One central and defining characteristic of land plants is the alternation of multicellular generations: a haploid entity (the gametophyte) differentiates gametes, gametes fuse to form a zygote, and the zygote gives rise to a diploid entity (the sporophyte), which forms haploid spores by meiosis. In contrast, in the aquatic ancestors of land plants (the charophyte algae) the fertilized zygote is the only diploid cell and there is no multicellular spore-producing generation. Even in derived and complex charophytes, such as Coleochaete (where egg cells are enveloped and protected by a layer of gametophytic cells) the fertilized zygote directly undergoes meiosis to form free-dispersing haploid spores (1).

One of the crucial innovations of the first land plants was an intercalation of mitotic divisions in the zygote before meiosis, causing the development of a multicellular embryo (a young sporophyte) within gametophytic tissue (2, 3). The sheltered multicellular sporophyte increased the number of spores that could be produced from a single water-dependent fertilization event, offering a huge competitive advantage on land. The rapid increase in the size and complexity of sporophytes that followed led to an explosion of land plant forms in the Devonian period (4), perhaps as soon as 50 million years after the transition to land.

While the sporophytes grew, the gametophytes (which still nourished and protected the embryos) became smaller. Eventually, female gametophytes were themselves retained within parental sporophytic tissue (Figure 1). The oldest evidence for these changes is 385 million year-old fossils of Runcaria (5). Later, the evolution of specialized integuments resulted in the appearance of the first true ovules and seeds around 365 million years old (6).

Figure 1

Progressive retention of the diploid embryo within parental tissues during land plant evolution.

Photograph credits: Sandy__RR, ‘Moss 1’, licensed under Creative Commons, downloaded from http://www.flickr.com/photos/35142635@N05/5475014490/ on 27 October 2013; GaggieITMI, ‘Fern’, licensed under Creative Commons, downloaded from http://www.flickr.com/photos/gaggieitmi/8154028076/ on 27 October 2013; Scott Robinson, ‘Pine Cones’, licensed under Creative Commons, downloaded from http://www.flickr.com/photos/clearlyambiguous/17204706/on 2 October 2013; Mr. Tonreg, ‘Nuphar lutea’, licensed under Creative Commons, downloaded from http://www.flickr.com/photos/63169246@N00/8444876794/ on 27 October 2013; Tom Hilton, ‘Polygonum paronychia’, licensed under Creative Commons, downloaded from http://commons.wikimedia.org/wiki/File:Polygonum_paronychia.jpg on 27 October 2013.

These first seeds (and the seeds of modern gymnosperms) consisted of three distinct genetic generations: 1) a sporophytic embryo, nourished by 2) a female gametophyte enveloped in 3) a maternal sporophytic coat (Figure 1). With the retention of female gametophytes within sporophytic tissues, vascular plants achieved a greater independence from water for fertilization, while the production of seeds provided a sophisticated and effective method of dispersal. Seeds are usually very resistant structures that can stay dormant and travel long distances before germination, a process that is supported by nutritional reserves accumulated in the female gametophyte. The success attained by seed-producing plants is attested by the dominance of gymnosperms in the world flora for most of the last 300 million years (7).

In the Early Cretaceous period (100–145 million years ago), a group of seed plants evolved another set of extremely successful reproductive innovations that made them the dominant plant group in terrestrial environments for the last 100 million years (7). The most obvious innovation of this group – the flowering plants or angiosperms – is the flower, a sophisticated complex of reproductive organs that promotes pollination and fertilization. A second major innovation of flowering plants is that their ovules are harbored in an ovary that develops into a fruit after fertilization, offering another layer of protection to the embryo and greatly increasing the potential for effective seed dispersal. A third more subtle but equally revolutionary innovation was the evolution of the endosperm, a biparental entity that acts as an embryo-nourishing tissue.

The endosperm evolved in parallel with a double fertilization mechanism that is universal in flowering plants. The pollen tube (male gametophyte) releases two sperm cells into the embryo sac (female gametophyte): one sperm cell fertilizes the egg cell to give rise to a zygote, while the second sperm cell fuses with the central cell of the embryo sac to give rise to the endosperm. The endosperm develops together with the sibling embryo, nourishing it and carrying food reserves that are often used later during seed germination.

The emergence of a biparental nourishing endosperm in the seeds of flowering plants is a pinnacle of an evolutionary trend towards increased parental control over plant embryo development. Developing angiosperm seeds are complex multi-generational structures that contain 1) a sporophytic embryo embedded in 2) sibling endosperm, 3) the receding female gametophyte [which can persist during early seed development in a few species (8)] and 4) a maternal sporophytic seed coat (Figure 1).

With the evolution of double fertilization, the control over embryogenesis shifted from the female gametophyte to the biparental endosperm, breaking maternal hegemony over the control of embryogenesis. The intrusion of fathers on the control of the embryogenesis process set the scene for potential conflicts between maternal and paternal genomes over the allocation of resources during embryogenesis. The clash of paternal, maternal and offspring interests that ensued will be discussed later in this review.

The evolution of flowering plants in the Cretaceous period was an extremely dynamic and innovative time in the history of eukaryotic life. Reconstructing the evolutionary steps that resulted in the appearance of flowers and the radiation of angiosperms has been a puzzle for botanists since the time of Darwin (‘an abominable mystery’). Tracing back the evolutionary origins of the endosperm has similarly been an arduous task since the discovery of double fertilization at the end of the 19th century (9, 10).

Two competing hypotheses were soon advanced to explain the origin of the endosperm. It was proposed to be either homolog to a supernumerary embryo that fails to develop into a plant (11) or homolog with the female gametophyte (12, 13, reviewed in 14). This discussion went on through the 20th century, and until today the mechanisms that drove the evolution of double fertilization are not fully understood. Surprisingly, the Gnetales (a derived lineage of gymnosperms) were found to also undergo a well-defined double fertilization event, where a second sperm cell nucleus fuses with a sister egg cell nucleus, giving rise to a second embryo (15, 16). The discussion over the phylogenetic positioning of the Gnetales (nested within other gymnosperms vs. sister to the flowering plants) is not yet closed (17). Nevertheless, the presence of a double fertilization mechanism in the Gnetales suggests that the ancestors of flowering plants (or indeed the first seed plants) were experimenting with different types of multiple fertilization, and that a pro-endosperm could have potentially evolved from a second altruistic (or subjugated) embryo.

Further circumstantial evidence supporting the supernumerary embryo hypothesis comes from a comparative analysis of endosperm ontogeny. In many flowering plants, including the model plant Arabidopsis thaliana, endosperm development is initially free-nuclear (i.e., no cell walls are laid during the first nuclear divisions); whereas embryo development is always cellular (cell wall formation follows nuclear divisions). However, in basal flowering plants this distinction is not so clear: in a similar way to the embryo, endosperm development is initially cellular and forms two distinct chalazal and mycropylar domains (18, 19). Interestingly, embryo development in most gymnosperms is itself free-nuclear (20).

The process of double fertilization and the resulting production of a biparental endosperm is a hallmark of sexual reproduction in flowering plants. The evolution of an endosperm drastically changed the regulation of embryo development from being an exclusive maternal affair (controlled through the female gametophyte and sporophyte) to allowing a substantial degree of paternal control. Different hypotheses have been advanced to explain the advantages conferred by an endosperm (reviewed in 21). Higher levels of heterozygosity and ploidy in the endosperm were initially suggested to allow a more vigorous role in embryo nutrition than the female gametophytes of gymnosperms could provide. Another set of theories viewed the rise of the endosperm as the outcome of a conflict of interests between mother, fathers and offspring in the allocation of resources from the maternal sporophyte [kin conflict (22–24)]. These conflicts arise because each individual offspring competes with its siblings for resources from the maternal sporophyte. Each mother is equally genetically related to all its siblings, so it is in its interest to provide equal levels of nourishment to all its embryos. By contrast, the offspring of a father is typically in direct competition with offspring from other fathers. It is in the interest of the father to maximize the allocation of resources to its own offspring, at the expense of other progeny from the same mother. Therefore, fathers are predicted to try to maximize nutrient allocation and the growth of seeds, whereas mothers are predicted to equitably allocate resources and constrain seed growth. In earlier seed plants (and in extant gymnosperms), the only option for male progenitors to increase the success of their offspring was to improve the fitness of the male gametophytes, gametes and embryos. These were often in direct competition with those of other males, because in gymnosperms multiple embryos develop within each female gametophyte until only one embryo becomes dominant (25). With the evolution of a biparental nourishing tissue, angiosperm fathers could have a more direct role in the nourishing and development of their offspring.

Soon after the evolution of a biparental endosperm, some lineages of early flowering plants doubled the ploidy of the central cell (the maternal precursor of the endosperm). This resulted in a doubling of the maternal chromosome contribution to the endosperm, allowing mothers to regain privileged control over embryo development.

Endosperm and embryo sac genetics are intimately related. There is a high diversity in embryo sacs types, particularly in the ploidy and genetic composition of central cells, among different flowering plants (8). This diversity can be easily understood in the context of embryo sac ontogenesis (Figure 2). Embryo sacs derive from haploid megaspores that are produced through meiosis from a diploid megaspore mother cell. In most seed plants, three of the four megaspores degenerate. The surviving one (the functional megaspore, situated at the chalazal end) undergoes three free-nuclear divisions, forming eight nuclei within a syncytium. The syncytium later cellularizes to give rise to two synergids and an egg cell at the micropylar end, a central cell with two nuclei, and three antipodals at the chalazal end. This type of monosporic seven-celled/eight-nucleate embryo sac, known as the Polygonum-type, gives rise to a triploid endosperm after fertilization; because the endosperm is derived from a central cell with two nuclei it has a 2:1 ratio of maternal to paternal chromosomal contributions. However, embryo sacs can be derived not only from a single functional megaspore (monosporic type), but also from two megaspores (bisporic type) or even from all four meiotic products (tetrasporic type). Further variation occurs in the number and spatial organization of mitotic nuclei in the mature female gametophyte. A comparative analysis in different types of flowering plants suggests that embryo sacs are modular structures based on repetitions of developmental motifs (26, 27). A basic module is a quartet, formed by a nucleus that undergoes two free-nuclear mitoses to yield four nuclei (Figure 2). Three of these nuclei are partitioned into a pole of the embryo sac while the fourth contributes to the common cytoplasm of the central cell. In Polygonum-type embryo sacs one quartet forms the micropylar egg apparatus (egg cell plus two synergids) and one of the polar nucleus of the central cell; the second quartet corresponds to the three antipodals plus the second polar nucleus of the central cell.

Figure 2

Development of Polygonum-type and Nuphar-type embryo sacs, showing how a duplication of a quartet module could give rise to the doubling in ploidy of the central cell.

ch, chalazal pole; mi, micropylar pole; cc, central cell; sy, synergid; ec, egg cell; an, antipodals.

Unlike the vast majority of angiosperms, the basal flowering plant lineages Nymphaeales and Austrobaileyales have a single basic module that gives rise to a monosporic four-celled/four-nucleate embryo sac (Nuphar-type) (26). The central cell of these embryo sacs have a single nucleus, and therefore gives rise to diploid endosperms. The single quartet module giving rise to the four-celled embryo sac and diploid endosperm of the Nymphaeales and Austrobaileyales is possibly the ancestral condition, present in the first flowering plants. If this hypothesis is correct, a duplication of this module appears to have occurred independently at least twice: once in the Amborella lineage (the most basal of living flowering plants, and which has eight-celled embryo sacs) (28) and also in the common ancestor of monocots, eudicots and magnoliids (26, 27). Later, in different groups of flowering plants, further duplications of the basic quartet module gave rise to bisporic embryo sacs (which also originate triploid endosperms), and tetrasporic embryo sacs (which originate triploid, pentaploid, nonaploid or decapentaploid endosperms) (8, 29).

After the evolution of triploid endosperms there were practically no reversals to diploid endosperms or to a strategy of embryo nourishment directly through the female gametophyte [one possible exception are the Nymphaeales, where the diploid endosperm is reduced and the nourishment of the embryo is provided by a perisperm derived from the maternal sporophytic nucellus (30, 31)]. An endosperm with an asymmetric contribution of parental genomes became an integral part of the embryogenesis process of flowering plants.

A correct balance of parental genome contributions on the endosperm is crucial for correct endosperm development. When individuals of different ploidies are crossed, the endosperm often fails to develop properly. This phenomenon, called triploid block, was initially puzzling to plant breeders because while interploidy crosses failed, crosses between tetraploids or hexaploids (resulting in progeny with various levels of ploidy) were viable. Triploid block was eventually explained as a requirement for a correct balance of 2:1 maternal to paternal genomes in the endosperm (32–35).

If the paternal genome contribution is in excess (e.g., when crossing a diploid female with a tetraploid male, resulting in a 2:2 maternal to paternal ratio of chromosomes in the endosperm), the endosperm typically over-proliferates because of an accelerated rate of mitotic division and delayed cellularization (36, 37), associated with an altered timing of cell cycle progression (38). In cases where viable seeds can be produced from paternal excess crosses, seeds are larger and heavier than normal seeds. Conversely, if the maternal contribution is in excess (e.g., in the reciprocal cross of a tetraploid female with a diploid male, resulting in a 4:1 maternal:paternal ratio in the endosperm), the endosperm exhibits reduced mitotic divisions and precocious cellularization, resulting in seed abortion or the production of smaller and lighter seeds.

The opposite effects of reciprocal interploidy crosses on endosperm development demonstrate that the parental genomes are not equivalent and that mothers and fathers have opposite impacts on the growth of their offspring. Interestingly, many interspecific crosses result in reciprocal endosperm failure phenotypes that resemble the failure of interploidy crosses (reviewed in 39–41). Increasing the ploidy of one of the parents can sometimes rescue otherwise unsuccessful crosses (40, 42). This suggests that the common cause of endosperm failure following interspecific and interploidy crosses is a disruption of the balance of maternal and paternal gene expression (39, 43).

A simplistic interpretation of the effects of interploidy crosses on seed development is that the maternal genome restricts endosperm development, while the paternal genome promotes it, thereby increasing the potential for the formation of larger and better nourished embryos. Maternal control of seed size can also be exerted directly through the (maternal) seed coat. In contrast, endosperm vigor is controlled by both maternal and paternal genomes. The transition from the syncytial phase (during which endosperm nuclei proliferate through multiple rounds of mitosis without cytokinesis) to the cellularized phase (during which cell walls are formed) is an important developmental transition during seed development (44–46). Accordingly, the timing of endosperm cellularization is particularly sensitive to the balance of parental genomes, with maternal excess crosses typically promoting premature endosperm cellularization and paternal excess promoting a delaying of endosperm cellularization (47). Nevertheless, the maternal nature of seed coats and the higher genome dosage in the endosperm allows the maternal side to have a strong control on seed size. In Arabidopsis, seed size has been shown to be predominantly determined by the maternal genotype, while the paternal genotype explains only around 10% of the variation (48).

The balance of paternal and maternal gene expression can be regulated in two main ways: by the relative ploidy of the central cell (discussed above) and by the differential expression of alleles from each parent. One extreme case of differential allelic expression is imprinting, where only one of the parental alleles is expressed. The maize R gene was the first imprinted locus to be discovered (49) and today several imprinted genes are known in flowering plants and mammals (reviewed in 50–52). The evolutionary origins of imprinting are not well understood, but a theory proposed by David Haig and Mark Westoby (the parental-conflict theory of the evolution of genomic imprinting) offers a popular (but sometimes misinterpreted) explanation for some types of imprinting (39, 53). The assumption is that in a maternal plant with progeny from different fathers, paternally-derived genes in any given seed have an interest in maximizing the success of that seed (at the expense of seeds from other fathers), whereas maternally-derived genes have an interest in also promoting the success of other seeds. This hypothesis predicts that this conflict favors the expression of maternal genes that silence paternal genes that would otherwise increase nutrient acquisition demands; this eventually gives rise to mono-allelic expression (imprinting).

It is important to highlight that this theory does not explain all types of imprinting, and that imprinting can explain some, but not all, parent-of-origin effects. Parent-of-origin effects are perhaps better explained by the differential dosage hypothesis that involve differential biallelic expression (and not just monoallellic expression) of dosage-sensitive regulators (43). In plants, most imprinted genes are expressed in the endosperm, but there are also some genes imprinted in the embryo (54, 55). Interestingly, early embryo development may be predominantly controlled by maternal transcripts, although different studies have come to different conclusions (56–60).

In the next section the two main epigenetic mechanisms that control imprinting will be discussed: DNA methylation (particularly CG methylation) and histone methylation, particularly repressive H3K27me3 marks catalyzed by polycomb group proteins.

DNA methylation

In Arabidopsis, the DNA methyltransferase MET1 is responsible for the maintenance of genome-wide CpG methylation (61). Crosses between a hypomethylated met1 mutant pollen donor and a normal seed parent mimic the phenotype of a maternal excess cross between a diploid father and a tetraploid mother, resulting in small seeds and in a reduction of endosperm mitotic divisions (62, 63). This suggests that parent-of origin effects require DNA methylation. One explanation is that the release of repressive methylation marks from paternally silenced genes originates extra transcription of otherwise maternal-specific genes, originating a maternal excess phenotype. The reciprocal cross of a hypomethylated met1 seed parent with normal pollen mimics the phenotype of a paternal excess cross (40, 62, 63), although this appears to rather be caused by a sporophytic effect in the cell proliferation in the seed integuments, and not by a gametophytic effect (64). The maintenance of MET1 CpG methylation in the gametophytic phase has been shown to be essential for the inheritance of epigenetic marks (65). Together, these observations suggest that CpG methylation mediated by MET1 plays an important role in the maternal control of seed development by restricting cell proliferation in the sporophytic integuments and by restricting the expression of paternal genes in the endosperm.

The imprinting of several genes is mediated by DNA methylation. In Arabidopsis, DNA methylation is predominantly controlled by the antagonist action of the DNA methyltransferase MET1 and the DNA demethylase demeter (DME). DME is expressed in the central cell of the embryo sac before fertilization but not in pollen sperm cells (66, 67), whereas MET1 is strongly expressed in sperm cells but repressed in the central cell by the retinoblastoma pathway (68–70) (Figure 3). As a consequence, the maternal genomes that are contributed to the endosperm are demethylated relative to the paternal genomes and embryo genomes (71–73). The DNA demethylation that occurs in the central cell is partially responsible for the maternal expression of imprinted genes such as MEA, FWA, FIS2 in Arabidopsis and FIE2 in maize (74–78).

Figure 3

Highly simplified diagram showing the genetic and epigenetic mechanisms that balance paternal and maternal information during gametogenesis and seed development.

The failure of endosperm development following incompatible interspecific/interploidy crosses and in PRC2 mutants is mediated by several type-1 MADS box proteins. Many of these transcription factors have recently been identified as central regulators of plant reproductive development, from patterning the female gametophyte to controlling the timing of endosperm cellularization (112).

PHERES1 (PHE1) is maternally silenced by H3K27me3 marks deposited by the maternally active FIS-PRC2, so that only the paternal allele is expressed (113–115). The imprinting of PHE1 is lost in fis mutants (114) and following incompatible crosses between A. thaliana and A. arenosa (116). The mea endosperm over-proliferation phenotype is partially mediated by PHE1, as reducing PHE1 expression can reduce mea seed abortion (113). Another key type-1 MADS box gene regulating endosperm development is AGL62. AGL62 is expressed during the proliferating phase of endosperm development and drops just before cellularization because of FIS-PRC2 activity (46, 117). agl62 mutants undergo precocious endosperm cellularization, suggesting that AGL62 suppresses cellularization during syncytial development (117). Other AGLs (PHE2, AGL35, AGL36, AGL40, and AGL90) are also upregulated following incompatible crosses between A. thaliana and A. arenosa that resemble paternal excess crosses (118), and several AGL genes are downregulated in uniparental endosperm (119). Conversely, in maternal excess crosses these AGL genes are downregulated (102). In A. thaliana, AGL36 has been shown to be expressed exclusively from the maternal allele (119).

These results suggest that AGL genes act downstream of FIS-PRC2 to positively regulate endosperm proliferation and/or negatively regulate endosperm cellularization (Figure 3). Supporting this hypothesis, inactivating PHE1, AGL62 or AGL90 reduces endosperm over-proliferation and reduces seed failure in fis mutants and in interspecific crosses between A. thaliana and A. arenosa (46, 113, 118). This hypothesis suggests that the defects in endosperm development associated with loss of FIS-PRC2 function, parental genome unbalance and interspecific crosses are at least partially caused by misexpression of AGL genes. Indeed, the role of AGL genes in controlling endosperm development appears to be quite ancient. In rice, at least one type-1 MADS box gene is maternally imprinted. Following interspecific crosses imprinting is lost and other rice MADS box genes also become deregulated (41). Interestingly, a genomic widespread disruption of imprinting has been reported in interspecific crosses in rodents (120).

MADS proteins are not the only mediators of the balance of parental genomes and FIS-PRC2 activity. Loss-of-function of a paternally expressed J-domain molecular chaperone, ADMETOS, can partially rescue the seeds derived from paternal excess crosses and from a mea mutant (121).

In the last years small RNAs were shown to play a central role in regulating the epigenetic state of the gametes and of the endosperm. In Arabidopsis, the production of 24nt PolIV-derived siRNAs (p4-siRNAs) in the endosperm is under strict maternal control (122). The maternal p4-siRNA population accumulates at high levels in the central cell and during endosperm development, suggesting that these small RNAs may move to the egg and embryo to reinforce silencing of TE through non-CG methylation (Figure 3) (123–125). This mechanism is still very unclear, and the 24nt RNA-directed DNA methylation (RdDM) mechanism was conversely suggested to be repressed in the female gametes, leading to activation of the maternal alleles and causing DME-independent imprinted expression in some loci (126).

A similar mechanism whereby an accessory cell loses its genome integrity in order to reinforce silencing in the gametes was proposed to occur in pollen. The chromatin remodeling ATPase DECREASE IN DNA METHYLATION 1 (DDM1) is a central regulator of TE activity in Arabidopsis. DDM1 accumulates in pollen sperm cells, but not in the pollen vegetative nucleus (127). The absence of DDM1 protein in the pollen vegetative nucleus causes a loss of heterochromatin and a massive activation of TE, which leads to the production of 21nt siRNAs. These siRNAs are then exported to the sperm nuclei to reinforce TE silencing (127).

The changes in the chromatin status of the vegetative nucleus of pollen and of endosperm nuclei (127–129), associated with the massive maternal and paternal production of 24nt and 21nt siRNAs are evidence of the existence of large epigenetic changes during gametogenesis and embryogenesis. This led to the proposal that a ‘genome shock’ causing TE to escape methylation is partially responsible for hybrid incompatibility (hybrid dysgenesis) following interspecific and interploidy crosses (130). However, it is not obvious how TE activation alone could lead to the characteristic over-proliferation and/or precocious cellularization phenotypes of parentally unbalanced endosperms. Furthermore, nrpd1a mutants, which are impaired in the production of p4-siRNAs, do not show an obvious difference in endosperm growth (122). Nevertheless, lack of p4-siRNAs induces an upregulation of AGL genes (131). In other words, the loss of the maternal p4-siRNAs paternalizes the endosperm, at a transcriptomic level. One possible explanation for the lack of a requirement for p4-siRNA production for normal seed development Arabidopsis may be that these mechanisms are relaxed in a predominantly inbreeding species. Interestingly, the 24nt p4-siRNA mechanism may have evolved with the flowering plants (132).

In incompatible interspecific crosses between A. thaliana and A. arenosa, ATHILA, a major Arabidopsis retrotransposon that is typically silenced, becomes expressed from the paternal alleles (116). This may be because maternal siRNAs are unable to repress the paternally-derived retrotransposons. However, other transposons than ATHILA do not appear to become activated in the A. thaliana x A. arenosa cross (116, 133), suggesting that retrotransposon activation is not a general consequence of interspecific crosses.

Seeds are multi-generational structures enveloped by a seed coat of maternal sporophytic origin. The seed coat offers an effective means through which mothers can regulate seed development, particularly by determining seed size. In Arabidopsis, several maternal sporophytic mutations that affect seed size and shape through integument (testa) development have been isolated (134–138). One important maternal sporophytic regulator of seed size in Arabidopsis is the WRKY transcription factor TRANSPARENT TESTA GLABRA2 (TTG2). The endosperm of seeds developing from maternal homozygous ttg2/ttg2 mutants cellularize precociously and seeds are small; however, maternal heterozygous ttg2/TTG2 mutants generate normal seeds, suggesting that TTG2 controls seed development as a maternal seed coat sporophytic factor, possibly by controlling integument cell elongation (138). Natural allelic variation at the TTG2 locus is also responsible for variation in the tolerance to interploidy crosses in Arabidopsis, and ttg2 mutants can reduce seed lethality in paternal excess and in interspecific crosses (133, 139).

Integument growth can nevertheless be regulated by the rate of endosperm growth. A putative pathway formed by the VQ motif protein HAIKU1, the leucine-rich kinase HAIKU2, the WRKY transcription factor MINI3 and the cytokinin oxidase CKX2 regulates endosperm proliferation; mutations in these genes cause reduced endosperm growth and precocious cellularization, and this negatively impacts on integument elongation (140–143). The initiation of seed coat development from ovule integuments upon fertilization is controlled by signaling from the endosperm. fie and msi1 mutant seeds can develop an endosperm in the absence of fertilization, and this autonomous endosperm development is sufficient to initiate seed coat development (87, 144, 145). Nevertheless, it has also been proposed that a correct signal initiating seed coat development requires a sexual, fertilized endosperm (145). Other non-FIS polycomb group proteins such as VERNALIZATION2, EMBRYONIC FLOWER2 and SWINGER act in the sporophytic maternal integuments to restrict autonomous seed development (145).

The last 15 years were a particularly productive period in which we greatly increased our understanding of how seeds evolved and how they develop. Sophisticated phylogenetic analysis and detailed comparative analyses at the base of the angiosperm family gave us a glimpse into some of the steps that resulted in the appearance of an endosperm. However, it is still far from clear how the transition from the polyembryonic seeds of gymnosperms gave rise to double fertilization and the endosperm in angiosperms, and how the diploid endosperm of early angiosperms gave rise to the triploid endosperm of Amborella and most angiosperm lineages. Over the next years it is expected that further comparative analyses will help to clarify this. In addition, the increasing power and availability of genomics and transcriptomics should also help us to understand how seed development in basal angiosperms (and indirectly in early angiosperms), relates to seed development in more derived lineages such as A. thaliana and the grasses.

Much of what we learned from seed and endosperm development in these 15 years was made possible by the many resources and tools available for Arabidopsis. These are continuously expanding, and large-scale analyses made possible by next-generation sequencing are already painting a detailed picture of the transcriptomic dynamics that occur throughout seed development. Omics approaches will become more recurrent in the future, and the massive amount of data accumulated will require the adoption of computational and mathematical modeling.

Conversely, some of the traits that make A. thaliana such a good tool for experimental biologists, particularly its inbred character, also make it a poor choice for studying parental conflicts and imprinting, as these are less intense in self-pollinating plants than in outcrossers (146). A. thaliana atypically tolerates crosses where the genome contribution of one of the progenitors is doubled (36), while MEA has been shown to evolve faster in the outcrosser A. lyrata than in A. thaliana (147, 148). Endosperm and epigenetics research in rice and maize have provided an invaluable complement to Arabidopsis research, but parental-conflict in these crop species can also be predicted to be distorted by the very strong artificial selection that these species underwent over the last thousands of years. A strategic sampling and study of non-model species will help us to have a clearer picture of how the interaction of paternal and maternal genomes controls seed development.

• Embryos of land plants were progressively embedded within parental tissues, a process that culminated in the evolution of seeds and the endosperm.

• The evolution of an endosperm allowed paternal genes to participate in seed development, setting the conditions for the manifestation of conflicts of interest between maternal and parental genomes.

• A correct balance of maternal and paternal gene dosage in the endosperm is required for correct seed development.

• The balance of parental gene dosages in the endosperm can be adjusted by changes in the ploidy of the central cell or by differential allelic expression (e.g., imprinting).

• MADS transcription factors are central regulators of the transition to endosperm cellularization.

• Similar mechanisms underlie the endosperm development failure phenotypes caused by PRC2 mutants, interploidy and interspecific crosses.

• The role of PRC2 complex in restricting autonomous endosperm development and over-proliferation may have evolved from a more ancient role in restricting apogamy.

• Small RNAs undergo massive de-repression in the endosperm and in the vegetative nucleus.

Many thanks to Ueli Grossniklaus and members of the Grossniklaus lab for invaluable discussions, and to two anonymous reviewers for constructive comments that improved the quality of this manuscript. The author declares no conflict of interest.