Epigenetic processes in flowering plant reproduction

Abstract

Seeds provide up to 70% of the energy intake of the human population, emphasizing the relevance of understanding the genetic and epigenetic mechanisms controlling seed formation. In flowering plants, seeds are the product of a double fertilization event, leading to the formation of the embryo and the endosperm surrounded by maternal tissues. Analogous to mammals, plants undergo extensive epigenetic reprogramming during both gamete formation and early seed development, a process that is supposed to be required to enforce silencing of transposable elements and thus to maintain genome stability. Global changes of DNA methylation, histone modifications, and small RNAs are closely associated with epigenome programming during plant reproduction. Here, we review current knowledge on chromatin changes occurring during sporogenesis and gametogenesis, as well as early seed development in major flowering plant models.

DNA methylation, epigenetics, histone modifications, plant reproduction, reprogramming, small RNAs.

Introduction

The formation of seeds is the evolutionary most successful reproductive strategy found in vascular plants. With more than 260 000 species, seed-forming flowering plants are the most species-rich lineage within the vascular plants (Soltis and Soltis, 2004). A seed consists of an embryo, a nutritive endosperm tissue supporting embryo growth and/or germination, and surrounding maternal tissues. The maternal tissues can be derived from the integuments forming a seed coat, as found in most dicots, or from transformed ovary walls forming a pericarp, as found in most monocots (reviewed in Olsen, 2004; Radchuk and Borisjuk, 2014).

Seeds synthesize and deposit storage substances either in the embryo or the surrounding nutritive tissue, ranging from starch to storage proteins and oil. These reserves not only nourish the embryo, they also provide up to 70% of the energy intake of the human population via food and animal feed (reviewed in Sreenivasulu and Wobus, 2013).

Double-fertilization occurs in angiosperms and in the group Gnetales, the closest living relatives of angiosperms (reviewed in Friedman, 1998). In angiosperms, two haploid sperm cells (1n) are delivered through the pollen tube to the female gametophyte and simultaneously fertilize the haploid egg cell (1n) and the homodiploid central cell (2n), generating a diploid embryo (2n) and a triploid endosperm (3n), respectively (reviewed in Bleckmann et al., 2014). In the dicot Arabidopsis, the embryo follows a predictable series of cell divisions leading to a mature embryo with shoot and root apical meristems, two cotyledons, and the hypocotyl (reviewed in Wendrich and Weijers, 2013). The endosperm differentiates into micropylar, peripheral/central, and chalazal domains and is then gradually consumed by the embryo, except for a single peripheral layer that remains at the mature stage and that controls seed germination (Penfield et al., 2006; reviewed in Li and Berger, 2012). By contrast, in monocots such as maize and rice, the endosperm is maintained after cellularization and differentiates into four major tissue types: the starchy endosperm, the aleurone, the transfer cells, and the embryo-surrounding region (reviewed in Olsen, 2004; Sabelli and Larkins, 2009). The fully developed maize embryo consists of shoot and root apical meristems, leaf primordia, and the monocot-specific coleoptile and scutellum, which are functional homologs to the dicot cotyledons (Takacs et al., 2012). Thus, monocot and dicot embryo and endosperm development follow distinctively different strategies that nevertheless lead to viable seed formation. Identifying common mechanisms governing this process is of utmost importance to reveal the general principles for viable angiosperm seed formation.

Accumulating evidence over recent years has revealed that gamete formation and early seed development in monocots and dicots are under epigenetic regulatory control. Covalent modifications of the core components of chromatin, DNA, and histones provide a heritable mechanism for the regulation of gene expression. Depending on the combination of histone modifications and DNA methylation, four distinct chromatin states can be distinguished: those that mark active or repressed genes, transposable elements (TEs), and intergenic regions (Roudier et al., 2011). Repression of TEs is required to ensure genome stability; therefore, TEs are generally located in transcriptionally silenced heterochromatic regions marked by DNA methylation and repressive histone modifications, with dimethylation on histone H3 at lysine 9 (H3K9me2) being the most prominent one. In plants, cytosine DNA methylation can occur in symmetrical (CG and CHG; where H corresponds to A, T, or C) and asymmetrical (CHH) DNA contexts (reviewed in Law and Jacobsen, 2010). Symmetric CG methylation is maintained by the METHYLTRANSFERASE 1 (MET1) enzyme, while non-CG context methylation (CHG and CHH) is dependent on H3K9me2 methylation established by KRYPTONITE (KYP/SUVH4), SUVH5, and SUVH6 proteins (reviewed in Law and Jacobsen, 2010; Du et al., 2012, 2015; Stroud et al., 2013). This modification is bound by the CHROMOMETHYLASE (CMT) proteins CMT2 and CMT3, which methylate adjacent non-CG sites of the newly replicated DNA (Du et al., 2012). CMT2 and CMT3 have overlapping functions, but CMT2 predominantly targets CHH sites, while CMT3 preferentially methylates CHG sites (Stroud et al., 2013). Typical targets of both enzymes are long TEs and gene-distal TEs (Stroud et al., 2013). Maintenance of CHH methylation at short, gene-proximal TEs as well as at the edges of long TEs requires the RNA-dependent DNA methylation (RdDM) pathway involving the plant-specific DNA-dependent RNA polymerases IV and V (Pol IV and Pol V). Pol IV is recruited via methyl-DNA-binding proteins (Law et al., 2013) and produces short precursor RNAs that are processed into 24-nt small interfering RNAs (siRNAs) (Blevins et al., 2015; Zhai et al., 2015a; Yang et al., 2016). In contrast, Pol V produces chromatin-associated scaffold transcripts at sites of DNA methylation (Haag and Pikaard, 2011). These scaffold transcripts are bound by Pol IV-dependent 24-nt siRNAs that recruit DOMAINS REARRANGED METHYLTRANSFEREASES 1 and 2 (DRM1 and DRM2) to maintain DNA methylation in all sequence contexts (reviewed in Kim and Zilberman, 2014; Zhong et al., 2014). For reasons not yet completely understood, plant gamete formation is associated with reduction of heterochromatin and the transcriptional activation of TEs. Genome stability during this sensitive phase is ensured by cleaving of TE transcripts, leading to the formation of 21–22-nt siRNAs that potentially can initiate de novo DNA methylation (reviewed in Cuerda-Gil and Slotkin, 2016).

In this review, we will discuss changes in chromatin structure associated with male and female gametogenesis, double-fertilization, and early seed development in Arabidopsis, rice, and maize, emphasizing the potential regulatory importance of changes in chromatin structure to ensure reproductive success.

Epigenetic changes associated with male lineage formation

The formation of the male lineage initiates with the differentiation of meiotic-competent pollen mother cells (PMCs or meiocytes) in the developing anthers. The meiotic products of PMCs are four microspores (1n). After completion of sporogenesis, each microspore enters gametogenesis, where it undergoes an asymmetric division to produce one vegetative cell (1n) and one generative cell (1n). The generative cell forms two sperm cells by undergoing one additional mitotic division either during pollen formation or pollen germination (reviewed in Feng et al., 2013; Schmidt et al., 2015) (Fig. 1).

Fig. 1.

Chromatin dynamics during sexual reproduction in Arabidopsis. Chromatin configurations during different stages of reproduction are color-coded, with red symbolizing compacted chromatin (heterochromatin content similar or higher compared to somatic tissues) while green symbolizes relaxed chromatin (lower heterochromatin content compared to somatic tissues). In the flower, pollen mother cells (PMCs) and megaspore mother cells (MMCs) are generated from somatic cells in the male and female reproductive tissues, respectively. PMCs and MMCs have reduced heterochromatin content compared to the surrounding cells. PMCs and MMCs will undergo meiosis, forming microspores and megaspores, respectively, that both have a relaxed chromatin configuration. The microspore undergoes an asymmetrical division, generating a vegetative cell and a generative cell that possibly differ in their chromatin compaction, with the generative cell being more compacted compared to the vegetative cell. The generative cell divides to form two sperm cells with highly compacted chromatin within the vegetative cell that will form the pollen tube. In contrast, the one surviving megaspore undergoes three rounds of nuclear divisions to generate an eight-nuclei-containing syncytial female gametophyte. The heterochromatin content in those nuclei is high compared to that of the megaspore. After cytokinesis, the mature female gametophyte consists of the egg cell, the central cell, and accessory cells (antipodals and synergids). The egg and central cells differ in heterochromatin content, with the central cell having reduced heterochromatin content compared to the egg cell. The egg cell and the central cell are each fertilized by one sperm cell, producing the zygote and the endosperm, respectively. Chromatin differences present between the egg and central cells are maintained after fertilization, with the embryo having a higher heterochromatin content compared to the endosperm. The movement of 21-nt easiRNA from the vegetative cell to sperm cells is symbolized by a solid arrow. The proposed movement of 21-nt siRNA from the central cell to the egg cell is symbolized by a dashed arrow.

Whole-genome expression profiles indicate that a subset of normally silenced TEs become active in Arabidopsis and maize PMCs (Chen et al., 2010; Yang et al., 2011; Dukowic-Schulze et al., 2014), suggesting reduced epigenetic silencing in PMCs. Consistently, PMCs establish a distinct chromatin state, characterized by a reduction of heterochromatin, removal of the linker histone H1, and changes in the pattern of histone modifications compared to that in somatic cells (She and Baroux, 2015). Sporogenesis in maize and rice is accompanied by the production of 21-nt and 24-nt phased siRNA (phasiRNA), triggered by microRNAs (miRNAs) miR2118 and miR2275 targeting PHAS loci, respectively. The formation of phasiRNAs depends on RDR6 (RNA DEPENDENT RNA POLYMERASE6) and DCL4/DCL5 (DICER LIKE 4/5) (Johnson et al., 2009; Song et al., 2012; Zhai et al., 2015b). While 21-nt phasiRNAs accumulate during PMC specification and then diminish, 24-nt phasiRNAs accumulate during meiosis and persist during gametogenesis (Zhai et al., 2015b). Premeiotic 21-nt phasiRNAs lack obvious targets (Song et al., 2012; Zhai et al., 2015b). Nevertheless, loss of MEL1 (MEIOSIS ARRESTED AT LEPTOTENE 1), an ARGONAUTE (AGO) protein that selectively binds to 21-nt phasiRNAs, causes abnormal tapetum formation and aberrant early-arrested PMCs (Nonomura et al., 2007; Komiya et al., 2014), suggesting an important function of 21-nt phasiRNAs for male meiosis. Thus, differentiation of the PMC is associated with general changes in chromatin structure and in particular relaxation of heterochromatin, causing TE activation. This may necessitate the activation of post-transcriptional silencing mechanisms acting via 21-nt phasiRNAs possibly targeting TE transcripts, ensuring genome stability.

The closest orthologue of MEL1 in maize, AGO5c, is coordinately expressed with 21-nt phasiRNAs, suggesting that AGO5c is the binding protein of 21-nt premeiotic phasiRNAs in maize (Zhai et al., 2015b). The recently evolved maize AGO18b gene is enriched in tapetal and meiotic cells and most likely loads meiotic 24-nt phasiRNAs (Zhai et al., 2015b). AGO18b is absent in dicots, and miR2275 and meiotic phasiRNAs have only been reported in grass species, suggesting recent evolution of meiotic phasiRNAs in grasses or related monocots (Zhai et al., 2015b). While miR2118 is present in dicots, the primary targets of miR2118 in dicots are NB-LRR pathogen-defense genes, suggesting that miR2118 has evolved distinct functions in monocots and dicots (Zhai et al., 2015b). Consistently, mutations in the closest Arabidopsis homolog of MEL1, AGO5, cause no defect during male meiosis, but a semi-dominant ago5-4 mutant has defects in the initiation of megagametogenesis and is female-sterile (Nonomura et al., 2007; Tucker et al., 2012). Functional evidence for a role of 24-nt siRNAs in regulating meiosis has been derived from the phenotypic analysis of the ago104 mutant in maize (Singh et al., 2011). AGO104 is required for establishing or maintaining DNA methylation at non-CG sites in heterochromatin, like its orthologue AGO9 in Arabidopsis. Mutations in AGO104 cause reduced chromatin condensation of meiotic chromosomes and the formation of unreduced gametes, revealing the functional requirement of siRNAs in meiosis. While the siRNAs bound by AGO104 remain to be identified, it seems most likely that AGO104 binds to 24-nt siRNAs like its orthologue AGO9 (Havecker et al., 2010). Despite the apparent absence of 24-nt phasiRNAs in Arabidopsis, the 24-nt siRNA-binding AGO4 protein is required for male meiosis (Oliver et al., 2016), suggesting that the requirement of 24-nt siRNAs and the RdDM pathway for regulating male meiosis is a conserved feature in angiosperms.

After meiosis, the resulting microspores in Arabidopsis lose CHH methylation from pericentromeric retrotransposons and satellite repeats, while CG and CHG methylation remains unchanged (Calarco et al., 2012). A similar pattern is observed in sperm cells, contrasting with restoration of CHH methylation in the vegetative nucleus (VN) (Calarco et al., 2012). Expression of DRM2 is low in microspores and sperm cells but present in the VN, correlating with the CHH methylation pattern (Calarco et al., 2012). Despite the presence of CHH methylation, TEs are strongly activated in the VN, most likely because the chromatin remodeler DDM1 in not active in the VN (Slotkin et al., 2009). Reactivation of TEs has also been reported in maize and rice pollen, revealing conservation of this phenomenon among flowering plants (Nobuta et al., 2007; Slotkin et al., 2009). Furthermore, the DNA glycosylase DEMETER (DME) is active in the VN but not in the sperm cells (Schoft et al., 2011), and is responsible for removing CG methylation at small euchromatic TEs (Ibarra et al., 2012). Activated TEs in the pollen vegetative cell give rise to epigenetically activated 21- and 22-nt siRNAs (easiRNAs) instead of the canonical 24-nt siRNAs (Slotkin et al., 2009). In somatic tissues of the ddm1 mutant, formation of 21/22-nt easiRNAs is triggered by miRNAs targeting active TEs and depends on RDR6, DCL1, and DCL4 (Nuthikattu et al., 2013; Creasey et al., 2014; McCue et al., 2015) and it seems likely that a similar pathway is active in the vegetative cell of pollen.

easiRNAs are mobile and can migrate to the sperm cells (Martínez et al., 2016), suggesting transport of 21/22-nt easiRNAs to the female gametophyte via sperm cells. Importantly, miRNAs cannot move into sperm cells (Grant-Downton et al., 2013), suggesting specific recognition of AGO proteins by the transport machinery.

In addition to DNA methylation changes, the VN undergoes extensive histone replacement, and many canonical histones including the centromeric histone CENH3 are lost, which may contribute to TE activation (Ingouff et al., 2007; Schoft et al., 2009).

Together, functional male gamete formation is associated with chromatin reconfiguration and relaxation of heterochromatin, leading to the activation of post-transcriptional silencing pathways targeting active TEs, ensuring genome integrity. The functional role of sperm-loaded easiRNAs remains to be identified. Since CHH methylation levels are low in sperm (Calarco et al., 2012), it is unlikely that easiRNAs enforce DNA methylation levels in sperm. One intriguing hypothesis is that easiRNAs are transported to the female gametophyte and act during or after fertilization to measure genetic compatibility between the parents, which in case of incompatibilities will lead to a phenomenon like that of hybrid dysgenesis in Drosophila (reviewed in Martienssen, 2010).

Epigenetic changes associated with female lineage formation

The formation of the female lineage initiates with the differentiation of a megaspore mother cell (MMC) from hypodermal sporophytic cells in ovule primordia. Distinctively, after meiosis only one cell will differentiate into a functional megaspore (FMS) (1n); the other meiotic products will undergo programmed cell death. The FMS undergoes three rounds of mitosis followed by cellularization, typically generating a seven-celled mature embryo sac that harbors four cell types: one egg cell (1n), one central cell (2n), two synergids (1n), and three antipodals (1n) (reviewed in Feng et al., 2013) (Fig. 1).

As in PMCs, the MMC in Arabidopsis undergoes dramatic chromatin changes, characterized by a reduction in heterochromatin, eviction of the linker histone H1, and changes in histone modifications (Figs 1 and 2; She et al., 2013; reviewed in Baroux and Autran, 2015). Importantly, chromatin decondensation seems primarily a consequence of H1 eviction, since remaining heterochromatin remains enriched for H3K9me2 (She et al., 2013). Therefore, silencing of TEs is unlikely to be impaired at this stage and thus the activation of post-transcriptional TE silencing is most likely not required.

Fig. 2.

Epigenetic changes during Arabidopsis reproductive development. (A) Mature pollen grain containing two condensed sperm cells (sc, 1n) and one relaxed vegetative cell (vc, 1n), visualized by DAPI staining. (B) Mature female gametophyte harboring one egg cell (ec, 1n), one central cell (cc, 2n), two synergids (sy, 1n) and three antipodals (1n). (C) Epigenetic modifications, siRNAs, and expression of genes involved in epigenetic pathways in the male and female reproductive lineage in Arabidopsis. A plus sign (+) indicates the presence of epigenetic modification or expression of the indicated genes. A minus sign (–) indicates either decreased or no detectable modification or absence of expression of the indicated genes. An asterisk (*) indicates that the presence or absence of certain modifications or gene expression is based on indirect evidence. ND denotes ‘not determined’. The red solid arrow symbolizes siRNA movement from the vegetative cell to a sperm cell. The dashed red arrow symbolizes the proposed siRNA movement from the central cell to the egg cell. PMC, pollen mother cell; MIS, microspore; VC, vegetative cell; SC, sperm cell; MMC, megaspore mother cell; MES, megaspore, CC, central cell; EC, egg cell; CR, chromatin remodeler. The data have been derived from Ingouff et al. (2007), Slotkin et al. (2009), Pillot et al. (2010), Schoft et al. (2011), Calarco et al. (2012), Ibarra et al. (2012), Jullien et al. (2012), She et al. (2013), and She and Baroux (2015).

Expression of AGO9 and RDR6 in MMC-surrounding cells is required to restrict MMC-specification, since mutants in both genes form additional MMCs in Arabidopsis (Olmedo-Monfil et al., 2010). However, the means of movement as well as the downstream action of mobile siRNAs in cell fate specification remains to be addressed. The maize orthologue of AGO9, AGO104, is also specifically expressed in ovule somatic cells surrounding MMCs and is necessary for non-CG methylation in heterochromatin (Singh et al., 2011). However, mutations in AGO104 cause the production of unreduced male and female gametes (Singh et al., 2011), suggesting a non-cell autonomous function of AGO104 in regulating meiosis. Meiosis is associated with re-establishment of heterochromatin by re-loading H1 and enforcing of H3K9me2 (She et al., 2013); it thus seems possible that this process is guided by siRNAs derived from surrounding tissues. Unreduced gamete production and formation of multiple gametophytes within one ovule was also reported in maize mutants deficient in DNA methyltransferases of the RdDM and CMT3 pathways (Garcia-Aguilar et al., 2010), highlighting the requirement of heterochromatin maintenance in regulating meiotic progression. Natural apomicts bypass meiosis and form viable seeds, thus resembling mutants impaired in epigenetic pathways of heterochromatin formation. It is therefore tempting to speculate that apomixis is under epigenetic control. Together, the striking effects of mutants in siRNA production and DNA methylation reveal the potential of manipulating epigenetic pathways for engineering apomixis in crops for hybrid vigor fixation (reviewed in Borges and Martienssen, 2015). For further details on chromatin changes during meiosis we refer to recent reviews on this topic by Yelina et al. (2015) and Mainiero and Pawlowski (2014).

After meiosis, the FMS will go through another wave of chromatin reconfiguration leading to a strong reduction of heterochromatin by H1 eviction and loss of H3K9me2 (She et al., 2013). During the three mitotic cycles, high levels of H3K9me2 are re-established, with no apparent differences in the eight nuclei of the gametophytic syncytium (Pillot et al., 2010). Mitotic progression of the FMS seems to require the action of mobile siRNAs from surrounding tissues, as suggested by the developmental arrest of the FMSs in the semi-dominant ago5-4 mutant (Tucker et al., 2012).

Epigenetic processes also regulate cell fate specification in the female gametophyte. After cellularization, the egg cell and central cell have strikingly different chromatin configurations (Fig. 2; Pillot et al., 2010). Similar to the pollen vegetative cell (Schoft et al., 2009), a severe depletion of heterochromatin connected with loss of the heterochromatic mark H3K9me2 occurs in the central cell of Arabidopsis (Pillot et al., 2010), correlating with the transcriptional activation of TEs observed in the Arabidopsis central cell and the female gametophyte of maize (Schmid et al., 2012; Chettoor et al., 2014). By contrast, heterochromatin is restored in the egg cell, as visualized by well-defined chromocenters marked by H3K9me2 (Pillot et al., 2010). Mutations in DME-LIKE (DML) genes restore heterochromatin formation in the central cell while not affecting the egg cell (Pillot et al., 2010), suggesting that DNA demethylation mediated by DME and DML enzymes in the central cell prevents heterochromatin establishment. Furthermore, there is almost no detectable expression of maintenance as well as de novo DNA methyltransferases in the central cell, contributing to reduced heterochromatin formation (Jullien et al., 2012; Belmonte et al., 2013). Similar to the transport of siRNAs occurring between vegetative cells and sperm cells (Martínez et al., 2016), it has been proposed that siRNAs can migrate from the central cell to the egg to silence TEs by either transcriptional or post-transcriptional mechanisms (Ibarra et al., 2012). This scenario bears striking parallels to the silencing of TEs in the germline of Drosophila by PIWI-interacting RNAs (piRNAs) formed in the surrounding nurse cells (Siomi et al., 2011), suggesting that nurse cell-assisted silencing of TEs in the germ cells is an essential mechanism accompanying reproduction in animals and plants. An important question concerns whether distinct chromatin states in female gametes are causing the divergent fates of egg and central cells or are rather a consequence of cell fate establishment. The differential expression of DNA-modifying enzymes in the egg and central cell together with the fact that there are no apparent differences in heterochromatin modifications before cell specification in the female gametophyte would favor the second scenario; nevertheless, the experimental evidence is still lacking.

DNA methylation changes occurring post-fertilization

In Arabidopsis, genome-wide DNA methylation studies of embryo and endosperm at around 7 d after pollination (DAP) have revealed a genome-wide hypomethylation of the endosperm compared to the embryo (Gehring et al., 2009; Hsieh et al., 2009). Global demethylation is a consequence of DME activity with a preference for small euchromatic TEs (Ibarra et al., 2012). DME is expressed in the central cell of the female gametophyte but apparently not in the egg cell and the fertilization products, suggesting that decreased methylation in the endosperm is a consequence of DME activity in the central cell before fertilization (Choi et al., 2002). Consistently, hypomethylation occurs specifically on the maternal genome and global methylation differences in CG context are restored in the dme mutant (Ibarra et al., 2012). Profiling of DNA methylation in endosperm at 4 DAP revealed a near absence of de novo CHH methylation (Moreno-Romero et al., 2016), consistent with de novo DNA methyltransferases DRM1 and DRM2 not being expressed in the early endosperm (Jullien et al., 2012). De novo DNA methylation activity and expression of the responsible genes is detected in the endosperm after cellularization (Hsieh et al., 2009; Ibarra et al., 2012; Belmonte et al., 2013), when the endosperm ceases to proliferate (Boisnard-Lorig et al., 2001). Thus, the absence of de novo DNA methylation in the early proliferative endosperm could explain how DME-mediated hypomethylation of the maternal genome is maintained after fertilization.

In rice, overall methylation levels in all DNA contexts are lower in the endosperm relative to the embryo and other tissues (Zemach et al., 2010). In contrast to Arabidopsis, there is a global reduction of non-CG methylation evenly across the genome, whereas CG hypomethylation is restricted to specific gene bodies, transcriptional start site-proximal regions, and short TEs. It has been proposed that rice lacks a DME orthologue, explaining the methylation differences compared to Arabidopsis (Zemach et al., 2010). However, phylogenetic studies have revealed putative DME orthologues present in the rice and maize genomes (Kapazoglou et al., 2013), and based on expression patterns and mutant phenotypes the rice DME homolog ROS1A is a good candidate for a functional rice DME analog (Ono et al., 2012). It therefore seems more likely that, as in Arabidopsis, the onset of de novo DNA methylation is delayed, accounting for the global reduction in non-CG methylation observed in rice. Consistently, both parental genomes are globally hypomethylated in a non-CG context, while reduction of CG methylation is particularly prominent at the maternal genome (Rodrigues et al., 2013). Furthermore, DNA methylation levels in rice endosperm at the syncytial stage (2–3 DAP) are substantially lower compared to later stages (Xing et al., 2015). In contrast to Arabidopsis and rice, there is no global reduction of CG DNA methylation in the endosperm of maize compared to other tissues, but there is a slight reduction in CHG methylation (Zhang et al., 2014). Furthermore, the global methylation levels of the maternal and paternal endosperm genomes are nearly equivalent in both CG and CHG contexts. Nevertheless, there are close to 7000 differentially methylated regions (DMRs) with increased methylation in the paternal compared to the maternal genome (Zhang et al., 2014). The presence of DMRs in Arabidopsis, rice, and maize is closely connected to parentally biased expression of genes, a phenomenon referred to as genomic imprinting (Gehring et al., 2009; Rodrigues et al., 2013; Zhang et al., 2014). Genomic imprinting is an epigenetic phenomenon occurring in mammals and flowering plants causing genes to adopt a parent-of-origin-specific expression pattern (reviewed in Rodrigues and Zilberman, 2015). The ‘imprint’ causing genetically identical alleles to be differentially expressed after fertilization is an epigenetic modification established during gamete formation. In flowering plants, genomic imprinting is mainly confined to the endosperm, and imprinted expression in the embryo is restricted to few loci and occurs only transiently (Jahnke and Scholten, 2009; Hsieh et al., 2011; Luo et al., 2011; Nodine and Bartel, 2012; Raissig et al., 2013; Pignatta et al., 2014). The presence of DMRs in the endosperm is closely connected to genes with maternally-biased expression (referred to as maternally-expressed imprinted genes, MEGs), associating hypomethylated regions with activation of gene expression (Gehring et al., 2009; Hsieh et al., 2011; Waters et al., 2011; Wolff et al., 2011; Rodrigues et al., 2013; Zhang et al., 2014). Several MEGs are essential for seed development, as will be discussed below.

Changes of histone modifications post-fertilization

As for MEGs, paternally-expressed imprinted genes (PEGs) are also closely associated with DMRs, indicating that maternally hypomethylated regions contribute to silencing of neighboring genes (Gehring et al., 2009; Hsieh et al., 2011; Waters et al., 2011; Wolff et al., 2011; Rodrigues et al., 2013; Zhang et al., 2014). Those DMRs flanking the silenced maternal alleles of PEGs are targeted by the Polycomb Repressive Complex 2 (PRC2), an evolutionary conserved chromatin-modifying complex applying a repressive trimethylation mark on lysine 27 of histone H3 (H3K27me3) (Zhang et al., 2014; Moreno-Romero et al., 2016). In maize, as in Arabidopsis, H3K27me3 is also deposited at hypomethylated regions in the maternal genome and determines the imprinted expression of PEGs (Makarevitch et al., 2013; Zhang et al., 2014), revealing an evolutionary conserved mechanism of imprinting regulation in angiosperms. PRC2 activity and DNA methylation are generally considered to be negatively correlated (Weinhofer et al., 2010; Deleris et al., 2012; Reddington et al., 2013); nevertheless, H3K27me3 is present at densely DNA-methylated pericentromeric regions in the Arabidopsis endosperm, indicating that H3K27me3 and DNA methylation are not necessarily exclusive marks (Moreno-Romero et al., 2016). There are increased levels of H3K27me3 at paternal compared to maternal pericentromeric regions, negatively correlating with the heterochromatic marks H3K9me2 and H3K27me1 that are reduced on paternal compared to maternal pericentromeric regions (Moreno-Romero et al., 2016). Strikingly, paternal pericentromeric regions of early mouse embryos are also marked by H3K27me3, which is replaced by the heterochromatic mark H3K9me3 after the first three cleavage divisions concomitantly with the exchange of the histone variant H3.3 by the H3.1 variant (Puschendorf et al., 2008; Santenard et al., 2010; Akiyama et al., 2011). Replication-coupled exchange of the paternal histone H3.3 by H3.1 does also occur in the early endosperm of Arabidopsis (Ingouff et al., 2007), causing maternal and paternal chromatin to have distinct histone H3 variants during the first nuclei divisions. The histone variant H3.3 cannot be targeted by enzymatic machineries applying the heterochromatic mark H3K27me1 (Jacob et al., 2014), providing a possible explanation for delayed heterochromatin formation at the paternal endosperm genome. In contrast, maternal and paternal H3 variants are removed from the zygote nucleus within a few hours after fertilization, resetting histone-mediated parental-specific information in the zygote (Ingouff et al., 2010). This rapid exchange of the parental histones and erasure of histone-mediated parental information seems in conflict with the proposed parental chromatin asymmetries in the zygote genome, causing delayed paternal genome activation (Autran et al., 2011). The chromatin assembly factor 1 (CAF1) that incorporates the H3.1 variant during replication has been proposed to be required for paternal genome activation (Autran et al., 2011). To resolve whether there are indeed parental-specific chromatin differences in the early zygote remains a challenge for future studies.

There are several PRC2 complexes acting during different stages of plant development. In Arabidopsis, the FIS-PRC2 that is active in the central cell of the female gametophyte and the descendent endosperm is composed of FERTILIZATION INDEPENDENT SEED2 (FIS2), FERTILIZATION INDEPENDENT ENDOSPERM (FIE), MEDEA (MEA), and MSI1 (Mozgova and Hennig, 2015). Loss of FIS-PRC2 function causes de-repression of the maternal alleles of PEGs, revealing that H3K27me3 deposition on maternal PEG alleles is required for imprinted expression (Hsieh et al., 2011; Wolff et al., 2011). The FIS-PRC2 genes MEA and FIS2 are only maternally expressed in the endosperm (reviewed in Rodrigues and Zilberman, 2015), but the functional relevance of imprinted expression of both genes remains to be determined. Specific expression of the FIS-PRC2 gene MSI1 after fertilization does not rescue msi1 seed lethality, suggesting a functional requirement of the FIS-PRC2 in the female gametophyte to determine seed development after fertilization (Leroy et al., 2007). There are no orthologs of FIS2 and MEA in rice and maize, or in any other species outside the Brassicaceae (Luo et al., 2009). Nevertheless, rice and maize have a functional PRC2 acting in the endosperm composed of subunits with homology to sporophytic PRC2 proteins in Arabidopsis (reviewed in Tonosaki and Kinoshita, 2015). PRC2 genes are also maternally expressed in the endosperm of rice and maize, revealing independent evolution of imprinted expression of PRC2 genes in monocots and dicots (Danilevskaya et al., 2003; Luo et al., 2009).

Potential role of siRNAs post-fertilization

Strong maternally-biased production of Pol IV-dependent 24-nt siRNAs (p4-siRNAs) has been reported in Arabidopsis seeds, with a subset of these p4-siRNAs being specific to flowers and young siliques, while a second type of p4-siRNAs are widely expressed throughout development (Mosher et al., 2009). It was later demonstrated that p4-siRNAs in Arabidopsis seeds are derived from both parental genomes (Pignatta et al., 2014), suggesting that the reported maternal bias of p4-siRNAs is either restricted to early stages of seed development or is a consequence of p4-siRNAs from maternal tissues biasing the detection of paternal p4-siRNAs.

The latter possibility seems more likely, since parental-specific p4-siRNAs derived from both parental genomes also accumulate in the endosperm of rice and maize (Rodrigues et al., 2013; Xin et al., 2014). Imprinted p4-siRNAs overlap with imprinted genes in rice and maize with siRNA and mRNA expression occurring from opposite parental alleles, suggesting that p4-siRNAs may regulate genomic imprinting (Rodrigues et al., 2013; Xin et al., 2014). Consistently, imprinted p4-siRNA-accumulating loci in rice are enriched at DMRs (Rodrigues et al., 2013) and the RdDM pathway has been shown to be required for repression of the MEGs SDC and MOP9.5 in Arabidopsis somatic tissues (Vu et al., 2013). Furthermore, p4-siRNAs accumulate in the flanking regions of MEGs, correlating with specific CHH methylation at the paternal alleles and paternal allele silencing (Calarco et al., 2012; Pignatta et al., 2014). Nevertheless, precisely how and when these p4-siRNAs act remains to be established; since the RdDM pathway is not active in the early endosperm (Jullien et al., 2012; Belmonte et al., 2013; Moreno-Romero et al., 2016) their action seems unlikely to take place via the RdDM pathway during the early phases of endosperm development.

Functional roles of imprinted genes in the endosperm

There have been many insightful reviews published recently about the evolution and function of genomic imprinting (Gehring, 2013; Bai and Settles, 2014; Rodrigues and Zilberman, 2015), and therefore we will concentrate here only on those imprinted genes where a potential functional role can be assigned based on mutant phenotypes. Among MEGs with known functional roles are the Arabidopsis FIS-PRC2 component-encoding genes MEA and FIS2 (Grossniklaus et al., 1998; Kiyosue et al., 1999; Luo et al., 1999) and the maize Meg1 that is required for transfer cell development (Costa et al., 2012). Mutations in MEA and FIS2 allow the formation of seeds without fertilization and cause seed abortion after fertilization (Grossniklaus et al., 1998; Kiyosue et al., 1999; Luo et al., 1999). PEGs encoding for the auxin biosynthesis enzymes YUCCA10 (YUC10) and TAR1 are highly induced in mea and fis2 mutants, leading to increased auxin production in the central cell and initiation of central cell replication and seed coat development (Figueiredo et al., 2015, 2016). Thus, maternal repression of YUC10 and TAR1 before fertilization by FIS-PRC2 is required to prevent autonomous endosperm and seed coat formation, while paternally-provided active alleles of both genes initiate central cell replication. The majority of identified imprinted genes do not share common imprinted expression in Arabidopsis, maize, and rice (Waters et al., 2013); it is therefore striking that homologs of YUC10 and TAR1 are imprinted in these three species (Hsieh et al., 2011; Luo et al., 2011; Xin et al., 2013). It thus seems likely that the coupling of fertilization to auxin production by the paternal expression of imprinted auxin biosynthesis genes is a conserved feature in flowering plants.

Hybridizations of plants that differ in ploidy (interploidy hybridizations) cause developmental defects in the endosperm. While increased maternal ploidy causes reduced endosperm growth and early endosperm cellularization, increased paternal ploidy causes the opposite phenotype with increased endosperm proliferation and delayed or failed endosperm cellularization (Scott et al., 1998; reviewed in Ramsey and Schemske, 1998; Pennington et al., 2008; Sekine et al., 2013). These parent-of-origin effects on seed development strongly suggest that deregulated imprinted genes underpin the response to interploidy hybridizations. Interploidy and interspecies crosses cause similar effects on endosperm development and deregulation of imprinted genes, suggesting a common mechanistic basis (Rebernig et al., 2015; Florez-Rueda et al., 2016; reviewed in Lafon-Placette and Köhler, 2016). A subset of PEGs has indeed been demonstrated to have functional roles in building interploidy hybridization barriers in Arabidopsis (Kradolfer et al., 2013; Wolff et al., 2015). Mutations in the PEGs ADMETOS, SUVH7, PEG2, and PEG9 restore viability of triploid seeds resulting from crosses of diploid plants with tetraploid pollen donors. Mutations in the aforementioned PEG genes restore seed viability by restoring endosperm cellularization (Kradolfer et al., 2013; Wolff et al., 2015), implicating a direct or indirect role of PEGs in regulating endosperm cellularization. Importantly, mutations in the same genes do not cause aberrations of diploid seed development, revealing that increased dosage but not loss of imprinted gene function is detrimental for viable seed formation (Kradolfer et al., 2013; Wolff et al., 2015).

Conclusions and perspectives

Undoubtedly, recent advances in high-throughput sequencing technologies allowing genome-wide profiling of DNA and histone methylation, RNAs, and sRNAs have greatly expanded our understanding of epigenetic reprogramming during plant reproduction. Furthermore, new technical tools such as the INTACT (isolation of nuclei tagged in specific cell types) system (Deal and Henikoff, 2010) open new avenues for the isolation of specific reproductive cell types in sufficient quantities suitable for epigenome studies. Despite the enormous progress made, there are many open questions that need to be addressed to fully understand the functional role of reprogramming events for functional gamete formation and seed development. Among others, we need to: (i) identify the targets of reproduction-associated siRNAs before and after fertilization and decipher their functional relevance; (ii) understand to what extend siRNAs can travel in reproductive tissues; and (iii) reveal the functional importance of genomic imprinting and the genes regulated by this phenomenon. Answering these questions will be facilitated by comparing knowledge gained from different model systems, which will allow us to discern between evolutionary conserved mechanisms and thus advance our understanding of plant reproductive development more comprehensively than by merely focusing on one particular model. Research in this direction promises to be fruitful and is expected to provide new candidate targets and approaches for crop improvement and breeding.

Acknowledgments

This research was supported by a European Research Council Starting Independent Researcher grant (to CK) and a grant from the Olle Engkvist Byggmästare Foundation (to CK). We thank Duarte Figueiredo for help with Fig. 2 and German Martinez Arias for critical comments on this manuscript.

Author notes

* Correspondence:

Editor: George Bassel, University of Birmingham