Endosperm: food for humankind and fodder for scientific discoveries

Abstract

Contents

	Summary	290
I.	Introduction	290
II.	Development of the endosperm	291
III.	Patterning the endosperm	294
IV.	Control of endosperm size and its influence on seed size	295
V.	Hormonal physiology of the A. thaliana endosperm	298
VI.	Epigenetic regulation of the endosperm	298
VII.	Conclusions	301
	Acknowledgements	301
	References	301

Summary

The endosperm is an essential constituent of seeds in flowering plants. It originates from a fertilization event parallel to the fertilization that gives rise to the embryo. The endosperm nurtures embryo development and, in some species including cereals, stores the seed reserves and represents a major source of food for humankind. Endosperm biology is characterized by specific features, including idiosyncratic cellular controls of cell division and epigenetic controls associated with parental genomic imprinting. This review attempts a comprehensive summary of our current knowledge of endosperm development and highlights recent advances in this field.

I. Introduction

With the increase in the human population and the rapid growth of demand for biofuel, food supply has become one of the most challenging problems confronting us in this century. In both developed and developing countries, high food prices have the potential to cause economical instability and social unrest. Plant seed is the most important renewable source of food for humans, both directly and indirectly as the main feed for livestock and as a source of industrial raw materials for the food industry.

Seed size is one of the main seed traits selected during domestication of grasses as crops in human history (Gepts, 2004; Fuller, 2007). However, as a consequence of the complex organization of seeds, our understanding of the mechanism of seed development and seed size control has remained limited until recent years.

The seed of angiosperms comprises three major components, the seed coat, the embryo and the endosperm, which supports embryogenesis (Berger & Chaudhury, 2009) (see Table 1 for a glossary of the terms used). The development of the endosperm and the embryo depends on both parental genomes, while the surrounding seed coat develops under the control of the maternal genome. In addition, there is evidence for cross-regulation between seed components. Thus, seed development, including seed growth, requires communication and coordination of distinct genetic programs that govern the development of each seed component. In this context, the endosperm plays a central role. The endosperm controls seed growth and allows transfer of maternal nutrients to the embryo. In certain species, including cereals, the endosperm stores reserves in the form of starch, proteins and lipids. The endosperm constitutes the edible part of the cereal seed and as such sustains directly or indirectly (via animal food) > 60% of human nutrition.

Table 1.   Glossary

Central cell	The cell that initiates the endosperm lineage after fertilization.
Double fertilization	Sexual reproduction in flowering plants requires two parallel fertilizations of the egg cell and the central cell by two sperm cells delivered by the pollen tube.
Egg cell	The female gamete that initiates the embryo lineage after fertilization.
Embryo sac	The female gametophyte that contains at maturity the egg cell and the central cell.
Endosperm: the embryo	Nurturing annex that develops from the fertilized central cell.
Gametophyte	The plant haploid life form that produces gametes. In plants, meiosis produces haploid spores that develop as haploid gametophytes. In flowering plants the gametophytic phase is reduced to a few cells.
Pollen	The male gametophyte which contains at maturity two sperm cells and a vegetative cell. The vegetative cell elongates the pollen tube which delivers the two sperm cells to the embryo sac.
Seed coat	The four/five cell layers derived from the ovule integuments, which surround the endosperm.
Sporophyte	The diploid organism that produces spores from meiosis. In flowering plants the life cycle is essentially represented by the sporophytic phase.

In Arabidopsis thaliana (dicot), the endosperm is transient and is consumed by the embryo, which stores the seed reserves. In this species, many advances have been made in our understanding of endosperm development, in addition to the wealth of data obtained from the study of cereals. This review provides an overview of our current knowledge of endosperm development in plants. The review does not address the physiology and functions of the endosperm during seed maturation and germination. For further details of these topics, the reader should refer to previous reviews (Jolliffe et al., 2005; Penfield et al., 2005; Vicente-Carbajosa & Carbonero, 2005; Finch-Savage & Leubner-Metzger, 2006; Finkelstein et al., 2008; Angelovici et al., 2010; Kawakatsu & Takaiwa, 2010; Raghavendra et al., 2010; Zeeman et al., 2011). There have been several controversies regarding various aspects of endosperm biology. When controversies have been resolved, this review will not attempt to summarize historical debates but will focus only on the currently accepted models.

II. Development of the endosperm

1. Initiation of endosperm development

Plant reproduction is characterized by a double fertilization that leads to the production of the embryo and the endosperm. In plants, male and female gametes are produced after meiosis following a series of divisions of a haploid male or female spore (Fig. 1). In angiosperms, male gametogenesis takes place in stamens and leads to pollen containing two sperm cells (Fig. 1). Female gametogenesis takes place within the diploid tissues of the ovule. The haploid megaspore produced by meiosis undergoes a series of three syncytial divisions, followed by cellularization, producing the embryo sac. The embryo sac contains the haploid female gamete or egg cell, and the central cell (Fig. 1). The pollen tube delivers two sperm cells to the female gametophyte, one of which fuses with the egg cell and one of which fuses with the central cell. Fertilization of the egg cell leads to embryogenesis (Fig. 1). The endosperm is produced by the fusion of the central cell with the other sperm cell. In most plant species, the central cell inherits two haploid nuclei from the syncytial female gametophyte. The endosperm genome thus contains two doses of the maternal genome and one dose of the paternal genome. The ploidy of the endosperm varies amongst angiosperms, as a result of a varied dosage of the maternal genome. Hereafter, we shall focus solely on endosperm development in A. thaliana and cereals. Details of other modes of development can be found in extensive reviews (Maheshwari, 1950; Floyd & Friedman, 2000; Olsen, 2001).

Figure 1. Flowering plant gametogenesis. (a) Male meiosis produces a tetrad of four haploid microspores. Each microspore polarizes and experiences a first unequal mitosis leading to a bicellular stage comprising a generative cell (small nucleus) inside a vegetative cell. A second mitotic division of the generative cell produces two identical haploid sperm cells. (b) Female gametogenesis is also initiated by meiosis in the megaspore mother cell (MMC). The megaspore is the only surviving meiotic product. The megaspore experiences three syncytial nuclear divisions marking a series of specific stages of female gametophytic development (FG3 to FG5), and leading to the eight-celled female gametophyte. Cellularization at FG6 leads to the production of seven cells, three antipodals (AC) of unknown function, two synergids (SC), which attract the pollen tube, the egg cell (EC) and the central cell (CC), which contains two nuclei. These seven cells differentiate into the mature embryo sac (FG7).

2. Modes of endosperm development

Endosperm development follows three major phases. Once fertilized, the central cell immediately undergoes mitoses that are not followed by cell division, leading to the syncytial endosperm developmental phase. This phase ends with the partition of the multinucleated syncytial endosperm into individual cells. During the cellular phase that follows, cells divide and the various cell types are differentiated. The final maturation phase of endosperm development is marked by an arrest of cell division and a change of metabolism, leading to reserve accumulation in the seed. The duration and importance of these three phases vary amongst species.

Syncytial phase  This mode of syncytial division is typical of large rapidly growing cells and is observed in other tissues associated with nutrient transport function such as the vascular system of plants, feeding cells induced by parasitic nematodes (Favery et al., 2004) and some cells in the mammalian placenta (Cross, 2005).

In A. thaliana, a series of eight syncytial divisions leads to c. 200 nuclei mostly located at the periphery of the endosperm (Mansfield & Briarty, 1990a; Mansfield & Briarty, 1990b; Brown et al., 1999; Boisnard-Lorig et al., 2001) (Fig. 2). Semi-synchronicity of the syncytial divisions allows the classification of the syncytial phase into successive stages I to VIII (Ingouff et al., 2005b). The center of the syncytium is occupied by a single large vacuole. Each nucleus is surrounded by a mass of cytoplasm, the boundary of which is delimited by a dense cortical array of cytoskeleton. Such a unit has been defined as a nucleo-cytoplasmic domain (NCD) (Brown et al., 1999).

Figure 2. Developmental features of flowering plant seeds. (a) The double fertilization initiates seed development. Two sperm cells are released by the pollen tube into the female gametophyte. One sperm cell fuses with the egg cell and produces the embryo. The other sperm cell fuses with the central cell, initiating endosperm development. (b) A section of an Arabidopsis thaliana seed (4 d after fertilization) shows the embryo surrounded by the endosperm, itself surrounded by the maternal seed coat. The embryo-surrounding region of the endosperm is shown in dark blue, the peripheral endosperm in light blue, and the chalazal endosperm in light brown.

The pace of cell division is c. 2–3 h per cell cycle for the first three syncytial divisions and gradually increases to 12 h during subsequent cycles. The first divisions lead to eight endosperm nuclei evenly spaced along a curved tube bounded by the inner seed integument (Boisnard-Lorig et al., 2001; Ingouff et al., 2005b). During the following cycle of syncytial division, the nuclei at the chalazal pole, opposite to the site of sperm entry, do not enter a further round of mitosis and presumably undergo endoreduplication as they enlarge (Fig. 2). Larger nuclei are the first marker of the posterior endosperm pole, also termed the chalazal endosperm (CZE). The next round of syncytial division is marked by the definition of a second mitotic domain at the anterior pole composed of five to eight nuclei surrounding the embryo. Direct dynamic observation of mitosis as well as regional expression of the mitotic cyclins shows that these nuclei divide earlier than the nuclei in the peripheral mitotic domain. Thereafter, the cell cycle maintains an independent pace in each of the three domains. Further divisions in the peripheral domain lead to stages VII (c. 50 nuclei), VIII (100 nuclei) and IX (200 nuclei).

The mode of syncytial endosperm development in cereal species is similar to that described above and is detailed in Brown & Lemmon (2007).

Cellularization  The eighth cycle of nuclear division marks the end of the syncytial phase as it is followed by cellularization in the anterior and peripheral domains, leading to isolation of most nuclei in individual cells. In A. thaliana time-lapse recordings it has been shown that endosperm cellularization occurs immediately after the eighth mitotic division (Sorensen et al., 2002). Cellularization is absent in mutants affected in the basic mechanisms of cytokinesis also active in the embryo (Sorensen et al., 2002). Cellularization takes place early in mutants that have reduced endosperm growth; however, the pace of nuclear division remains unaffected, leading to a reduced number of cells in the mutant cellular endosperm (Garcia et al., 2003). This suggests that a critical nucleo-cytoplasmic ratio may represent a signal that triggers cellularization.

The type I MADS box transcription regulator AGAMOUSLIKE 62 (AGL62) plays an important role in the control of cellularization (Kang et al., 2008). AGL62 expression is confined to the endosperm after fertilization and drops abruptly before cellularization. Mutants deprived of AGL62 show very precocious cellularization, suggesting that AGL62 prevents the onset of cellularization during syncytial endosperm development.

The process of cellularization itself is a variation of conventional cytokinesis with notable differences in cytoskeletal arrangement (Otegui & Staehelin, 2000a; Olsen, 2001; Otegui et al., 2001). Endosperm cellularization depends mostly on the functions essential for cytokinesis in vegetative cells (microtubules, actin and associated proteins, and vesicle trafficking) (Sorensen et al., 2002). Formins, as actin polymerization-regulating factors (Blanchoin & Staiger, 2010; Bugyi & Carlier, 2010; Campellone & Welch, 2010), also play a role in endosperm cellularization (Ingouff et al., 2005a; Fitz Gerald et al., 2009). This suggests that the actin cytoskeleton plays a role in cellularization, as it does in plant cytokinesis. Hence it is very likely that endosperm cellularization mainly depends on the same molecular components as cytokinesis. However, de novo formation of cell walls around syncytial NCDs probably relies on distinct architectural organization of these conserved components.

Each NCD is delimited by a dense array of microtubules defining the future division sites. Unlike cytokinesis, no preprophase band anticipates the spatial position of the new cell wall during cellularization (Otegui & Staehelin, 2000b). After the critical last syncytial mitosis, a series of cell plates are formed between the sister nuclei as a result of a rather complex mechanism based on vesicles derived from the endoplasmic reticulum (Otegui et al., 2001). Other de novo assembly of cell plates takes place between non-sister nuclei defining hexagonal or pentagonal shaped boundaries around each NCD. Endosperm cell plates arise simultaneously around all NCDs (Sorensen et al., 2002). This implies that some unknown coordination mechanisms ensure that each cell plate connects only once to its multiple neighbors, forming three-way junctions, which is uncommon in conventional cytokinesis. Another idiosyncratic feature of cellularization lies in the centripetal growth of cell plates (Olsen, 2001). After cellularization, several rounds of cell division take place and the cell plates form from the center to the periphery in all directions, as during conventional cytokinesis (Olsen, 2001, 2004; Nguyen et al., 2002).

3. Cellular phase and maturation

After cellularization, the cellular developmental phase comprises a small series of synchronous cell divisions leading to formation of four to six layers of endosperm cells, depending on the position along the micropylar–chalazal (MC) axis. Divisions persist in the central part of the peripheral endosperm and progresses toward the chalazal pole, while the central vacuole recesses progressively until the entire peripheral endosperm becomes cellular.

In A. thaliana and many dicotyledonous species, which produce seeds with a morphology comparable to that of legumes, the endosperm does not persist, with the exception of the cell layer immediately in contact with the seed coat. The function of this layer is linked to the control of seed dormancy, and its function has been thoroughly reviewed (Koornneef et al., 2002; Finch-Savage & Leubner-Metzger, 2006; Finkelstein et al., 2008; Angelovici et al., 2010; Raghavendra et al., 2010). Reserves are stored in the embryo, which proliferates rapidly at the expense of the cellular endosperm after the heart stage. Starch synthesis takes place in the cellular endosperm and amyloplasts differentiate. This phase has received very limited attention in A. thaliana (Mansfield & Briarty, 1990a,b), but recent investigations in other species have provided important physiological insights (see Section IV. 1. Physiological control of endosperm growth).

In cereals, maturation of the endosperm is marked by endoreduplication and programmed cell death. Endoreduplication coincides with the accumulation of reserves in the starchy endosperm, although the aleurone cell can also endoreduplicate in some species. Endoreduplication is gradual and probably involves RETINOBLASTOMA RELATED 3 (RBR3), which controls the G1/S transition (Sabelli et al., 2005; Nguyen et al., 2007).

The development of the endosperm, like the other seed components, ends with dessication. Programmed cell death affects the starchy endosperm in cereals. It is synchronous with peaks of ethylene and elevated concentrations of abscisic acid. Most of the endosperm cells loaded with reserves probably die, with the exception of the outer cell layer, called the aleurone layer in cereals (Sabelli & Larkins, 2009).

4. Varied modes of endosperm development

Although the sequence of endosperm development described in A. thaliana and cereals is observed in many other plant genera, the duration and importance of the syncytial and cellular phases of development may vary from one species to another, leading to extreme cases of pure syncytial endosperm development or pure cellular development. Initial division of the central cell into two compartments followed by syncytial division in each or only one of the compartments has also been described as ‘helobial’ endosperm development. (Maheshwari, 1950; Vijayaraghavan & Prabhakar, 1984).

Most basal angiosperms show the prevalence of cellular endosperm development (Floyd & Friedman, 2000, 2001). In all basal angiosperm species examined the first division takes place along the axis defined by the micropylar and chalazal domains, as shown in most recent angiosperms (see the paragraph below for a molecular definition of this axis). In all basal taxa, the endosperm stores the seed reserves. It is thus considered that syncytial endosperm development and storage of reserves in the embryo represent further steps of endosperm evolution from an ancestral helobial or cellular mode (Floyd & Friedman, 2000, 2001).

III. Patterning the endosperm

The endosperm is not a uniform tissue. The cellularized endosperm contains several cell types based on expression patterns and cytological differences. These structures are arranged along the axis running from the anterior pole where the pollen tube delivers the sperm cells (the micropyle) to the posterior pole where maternal nutrients transit (the chalaza). Other cell types are arranged according to radial symmetry. These two axes of patterning are conserved amongst flowering plants.

1. Polar organization

An antero-posterior axis pre-exists in the embryo sac and it is not known if this axis influences the corresponding MC axis in the endosperm. The micropyle is the site of entry of the pollen tube in the ovule. The chalazal pole is opposite the micropyle and represents the area where the maternal nutrients are deposited in the endosperm. In many species, including A. thaliana and maize (Zea mays), this axis is curved, such that the micropyle and the chalaza become positioned side by side at the end of ovule development. During the syncytial stage, the influence of the MC axis is observed in the orientation of syncytial divisions, the mitotic domains and the migration of NCDs (Fig. 2). The first two divisions take place along the MC axis of the endosperm. The third division is perpendicular to the MC axis. Subsequently, the mitotic domains are established first in the chalazal endosperm (CZE), with endoreduplication replacing nuclei division. Later nuclei division in the micropylar endosperm (MCE) no longer takes place in synchrony with divisions in the peripheral endosperm (PEN) (Fig. 2). This general organization is also marked by a distinct cytological organization of the cytoskeleton in the three domains (Brown et al., 1999; Nguyen et al., 2002). The CZE, initially defined by two to four large nuclei in a pool of cytoplasm forming the cyst, is later marked by a posterior-directed migration of NCDs (Guitton et al., 2004). After developmental stage VI, each syncytial division is followed by gradual migration of NCDs from the posterior part of the peripheral endosperm toward the cyst. Theses NCDs fuse to generate multinucleate nodules, which appear as large NCDs (Fig. 2). Over time, individual NCDs and nodules migrate and fuse with the chalazal cyst. As cellularization takes place, all remaining NCDs fuse into the cyst, which gradually recedes, while the entire peripheral endosperm becomes cellular.

The identity of the three domains along the MC axis is established by specific expression of genes and other markers. In A. thaliana, the micropylar endosperm, which occupies a domain called the embryo-surrounding region (ESR), is marked by the expression of GFP in the enhancer trap line N9185 (Ingouff et al., 2005b), the sucrose transporter AtSUC5 (Baud et al., 2005), the subtilisin-like serin protease ABNORMAL LEAF SHAPE 1 (Tanaka et al., 2001), the Basic helix loop helix factor ZHOUPI (Yang et al., 2008), the transcription factor MINISEED3 (Wang et al., 2010a) and several other genes that remain to be catalogued from an extensive transcriptome study of dissected seed components (Le et al., 2010). In cereals, several other genes have been characterized by their localized expression at the embryo-surrounding pole (Cossegal et al., 2007) and at the posterior (basal) pole (Royo et al., 2007). The MC axis is thus potentially conserved but homologous functions in the various domains between cereals and A. thaliana remain to be identified. The three domains of the A. thaliana endosperm have been dissected and the transcriptome analyzed at different developmental stages (Le et al., 2010).

The definition of MC polarity is compromised in mutants for the polycomb group (PRC2) pathway FIS, including MEDEA (MEA), FERTILIZATION INDEPENDENT SEED 2 (FIS2), FERTILIZATION INDEPENDENT ENDOSPERM (FIE), MULTICOPY SUPPRESSOR OF IRA 1 (MSI1) and DEMETER (Grossniklaus et al., 1998; Luo et al., 1999, 2000; Choi et al., 2002; Guitton & Berger, 2005). The expression of MEA and FIS2 is limited to the endosperm, providing a specific tissue identity for the FIS polycomb group complex (Luo et al., 2000; Wang et al., 2006), while FIE and MSI1 are expressed in other tissues (Yadegari et al., 2000; Hennig et al., 2003) and participate in other PRC2 protein complexes (Chanvivattana et al., 2004; Katz et al., 2004).

Mutations in FIS genes disrupt mitotic domain organization (Ingouff et al., 2005b). The FIS complex also prevents the down-regulation of AGL62 expression (Kang et al., 2008), presumably leading to the absence of cellularization in fis mutant endosperm (Chaudhury et al., 1997; Kiyosue et al., 1999; Sorensen et al., 2002). FIS genes presumably control the transition between developmental phases during endosperm development and fis loss of function prevents exit from the juvenile stage before the definition of mitotic domains. This hypothesis has been supported by genome-wide analyses (Weinhofer et al., 2010). Using tagged fluorescent endosperm nuclei it was possible to isolate a fraction enriched in endosperm chromatin and to study the genome-wide profile of H3K27me3. The repressive marks H3K27me3 were present over a large set of genes that are not expressed during early endosperm development, suggesting that the FIS PcG prevents the acquisition of somatic traits in the endosperm and thereby preserves the endosperm identity. A group of genes specifically expressed during early wild-type endosperm development became repressed around the heart stage by the FIS complex (Weinhofer et al., 2010), supporting the idea that the FIS PcG complex is required for the repression of a defined set of genes around endosperm cellularization (Ingouff et al., 2005b; Berger et al., 2006). Beside this global molecular control by the FIS complex, the molecular and cellular mechanisms responsible for the definition of the mitotic domains in the syncytial endosperm are not understood.

2. Radial organization

After cellularization, the outer endosperm cell layer undergoes a specific series of events leading to its differentiation into a specialized structure, the aleurone layer (Becraft & Yi, 2011). In cereals, the aleurone layer shows specific storage of phytic acids that chelate mineral ions such as magnesium and phosphate. Aleurone cells are involved in cereal grain maturation (Vicente-Carbajosa & Carbonero, 2005). During germination, enzymes produced by aleurone cells are important for remobilization of the reserves of the starchy endosperm. This differentiation has been documented in detail in cereals (Becraft, 2007). After cellularization, repetitive periclinal divisions form cell files, and the outermost of these cells undergoes a series of anticlinal divisions leading to the aleurone layer of cuboidal cells. Defective endosperm kernel 1 (Dek1), encoding a cysteine proteinase (Lid et al., 2002), and Crinkly 4 (Cr4) (Becraft et al., 1996; Becraft & Asuncion-Crabb, 2000), encoding a leucine-rich repeat kinase, are important for aleurone differentiation in maize. The function of these genes is controlled or regulated by Supernumerary aleurone layer 1 (Sal1), which may be involved in receptor recycling (Shen et al., 2003). The signaling from Dek1 is mediated by the locus defining the negative regulator Thick Aleurone 1 of unknown identity (Yi et al., 2011). Phylogenetic and functional analyses of the four genes encoding serine/threonine receptor kinases in rice (Oryza sativa) and A. thaliana identified an ortholog in rice but not in A. thaliana, suggesting that this family acquired different functions in endosperm development in dicots and monocots (Cao et al., 2005). The Dek1 homolog in A. thaliana may have a general role in cell differentiation and the disorganized outer endosperm cell layer in dek1/dek1 seeds does not necessary reflect a specific function in aleurone differentiation (Lid et al., 2005). In conclusion, it remains undetermined whether the aleurone layer in the cereal endosperm is equivalent to the outer endosperm layer in Arabidopsis.

In cereals, cells of the future aleurone layer that are adjacent to the site of delivery of maternal nutrients through the nucellus differentiate as transfer cells (TCs). TCs are very elongated, develop cell wall in-growths and express sugar and amino acid transporters (reviewed in Royo et al., 2007). TCs are conspicuous in maize, less numerous in wheat (Triticum aestivum) and barley (Hordeum vulgare) and hardly distinguishable in rice. These cells are responsible for the intake of maternally provided nutrients. TCs also express cysteine-rich polypeptides involved in defense (Gomez et al., 2009). Maize TCs express MYB-related protein 1 (ZmMPR-1), which activates expression of several genes (Gomez et al., 2002). This transcription factor is sufficient to cause TC differentiation from other domains of the aleurone, suggesting that TC differentiation results from a sequential series of activations that take place locally and redirect aleurone differentiation (Gomez et al., 2009). The signals that cause this local activation of ZmMPR-1 expression are unknown but they could involve sensing sugar or other products delivered from the phloem.

IV. Control of endosperm size and its influence on seed size

In A. thaliana, the syncytial phase of endosperm development corresponds to rapid growth of the seed (Garcia et al., 2003). After cellularization, the endosperm still experiences cell division but at a much lower rate than during the syncytial phase and growth is greatly reduced (Scott et al., 1998). Eventually the endosperm experiences cell death, and the space it occupies is gradually taken over by the embryo (Mansfield & Briarty, 1990a,b, 1991). Thus, the volume of endosperm achieved at the time of cellularization is critical in determining the final seed size.

1. Physiological control of endosperm growth

Most of the carbon for seed growth is supplied as sucrose. In A. thaliana, the phloem terminates in a region of the integument lying below the chalazal endosperm. Symplastic unloading of sucrose into the integuments of the seed occurs at this point (Stadler et al., 2005). There is little information about the routes by which sucrose subsequently reaches the embryo. In oilseed rape (Brassica napus) and legumes, much of the sucrose entering the seed is converted to hexoses, which accumulate in the endosperm, coinciding with a rapid increase in endosperm volume. The hydrolysis of sucrose to hexoses probably contributes to this increase, by providing a high water potential leading to water uptake by the endosperm. During a second phase, hexoses are presumably transported to the embryo and converted to sucrose, which is later stored as starch and oil (Hill et al., 2003; Weber et al., 2005; Morley-Smith et al., 2008). The position of the endosperm between the embryo and maternal tissues of the chalaza, where the maternal nutrients are delivered, intuitively suggests that the endosperm plays a role in maternal nutrient transfer to the embryo. However, this idea has been debated and only a recent direct investigation using nuclear magnetic resonance (NMR) in pea (Pisum sativum) has provided convincing evidence supporting the hypothesis that maternal nutrients transit through the endosperm to the embryo (Melkus et al., 2009). NMR enables imaging of the concentrations of sucrose and the two major amino acids delivered from the phloem, alanine and glutamine, during endosperm and embryo development. There is a marked increase in the sucrose concentration in the endosperm vacuole long before any increase can be observed in the embryo. This increase is accompanied by high expression of sucrose transporters, which enables prolonged uptake by the embryo to be sustained. By contrast, in Brassica rapa, NMR studies have suggested that the embryo is not surrounded by an environment that initially contains a high hexose content and later a high sucrose content (Morley-Smith et al., 2008). It is thus unlikely that the hexoses contained in the endosperm vacuole are transferred directly to the embryo in this species representative of oilseeds (Morley-Smith et al., 2008). In A. thaliana, the sucrose transporter AtSUC5 is expressed in the endosperm surrounding the embryo and plays a role in reserve storage in the embryo (Baud et al., 2005). According to NMR studies in B. rapa, the young embryo may receive its sugar via localized transport of sucrose from the integuments into the chalazal and possibly also the micropylar endosperm, and sucrose will be continuously delivered to the embryo apoplastic space (Morley-Smith et al., 2008). However, how the sucrose crosses the ESR and the cuticle of the embryo epidermis remains unknown.

In pea, NMR studies have also shown that amino acid permeases are strongly expressed during early endosperm development and the endosperm vacuole accumulates high concentrations of amino acids that can be transferred to the embryo (Melkus et al., 2009). It is thus likely that, in dicotyledon species, where the embryo stores seed reserves, the endosperm is transient and serves as an intermediate sink for maternal nutrients (Hill et al., 2003; Weber et al., 2005). In cereals, the endosperm stores reserves at maturation. Some data indicate that the role of the syncytial endosperm is essentially the same in cereals as described for dicotyledons (Olsen et al., 1999; Olsen, 2004). In cereals, specific TCs differentiate in the endosperm, in the vicinity of the maternal tissues where nutrients are unloaded. In barley, the endosperm TCs have been isolated by laser-capture and the transcriptome indicates increased expression of genes with functions related to transport (Thiel et al., 2008). Aquaporins, sugar transporters, amino acid transporters and ion channels are all expressed at high levels in barley endosperm TCs. Similar observations have been made in studies in maize (Royo et al., 2007). In maize, cell wall invertases INCW1 and INCW2 are expressed in TCs during endosperm development and probably ensure the cleavage of sucrose delivered from maternal tissues in the apoplast (Miller & Chourey, 1992; Chourey et al., 2006). Loss of function of the genes Incw1 and Incw2 causes a reduction in endosperm and seed size (Olsen et al., 1999), demonstrating the importance of TCs in the intake of maternally derived sucrose and its control of seed growth.

2. Genetic control of endosperm size

A few A. thaliana mutants show impaired endosperm growth during seed development (Garcia et al., 2003; Luo et al., 2005). All these mutants are characterized by a reduced seed size and share the same haiku (iku) phenotype. The effect of iku mutations is sporophytic recessive. Heterozygous iku mutant plants do not show any obvious phenotypes in tissues other than the endosperm in 25% of the seeds. Homozygous iku1, iku2 or miniseed 3 (mini3) endosperm growth is arrested prematurely and gives rise to seeds smaller than those of the wild type (Garcia et al., 2003, 2005; Luo et al., 2005). An identical phenotype was observed in the mutant short hypocotyl under blue 1 (shb1) (Zhou et al., 2009). The rate of nuclear division and the patterning of the endosperm are not affected in iku mutants. Only endosperm cellularization initiates too early. Reduction of endosperm growth caused by iku mutations does not usually cause seed abortion and most iku/iku seeds are viable and give rise to iku/iku plants that do not show any obvious vegetative phenotype. Combinations of iku mutations do not cause phenotypes different from those caused by a single mutation and thus the four mutations constitute a genetic pathway.

Four genes of the IKU pathway have been identified. IKU1 encodes a VQ domain protein of unknown function (Wang et al., 2010a,b). IKU2 and MINI3 encode a leucine-rich repeat transmembrane kinase and a WRKY10 transcription factor, respectively. MINI3, HAIKU2 and HAIKU1 are expressed in the syncytial endosperm after fertilization (Luo et al., 2005; Wang et al., 2010a,b). SHB1 expression is not confined to the endosperm and thus probably does not confer the endosperm specificity related to the iku phenotype (Zhou et al., 2009). The expression of IKU2 depends on MINI3 (Luo et al., 2005). The expression of MINI3 and IKU2 depends on SHB1 but the mechanism of transcriptional activation involved remains unclear (Zhou et al., 2009). The expression of IKU2 and that of MINI3 also appear to depend on each other and it is possible that the transcription factor MINI3 (WRK10) is required to activate IKU2 expression (Luo et al., 2005). The signal that activates the transmembrane kinase IKU2 is not know but some clues may be provided by the functional study of CLAVATA3/EMBRYO SURROUNDING REGION-RELATED 8 (CLE8), which encodes a small peptide (Fiume & Fletcher, 2012). CLE8 is expressed both in the embryo and in the endosperm. It regulates expression of the transcription factor gene WUSCHEL-LIKE HOMEOBOX8 (WOX8), and together CLE8 and WOX8 form a signaling module that promotes seed growth and controls overall seed size (Fiume & Fletcher, 2012). We propose that CLE8 could be secreted outside the endosperm and initiate the IKU signaling pathway through the recognition of the IKU2 receptor kinase.

The molecular functions of IKU genes do not show a direct connection with cell growth and the mechanism by which the IKU pathway controls growth during syncytial endosperm development remains unexplained. The relationships among the different genes identified remain to be clarified in order to establish whether they form a single molecular pathway or several pathways that would converge on shared targets.

3. Parental control of endosperm growth

In contrast to genetic controls, epigenetic controls involve mechanisms that affect gene expression without changes in genome DNA sequence. Seed growth is regulated by epigenetic mechanisms leading to parent-of-origin effects (Spielman et al., 2001; Baroux et al., 2002; Berger & Chaudhury, 2009). In most plants, the ratio of maternal (m)/paternal (p) genomes in the endosperm is 2 m per 1 p because the central cell is homodiploid (Floyd & Friedman, 2000; Dilkes & Comai, 2004). When the normal 2 m per 1 p ratio is changed, expression of maternal and paternal genomes becomes imbalanced, which influences the development of the endosperm and seed (Scott et al., 1998; Li & Dickinson, 2010). Reciprocal interploidy crosses between diploid and polyploid plants produce seeds with an excess of maternal genome dosage characterized by precocious cellularization and reduced endosperm growth. An excess of the paternal dosage has opposite effects, with enhanced growth and delayed cellularization of the endosperm. These results have been interpreted using the parental conflict theory. This theory proposes that when offspring from different fathers develop inside the same mother (as in some mammalian species and some flowering plants) the father tries to derive as much maternal nutrients as possible for its own offspring, but the mother tends to distribute her nutrition evenly to all offspring (Haig, 1997, 2004; Haig & Wilczek, 2006). In plants, this theory has received limited support (Berger & Chaudhury, 2009; Wollmann & Berger, 2012) but still provides a valuable incentive to promote research on maternal effects and parental genomic imprinting.

4. The coordinated growth of maternal and zygotic compartments of seeds

As a consequence of the distinct genetic origins of the three main seed components, the genetic control of seed development is more complex than the development of other plant structures. The embryo is diploid and its genome contains one copy of the maternal and one copy of the paternal genome. However, the embryo nutrient tissue, the endosperm, is triploid with two identical copies of the maternal genome and one copy of the paternal genome. The seed coat is diploid and purely of maternal origin. Seed growth results from the coordinated growth of the three seed components and thus is controlled by three different genetic programs (Berger & Chaudhury, 2009).

The maternal seed coat that encloses the zygotic embryo and endosperm develops from the ovule integument. Fertilization promotes cell division in the integuments (Ingouff et al., 2006) and it was further shown that endosperm development triggers a signal that promotes cell division and elongation in the seed coat (Ingouff et al., 2006; Roszak & Kohler, 2011). Impairment of endosperm development prevents proper seed coat elongation (Roszak & Kohler, 2011) and, conversely, an engineered reduction of cell division in the integuments reduces the cell number and endosperm size (Garcia et al., 2005). Similarly, a reduction of cell growth in the integuments caused by loss of function in the TRANSPARENT TESTA GLABRA (TTG) pathway results in smaller seed size (Debeaujon et al., 2000; Garcia et al., 2005). Quantitative trait locus (QTL) analysis of reciprocal crosses between natural accessions Landsberg erecta (Ler) and Cape Verde islands (Cvi) showed that both maternal and nonmaternal genetic factors are involved in seed size variation and there are interactions between the two types of factor during seed growth (Alonso-Blanco et al., 1999). The number and size of cells in the seed coat vary between ecotypes and parallel variations in seed size. One QTL was linked to TTG2 (Dilkes et al., 2008), which affects seed coat elongation (Garcia et al., 2005). This trait may account for some degree of maternal control associated with natural variation (Alonso-Blanco et al., 1999).

The transcription factors APETALA2 (AP2) (Jofuku et al., 2005; Ohto et al., 2005) and AINTEGUMENTA (ANT) (Klucher et al., 1996; Mizukami & Fischer, 2000) are also involved in the coordination of seed growth in A. thaliana. Several other factors that regulate growth or cell proliferation in the seed coat also control seed size (Canales et al., 2002; Li et al., 2005; Anastasiou et al., 2007; Adamski et al., 2009).

In addition, DNA METHYLTRANSFERASE1 (MET1) prevents proliferation and elongation of ovule integuments. This explains the dominant maternal effect of the loss of MET1 on seed size (Xiao et al., 2006; FitzGerald et al., 2008). MET1 probably regulates the expression of several genes that control cell proliferation and elongation during seed coat development. The identity of these genes remains to be established.

It has become clear that the endosperm controls embryo development. The secreted subtilisin-like serine protease ALE1 is expressed predominantly in the ESR and is required for normal cuticle production in the embryo (Tanaka et al., 2001). The transcription factor ZHOUPI (ZOU), also called RETARDED GROWTH OF EMBRYO1 (RGE1) (Kondou et al., 2008), is expressed exclusively in the endosperm of developing seeds. After fertilization, ZOU is initially expressed uniformly in the endosperm, subsequently becoming restricted to the ESR together with ALE1. zou mutant embryos show severe defects in cuticle formation and in epidermal cell adhesion, suggesting that ZOU functions nonautonomously to regulate embryonic development (Yang et al., 2008). ZOU is required for ALE1 expression but not the expression of other genes that are also expressed specifically in the ESR. Hence ZOU and ALE1 define a pathway active in the endosperm. (Tanaka et al., 2002, 2007; Watanabe et al., 2004). This pathway also controls proper differentiation of the embryo epidermis (Yang et al., 2008). These observations support the theory that signaling from the endosperm is essential for embryogenesis. However, a feedback signal from the embryo to the endosperm has not been identified in A. thaliana.

In total, our current knowledge indicates that the endosperm acts as the central controller of seed growth and that its action is modulated by developing seed integuments, defining the final space that the embryo fills. In addition, the endosperm also regulates developmental aspects of embryogenesis and may also affect seed coat differentiation. The signals involved in these dialogues among the endosperm, the embryo and the seed coat remain unknown.

V. Hormonal physiology of the A. thaliana endosperm

The endosperm is known to be a source of phytohormones (Lopes & Larkins, 1993). Cytokinins have been identified in the maize endosperm (Yang et al., 2002). Coconut (Cocos nucifera) liquid syncytial endosperm is used to provide phytohormones for in vitro cell cultures (Ge et al., 2005). The expression of cytokinin-biosynthetic genes has been detected in the endosperm posterior pole, suggesting a localized production of cytokinin in the cyst (Miyawaki et al., 2004).

The high level of cytokinin activity found in early developing seeds indicates that cytokinins could be involved in the growth of the seed component and control seed mass/yield. However, multiple mutants of the gene families involved in cytokinin signal transduction were obtained and found to affect seed size (Hutchison & Kieber, 2002; Riefler et al., 2006). Bigger seeds are produced by a combination of mutations that block cytokinin signaling, cytokinin (Riefler et al., 2006; Hutchison & Kieber, 2002; Argyros et al., 2008). The idea that cytokinin could inhibit seed growth in some way was further confirmed by ectopic over-expression of cytokinin oxidases which gave rise to big seeds (Werner et al., 2003). However, these results are difficult to interpret for the following reasons. Mutants affected in cytokinin signaling display many vegetative defects. These defects also impair ovule development, leading to a reduction in the number of seeds per silique. As a result, the amount of maternal reserves allocated per seed increases, resulting in larger seed size. Although the high abundance of cytokinin in early endosperm development was recognized nearly half a century ago, its origin and its functions in this tissue are still unclear.

1. Other hormones

Auxin plays a role in the control of cell division in the maize endosperm (Lur & Setter, 1993). The production of auxin and its transport have been localized in the maize endosperm to the ESR and TCs (Forestan et al., 2012) and two AUXIN RESPONSE FACTORS are specifically expressed in the ESR (D. Weijers, pers. comm. cited in Yang et al., 2008), suggesting a potential gradient of auxin activity along the MC axis in the endosperm. Together with gibberelins the ethylene signaling pathway is required for proper differentiation of barley endosperm TCs (Thiel et al., 2008).

At the end of seed maturation, functional analyses of abscisic acid (ABA)-biosynthetic genes showed that, in addition to the embryo, the endosperm contributes to the production of ABA which is involved in the induction of seed dormancy (Lefebvre et al., 2006). This function is probably linked to the ABA-dependent pathway affecting dormancy and causing vivipary in maize (Hattori et al., 1992). ABA operates through a dynamic balance with gibberellic acid (GA). Regulation of dormancy status results from the response to this balance through hormone-signaling networks that influence sensitivity to ABA and GA. This topic is addressed in several reviews (Finch-Savage & Leubner-Metzger, 2006; Finkelstein et al., 2008; Holdsworth et al., 2008) and a recent study has shown the relevance of the role of hormonal pathways in the field (Footitt et al., 2011).

VI. Epigenetic regulation of the endosperm

1. The unusual epigenetic state of the endosperm

Low methylation of endosperm DNA  DNA methylation at the 5-position of cytosine (m5C) is the archetype of the heritable epigenetic mark and participates in developmental regulation in both animals and plants, mainly through transcriptional regulation (Law & Jacobsen, 2010). DNA methylation also protects the genome against the activity of transposable elements (TEs) and other repetitive sequences. In plants, in addition to symmetrical CG sequences, m5C is found also in symmetrical CHG and nonsymmetrical CHH sequences. Both the maintenance DNA METHYLTRANSFERASE 1 (MET1) VARIANT IN METHYLATION 1 (VIM1) and DECREASE IN DNA METHYLATION 1 (DDM1), a SWITCH/SUCROSE NON FERMENTABLE 2 (SWI2/SNF2) DOMAIN chromatin-remodeling factor, are required for maintenance of CG methylation in A. thaliana (Finnegan et al., 1996; Kankel et al., 2003). CHROMOMETHYLASE 3 (CMT3) methylates DNA in the CHG context (Lindroth et al., 2001). De novo DOMAINS REARRANGED DNA METHYLTRANSFERASES 1 and 2 (DRM1/DRM2) are involved in DNA methylation in CHH contexts, including CG and CHG contexts (Cao & Jacobsen, 2002a,b).

DNA methylation is removed by four glycosylases of the REPRESSOR OF SILENCING 1 (ROS1) family (Agius et al., 2006; Morales-Ruiz et al., 2006; Zhu et al., 2007; Ikeda & Kinoshita, 2009). This family includes DEMETER (DME), which is expressed in the central cell (Choi et al., 2002).

In both A. thaliana and rice, endosperm DNA is globally less methylated than that of the embryo (Gehring et al., 2009; Hsieh et al., 2009; Zemach et al., 2010). DNA methylation at CG and CHG sites is maintained through cell division in a semiconservative manner. This suggests that the low DNA methylation in the endosperm may originate from a template of low methylation in the central cell.

The pattern of DNA methylation during female gametogenesis has not been established directly. However, in A. thaliana, the expression of MET1 and DME strongly suggests that the central cell experiences genome-wide DNA demethylation. MET1 expression is repressed in the central cell by the retinoblastoma pathway (Jullien et al., 2008). In addition, DME removes methylated cytosine residues via a mechanism involving single-strand break DNA repair (Choi et al., 2004). In maize, there is direct evidence that DNA methylation is reduced in the central cell at certain loci (Gutierrez-Marcos et al., 2006; Jahnke & Scholten, 2009). The idea that the central cell genome is demethylated is also supported by cytological evidence. Constitutive heterochromatin which forms chromocenters is absent from the central cell, which suggests global DNA demethylation (Jullien & Berger, 2010a). Taken together, these results suggest that the central cell chromatin is less methylated than that of somatic cells, but the degree of DNA demethylation and whether it affects only CG or also non-CG contexts remain to be established.

Chromatin modifications in the endosperm  In addition to DNA methylation, chromatin activity and transcription are regulated by modifications of histones. The chromatin of endosperm nuclei does not show the condensed heterochromatic bodies (chromocenters) observed in somatic nuclei. As noted in the previous paragraph, this may result from a low level of DNA methylation. Cytological investigations suggest that endosperm chromatin is enriched in histone H3 lysine 9 monomethylation (H3K9me) (Baroux et al., 2007). The functional significance of this mark remains unknown.

H3 lysine 27 trimethylation (H3K27me3) has a strong impact on endosperm development. In A. thaliana, the four FIS genes encode the members of the PRC2 family, which methylates the lysine residue 27 from histone H3 N-terminal tail. Loss of function in any of the FIS genes causes a maternal gametophytic effect on endosperm development, as detailed in the above section reviewing endosperm development.

2. Imprinting in the endosperm

In contrast with most genes expressed equally by both parental alleles, imprinted genes are differentially expressed depending on their parental origin.

The first example of imprinted expression of a gene was identified by Jerry Kermicle in a study of pigmentation of the outer layers of the endosperm in maize (Kermicle, 1970, 1978). Irregular anthocyanin pigmentation was linked to certain alleles of the R gene and was conferred only when the mutation was inherited from the mother. J. Kermicle proposed that expression of the r allele depended on its parental origin, and he further showed that the pigmentation defects associated with the r mutation did not depend on gene dosage. Genetic studies led to the identification of five imprinted genes in A. thaliana from 1999 to 2010. Genome-wide sequencing techniques have since led to the identification of > 50 new imprinted genes in A. thaliana (Berger & Chaudhury, 2009; Gehring et al., 2011; Hsieh et al., 2011; Wolff et al., 2011), at least 56 loci in rice (Luo et al., 2011) and > 200 loci in maize (Waters et al., 2011; Zhang et al., 2011). Although these studies show that imprinting is largely confined to the endosperm, a few imprinted genes are expressed in the embryo. Further work is required to confirm the imprinted status of the candidate imprinted genes and to determine the roles they play in the endosperm, if any.

Molecular mechanisms controlling imprinting  Parental genomic imprinting originates in epigenetic mechanisms acting during gametogenesis which differentiate the transcriptional states of the two prospective parental alleles (Fig. 3). During vegetative development, most imprinted genes are silenced by either DNA methylation or histone3 lysine 27 (H3K27) methylation. It is predicted that chromatin modifications (cytosine or H3K27 methylation), which silence gene expression, are removed by the end of female gametogenesis but are maintained in male gametes. After fertilization, the difference between the transcriptional statuses of the two parental alleles persists, through semiconservative mechanisms, leading to stable imprinted expression in the endosperm.

Figure 3. DNA methylation-dependent mechanisms leading to imprinting of maternally expressed genes in Arabidopsis thaliana. METHYLTRANSFERASE1 (MET1) maintains CpG methylation silencing marks on the parental alleles of imprinted genes (gray triangle). During male gametogenesis, CpG methylation is maintained in sperm cells. The methylation mark is removed in the central cell and the gene becomes expressed. Hence the endosperm inherits a silenced paternal allele (p) and an active maternal allele (m), resulting in monoparental imprinted expression in the endosperm.

Control by DNA methylation  Silencing of many imprinted genes is mediated by MET1, which maintains DNA methylation of CpG sites. In maize, analysis of the locus Fertilization independent endosperm1 (Fie1) showed that DNA methylation is present in sperm cells and is specifically removed in the central cell, but not in the egg cell (Gutierrez-Marcos et al., 2006). This direct assessment of DNA methylation in isolated central cells shows that epigenetic marks differ between gametes and foreshadow the imprinted expression after fertilization in the endosperm.

In A. thaliana, the synergistic action of passive demethylation by repression of MET1 activity followed by active demethylation by DME may completely demethylate the cis elements in promoters of imprinted genes, causing expression of these genes in the central cell (Fig. 3). After fertilization, the active maternal allele is inherited with a demethylated cis element while the inactive paternal allele is inherited with a fully methylated cis element. This imbalance results in monoparental expression (Kinoshita et al., 2004; Jullien et al., 2006b; Tiwari et al., 2008; Berger & Chaudhury, 2009; Gehring et al., 2009; Jullien & Berger, 2009). Such a mechanism is likely to apply to most maternally expressed imprinted genes silenced by MET1 in sperm cells.

Control of imprinting by histone methylation  MEDEA (MEA) was the first imprinted gene identified in A. thaliana (Grossniklaus et al., 1998; Kinoshita et al., 1999; Vielle-Calzada et al., 1999). In vegetative tissues, both alleles of MEA are silenced by H3K27 tri-methylation mediated by PRC2 complexes (Gehring et al., 2006; Jullien et al., 2006a). Compromising H3K27 trimethylation in mutants for PRC2 activity causes MEA ectopic expression in pollen and affects MEA imprinting in the endosperm (Gehring et al., 2006; Jullien et al., 2006a).

The transcription of the gene AtFH5 is also directly controlled by PRC2 complex activity (Fitz Gerald et al., 2009). AtFH5 expression is silenced by PRC2 complexes active in vegetative tissues before gametogenesis. Unlike MEA, AtFH5 expression is not activated in the central cell, but only the maternal allele of AtFH5 is expressed after fertilization. AtFH5 expression is also confined by PRC2 activity to the posterior pole of the endosperm, suggesting additional transcriptional controls.

Genome-wide studies have provided additional examples of imprinted genes that are probably controlled by PRC2 (Weinhofer et al., 2010; Gehring et al., 2011; Hsieh et al., 2011).

A major challenge is to understand what mechanisms remove the H3K27 methylation mark from the expressed allele of maternally expressed imprinted genes silenced by PRC2.

Paternally expressed imprinted genes  The gene PHERES1 (PHE1) is expressed primarily by the paternal allele in the endosperm (Kohler et al., 2005). PHERES1 is silenced in vegetative tissues by PRC2. Long-distance regulatory elements are essential in mammalian imprinting regulation (Bartolomei, 2009). In plants a comparable regulatory mechanism affects PHE1 imprinting (Makarevich et al., 2008). The mechanism responsible for removal of the silencing H3K27 methylation marks from the PHE1 locus in sperm cells, despite the presence of functional PcG, remains unknown. However, MET1 also appears to regulate PHE1 imprinting. A methylated repeat region is located 2.6 kb from the 3’ end of the PHE1 coding sequence and its methylation is required for the maintenance of the expression of the paternal PHE1 allele.

New paternally expressed imprinted genes have been identified using genome-wide strategies. In rice and maize, paternally expressed imprinted genes are more frequent than maternally expressed imprinted genes (Luo et al., 2011; Zhang et al., 2011) but the mechanisms that cause imprinting are not known.

Function of imprinted genes  As DNA methylation by MET1 has a strong impact on silencing of the paternal allele of imprinted genes, removal of paternal silencing from all loci seemed to be a valuable approach to assess the global function of imprinted genes. The loss of MET1 during male gametogenesis inhibits endosperm growth and results in smaller seeds (Xiao et al., 2006; FitzGerald et al., 2008). By contrast, the inheritance of met1 by the female gamete has no effect on endosperm and seed development (Xiao et al., 2006; FitzGerald et al., 2008). This result could be explained by the reduced expression of MET1 in female gametes (Jullien et al., 2008). Hence it is likely that the maternally expressed imprinted genes act together to repress endosperm growth. This might explain the similar small seed size phenotype caused by the increased dosage of the maternal genome in crosses involving a tetraploid mother (Scott et al., 1998). However, this simple interpretation does not reflect the real mechanism involved, as it has been shown that an imbalance in the parental genome dosage deregulates imprinting more globally, probably through the action of FIS genes (Walia et al., 2009; Jullien & Berger, 2010b; Wolff et al., 2011).

Besides the roles already described for FIS2 and MEA in the endosperm, only a few other imprinted genes have been associated with specific functions in the endosperm (Wollmann & Berger, 2012). It is also important to note that several imprinted genes such as FLOWERING WAGENINGEN A (FWA), MATERNALLY EXPRESSED PAB C TERMINAL (MPC), and SUPPRESSOR OF DRM1, CMT3 (SDC) do not appear to play a major role in the endosperm. Many candidate imprinted loci are expressed by both parental alleles, although with higher expression by one parental allele (Gehring et al., 2011; Hsieh et al., 2011; Luo et al., 2011; Waters et al., 2011; Wolff et al., 2011; Zhang et al., 2011). It is thus difficult to test to what extent monoparental expression matters for these loci.

Evolution of imprinted genes  Imprinting arose independently in plants and mammals, which are both characterized by a mode of reproduction involving maternal nutrition of the developing embryo through specialized tissues. According to a scenario where a mother produces offspring by different fathers, embryos carrying different paternal genomes compete for resource allocation from the mother. It is beneficial for each father to divert as many maternal resources as possible to his own embryos. When kin selection is considered, it is advantageous for the mother to down-regulate such interference from the father to ensure equitable distribution of nutrients to each of her offspring. These considerations led to the parental conflict theory, which predicts positive selection of maternally expressed growth inhibitors and paternally expressed growth enhancers (Wilkins & Haig, 2003; Haig, 2004; Haig & Wilczek, 2006).

This reproductive scenario associated with the parental conflict theory applies to certain mammalian and out-crossing plant species but does not apply to self-fertilizing A. thaliana, in which imprinting, nonetheless, is found to be active. Even if one assumes that ancestors of mammals and flowering plants were obligate out-breeders with no restriction on the numbers of male partners, the parental conflict theory also suggests that imprinted genes should no longer be under positive selection in species that always self-fertilize. It appears that MEA has not been subjected to strong positive selection in A. thaliana (Kawabe et al., 2007; Spillane et al., 2007; Miyake et al., 2009). Similarly, no positive selection was detected for the MEA maize homolog Maize enhancer of zeste 1 (Mez1) (Haun et al., 2007). In addition, the functional analysis of Maternally expressed gene1 (Meg1) in maize (Gutierrez-Marcos et al., 2004) provides a case for a maternally expressed gene that promotes growth, against the prediction of the parental conflict hypothesis (Costa et al., 2012). Meg1 is required for the establishment and differentiation of the endosperm nutrient TCs which allow the intake of maternal nutrients. Importantly, increasing the Meg1 dosage results in an unequal investment of maternal resources into the endosperm, leading to larger seeds. Together, these studies do not provide strong support for the conflict theory.

A genome-wide census of imprinted genes in A. thaliana, rice and maize showed limited conservation among related genes or even functional classes (Waters et al., 2011; Zhang et al., 2011; Wollmann & Berger, 2012). Together with the lack of an obvious function of several imprinted genes identified in A. thaliana, these findings raise the question of whether their functions in the endosperm have led to positive selection of imprinted genes or whether other scenarios must be envisaged. However, there is a strong conservation of imprinting of genes encoding subunits of PRC2 in plants between dicots and monocots (Haun et al., 2007; Luo et al., 2009). It remains to be established whether a high degree of conservation of imprinting on genes encoding proteins with specific function such as PRC2 is the exception rather than the rule.

In plant species with an obligate out-crossing reproductive strategy, imprinting may play a role in the prevention of hybridity (Josefsson et al., 2006; Walia et al., 2009). Interspecific crosses are possible in related plant species but often lead to reduced seed fertility. It is thus possible that imprinting has been selected as a mechanism regulating hybrid seed viability and as such would be crucial for speciation.

VII. Conclusions

The last 10 yr of research on the endosperm have led to remarkable advances in our understanding of the impact of epigenetic controls on reproduction and to the definition of key features of seed biology. Endosperm cellular biology is equally fascinating, although we are still far from understanding the basis of syncytial divisions and growth. Although apparently dissimilar, the short-lived endosperm of A. thaliana and the endosperm of cereals, which stores seed reserves, share a number of molecular and developmental features. Based on these similarities, next-generation sequencing technologies and new noninvasive techniques to image and measure physiological parameters will play an essential role in furthering our mechanistic understanding of endosperm biology. Although early encouraging results were obtained using natural variation, this resource remains to be used fully to probe how the endosperm controls seed growth and reserves. Ultimately, the relative conservation of developmental mechanisms and physiological regulation certainly offers hope that the conceptual framework derived from basic studies will provide useful tools and indicators for breeders seeking to develop crops with higher performance.

Acknowledgements

This work was funded by Temasek Life Sciences Laboratory. We thank the anonymous reviewers of this work for their input. We apologize for not having cited as many original references as we would have liked as a consequence of space constraints.