Sexual and asexual (apomictic) seed development in flowering plants: molecular, morphological and evolutionary relationships

Additional keywords: apomixis, evolution, Hieracium, ovule, reproduction.

The angiosperms, or flowering plants, make up the largest number of species in the plant kingdom and dominate most terrestrial environments. They span diverse body plans and growth forms and are generally distinguished from other seed bearing plants (gymnosperms) by several key features. These include the presence of flowers, which comprise a perianth of attractive structures (e.g. petals) around the reproductive organs, and ovules that are enclosed in carpels (female sporophylls that after fertilisation of the ovule form part of the fruit) rather than lying bare on the surface of a cone. Many angiosperm flowers are hermaphroditic, or bisexual, meaning that the female and male reproductive organs are united in one structure instead of being located in different structures. Furthermore, during seed formation, double fertilisation leads to the production of the embryo that is the progenitor of the seedling, and also the triploid endosperm, which is often considered to be an important feature of the angiosperms since it provides additional nutrition to the developing embryo.

The reason for the success of the angiosperms, however, has been a remarkably controversial topic, mainly because it is still unclear when and from what ancient plant lineages they evolved (Theissen and Melzer 2007; Crepet and Niklas 2009). Interestingly, Darwin considered this was an ‘abominable mystery’ (Darwin 1903), and it is yet to be solved despite many years of research and numerous hypotheses. Although the most favoured hypothesis suggests that angiosperm diversification and success is linked to the relationship between flowering plants and the animal species that facilitate pollination or seed/fruit dispersal, there are alternatives, elegantly summarised by Crepet and Niklas (2009). These consider features such as vegetative and reproductive attributes, phenotypic plasticity and combinations of structural, genetic and chemical characteristics. In any case, the advent of angiosperms and their subsequent ecological success most likely reflects the synergy among many functional traits that allowed them to break free from the developmental constraints experienced by their gymnosperm progenitors (Crepet and Niklas 2009).

Seed development in the angiosperms requires a series of defined steps. First, a transition from a vegetative plant growth mode to a reproductive phase results in the formation of the floral organs such as petals, and the male and female reproductive organs, the stamens and carpels (Fig. 1). Meiosis and gamete formation occur in the ovules of the carpels and anthers of the stamen. Fertilisation begins with pollination of the stigma and concludes with double fertilisation within the embryo sac of the ovule. In this process, two sperm cells are transmitted to the ovule via a pollen tube and released into the embryo sac (female gametophyte) where they fuse with the egg cell to give rise to the embryo and with the central cell to produce the endosperm (Fig. 1).

Of all the structures involved in seed development, the ovule is probably the most vital, since it is physically connected to the maternal plant and thus a supply of hormones and nutrients, and contains specific tissues called the integuments, which eventually form the seed coat and protect the seed during its development and dispersal. The ovule also supports growth of the embryo sac, which is formed from a specific ovule cell that transits through multiple developmental states including cell differentiation, meiotic and mitotic division, cell selection and cell death (megasporogenesis and megagametogenesis; Fig. 1). Most angiosperms follow the Polygonum-type of embryo sac development to produce a seven-celled, eight-nucleate embryo sac (Willemse and van Went 1984), and although the details may vary between species, the basic sexual pathway requiring meiosis and fertilisation for seed development is highly conserved. Sexual reproduction supports the generation of diverse progeny through meiotic recombination and the union of male and female gametes. Combined with the evolution of flowers that help to attract pollinators, this is likely to be another contributing factor to the success of the angiosperms.

While the majority of flowering plants rely upon the events of sexual reproduction to form a seed, over 400 species from 40 different families including both monocots and eudicots, have evolved an asexual form of seed formation termed apomixis. Despite deviating significantly from the sexual pathway, apomixis occurs without compromising seed viability. During apomictic reproduction, the embryo is formed from a somatic cell, independently of meiosis and fertilisation, thus resulting in seedling progeny that are identical to the maternal plant. Several types of apomixis have been identified in plants that differ based on the origin and location of the somatic cell that gives rise to the embryo (Fig. 1). The embryo can form within an embryo sac produced entirely from mitotic events (apospory or diplospory) or directly from somatic cells surrounding the developing sexual embryo sac (adventitious embryony). In some cases, fertilisation is required for endosperm development and subsequent seed viability, but in others it is not, and the endosperm, similar to the embryo, can develop autonomously. Furthermore, most apomictic plants are facultative, meaning that sexual reproduction is not completely eliminated. Apomixis and sexual reproduction can take place in the same plant, and even in the same ovule, giving rise to both maternal apomictic and hybrid progeny. The occurrence of apomixis in different forms and in unrelated families suggests that the apomictic pathway has evolved multiple times during plant evolution.

Apomixis is also a trait of considerable interest for agriculture, since in contrast to sexual reproduction, which typically results in the segregation of traits, apomixis results in the production of clonal progeny through seed. Theoretically, if apomixis could be employed in hybrid seed production it could fix hybrid vigour and allow for the stable fixation of many advantageous traits over many seed generations. Apomixis is noticeably absent from most crop plants, and strategies aimed at integrating apomixis into crops via hybridisation from the few known wild relatives have failed. Therefore, the molecular characterisation of the apomictic process in natural species has attracted considerable interest.

Recent reviews have focussed on the genetic inheritance of apomixis (Ozias-Akins and van Dijk 2007), the polyploid nature of most apomictic plants (Mogie et al. 2007) and the costs and benefits of apomixis and sexual reproduction from ecological perspectives (van Dijk 2007). Other reviews have focussed on the maternal and paternal genomic contributions to sexual seed formation and viability and speculated on how apomixis might overcome these requirements (Grossniklaus et al. 2001; Köhler and Grossniklaus 2005). Here, we consider apomixis and its origin by comparing the morphological and molecular relationships of sexual and apomictic pathways from a developmental perspective. We will focus our attention predominantly towards the earliest stages of sexual and apomictic reproduction in the ovule. These are the stages in sexual reproduction where the cells competent to undergo meiosis differentiate and the embryo sac, which houses the egg cell, is formed and fertilised, thus perpetuating the processes of angiosperm evolution and adaptation. At these same ovule stages in most apomictic plants, reproductive development is diverted towards the mitotic production of embryo sacs from somatic cells and the differentiation of eggs that are identical in genotype to the maternal plant; steps that are essential for the generation of clonal progeny.

In the case of Arabidopsis, a sexually reproducing model angiosperm, the ovules are aligned in linear files within the developing carpels and are relatively small, easily accessible to mechanical dissection and compatible with simple clearing methods (Schneitz et al. 1995). Ovule structure often varies between species, but typically consists of only a few pattern elements or cell types on the proximal-distal axis. In Arabidopsis, these include the nucellus, which gives rise to the germ-cell progenitors, the chalaza (meaning ‘hard lump’ or ‘hailstone’ in Greek), which separates the nucellus from the funiculus and gives rise to the inner and outer integuments, and the funiculus, which is a stalk-like structure that contains vascular strands and connects the ovule to the placenta of the flower (Fig. 1). Similarly in the model monocot rice (Oryza sativa L.), two integuments form from the chalaza, but the funiculus is rudimentary and the chalaza seems to be attached directly to one side of the ovary wall (Itoh et al. 2005). A recent review summarises early stages of ovule development and its evolutionary conservation (Colombo et al. 2008).

The first stage of female reproductive development takes place within the ovule when an archesporial cell differentiates from sub-epidermal cells of the nucellus (Fig. 1). In the majority of angiosperms, this cell can be recognised by its elongated size relative to the surrounding cells, a large nucleus and vacuole and its position neighbouring the epidermis. Although a distinction is made between the archesporial cell and megasporocyte/megaspore mother cell (MMC), in Arabidopsis the archesporial cell directly functions as the MMC (Webb and Gunning 1990; Reiser and Fischer 1993). This is in contrast to the majority of angiosperms where the archesporial cell often undergoes a periclinal division, and subsequently the inner cell differentiates into the MMC.

Once the MMC has differentiated from the archesporial cell, it continues to expand and, in the case of Arabidopsis, eventually occupies much of the space at the apex of the nucellus. Prior to the initiation of meiosis, the cell exhibits a large nucleus and polarisation of organelles, such as the plastids, which are predominantly located at the chalazal end of the cell, and the rough endoplasmic reticulum, which are predominantly located in the cytoplasm at the micropylar end of the cell (Bajon et al. 1999). Although the reason for this polarisation is unclear, it may lead to the differential partitioning of cellular components in the megaspores after meiosis, hence influencing their survival (see below). Intimate details of female meiosis in Arabidopsis have been described in several studies (Webb and Gunning 1990, 1991, 1994). During meiosis I, the MMC nucleus divides transversely giving rise to a binucleate intermediate, which depending on the species and type of embryo sac development (Bouman 1984), typically comprises two dyad cells. Meiosis II proceeds rapidly and leads to the production of four haploid megaspores that can be arranged in a linear (Fig. 1) and/or multiplanar tetrad. In Arabidopsis, which develops the most common form of embryo sac structure found in angiosperms, the monosporic Polygonum-type, the spatial arrangement of megaspores varies between ovules, but in all cases the megaspore located closest to the chalazal region of the ovule enlarges while the other three rapidly degenerate. At maturity, the functional megaspore is tear-shaped, rich in organelles that are dispersed throughout the cytoplasm and contains small vacuoles (Webb and Gunning 1991). By contrast, the degeneration of the non-functional megaspores is accompanied by an accumulation of ribosomes, dicotysomes and lipid granules, vesicles and autophagic vacuoles and low cellular activity (Willemse and De Boer-De Jeu 1981). Expansion of the functional megaspore, sometimes referred to as the one-nucleate embryo sac, progressively displaces the degenerating spores towards the micropylar end of the ovule before initiating the first mitotic division of megagametogenesis (Fig. 1).

Most molecular studies of ovule development have focussed on dicot species such as Arabidopsis and Petunia, and these have identified several genes associated with primordium differentiation and ovule identity (for a summary, see reviews by Grossniklaus and Schneitz 1998; Skinner et al. 2004; Colombo et al. 2008). More recently, studies in rice have revealed novel details of the female reproductive pathway in monocots. Combined, results from these different systems have elaborated specific details of early ovule development and reproductive cell identity (Fig. 2).

Development of the MMC appears to be under the control of several pathways that either restrict or promote its differentiation. In Arabidopsis mutants lacking function of the SPOROCYTELESS/NOZZLE (SPL) gene, a proportion of ovules fail to produce a mature MMC from the archesporial cell (Schiefthaler et al. 1999; Yang et al. 1999), suggesting that SPL acts to promote MMC differentiation. SPL encodes a plant-specific nuclear protein and is expressed in various plant tissues, including the nucellus, archesporial cell and MMC. Since spl mutant ovules also show defects in nucellar identity and integument growth (Schiefthaler et al. 1999; Balasubramanian and Schneitz 2000; Sieber et al. 2004), it is unclear whether SPL functions in multiple independent pathways or primarily in ovule patterning thereby influencing MMC specification. Interestingly, ovules lacking function of the WUSCHEL (WUS) gene, which encodes a homeodomain transcription factor that acts downstream of SPL, also display defects in MMC differentiation and integument growth, but in contrast to spl ovules, appear to develop a normal nucellus (Groß-Hardt et al. 2002; Sieber et al. 2004). Therefore, WUS, which is a key regulator of cell-differentiation in meristems and anthers (Laux et al. 1996; Deyhle et al. 2007), may act as a specific component of a MMC differentiation pathway promoted by SPL. Recent studies of a dominant spl mutant (spl-D) suggest that SPL also acts to repress expression of YUCCA genes, which encode for flavin mono-oxygenases and are key players in synthesis of the plant hormone auxin (Li et al. 2008). Although this function is yet to be shown in ovules, auxin responses have previously been shown to influence ovule integument growth and female fertility (Schruff et al. 2006; Wu et al. 2006). Thus, a link is emerging between known regulators of cell differentiation and region-specific auxin responses that might influence reproductive cell fate during ovule development.

In angiosperm ovules, usually only one cell differentiates from the nucellus to undergo meiosis and give rise to the embryo sac. Recent studies in rice suggest that a signalling mechanism acts to restrict MMC competence to this one cell. The multiple sporocyte (msp1) mutant was identified as a male sterile mutant and found to contain an insertion of the Tos17 endogenous retrotransposon in a gene encoding for a leucine-rich repeat receptor–like protein kinase (LRR-RLK; Nonomura et al. 2003). msp1 mutants display an increased number of both male and female sporocytes, which in the case of the ovule can give rise to multiple disorganised, but occasionally viable, female gametophytes that produce germinable seeds after fertilisation with wild-type pollen. In the ovule, MSP1 mRNA is detected in cells surrounding the MMC, but not in the MMC itself, suggesting it is part of an intercellular signalling mechanism regulating sporogenic cell fate. This mechanism may be conserved in other monocots, since the male sterile multiple archesporial cell (mac1) mutant in maize (Zea mays L.) also leads to the production of extra sporocytes in both ovules and anthers (Sheridan et al. 1996; Sheridan et al. 1999), which in the case of the ovule, are capable of undergoing meiosis and embryo sac development. The identity of the mac1 gene is yet to be reported, but the recessive, sporophytic nature of the mutation suggests that it is also likely to act as part of a signalling mechanism from surrounding tissues that restricts MMC fate. Homologues of MSP1 (EXTRA SPOROGENOUS CELLS/EXCESS MICROSPOROCYTES1; EXS/EMS1; Canales et al. 2002; Zhao et al. 2002) have also been identified in the Arabidopsis genome and were shown to regulate aspects of microspore fate during pollen development similar to that reported in maize and rice. However, it is not known if these mutants display a multiple-MMC phenotype. Therefore, the conservation of RLK/MSP1-like signalling pathways regulating multiple MMC formation in monocot and eudicot species remains to be determined.

How might the MSP1 signalling mechanism function? LRR-RLKs are highly represented in plants (Shiu and Bleecker 2001) and are identified as transmembrane proteins with putative N-terminal extracellular domains and carboxyl-terminal intracellular kinase domains. RLKs are usually the starting point of a complex array of signalling pathways that depend on the binding of a ligand to the LRR extracellular domain, and have been shown to function in diverse developmental events including cell differentiation and pathogen defence (Diévart and Clark 2004). One well characterised example is the Arabidopsis CLAVATA1 (CLV1) LRR-RLK that binds the stem cell specific CLAVATA3 (CLV3) signal peptide and regulates meristem development (Brand et al. 2000; Schoof et al. 2000; Ogawa et al. 2008). In the absence of CLV1 function, or that of its CLV3 ligand, the size of the Arabidopsis meristem increases drastically due to an over-accumulation of stem cells and increased activity of WUSCHEL (Brand et al. 2000; Schoof et al. 2000). In Arabidopsis, the EXS/EMS RLK protein binds the putative ligand TAPETUM DETERMINANT (TPD1; (Ma 2005; Jia et al. 2008) and functions to restrict the number of male sporocytes. A recent report suggests that the homologue of TPD1 in rice, TAPETUM DETERMINANT-LIKE (TDL1A), may be the ligand for MSP1, since MSP1 and TDL1A interact in yeast-2-hybrid and bimolecular fluorescence experiments and TDL1A RNAi lines show a similar phenotype to msp1 mutants in the ovule (Zhao et al. 2008). TDL1A mRNA is also detected in the ovule cells surrounding the developing MMC.

Collectively, these findings identify the importance of inter-regional communication in specification of the MMC. At least two pathways appear to influence MMC specification in angiosperm ovules, one of which promotes the differentiation of an MMC from the archesporial cell and another that restricts sporogenous identity to a single nucellar cell (Fig. 2). The latter, in monocot rice and possibly maize, appears to comprise a LRR-RLK signalling pathway. Further details of the components of this signalling mechanism, particularly the final downstream targets and whether they are expressed in the MMC, will be useful in characterising the first step of sporogenous development. Although this pathway has yet to be extended to eudicot ovules, the observation that multiple MMCs can occasionally be detected in Arabidopsis ovules (U. Grossniklaus, pers. comm.) suggests that a similar mechanism may be in place.

Once the MMC has differentiated in the nucellus, it enters a phase of meiotic division leading to the production of four reduced megaspores. Although it is unclear on a molecular level why the MMC alone enters a meiotic pathway, several genes have been described that influence the progression of megasporogenesis. Most of these also play a role during early pollen development (microsporogenesis), suggesting they play a conserved role during meiotic reduction. Accordingly, some show homology to factors important for meiosis in yeast, but others are plant specific, and recent reviews provide a detailed summary (Yang 2006; Mercier and Grelon 2008). Several genes that function during megasporogenesis in rice and Arabidopsis are of particular interest, based on the identity of the proteins they encode and their loss-of-function phenotypes.

The MEIOSIS ARRESTED AT LEPTOTENE1 (MEL1) gene from rice was identified in a screen for seed-sterile rice plants (Nonomura et al. 2007). In mel1 mutants, both female and male sporogenesis is disrupted; pollen mother cells (PMCs) arrest during meiosis I and MMCs arrest at the pre-meiosis, meiosis or tetrad stages. In the ovule, MEL1 mRNA is first detected in a pool of hypodermal cells within the nucellus before becoming restricted to the MMC, but is notably absent during meiosis. Similar expression is observed in the anthers. Therefore, the initial expression of MEL1 marks a group of cells that, taken together with the MSP1 data discussed above, are likely to have competence to initiate sporogenous development. Interestingly, based on sequence homology, MEL1 encodes a putative member of the ARGONAUTE (AGO) protein family, which is conserved in plants and animals and has been implicated in many developmental processes. AGO proteins function to bind small RNA (sRNA) molecules and regulate expression of complementary RNA targets through either mRNA cleavage, translational repression or chromatin modification (Vaucheret 2008). The rice genome encodes 18 AGO proteins and the Arabidopsis genome encodes 10, of which AGO5 shows highest homology to MEL1. Publicly available microarray data (Arabidopsis eFP browser; http://bbc.botany.utoronto.ca/efp/cgi-bin/efpWeb.cgi) show that AGO5 expression is restricted mainly to reproductive tissues in Arabidopsis, suggesting that it may have a similar function to MEL1. Although the developmental function and expression of AGO5 on the cellular level remain unknown, two closely related characterised Arabidopsis genes are AGO1, which regulates leaf polarity, resistance against pathogens and shoot development (Bohmert et al. 1998) and AGO10/ZWILLE/PINHEAD (Moussian et al. 1998; Lynn et al. 1999), which regulates stem cell maintenance by promoting WUS function from embryonic vascular cells (Tucker et al. 2008). Based on the nature of the MEL1 protein and the phenotype of mel1 mutants, it is likely that MEL1 regulates the expression of mRNA targets via small RNAs in young sporogenous tissues and this is critical for the completion of normal sporogenesis and meiosis. The further characterisation of MEL1, its targets and the sRNA molecules through which it functions will provide novel insight into the mechanisms of megasporogenesis in plants.

Another gene required for megasporogenesis (and microsporogenesis) is the Arabidopsis gene DYAD/SWITCH1 (SWI1), which is expressed in female and male meiotic cells and functions in sister chromatid cohesion and centromere organisation (Mercier et al. 2001). In swi1 mutants, a single equational division of the MMC occurs in place of normal female meiosis, but subsequent development is aborted. In the majority of swi1 alleles, pollen development is also disrupted leading to male sterility. In this respect, the SWI1 gene is not so different from other factors that are crucial for the progression of normal plant meiosis. However, one specific allele of swi1, called dyad, shows a less extreme phenotype and while female meiosis is similar to that in other swi1 alleles, male gamete formation is normal and female gametes can be produced at low frequency (Siddiqi et al. 2000). This leads to the production of viable seeds in a few cases. Subsequent analysis of these seeds showed that a small proportion of the few functional female gametes produced in dyad do so without meiotic reduction, as evidenced by the presence of triploid plants amongst the progeny of selfed dyad mutants and maintained parental heterozygosity (Ravi et al. 2008). This phenotype is particularly interesting because the bypass of meiosis and production of unreduced gametes resembles a mechanism called diplospory found in some natural apomictic plants (see below; Koltunow et al. 1995a; Koltunow and Tucker 2008).

Although megasporogenesis has been the topic of many morphological and ultrastructural studies, the regulation of megaspore selection and fate is not clearly understood, mainly due to the inaccessible nature of the megaspores in the young ovule. In the majority of angiosperms, only one of the megaspores produced during meiosis continues to develop and form an embryo sac while the others degenerate. One report suggests that megaspore survival is influenced by function of the ANTIKEVORKIAN (AKV) gene in Arabidopsis, because akv mutants show a proportion of ovules (~10%) with all four megaspores surviving (Yang and Sundaresan 2000). However, the molecular identity of the AKV gene has yet to be revealed.

Ultrastructural studies have shown that cellular components such as organelles and plastids are polarised during MMC development and meiosis (Bajon et al. 1999), which may contribute to differential patterning of the megaspore tetrad during, or even before meiosis. Alternatively, the functional megaspore may simply reside in a privileged position within the ovule, thereby gaining access to higher concentrations of nutritional or regulatory molecules that enhance its survival. In either case, some positional information seems to play an important role.

It has also been suggested that the cause of megaspore degeneration is a type of programmed cell death (PCD) or apoptosis (Wu and Cheung 2000; Yang and Sundaresan 2000). Plant cells undergoing PCD show distinguishing features such as cell shrinkage, nuclear DNA degradation, loss of mitochondrial membrane potential, cytochrome c release from mitochondria and induction of caspase-like activity (García-Heredia et al. 2008). The three degrading megaspores show some of these characteristics. TUNEL (TdT-mediated dUTP nick-end labelling) assays are a common method for detecting DNA fragmentation resulting from apoptotic signalling cascades (Gavrieli et al. 1992). In alfalfa (Medicago sativa L.) ovules, TUNEL assays showed that DNA fragmentation was mainly localised to the degenerating megaspores during megasporogenesis (Citterio et al. 2005). In the same study, several MPS-ONE-BINDER (MOB1) genes that encode cell-cycle associated proteins were identified and found to be expressed in the degenerating megaspores (Citterio et al. 2005). In Drosophila and mammalian cells, MOB-like proteins regulate several factors that are involved in cell proliferation and apoptosis (Hirabayashi Nakagawa et al. 2008). Another protein associated with cell death during reproduction and disease resistance is aspartic protease (Xia et al. 2004; Ge et al. 2005). In rice ovules, expression of the S5 aspartic protease is detected in the megaspores and nucellar cells adjoining the megaspores, and loss of function leads to a reduction in female fertility (Chen et al. 2008), suggesting that S5 may regulate megaspore fate.

Changes in dynamic calcium (Ca²⁺) concentrations have also been implicated in PCD of plant and animal cells (Yamaguchi et al. 1999; Canzoniero et al. 2004). To determine the relationship between calcium levels and megaspore death in lettuce (Lactuca sativa L.) ovules, free Ca²⁺ levels were examined by histological staining with potassium pyroantimonate (Qiu et al. 2008). Calcium precipitates were rarely detected in the cytoplasm and nucleus of the MMC, but accumulated to similar levels in the linear tetrad of megaspores after meiosis. Interestingly, during the sequential degeneration of the three most micropylar megaspores, calcium precipitates were observed in the cytoplasm but not the nucleus of the degenerating spore, followed by an increase in calcium levels in the nucleolar vacuole of the adjoining megaspores. After degeneration of the last non-functional megaspore, numerous calcium precipitates were noted in the cytoplasm and nucleus of the functional megaspore (Qiu et al. 2008). Although the functional relevance of these changes in calcium distribution is uncertain, the flux of free calcium into the degenerating spores appears to foreshadow imminent degeneration, suggesting that Ca²⁺ may play a role in signalling between the megaspores.

Finally, one of the most distinguishing morphological features observed during megasporogenesis in the majority of flowering plants is the accumulation of callose (β-1,3-glucan), a plant polysaccharide, in the cell walls of the MMC and megaspores. Callose deposition patterns vary between species (Rodkiewicz 1970) but in most cases, callose accumulates at one pole of the mature MMC and then in the transverse walls that separate the megaspores during meiosis. After the completion of meiosis, callose tends to persist in the part of the cell wall of the functional megaspore that is closest to the degenerating megaspores and diminishes elsewhere, possibly due to the activity of β-1,3-glucanase enzymes that target callose for degradation (Hird et al. 1993; Tucker et al. 2001; Levy et al. 2007). Whether the presence of callose or its deposition in a particular pattern around the megaspores influences their development remains unclear. In the case of pollen development, callose also accumulates around the developing microspores, and alterations in callose deposition due to the precocious activity of β-1,3-glucanase enzymes affect pollen viability but not necessarily meiosis or cell division (Worrall et al. 1992; Scott et al. 2004). Since callose is also deposited in plasmodesmata, pollen tubes and other plant cell types it is not a specific marker for the events of meiosis (Stone and Clarke 1992; Donofrio and Delaney 2001; Rinne et al. 2001), but it may play a common role by acting as a physical barrier to reduce the free space available for the passage of molecules and/or microbes (Stone and Clarke 1992; Radford et al. 1998). In the ovule, callose may act as a molecular or nutritional filter decreasing the permeability of the cell wall, thus enabling the megaspores to embark upon an independent course of development compared with diploid sporophytic surrounding tissues (Bouman 1984). If this is the case, what are the molecules that callose acts to restrict? Is it keeping signals contained within the megaspores or acting to prevent sporophytic signals from gaining entry? Finally, if callose was not present around the MMC or megaspores, what effect would this have on the fate of the megaspores and/or adjoining cells? These questions have yet to be answered in any detail, but molecular techniques now make it possible to modify callose deposition patterns in a temporally and spatially restricted manner and to characterise the downstream effects.

So far in this review, we have considered the initial steps of the female sexual pathway from a morphological and molecular perspective. Although it is clear that many regulatory details of this pathway are missing, considerable progress has been made towards identifying key signalling processes that influence reproductive cell fate. These details are of particular interest from a developmental perspective because they provide starting points in the search for upstream regulators and downstream targets. They are also of interest because in several cases, the loss of these signalling pathways leads to phenotypes that resemble early steps in apomictic seed development. It is possible that similar genetic changes may facilitate asexual seed development in wild apomicts. In the next sections of this review, we will address the apomictic process by asking several basic questions: what are the key events that take place during the initiation of apomixis, what is the relationship between apomictic and sexual reproduction and what did apomixis evolve from?

In the simplest form of apomixis, referred to as adventitious embryony, sexual seed development initiates normally in the ovule and leads to the production of an embryo sac via meiosis and mitosis. However, concomitant with sexual development, multiple somatic nucellar or integumentary cells surrounding the embryo sac spontaneously initiate embryogenesis. Growth of these embryos to maturity appears to be dependent upon fertilisation of the sexual embryo sac, possibly because the fertilised endosperm provides important nutrients and/or growth signals (Lakshmanan and Ambegaokar 1984; Asker and Jerling 1992; Naumova 1993; Koltunow et al. 1995b). Therefore, seeds produced in this manner in plants such as mango and citrus often contain multiple embryos of both maternal and hybrid origin.

Species such as Taraxacum officinale L. (van Dijk et al. 1999), Boechera holboellii Hornem. (Schranz et al. 2005), Tripsacum dactyloides L. (Grimanelli et al. 1998) and Erigeron annus L. (Noyes 2000) follow the type of apomixis referred to as diplospory. In these plants, female sexual development appears to initiate normally with the differentiation of a MMC from the nucellus. However, dependent on the species, this MMC fails to initiate the events of meiosis and immediately begins mitosis (referred to as mitotic diplospory; e.g. T. dactyloides) or enters the first steps of meiosis before aborting and restarting development on a mitotic program (referred to as meiotic diplospory; e.g. E. annus). In both cases, meiotic crossovers between chromosomes are avoided and mitotic divisions of the MMC or its product give rise to an embryo sac without meiosis (apomeiosis) containing an egg-like cell that initiates embryogenesis without fertilisation. Endosperm development in diplosporous species is either autonomous, meaning that it initiates without fertilisation, or more commonly pseudogamous, meaning that it requires fertilisation of the central cell. Because diplospory initiates from cells that were previously ‘sexual’ in origin, there is essentially no interaction between sexual and apomictic cell types in diplosporous plants. However, this does not preclude diplosporous apomicts from producing seeds via the sexual pathway; facultative apomixis is clearly observed in diplosporous Tripsacum and Boechera species, presumably because apomixis fails to initiate in some ovules (Grimanelli et al. 1998; Schranz et al. 2005).

A third type of apomixis, referred to as apospory, occurs in species such as Ranunculus auricomus L. (Nogler 1984), Pennisetum squamulatum Fresen. (Roche et al. 1999) and Hieracium piloselloides Vill. (Koltunow et al. 1998). In ovules from aposporous apomicts, the sexual process typically initiates with differentiation of the MMC and, in the case of Hieracium, often continues to the completion of meiosis. Concurrent with these early stages of sexual development, one or more somatic cells in close proximity to the MMC or megaspores differentiate into aposporous initial (AI) cells, bypass meiosis and directly undergo mitosis (apomeiosis) to give rise to an embryo sac. Similar to diplospory, egg-like cells within these embryo sacs initiate embryogenesis without fertilisation and, depending on the species, the endosperm can initiate with or without fertilisation. In aposporous apomicts, AI cells can appear at different times and with different frequencies during ovule development. In apomictic H. piloselloides for example, between one and four aposporous initial cells differentiate at the chalazal end of the ovule during late meiosis and megaspore selection, and are clearly distinguished by their large nucleus, dense cytoplasm and their close proximity to the sexual structures (Koltunow et al. 1998, 2000). By contrast, in apomictic Hieracium aurantiacum L., multiple AI cells can be detected in ovules at the same time as the MMC, indicating that although the site of AI formation is similar, the timing of AI initiation may be influenced by species specific genetic modifiers. The presence of multiple AI cells, and in some cases the functional megaspore, can lead to the production of more than one embryo sac in ovules of aposporous apomicts. For example, in some apomictic species of Brachiaria (Araujo et al. 2000, 2005) many ovules contain both sexual and aposporous embryo sacs, resulting in the production of seeds containing both hybrid and maternal embryos. In other aposporous apomictic species such as Pennisetum and Hieracium, however, the development of the sexual embryo sac usually terminates at the megaspore mother cell or megaspore stage and is replaced by development of aposporous embryo sac (Bray 1978; Koltunow et al. 1998). Therefore, communication between the aposporous and sexual cell types is likely to influence the fate of the sexual process.

In summary, the common features shared by the different types of apomixis are the ability to generate a cell capable of forming an embryo without having first undergone meiosis (apomeiosis), the capacity to produce an embryo without fertilisation, and the capacity to either produce an endosperm independent of fertilisation (autonomously) or to utilise an endosperm derived from fertilisation. Although the specific details of apomictic reproduction vary from one species to another, all three types of apomixis interact with the sexual process. Diplospory directly replaces the sexual process, adventitious embryony depends on the sexual process for viability and apospory appears to interact with the sexual process on a cellular level to allow for the co-existence or degeneration of sexual cells. In all cases, the events of sexual reproduction appear to initiate in ovules before development deviates towards apomixis. This suggests that the initiation of sexual reproduction may be a prerequisite for apomixis and also that the apomictic process may be superimposed over a basic sexual framework (Fig. 3).

During the initiation of diplospory, reproductive development deviates from the normal sexual processes when meiosis of the MMC is inhibited and mitosis is precociously stimulated from an unreduced, previously meiotic cell. Although several meiotic mutants have been identified in sexual plants, only the dyad mutant of Arabidopsis has been reported to fail meiosis yet continue development to produce unreduced female gametes, suggesting that the inhibition of meiosis alone is not sufficient to stimulate apomeiosis or that unreduced gametes in other mutants have been overlooked. The elongate1 mutant of maize also produces unreduced female gametes by omitting the second meiotic division, but progeny are genetically diverse due to the success of meiotic crossovers in meiosis I (Barrell and Grossniklaus 2005). In dyad mutants (see above), normal meiosis is inhibited and an unreduced mitotic product of the MMC gives rise to an unreduced embryo sac, albeit at low frequency, containing a functional unreduced egg cell. After fertilisation, dyad eggs can give rise to viable embryos that retain parental heterozygosity. This suggests that dyad is a unique mutant that allows for a breakdown of meiosis while also allowing the precocious stimulation of mitosis from an unreduced cell. Since the phenotype of dyad appears superficially similar to that of diplospory, further studies of this mutant might be helpful in identifying factors that control the transition from meiotic to mitotic development. It is also possible that DYAD function, an important component of the sexual process in Arabidopsis, is altered in diplosporous apomicts thereby supporting development of the apomictic process. However, since apomixis appears to be controlled by one or few dominant loci in the majority of apomicts assessed to date, and the dyad mutation is recessive, it is unlikely that a similar mutation is the key dominant regulator of the diplosporous pathway in wild apomicts. Nonetheless, the characterisation of DYAD function in diplosporous Boechera, a relative of Arabidopsis may be informative. Furthermore, dyad may represent a useful tool for the de novo synthesis of apomixis.

During the initiation and progression of apospory, there are several features that bear a resemblance to modified sexual events. Morphologically, AI cells appear similar to both MMCs and functional megaspores, since they expand to a large size and contain a large nucleus and dense cytoplasm. In apomictic Hieracium, AI cells do not accumulate callose in their cell walls and unlike sexual Hieracium species fail to express MMC marker genes such as AtSPL:GUS and DMC1, and therefore are unlikely to share MMC identity (Tucker et al. 2001, 2003; Okada et al. 2007). However, this may not be the case in all aposporous species. The differentiation of multiple cells capable of giving rise to an embryo sac during apospory is somewhat similar to the ‘extra sporogenous cell’ phenotype of msp1, TDL1A-RNAi and mac1 mutants of rice and maize, respectively. Although this phenotype differs from apospory in that AI cells do not undergo meiosis, it identifies a region of reproductive competence in the developing ovule that surrounds the developing sexual structures. In aposporous apomicts, AI differentiation is predominantly restricted to cells in this zone, suggesting that a similar signalling mechanism may be compromised in these plants.

The development of AI cells and their subsequent morphogenesis into embryo sacs at the expense of the sexual cells in some aposporous apomicts such as Hieracium and Pennisetum bears a striking resemblance to the death of the three non-selected spores during normal sexual development. Termination of the sexual pathway may be promoted by AI cells inducing programmed cell death responses in the sexual cells, or the sexual cells may be crushed and physically displaced by the expanding aposporous embryo sac(s). Studies of megasporogenesis in apomictic Hieracium suggest that degeneration of the megaspores is not simply due to mechanical displacement (Tucker et al. 2001), because megasporogenesis occurs to completion in at least 96% of the ovules from H. piloselloides, which sets >97% apomictic seed. Furthermore, analysis of a H. aurantiacumloss of apospory1 (loa1) mutant (Okada et al. 2007) where the AI is defective in function in that it differentiates in a position removed from cells undergoing sexual events and frequently grows away from the sexual cells, meiosis occurs normally and mitotic events of sexual embryo sac formation proceed in the majority of ovules. Therefore at least in the case of Hieracium, an active mechanism induced by the presence of AI cells is likely to influence the fate of the sexual megaspores. Is this the same mechanism used by the functional megaspore? Molecular evidence indicates that apomictic and sexual reproduction are closely related pathways in Hieracium, since reproductive marker genes that are expressed in the MMC, embryo and endosperm show similar expression in apomictic and sexual plants (Tucker et al. 2003). This suggests that sexual and apomictic pathways in Hieracium share common molecular regulatory features, indicating that they are not distinct pathways. Rather, it seems that apomixis hijacks the process of sexual reproduction at key stages, thereby diverting the normal genetic program of sexual development towards apomictic processes and the production of maternal seed (Fig. 3).

The benefits of sex and recombination in promoting fitness and adaptation to changing environments are well documented (Otto 2003); how then is it possible for an aberrant process such as apomixis that avoids recombination to arise and be successful in over 40 different angiosperm families? There must be some ecological advantage over sex, and this is supported by the fact that asexual reproduction is not restricted only to the angiosperms but also occurs in other multicellular organisms such as fungi, aphids and sharks (Taylor et al. 1999; Delmotte et al. 2001; Chapman et al. 2007). Theories suggest that asexual reproduction may have benefits when a population is rapidly expanding in a stable environment, because it requires less energy than sexual reproduction (i.e. no fusion of gametes), but sexual reproduction holds the advantage in rapidly changing environments by offering organisms exposure to genetic variation and hence adaptation (Engelstädter 2008; Blachford and Doebeli 2009). Why then are all plants not facultative apomicts, enabling them to experience the best of both worlds? Along with the short-term benefits, there must be some underlying problems associated with apomixis, even if it is facultative, since most apomictic lineages are short-lived and sexually reproducing species are far more common in nature. These concepts are pursued in greater detail in recent reviews (van Dijk 2007; Whitton et al. 2008).

Several models and hypotheses have been proposed that explain the genetic basis and origin of apomixis. Considering that apomixis has been identified in various forms and in over 400 plant species, it is likely that apomixis has evolved multiple times and via multiple paths from sexual ancestors. Therefore, models explaining its appearance may not necessarily be mutually exclusive.

The close relationship between sexual and apomictic processes combined with the finding that apomixis is typically under the control of one or few dominant loci (Ozias-Akins and van Dijk 2007) has led to the hypothesis that apomixis may have evolved from mutation(s) in key sexual genes (Peacock et al. 1995). This is supported by the identification of factors in sexual species that show loss- or gain-of-function phenotypes similar to those seen in apomictic plants. For example, the msp1, TDL1A:RNAi rice and Arabidopsis dyad mutants (described above) show similar ovule phenotypes to plants initiating apospory or diplospory, while ectopic expression of various genes including WUSCHEL, BABYBOOM, SOMATIC EMBRYOGENESIS RECEPTOR-LIKE KINASE (SERK), LEAFY COTTYLEDON1 or LEAFYCOTYLEDON2 (reviewed in Curtis and Grossniklaus 2007) induces the formation of somatic embryos from vegetative tissues and could possibly act in a similar way during adventitious embryony or autonomous development of unreduced egg cells. Furthermore, Arabidopsis mutants of the FERTILISATION INDEPENDENT SEED (FIS) class can show initiation of embryo development without fertilisation (see multicopy suppressor of ira (msi1) mutants; Guitton and Berger 2005) or endosperm development without fertilisation (medea, fertilisation-independent endospern and fis2; Ohad et al. 1996; Chaudhury et al. 1997; Grossniklaus et al. 1998), similar to that seen in autonomous apomicts. Whether any of these genes play a primary role during the initiation of apomixis, autonomous embryogenesis or endosperm development in wild apomicts is sometimes difficult to assess, since time-consuming functional analysis is often required in complex polyploid sexual and apomictic systems to determine their relevance. In some cases, results are promising, such as those from diposporous Boechera and aposporous Poa pratensis L., where SERK homologues were shown to be differentially expressed between sexual and apomictic ovules, specifically in the cells initiating apomixis (Albertini et al. 2005; Sharbel et al. 2009). In other cases, results are not so promising, such as that of the Arabidopsis FERTILISATION INDEPENDENT ENDOSPERM (FIE) FIS-class gene that represses endosperm development in the absence of fertilisation. fie mutations in Arabidopsis lead to autonomous proliferation of the central cell, but in apomictic Hieracium, which naturally initiates endosperm without fertilisation, a functional FIE orthologueue is critical for the initiation of endosperm development and embryo viability (Rodrigues et al. 2008). Although these candidate genes may have different functions and relevance to apomixis in other species and may yet be useful for the de novo synthesis of apomixis in crops, the identification of ‘apomixis’ genes through mutagenesis or profiling of dicot sexual systems seems to be somewhat unpredictable. Further studies in monocot crops such as rice may prove to be more revealing.

To date, no single mutant has been identified in a mutagenesis screen that displays an intact form of functional apomixis. Despite this, the simple mutation hypothesis seems feasible in species such as Brachiaria brizantha Stapf. (Pessino et al. 1998) and P. squamulatum (Ozias-Akins et al. 1998) where apomeiosis and autonomous embryogenesis are controlled by a single dominant locus. However, it is difficult to rationalise when multiple independent loci control the different components, which is the case for species such as T. officinale (van Dijk et al. 1999), E. annus (Noyes 2005) and H. caespitosum (Catanach et al. 2006). The chance that two mutations could occur almost simultaneously in a plant or population and subsequently give rise to viable apomictic progeny is highly unlikely (Asker and Jerling 1992). Even if a single mutation occurred in a sexual plant that resulted in the formation of unreduced gametes, this would presumably lead to an increase in progeny ploidy and the formation of odd seed ploidy levels that may not be able to support viable seed formation in the F₁ generation. Furthermore, if a single mutation inducing parthenogenesis (the capacity to form an embryo autonomously from a reduced egg) occurred without apomeiosis in a sexual plant, then each subsequent reproductive cycle would lead to a halving of the genomic DNA content and severely reduced fitness. One factor that may enhance the longevity of such mutations is the predisposition of some plant families to support apomixis-like phenomena. Evidence to support this comes from the observation that the majority (~75%) of diplosporous and aposporous apomicts are restricted to three families: Asteraceae, Poaceae and Rosaceae. The reason for this bias is unknown, but may relate to the existence of reproductive features and/or genetic modifiers that are compatible with the stable induction of apomixis. Although apomixis typically appears to be controlled by few loci (Ozias-Akins and van Dijk 2007), a detailed flow cytometric seed screen in Paspalum determined that five loci control apomixis (Matzk et al. 2005). Interestingly, all five loci are not essential for apomixis, but different combinations can lead to a higher penetrance of apomictic seed set in a particular plant. Therefore, some of these loci may represent so-called ‘modifiers’ that, while not being essential for the apomictic process, enhance its progression at the expense of sexuality (Koltunow et al. 2000). It seems feasible that some of these loci may already be present in sexual plants, but without the complete set of ‘apomictic’ loci lead only to a base level of reproductive abnormalities.

The observation that virtually all natural apomicts are polyploid and highly heterozygous has led to the suggestion that hybridisation may play a role in the evolution of apomixis (Ernst 1918). Both hybridisation and polyploidy are important processes for the evolution of angiosperms because they have significant effects on genome organisation and gene expression that can increase adaptive potential (Chen 2007; Rieseberg and Willis 2007). The relative contributions of hybridisation and polyploidy to the evolution of apomixis, however, have been difficult to separate. Apomixis can be induced by the synthetic induction of polyploidy from sexual plants (Quarin and Hanna 1980; Quarin et al. 2001), suggesting that in some systems polyploidisation could initiate and maintain asexuality (Grimanelli et al. 2001). This might occur via genomic interactions, rearrangements or epigenetic phenomena that stabilise the genome and allow the manifestation of apomixis. However, apomixis can also occur in wild and synthetically induced diploid plants (Bicknell 1997; Schranz et al. 2005), suggesting that polyploidy is not necessarily a prerequisite but could be a consequence of asexual reproduction that confers some form of genomic stability.

One theory to explain the appearance of apomixis is referred to as the ‘hybridisation theory’ (Carman 1997, 2001, 2007). In this theory, the hybridisation of two sexual ecotypes or related species with different reproductive characters contributes to the induction of apomictic phenomena. On a mechanistic level, this results from hybrid plants containing two sets of parental genes that are involved in female embryo sac development; the asynchronous expression of these duplicated genes leads to precocious embryo sac initiation and embryogenesis at aberrant sites and times during reproduction. The attractive feature of the hybridisation model is that it relies only upon the additive effect of native gene expression, rather than mutations in genes involved in sexual reproduction, and thus overcomes the problems discussed above for independent mutations that have no immediate benefit to reproduction.

Support for this hypothesis comes from the analysis of gene expression in polyploids, where genes show asynchronous expression on a temporal and spatial level depending on the parental genome of origin (Adams et al. 2003; Gu et al. 2004), the appearance of apomictic traits in hybrids of related sexual plants, allo-polyploids or paleo-polyploids (for example, Antennaria, Sorghum and Arabidopsis; see Carman 2001, 2007), and the close morphological and molecular relationships observed during sexual and apomictic processes. Recent support also comes from a study of apomixis in Boechera where transcriptomic profiling was used to compare allele-specific gene expression in ovules from related sexual and apomictic plants (Sharbel et al. 2009). The results from this study show that several genes conserved in sexual and apomictic Boechera species are heterochronously expressed in the apomicts, and that altered levels of expression in the apomict are linked to gene duplication and parent-of-origin effects. Although further studies are required to assess the functional relevance of these expression changes, this study highlights the advantages of high-throughput transcriptomic approaches for the cross-species characterisation of apomixis and its evolutionary origin. If the hybridisation theory is correct, then the genomic regions controlling the initiation of apomixis might simply reflect key regulators of the sexual pathway that are mis-expressed in time and space. This may be an important point to consider as mapping and mutagenesis approaches in Pennisetum, Hieracium and Taraxacum and other apomictic species approach the core regions controlling apomixis in these plants (Vijverberg et al. 2004; Akiyama et al. 2005; Catanach et al. 2006).

In this review, we have explored morphological and molecular details of early stages in sexual and apomictic reproduction in flowering plants. The identification of novel apomixis-like mutants in sexual plants, the initiation of sexual reproduction preceding apomixis in most apomicts and the similar expression of reproductive marker genes in some sexual and apomictic species provide support for the theory that apomixis evolved and is manifested as a modified form of sexual reproduction, instead of an entirely novel reproductive pathway (Fig. 3). Whether this occurred via a combination of mutations in sexual genes and/or hybridisation of related sexual plants or by some other means remains unclear. Further characterisation of the sexual reproductive process, wild type apomicts and mutants lacking apomixis, combined with the identification of the genomic controlling regions and species- and allele-specific changes in gene expression on a genome wide scale will help to solve this puzzle, and bring understanding of apomixis one step closer to reality.