function and/or in attractive structures (Charlesworth and Charlesworth, 1981; Queller, 1984; West, 2009). Although, ideally, additional factors should be included, such as the investment per individual ovule or pollen grain, as well as in maternal care after fertilization, most empirical studies have focused on the P/O ratio within individual flowers. Thus, in the following we review available data on changes to the P/O ratio associated with the evolution of selfing.

Empirical studies of changes to sex allocation in selfing species

The reduced P/O ratio in selfing compared with outcrossing species has been confirmed in a large-scale study of >160 taxa in the Polemoniaceae (Plitmann and Levin, 1990). Self-compatible taxa showed a highly significant 5-fold reduction in the P/O ratio compared with the self-incompatible group. However, it is currently not clear whether the reduced P/O ratio was due mainly to a reduced investment in male function and/or also to a reallocation of resources towards female function.

Additional, more detailed studies have focused on individual genera. For example, self-compatible species in a monophyletic sub-group of Solanum section Basarthrum showed a significantly lower P/O ratio per flower than their self-incompatible sister species, which was due to both a reduced number of pollen grains and more ovules per flower (Moine and Anderson, 1992). Similarly, the reduced P/O ratio in autogamous taxa of the genus Clarkia compared with their outbreeding sister taxa is due to both increased ovule numbers and reduced pollen production (Delesalle et al., 2008; Mazer et al., 2009). Thus, these comparisons of species with strongly contrasting mating systems, such as self-incompatible, obligate outbreeders vs. self-compatible, predominant selfers, fully agree with the predictions from the trade-off between investment in male vs. female function and the predicted selection for reduced pollen production in the more efficiently self-pollinating autogamous taxa. This selection may be particularly intense in cases where selfing evolves under conditions of pollinator limitation or low plant density to provide reproductive assurance, as in these cases the fitness gain from investment in male function beyond what is necessary to achieve efficient self-fertilization should be minimal.

However, taxa with lesser differences in the extent of selfing vs. outcrossing complicate the picture. For example, a study of eight taxa of the Mimulus guttatus species complex found that higher selfing rates are associated with reduced P/O ratios, as expected (Ritland and Ritland, 1989); yet, ovule number per flower was also reduced in more strongly selfing taxa, albeit to a lesser extent than pollen production, arguing against a strong negative correlation between the investment in male and female function at the level of individual flowers. Unfortunately, to our knowledge, no data are available on ovule numbers per plant in highly selfing vs. outbreeding Mimulus taxa, so potential compensatory changes in the selfing species, such as a higher overall number of flowers, cannot be assessed. Furthermore, when studying variation of the P/O ratio and selfing rate within populations with mixed mating (instead of between different taxa as in the studies mentioned above), a positive correlation was observed between the P/O ratio and the rate of selfing (Damgaard and Abbott, 1995). These observations are in line with ecologically based mass–action models for the evolution of sex allocation in self-compatible species; assuming that self pollen and pollen from other individuals deposited on the stigma compete for successful fertilization, and that the proportion of pollen that is used for selfing is independent of the total amount of pollen produced, genotypes with more pollen overall would show a higher rate of self-fertilization (Ziehe, 1988; Holsinger, 1991).

Although the negative correlation between investment in male and female function is a mainstay of sex allocation theory, numerous empirical studies have found evidence to the contrary, i.e. a positive correlation between male and female allocation in flowers (see table 1 in Mazer et al., 2007). Indeed, a reversed sign of this correlation was predicted in autogamous compared with outbreeding taxa (Mazer and Delesalle, 1998; Guitián et al., 2003; Mazer et al., 2007). In outbreeding species, a negative correlation between male and female investment is expected because of both competition for limited resources and the ability of gender-biased (i.e. predominantly male or female) phenotypes to achieve a high fitness (Mazer and Delesalle, 1998). In contrast, in exclusively or at least largely selfing species, the relative investment in male vs. female function should be under strong stabilizing selection; for example, producing more pollen than that which is needed to ensure efficient self-fertilization would not significantly increase fitness, and the corresponding genotypes should be counterselected, provided the cost of pollen production is not negligible. Under such stabilizing selection, any change in the number of ovules per flower will cause a comparable change in the amount of pollen produced. Thus, a positive correlation between male and female gamete production is predicted in strongly autogamous lineages. These predictions were tested by artificial selection experiments on pollen and ovule numbers in two sister species of Clarkia with opposite mating systems (Mazer et al., 2007). Indeed, the sign of the genetic correlation between male and female gamete production was reversed in the selfing relative to the outbreeding species; while the selfing species showed a positive correlation, either a trade-off or independence between the investment in male and female function was observed in the outbreeder.

Table 1.

Summary of quantitative–genetic studies of the selfing syndrome

Genus	Crossed species/strains	Mating system (outcrossing rate)	Traits investigated	No. of QTLs found	% of the total phenotypic variance/phenotypic difference between parents explained by all the QTLs	% of the total phenotypic variance/phenotypic difference between parents explained by the strongest QTLs	References
Mimulus	M. platycalyx	Predominantly selfing (0·1–0·2)	Flower width	0			Lin and Ritland (1997)
			Flower length	1	7·8	7·8
			Pistil length	1	28·6	28·6
	M. guttatus	Predominantly outcrossing (0·6–0·9)	Long stamen length	3	38·3	17·7
			Short stamen length	3	33·6	12·4
			Anther–stigma separation	2	21·2	10·7
Solanum (Lycopersicon)	S. esculentum	Selfing	Self-incompatibility	4	46·8	30	Bernacchi and Tanksley (1997)
			Unilateral incongruity	6	72·5	34·3
	S. hirsutum	Self-incompatible obligate outcrosser	Bud type	7	95·7	17·5
			Corolla indentation	2	23·2	12·3
			Rachis length	3	62·2	37
			Flower size	3	29·8	11·4
			Inflorescence vegetative meristem	2	17·9	9·7
			Flower number	1	28·4	28·4
			Stigma exsertion	1	19·7	19·7
Solanum (Lycopersicon)	S. pimpinellifolium LA1237	Selfer (0)	Flowers per inflorescence	3	55·3	29·8	Georgiady et al. (2002)
			Petal length	2	34·6	22
			Total anther length	3	82·2	36·7
	S. pimpinellifolium LA1581	Outcrossing (0·37)	Fertile anther length	2	27·2	19·6
			Sterile anther length	5	42·8	31
			Style length	3	66·1	42·2
Mimulus	M. nasutus	Obligate selfing	Throat width	14	52*	15*	Fishman et al. (2002)
			Corolla width	14	55*	15*
			Tube length	13	71*	30*
	M. guttatus	Predominant outcrossing	Corolla length	11	74*	20*
			Style length	12	78*	20*
			Stamen length	13	116*	55*
			Stigma	15	77*	20*
Leptosiphon	L. bicolor	Predominant selfing (0·01–0·07)	Corolla tube	8	116·54†	37·6†	Goodwillie et al. (2006)
			Corolla lobe length	8	75·26†	21·18†
			Corolla lobe width	6	88·04†	18·86†
	L. jepsonii	Predominant outcrossing (0·01–0·71)	Anther length	7	121·98†	26·56†
			Stigma length	4	115·6†	50·38†

* The percentage of phenotypic difference between parents explained.

^† The percentage of phenotypic difference between parents explained based on F₂ data.

The above notion by Mazer and Delesalle (1998) also implies that firstly, there should be little standing genetic variation in the P/O ratio in selfing species, as this would have been largely purged by stabilizing selection; and, secondly, that within individuals of autogamous species, the P/O ratio will be more stable than for individuals of outcrossing species, as the fitness return on investment in male vs. female function would be largely constant throughout the flowering period of an individual. In contrast, ecological factors such as reduced pollinator service towards the end of the flowering season could select for a temporal shift in sex allocation in outbreeding plants, such as increased male investment in late flowers (Brunet and Charlesworth, 1995). These two implications have been tested in the genus Clarkia by artificial selection and by comparing the temporal variation in ovule and pollen production between selfing and outcrossing sister taxa (Mazer et al., 2007, 2009; Delesalle et al., 2008). Artificial selection on the P/O ratio led to larger changes in the outcrossing compared with the selfing subspecies, suggesting the predicted larger amount of standing genetic variation for the P/O ratio in the former. However, whether variation in the P/O ratio in the selfing subspecies was reduced more strongly than the expected general reduction in genetic variation because of its history of selfing is presently unclear. Also, larger within-individual variation of the P/O ratio in outbreeders was not strongly supported.

Genetic basis of sex allocation

In contrast to the relative wealth of taxonomic data and theoretical work on changes of sex allocation with the transition from outbreeding to selfing, little is known about the genetic basis of the P/O ratio and its modifications. Studies in genetic models such as Arabidopsis thaliana are beginning to shed some light on this issue.

Ovules are initiated as individual primordia from a specialized meristematic tissue within the gynoecium known as the placenta. The number of ovules per gynoecium depends on patterning events that specify the placenta and position primordia outgrowth (reviewed in Skinner et al., 2004). These patterning events resemble those occurring in shoot and floral meristems and involve partly the same genetic components. Several transcription factor genes such as AINTEGUMENTA (ANT), SHOOT MERISTEMLESS and CUP-SHAPED COTYLEDON1 (CUC1) and CUC2 are expressed in the medial domain of the gynoecium that will give rise to the placenta (reviewed in Skinner et al., 2004). Other transcription factors, such as LEUNIG, CRABS CLAW, SPATULA and REVOLUTA, but also the protein kinase TOUSLED or the mitochondrial GTPase SHORT INTEGUMENTS 2 (SIN2), play essential roles in the outgrowth of ovule primordia (Broadhvest et al., 2000; Skinner et al., 2004; Hill et al., 2006). These genes are candidates for underlying the evolution of ovule number per fruit. In particular, disruption of sin2 or ant function increases the distance between ovule primordia and thus decreases ovule number per fruit by 40 and 60 %, respectively (Elliot et al., 1996; Broadhvest et al., 2000).

Pollen grains develop from microsporocytes by meiotic and mitotic divisions. Specification of microsporocytes and surrounding somatic tissues requires extensive cell–cell communication involving various transmembrane receptors; disruption of this signalling process often leads to an excess of microsporocytes at the expense of somatic tissues, suggesting that microsporocyte fate needs to be actively repressed in the prospective somatic supporting tissue (reviewed in Ge et al., 2010). Also, brassinosteroid signalling and chromatin regulation have been implicated in anther and pollen development (Grini et al., 2009; Ye et al., 2010). However, to our knowledge, no mutants have been described that show a higher or lower number of pollen grains without also affecting pollen viability and aspects of anther differentiation.

Thus, elucidating the genetic basis for changes in the P/O ratio between outbreeding and selfing sister taxa promises not only to provide insight into the evolution of this trait, but also to identify novel gene functions in plants.

EVOLUTION OF FLORAL ARCHITECTURE IN THE SELFING SYNDROME

Why does flower morphology differ between related selfing and outbreeding taxa?

Flower display strongly influences the frequency of pollinator visitation and pollen transfer (Sato and Yahara, 1999; Franceschinelli and Bawa, 2000; Hegland and Totland, 2005; Sanchez-Lafuente et al., 2005; Glaettli and Barrett, 2008). For instance, large flowers of the self-incompatible Impatiens hypophylla are visited more often by pollinators than the small flowers of the self-compatible I. microhypophylla (Sato and Yahara, 1999). Also the taxonomic distribution of the types of pollinators visiting the flowers can be affected by flower size and numbers (Glaettli and Barrett, 2008). Therefore, investment in flower display is critical to ensure efficient reproduction in animal-pollinated outbreeding species. Which ecological factors and genetic or developmental constraints drive the reduction in floral display in derived selfing species? At least four, not mutually exclusive, scenarios may explain floral evolution during and after the transition to selfing. These fall into two broad classes outlined in the Introduction: reduced floral display could result either as a by-product of selection primarily for increased fitness through selfing (scenarios one and two); or it could be the primary target of selection, with increased selfing as a consequence (scenarios three and four).

The first scenario is based on the theory of resource allocation. Theoretical models of life history assume limited resources for organisms (Roff, 2002). Thus, after the transition to selfing, particularly when selfing evolves under conditions of pollinator limitation, the energy used to build large flowers would be reallocated to other purposes, for example increased ovule production (for a review, see Brunet, 1992). Models of the evolution of flower display also predict that the finite amount of resources available for floral function will lead to a trade-off between flower number and the investment in each individual flower (Cohen and Dukas, 1990; Sakai, 2000; Sato, 2002). Although it may not be universal, several recent studies have detected this trade-off, which was in some cases correlated with increased selfing rates (Sato and Yahara, 1999; Worley and Barrett, 2000; Caruso, 2004; Sargent et al., 2007; Goodwillie et al., 2010). Other trade-offs, for example between flower size and seed size or number, are also likely to influence the evolution of floral display (Primack, 1987; Sargent et al., 2007). Thus, reallocation of resources could explain the selection of small flowers after the transition to selfing.

The second explanation proposes that the evolution of flower morphology during the transition to selfing, or after the breakdown of self-incompatibility, may have been largely driven by the selection for more efficient self-pollination. Again, this may have been particularly relevant in an ecological context where outcrossing opportunities are limited. For traits such as herkogamy, a direct link to the efficiency of selfing seems plausible. Indeed, among two groups of Gilia achileifolia individuals that differ in the extent of herkogamy, the group with the smaller distance between anthers and stigma showed a higher rate of selfing than the group with strong herkogamy (Takebayashi et al., 2006). A positive correlation between flower size and selfing efficiency could then be explained by a common genetic basis for the control of herkogamy and flower size. Although herkogamy and flower size were only weakly correlated in a study of M. guttatus and M. nasutus, several quantitative trait loci (QTLs) were identified that influenced both traits, hinting at a common genetic basis (Fishman et al., 2002).

Thirdly, the reduced flower size and herkogamy of selfing taxa have been proposed to result from a selection primarily for rapid maturation at the whole-plant and the flower level in ecologically marginal areas (Guerrant, 1989; Snell and Aarssen, 2005). Under this assumption, increased selfing may result from the reduced separation between anthers and stigma, as well as the reduced attraction of the smaller flowers to pollinators. This notion is based on the observation that selfing species often occupy ecologically more challenging environments than their outbreeding sister taxa. For example, selfing species in the genus Leavenworthia occupy drier, more marginal habitats with a shorter vegetation phase than outbreeding species (Solbrig and Rollins, 1977). Similarly, the selfing subspecies of Clarkia xantiana tends to occur in more arid environments than the outbreeding subspecies, and when growing in parallel the selfers will flower earlier and as smaller plants (Eckhart and Geber, 2000). However, at least in the case of Clarkia, subsequent ecological studies concluded that selection for reproductive assurance, not for rapid maturation, may have been the central driving force for the evolution of selfing (Moeller and Geber, 2005).

The fourth scenario considers the influence of herbivory on flowers (florivory) on survival and reproduction (Eckert et al., 2006). Florivory alters floral display, reduces the frequency of pollinator visitation and increases the selfing rate (Krupnick and Weiss, 1999; McCall and Irwin, 2006; Penet et al., 2009). Its influence on plant reproduction varies from neutral to negative (reviewed by McCall and Irwin, 2006). To our knowledge, few data are available on the influence of the flower phenotype on the susceptibility to florivory, and it is unclear whether smaller flowers are indeed less likely to be attacked than big ones. However, it has been hypothesized that the observed higher susceptibility of male and hermaphrodite flowers to florivory may be due to their large attractive display, to the nutritive value of pollen or to a less efficient defence system (for a review, see McCall and Irwin, 2006). Assuming that the first two notions are correct, large flowers full of pollen would be more susceptible to herbivory than the small flowers with low pollen content typical of selfing species.

Answering which of these factors is more likely to have been important in individual cases will require extensive studies of the species’ ecology. In the long run, as more genes are isolated that contribute to the evolution of different selfing syndrome traits it should become possible to study the effects of the relevant trait changes on fitness under different ecological scenarios using more rigorous experiments.

Developmental basis of the reduced flower morphology in selfing lineages

How does flower development differ between related outbreeding and selfing species to generate the differences in final flower size and architecture? In the simplest scenario, flowers of selfing species could either grow at the same rate as the flowers of the related outbreeders, but stop growing earlier, or they could grow for the same period of time, yet with a reduced rate. Several studies have addressed this question. For example, flowers of the selfing subspecies of C. xantiana grow for a considerably shorter period than flowers of the outbreeding subspecies (Runions and Geber, 2000). The rate of floral organ growth was actually faster in the selfers, such that the final difference in flower size was less than what would be expected just from the shortening of the growth period. The same pattern of a shorter growth period, but a faster growth rate in flowers of selfing compared with outbreeding sister taxa was also observed in Limnanthes alba (Guerrant, 1988) and between the outbreeding M. guttatus and the selfing M. micranthus (Fenster et al., 1995). In Mimulus, genetic analysis showed the two effects to be genetically independent, suggesting that selection has acted directly both on the rate and on the duration of growth. However, comparing flower growth in selfing and outbreeding populations of Arenaria uniflora revealed the opposite combination of parameter changes. Here, in the selfers, the rate of flower growth is reduced by almost 2-fold, and these flowers develop over a longer period of time than in the outbreeders (Hill et al., 1992). Thus, in this case also, the effect of the change to one of the two parameters is partly offset by a compensatory change to the other. These two opposing developmental strategies for generating smaller flowers have been suggested to reflect different ecological and evolutionary pressures promoting the outbreeding to selfing transition. A drastic shortening of flower growth and development is consistent with selection primarily for rapid maturation (but see above). In contrast, the evolution of selfing in A. uniflora has been proposed to reflect selection for an escape from pollinator competition with the related A. glabra (Wyatt, 1986).

The genetic basis of the adaptation of flower architecture under selfing

Whether adaptation occurs predominantly through mutations of large or small effect has been debated by evolutionary biologists since Darwin (1859). Based on theoretical work and geometrical arguments about a species’ path to its fitness maximum, Fisher proposed that adaptation generally occurs through the fixation of mutations with small effects, since the limited phenotypic modifications they cause are more likely to be advantageous than the large phenotypic consequences of major mutations (Fisher, 1930). This hypothesis of gradual evolution has been challenged, for example by Gottlieb who reviewed previous work analysing the number of loci involved in the evolution of plant shape and architecture (Gottlieb, 1984). In numerous cases, morphological differences of potential adaptive significance are caused by variation at only one or two loci, suggesting an important role for mutations with large phenotypic effects. More elaborate models reconcile both extreme viewpoints (Orr, 1998). For example, after a large shift in the fitness optimum for a given population, adaptation is expected to occur initially through mutations of large effect that would rapidly move the population towards this new optimum; in a second phase, small-effect mutations would accumulate to refine the population's position relative to the optimum. What can studying the quantitative genetic basis for the evolution of the selfing syndrome teach us about the mutations underlying these floral adaptations?

Variation in floral traits shows high heritabilities, both in outcrossing and in selfing lineages, indicating a strong genetic basis for this variation (Ashman and Majetic, 2006). Initially, biometric studies were used to estimate the effective number of loci contributing to the variation in floral traits between outbreeding and selfing lineages. In several studies, differences in flower size and architecture were found to be due to genetic variation at many loci (e.g. Shore and Barrett, 1990; Holtsford and Ellstrand, 1992). In contrast, it has been estimated that as few as two effective loci can account for the variation in flower size between outbreeding and selfing populations in A. uniflora (Fishman and Stratton, 2004). However, such estimates suffer from a number of limitations, and more accurate information is available from QTL mapping studies (summarized in Table 1).

Lin and Ritland (1997) studied the genetic basis of floral variation between the predominant outbreeder M. guttatus and the predominant selfer M. platycalyx (Lin and Ritland, 1997). In total, ten significant QTLs affecting six different floral traits were detected, accounting for up to 38 % of the phenotypic variance for each trait. While this suggests that most of the variation is explained by a large number of loci with small effect, individual QTLs with a large effect were also detected. For example, one QTL affecting pistil length, a major adaptation to the contrasting mating systems of the two species, explained almost 30 % of the phenotypic variance.

The genetic basis of floral evolution between the large-flowered, mainly outbreeding M. guttatus and another related selfer, M. nasutus, was also studied (Fishman et al., 2002). Selfing in M. natusus is associated with a reduction in corolla size, in the distance between anthers and stigma, and with a reduced P/O ratio. In an interspecific F₂ population, none of the parental phenotypes was recovered, indicating a highly polygenic control of the floral trait variation. This was supported by the identification of 24 QTLs for seven investigated traits, with each floral trait influenced by at least 11 loci. Notably, the two strongest QTLs for stamen length explained 51 and 27 % of the species difference. Thus, except for stamen length, a large number of QTLs with small effects appears to underlie the floral divergence between M. guttatus and M. nasutus.

Work on the highly selfing Leptosiphon bicolor and the partially outcrossing L. jepsonii also supports a complex genetic basis for the evolution of the selfing syndrome (Goodwillie et al., 2006). Leptosiphon jepsonii forms up to 90 % larger flowers than L. bicolor. Although the parental phenotypes could be recovered in the F₂ population, QTL mapping still identified around 25 loci affecting the traits investigated. Individual QTL effects explained between 2 and 28 % of the phenotypic variance, and each floral character studied was influenced by at least six QTLs.

A similar picture – polygenic control of selfing syndrome traits, yet some individual QTLs of large effect – also emerges from QTL mapping in tomato. In the progeny of a cross between the cultivated selfer Solanum lycopersicum (formerly Lycopersicon esculentum) and the self-incompatible obligate outcrosser S. hirsutum a single QTL accounted for 19 % of the variance in stigma exsertion (see below), while a QTL explaining 28 % of the variance was detected for flower number (Bernacchi and Tanksley, 1997). In contrast, for corolla indentation and flower size, only two and three minor-effect QTLs were detected, respectively. A second study dissected the genetic basis for the evolution of flower number and morphology between two sister taxa of Solanum pimpinellifolium that differ in their rate of outcrossing between 0 and 37 % (Georgiady et al., 2002). In general, the parental phenotypes were recovered in the F₂ progeny, suggesting a limited number of loci underlying the divergent flower morphology. Five significant QTLs were detected for the six traits investigated. Four of them were of major effect, each explaining >25 % of the phenotypic variance.

QTL mapping experiments generally underestimate the number of loci that affect a given trait and, as a consequence, overestimate their individual and combined effects, especially when only limited numbers of progeny individuals and markers are used (Mackay et al., 2009). This in turn can bias the distribution of estimated effect sizes for individual QTLs towards higher values (Mackay et al., 2009). Thus, the numbers of the QTLs described above are minimum estimates. Although there is no general consensus on what constitutes a major-effect QTL or mutation, an operational definition holds that major-effect QTLs explain >10 % of the phenotypic variance in a trait (Tanksley, 1993). Despite these caveats, flower morphology appears to have been adapted for increased selfing by the gradual accumulation of many genetic changes of small effect; this conclusion is based on the large numbers of QTLs that were detected and the absence – for many traits – of QTLs that individually explain a large part of the variance. Two plausible explanations for the conflict between this interpretation and models of adaptation involving major-effect mutations have been discussed (Fishman et al., 2002; Goodwillie et al., 2006). In the studies of Mimulus and Leptosiphon, the large flowered species each display strong phenotypic variation in the traits studied; thus, the changes in flower morphology in the related selfing taxa may have evolved by selection of existing alleles with a small effect from the standing genetic variation. Alternatively, the assumption of a fixed fitness optimum in Orr's model (Orr, 2005) may not hold for the studied cases of selfing syndrome evolution. For example, an initial modification of the flower that increased the selfing rate might also favour additional morphological changes that in turn increase the selfing rate further, and so on (Goodwillie et al., 2006). However, major-effect QTLs were detected for some traits, particularly ones such as stamen and style lengths, that would appear to be causally involved in controlling the selfing rate and may have contributed to the initial evolution of a predominantly selfing lineage (Bernacchi and Tanksley, 1997; Fishman et al., 2002; Georgiady et al., 2002). Also, the study of fully selfing and partially outbreeding S. pimpinellifolium varieties (Georgiady et al., 2002) found a small number of large-effect QTLs. These contrasting results may be due to the choice of traits, to the length of time since the chosen pairs of taxa diverged (the longer ago the divergence between two species occurred, the more minor mutations are likely to have accumulated, making it more difficult to detect major-effect QTLs) or to the circumstances under which predominant selfing evolved (e.g. breakdown of genetically encoded self-incompatibility with severe pollinator/mate limitation vs. gradual evolution of increased selfing from a self-compatible, but mainly outcrossing ancestor).

What is the dominance relationship between alleles underlying the evolution of the selfing syndrome?

Theoretical considerations suggest that in outbreeding populations advantageous dominant or semi-dominant mutations will tend to become fixed more easily than fully recessive mutations. However, as the rate of selfing rises, the fixation of a mutation would become increasingly independent of its dominance effect (Haldane, 1927; Charlesworth, 1992). Thus, alleles that increase selfing are expected to show some degree of dominance during the early stages of the transition to selfing, as long as the population still reproduces largely by outbreeding; in contrast, no specific pattern of dominance effects would be expected in later stages of the transition to selfing or when selfing evolves under conditions of severe pollinator/mate limitation such that most reproduction occurs by self-fertilization.

While early genetic studies in Mimulus observed dominance of the ancestral alleles (Macnair and Cumbes, 1989), later studies of selfing syndrome evolution did not observe any specific pattern of dominance (Bernacchi and Tanksley, 1997; Lin and Ritland, 1997; Fishman et al., 2002; Georgiady et al., 2002; Goodwillie et al., 2006). For instance, in the Mimulus genus, Fishman and colleagues observed on average partial dominance of M. guttatus alleles (Fishman et al., 2002), and the alleles from the selfer M. nasutus were recessive at the two directional stronger flower-size QTLs. However, detailed analysis of the dominance relationship at each QTL provided no support for a generalized directional dominance pattern. In contrast, Geordiady et al. (2002) provided evidence for a more general dominance of QTL alleles from the fully selfing parent in their study of contrasting S. pimpinellifolium varieties (Georgiady et al., 2002). Given the paucity of data and our limited knowledge about the circumstances under which selfing evolved in the studied examples, testing the predictions above must await further empirical studies.

Correlated evolution of floral traits: pleiotropy or linkage of causal genes?

As outlined above, the selfing syndrome is characterized by a set of co-occurring morphological and functional changes to flowers. A systematic survey of published studies indicates that there are generally strong genetic correlations between different floral traits (e.g. between primary sexual traits or between attraction traits) in both outcrossing and selfing species (Ashman and Majetic, 2006). Similar correlations were observed in the quantitative genetic studies described above, including correlations between traits affecting different organs, such as corolla width and stigma length or total pollen number (Bernacchi and Tanksley, 1997; Lin and Ritland, 1997; Fishman et al., 2002; Georgiady et al., 2002; Goodwillie et al., 2006). These pervasive genetic correlations could result from pleiotropy or from genetic linkage of the responsible loci. The best evidence for pleiotropic regulatory loci comes from the analysis of floral divergence between M. guttatus and M. nasutus (Fishman et al., 2002). Using multitrait composite-interval mapping, this study identified QTLs that influence several traits, with 19 of the 21 detected QTLs affecting at least two traits and only two QTLs influencing only a single character. However, specific modifications of individual organs are possible, and accordingly several QTLs have been identified that affect the size of only one floral organ, suggesting that organ-specific modifications also contribute to the evolution of the selfing syndrome (Bernacchi and Tanksley, 1997; Lin and Ritland, 1997; Fishman et al., 2002; Georgiady et al., 2002; Goodwillie et al., 2006). The positional cloning of a gene affecting style length in tomato in an organ-specific manner will be discussed below (Chen et al., 2007).

Compared with the strong correlations found amongst floral traits, the correlations between floral and vegetative characters, such as petal and leaf size, tend to be considerably weaker when compared systematically across outbreeding and selfing species (Ashman and Majetic, 2006). The evidence from the QTL studies described above comparing outbreeding and selfing taxa is equivocal. For example, leaf length is strongly correlated with floral tube length and corolla width in M. guttatus and M. nasutus (Fishman et al., 2002). In contrast, comparative QTL mapping studies for variation in leaf and floral organ size between the self-incompatible S. pennellii and the self-compatible cultivated tomato S. lycopersicum indicate that evolutionary variation in leaf and flower growth is controlled by independent genetic modules (Frary et al., 2004). The latter finding is in contrast to the results from mutational studies in model systems, such as A. thaliana and Antirrhinum majus, where almost all mutations that increase or decrease the size of floral organs affect vegetative organs in a similar manner (Breuninger and Lenhard, 2010). These results from model species indicate that growth of floral and vegetative organs is controlled by essentially the same toolkit of genetic components, raising the question of how evolution has specifically modified flower size, while – at least in some cases – leaving leaf size unchanged. Identifying some of the causal genes for the reduced flower size in selfing species will be necessary to determine whether this specificity reflects the modification of hitherto unknown organ-specific regulatory factors or organ-specific changes to the activity of shared growth regulators between leaves and flowers.

To what extent is morphological evolution in plants constrained?

Because of its frequent parallel evolution, the selfing syndrome represents an ideal model to address the extent to which morphological evolution in plants is constrained. Have the same genes and/or pathways been modified in all of the examples where the selfing syndrome has evolved? Or are there many different ways for natural evolution to bring about the morphological and functional changes in question? With only one of the causal genes identified up to now (see next section), insight into this question can only be gleaned from comparing the results of QTL mapping in related pairs of taxa. Independent mapping of QTLs affecting similar floral traits has been performed in the genera Mimulus and Solanum. Unfortunately, for Mimulus the absence of genome sequence information at the time of the studies precludes investigating orthology between the identified QTLs. Comparison of the QTL mapping experiments in Solanum suggests independent evolutionary pathways in different species. Both studies examined the evolution of style length in relation to different levels of selfing in two different crosses of Solanum species (Bernacchi and Tanksley, 1997; Georgiady et al., 2002). While one experiment localized three QTLs on chromosomes 4, 8 and 9, respectively, the other cross identified only one QTL on chromosome 2. Molecular identification of more of the affected loci will allow their involvement in different examples of selfing syndrome evolution to be studied.

Molecular mechanism and history of selfing syndrome evolution

So far only one gene contributing to the evolution of a key selfing syndrome trait has been identified. As mentioned above, a major locus influencing style length, termed style2·1, was found to be responsible for reduced herkogamy in cultivated tomatoes (Bernacchi and Tanksley, 1997; Fulton et al., 1997; Chen et al., 2007). The gene underlying this QTL was localized to a region of 20 kb using a fine-mapping approach (Chen et al., 2007). This region contained one entire gene and the regulatory region of a second gene. Transformation experiments showed that the causal mutation resided in the promoter of this second gene, which encodes a putative transcription factor of 92 amino acids with a helix–loop–helix motif. The long-style phenotype results from a stronger elongation of the cells in the distal part of the style and is associated with a stronger expression of the gene throughout style development. A 450 nucleotide insertion, located roughly 4 kb upstream of the coding sequence, is conserved among the short-style cultivars. Phylogenetic analysis suggested that the evolution of style length followed the breakdown of self-incompatibility.

As this is the only example describing the molecular basis for the evolution of a selfing syndrome trait, it is still too early to draw any general conclusions. However, this example suggests regulatory changes to gene expression as an important contributor to morphological evolution in plants, as has already been documented for several domestication-related morphological traits (e.g. Doebley et al., 1997; Cong et al., 2002).

Candidate genes from model species

Studies in genetic models such as A. thaliana or A. majus have identified a considerable number of genes affecting the size and shape of floral organs (reviewed in Weiss et al., 2005; Krizek, 2009; Breuninger and Lenhard, 2010). These include genes involved in various different cellular processes, such as transcriptional regulation, ubiquitin-dependent proteolysis, hormone responses and intercellular signalling. The majority of these factors act on cell proliferation and generally influence the duration, rather than the rate, of growth. As such, these genes represent candidates for being the targets of evolutionary modification during or after the transition to selfing, and this large body of knowledge will probably be very useful when attempting to isolate the causal genes underlying mapped QTLs for changes to flower size. However, at present it is not clear whether allelic differences in any of these candidates contribute to the changes in flower morphology between related outbreeding and selfing species.

As for herkogamy, little is known about its control in model species. A notable and intriguing exception is a recent study on the RDR6 gene, encoding RNA-dependent RNA polymerase 6 with a function in the trans-acting short interfering RNA (ta-siRNA) pathway (Tantikanjana et al., 2009). Mutations in RDR6 not only enhance the self-incompatibility response in A. thaliana transformed with a functional S-locus, but also increase style exertion and thus anther–stigma separation. Thus, modifications in the activity of RDR6 orthologues or the ta-siRNA pathway may have contributed both to the transition to self-compatibility and to more efficient selfing because of reduced herkogamy in other species.

CONCLUSIONS

Reductions of reproductive structures or functions after the transition to selfing are not restricted to the plant kingdom, but have also been observed in animals (Eberhard, 1996). Evolutionary consequences of autogamy in animals have, for example, been studied in nematode worms (for a review, see Cutter, 2008) and in hermaphroditic fish (Fischer, 1981), with an emphasis on patterns of sex allocation (reviewed in Scharer, 2009). As in plants, it is not entirely clear which selective forces drive the evolution of the selfing syndrome, i.e. the reduction of mating-related traits, such as mating behaviour or sperm size (Cutter, 2008). Another open question concerns the evolutionary path of trait modifications. The available evidence suggests that in nematodes genes with very specific functions and little pleiotropy have been modified in selfing lineages, as illustrated by the changes to copulatory plug formation (for a review, see Cutter, 2008). Whether this is a general trend and one that differentiates the evolution of selfing syndrome traits between plants and animals will require further studies.

Realizing the potential of the selfing syndrome as a model for morphological evolution in plants will require the molecular identification of more genes that have contributed to the relevant trait changes. This in turn will depend on the use of suitable model species, ideally ones where the divergence of a predominantly selfing from an ancestral outcrossing lineage has occurred only recently. Such species should be amenable to genetic analysis and promise to allow a clearer glimpse at the early stages of floral evolution after the transition to selfing. A pair of diploid species in the genus Capsella (Fig. 1) have recently been highlighted as a promising model (Foxe et al., 2009; Guo et al., 2009). The small-flowered, self-compatible C. rubella has evolved from the outcrossing, large-flowered C. grandiflora with an estimated divergence between 20 000 and 50 000 years ago, involving a severe bottleneck. The two sister species are still interfertile, allowing the use of segregating populations for genetic studies. Comparative genomics and the identification of genes underlying floral trait differences between C. rubella and C. grandiflora will help to understand the genetic basis and evolutionary path of selfing syndrome evolution. The knowledge gained should ultimately be compared with other genera to be able to draw general conclusions on the evolutionary trajectories of reproductive functions during and after the transition to selfing.