Parent-of-origin effects on seed development in Arabidopsis thaliana require DNA methylation

Parental imprinting plays an important role in the reproductive biology of mammals (Surani et al., 1990; Bartolomei and Tilghman, 1997) and flowering plants (Kermicle and Alleman, 1990; Haig and Westoby, 1991). For imprinted loci, the expression level of an allele depends on its parent of origin, due to differential epigenetic modifications imposed during male and female gametogenesis. Therefore male and female gametes are not equivalent, since each contributes a unique set of active alleles of imprinted genes to the offspring. As a consequence, in mammals, both a maternal and a paternal genome are required for development of a viable embryo, and a 1:1 ratio is necessary for normal development (Surani et al., 1990). The mechanisms by which imprinted loci are modified are still being uncovered but parent-specific DNA methylation has so far been associated with nearly all imprinted mammalian genes (reviewed by Neumann and Barlow, 1996; Jaenisch, 1997; Brannan and Bartolomei, 1999; Tilghman, 1999). Imprinting-associated methylation requires de novo modification of loci during gametogenesis, as well as post-fertilization propagation of imprints by maintenance methylation through many rounds of mitosis.

In flowering plants, the study of imprinting is complicated by the generation of two offspring – embryo and endosperm – in every seed, each with a different parental genome ratio. Endosperm plays an analogous role to the placenta in transferring maternal resources to the embryo (Brink and Cooper, 1947; Haig and Westoby, 1991; Lopes and Larkins, 1993; Berger, 1999), but unlike the placenta it is a separate fertilization product. Each pollen grain transmits two haploid sperm to the embryo sac, one of which fertilizes the egg to form a zygote with a ratio of 1 maternal to 1 paternal genome (1m:1p), while the other fertilizes a central cell containing two haploid polar nuclei (like the egg, derivatives of the female meiotic product) to form a primary endosperm cell with the constitution 2m:1p. There is strong though circumstantial evidence from many sources that imprinting directly effects endosperm development, with indirect consequences for embryo growth. Unlike the case in mammals, plant embryos can complete development and form viable adults with a constitution of 1m:0p or 2m:0p. In contrast, all sexually reproducing angiosperms need maternal and paternal contributions to the endosperm, and even seeds producing parthenogenetic embryos sometimes require fertilization of the central cell (Sarkar and Coe, 1966; Nogler, 1984; Kermicle and Alleman, 1990). Furthermore, the 2m:1p ratio of the endosperm rather than the 1m:1p embryo ratio appears to be critical for normal seed development (Lin, 1984; Kermicle and Alleman, 1990; Haig and Westoby, 1991).

Most of the known imprinted genes in plants are expressed late in development of the persistent endosperm of maize; for example some members of the multi-gene family encoding zein storage proteins (Lund et al., 1995), and the R locus which regulates endosperm pigmentation (Kermicle, 1970). To date one gene from Arabidopsis thaliana, MEDEA (MEA) (Grossniklaus et al., 1998), has been shown to be imprinted (Kinoshita et al., 1999; Vielle-Calzada et al., 1999). Seeds inheriting a maternal mea mutation produce abnormal endosperm and embryo and abort, but the observation that mutant embryos can be rescued by culture (Vielle-Calzada et al., 1999) suggests to us that embryo lethality is a consequence of endosperm defects. Both studies of MEA imprinting agreed that expression in the endosperm is from maternal MEA alleles only, though results on embryo expression differed.

Many hypotheses have been advanced to explain the evolution of imprinting (reviewed by Hurst, 1997), but the most widely accepted is the parental conflict theory (Haig and Westoby, 1989, 1991; Moore and Haig, 1991). This interprets imprinting as a battle between maternal and paternal genomes over resource allocation from the mother to the embryo, proposed to arise because the reproductive fitness of a mother is greatest when she distributes resources equally among all her offspring, while a father benefits when maternal resources are concentrated in his own offspring. Therefore, the model predicts that maternally and paternally derived alleles will be selected to have opposite effects on embryo growth, with some growth promoters being paternally active and maternally silenced, and some growth inhibitors showing the opposite expression patterns. In mammals, most known imprinted genes (more than two dozen have been identified) are uniparentally expressed in the placenta, and many of these fit the parental conflict theory in having the predicted opposite effects on growth depending on parent of origin (Tilghman, 1999). In flowering plants, imprinted genes are predicted to directly affect the growth of endosperm – as this has primary responsibility for acquiring maternal resources for the seed – with mainly indirect consequences for the embryo (Haig and Westoby, 1989, 1991).

The MEA locus provides the only direct evidence so far concerning the role of imprinting in seed morphogenesis. However, there are likely to be more imprinted genes with a function in seed development, and evidence for these has been inferred from the effects of altering the balance of maternal and paternal genomes in the seed through crossing parents of different ploidies. In many species, an excess of paternal relative to maternal genomes appears to promote early growth of the endosperm, while maternal excess has the opposite effect (reviewed by Haig and Westoby, 1991). We found that in Arabidopsis, crosses between diploid (2x) and tetraploid (4x) plants in either direction produced viable seeds containing triploid embryos, and these had reciprocal phenotypes as predicted, with [4x × 2x] crosses (i.e. between a 4x seed parent and 2x pollen parent) producing small, underdeveloped endosperms and small embryos, and [2x × 4x] crosses generating large endosperms and embryos (Scott et al., 1998). Crosses between diploid and hexaploid plants resulted in similar but more extreme phenotypes, followed by abortion. Our results were consistent with a model in which maternal genomes contributed active alleles of endosperm growth inhibitors, and paternal genomes contributed active growth promoters, with the observed parent-of-origin effects on seed development reflecting dosage imbalances of these alleles. Although the crosses also altered the balance of parental genomes in the embryo, we concluded that the effects on embryo growth and viability were likely to be indirect, partly because of previous work, cited above, showing that the embryo is relatively insensitive to parental genome imbalance, and partly because the major morphological effects we observed were on endosperm development.

Little is known about the parental imprinting mechanism in plants, although there is evidence that as in mammals DNA methylation is involved. In maize endosperm, imprinted zein genes are only expressed when inherited from the seed parent, and these loci are methylated at fewer sites on maternally than paternally derived chromosomes (Lund et al., 1995). Differential methylation also corresponds with parent-specific expression of the R locus (Kermicle and Alleman, 1990; Finnegan et al., 1998). The methylation patterns of MEA have not been reported, but in seeds homozygous for a decrease in DNA methylation 1 (ddm1) mutation, which reduces overall cytosine methylation by 70% (Vongs et al., 1993), a wild-type paternal MEA allele can rescue seeds carrying a normally lethal maternal mea mutation, implying that hypomethylation has activated the silenced paternal copy (Vielle-Calzada et al., 1999). However, ddm1 mutations do not affect methyltransferase activity (Kakutani et al., 1995), and DDM1 has recently been found to be a member of the SWI2/SNF2 family of chromatin remodelling proteins (Jeddeloh et al., 1999). Furthermore, single-copy DNA sequences only lose methylation gradually in ddm1 mutants through several generations of inbreeding (Kakutani et al., 1996). Therefore the ddm1 mutation is likely to have its primary effect on chromatin configuration with only indirect effects on methylation. In addition, effects of the ddm1 mutation alone on seed development have not been described beyond the observation that mutant seeds are viable.

In order to investigate the global role of methylation in parent-of-origin effects in seeds, we performed crosses using plants with DNA methylation reduced by a METHYLTRANSFERASE I antisense (METI a/s) transgene (METI is the predominant DNA methyltransferase in Arabidopsis) (Finnegan et al., 1996; Genger et al., 1999). If DNA methylation is essential to the imprinting mechanism in Arabidopsis, and if the antisense transgene prevents imprinting-specific methylation, we would expect hypomethylated plants to produce gametes in which imprinted alleles lose most or all of their silencing. For example, a hypomethylated pollen donor should provide sperm in which silencing is lifted from the alleles that are normally expressed from the maternal genome only, so the seed will contain extra active alleles of maternal-specific genes. Our results show that following reciprocal crosses between hypomethylated and wild-type plants, seed development is indeed affected as predicted: crosses between hypomethylated 2x seed parents and wild-type (normally methylated) 2x pollen parents phenocopy the [2x × 4x] cross (Scott et al., 1998) in seed size and morphology, while crosses between wild-type 2x seed parents and hypomethylated 2x pollen parents phenocopy [4x2x] crosses. We conclude that methylation plays an important role in parent-of-origin effects, and by inference imprinting, in flowering plants. The reciprocal phenotypes also suggest that in each cross imprints are propagated in the genome derived from the wild-type parent. From this we infer that the antisense methyltransferase prevents establishment or maintenance of imprinting in gametes rather than propagation of imprinting after fertilization.

Jaenisch (1997) proposed that removal of imprints or of imprinted genes themselves should have few developmental consequences, as they exist in “ ‘paired sets’ of genes involved in the same pathway” (e.g. of growth promoters and inhibitors, as predicted by Haig and colleagues). This has been difficult to test in mammals as embryos with reduced methylation die during gestation (Li et al., 1992). However, our results support a model in which imprinting is not essential to development, as we found that when both parents are hypomethylated seed phenotypes are less severe than the reciprocal phenotypes observed when only one is hypomethylated.

Plants were grown for 3-4 weeks at 22°C with a day length of 16 hours in a Fisons growth cabinet, then transferred to a glass house and grown at 24±2°C. Plants with wild-type methylation levels were C24 diploid (2x) A9-barnase (Paul et al., 1992), C24 tetraploid (4x), and Columbia hexaploid (6x) as described by Scott et al. (1998). Hypomethylated C24 plants were from the T3 generation of family 10.5, homozygous for the Arabidopsis methyltransferase I (METI) antisense construct under control of the cauliflower mosaic virus 35S promoter (Finnegan et al., 1996).

If the seed parent was male sterile (A9-barnase), open flowers were pollinated. If plants were male fertile, flower buds were emasculated 1 day prior to anthesis and pollinated 2 days later. Developing siliques were collected 2 to 8 days after pollination and processed as described below. Mature seeds were collected when pods were desiccated. Seeds were weighed using a Mettler UMT 2 microbalance (Mettler-Toledo, Leicester, UK).

Samples were prepared as in Braselton et al. (1996) and imaged at the University of Bath using an Axiovert 100M Zeiss LSM510 laser scanning microscope. Feulgen-stained samples were excited using an argon ion laser at 458 or 488 nm, and emissions detected at ≥515 nm. Images measuring 1024×1024 pixels were collected using a C-Apochromat 63×/1.2 water lens, saved in PSD format, and processed using Adobe Photoshop 4.0.1.

Genomic DNA was extracted from 0.1 g of leaf tissue using a Nucleon Phytopure Plant DNA Extraction kit (Nucleon, Biogenesis, Glasgow, UK) according to manufacturer’s instructions. 100 ng of genomic DNA was digested overnight with MspI or HpaII and separated by electrophoresis on a 1% agarose gel. Southern analysis was performed as described by Southern (1975) with 0.4 M NaOH replacing 20× SSC as the buffer solution. The probe used contains a 180 bp repeat sequence from Arabidopsis centromeric DNA (Martinez-Zapater et al., 1986). Probe DNA was digested with PvuII and a 500 bp fragment, consisting of the 180 bp repeat and 320 bp of pUC12 vector, was gel purified. The fragment was labelled by the random priming method using DIG11-UTP alkali label (Roche, Lewes, E. Sussex, UK) according to manufacturer’s instructions. Hybridization was carried out at 65°C. Filters were washed at room temperature in 0.1× SSC and 0.1% SDS, and developed using an anti-digoxigenin antibody (Roche) and CPD Star substrate (Promega, Southampton, UK), according to manufacturer’s instructions.

Genomic DNA was extracted from 0.1 g of leaf tissue according to the small scale method of Edwards et al. (1991). DNA template (10 ng) was added to a 20 ml reaction mix containing 1.8 ml of 11× buffer (500 mM Tris-HCl pH 8.8, 120 mM NH^₄SO^₄, 50 mM MgCl^₂, 75 Mm β-mercaptoethanol, 0.05 mM EDTA, 11 mM dNTPs, 1.25 mg/ml DNAse-free BSA), 10 pmol of each primer (see below), and 1 U of Taq polymerase (Advanced Biotechnologies, Surrey, UK). The primers used were METIdF (5′-TAT AGG CCT GAG GAT GTT TCT GC-3′) and METIbR (5′-AGG TCC ACC ATT GAT GAA GTC C-3′), which span an intron-containing sequence in the endogenous METI gene (Finnegan et al., 1993; accession no. L10692). Cycling conditions were 94°C for 3 minutes followed by 35 cycles of 94°C for 1 minute, 55°C for 1 minute, and 72°C for 1 minute, carried out in an MJ Research PTC-200 Peltier Thermal Cycler. The reaction amplified a 1 kb product from the endogenous METI gene, while the antisense transgene generated an additional product of 0.7 kb.

To test the effect of decreasing methylation levels on seed development, we performed a series of crosses using transgenic plants homozygous for a METHYLTRANSFERASE I antisense construct (METI a/s) (Finnegan et al., 1996) as one or both parents. The transgenic line used was previously reported to have approximately 13% of the wild-type level of DNA methylation (Finnegan et al., 1996).

Crosses between two METI a/s plants, as well as reciprocal crosses between METI a/s and wild-type 2x parents, all produced viable seeds. Crosses using a METI a/s plant as one parent gave rise to hemizygous F₁ plants with reduced methylation levels (Fig. 1 and Finnegan et al., 1996). Dry weight and germination frequencies of seeds from the METI a/s crosses and wild-type interploidy crosses are compared in Fig. 2. As previously reported (Scott et al., 1998), seeds from [4x × 2x] and [2x × 4x] crosses are nearly always viable. [4x2x] seeds, which have double the normal dose of maternal relative to paternal genomes, are lighter than [2x × 2x] seeds (mean 15.3 μg compared with 20.8 μg), while [2x X 4x] seeds, with a double dose of paternal genomes, are heavier (53.7 μg; Fig. 2A). Seeds from [6x × 2x] and [2x × 6x] crosses, with more extreme maternal and paternal excess respectively, are shrivelled and inviable. Fig. 2B shows that crosses using METI a/s plants follow a similar trend. [2x × METI a/s] seeds – in which we predicted that normally maternal-specific alleles would be derepressed on the paternal chromosomes (thus phenocopying maternal excess) – are lighter than [METI a/s × heavier (32.5 μg). In all three crosses, most seeds are viable.

In the [METI a/s] seeds (9.5 μg compared with 13.6 μg), while [METI a/s × 2x] seeds, in which paternally expressed genes could be activated on the maternal chromosomes (phenocopying paternal excess), are In the [4x × METI a/s] and [METI a/s × 4x] crosses, in which genomic imbalance is superimposed on hypomethylation, viability drops off sharply but is not reduced to 0 as for [2x × 6x] and [6x × 2x] crosses, and the mean seed weight is also higher than for these crosses.

In normal Arabidopsis seeds, the endosperm proliferates as a syncytium until the embryo reaches heart stage, and then begins to cellularize from the micropylar pole (Mansfield and Briarty, 1990a,b; Scott et al., 1998; Brown et al., 1999; Berger, 1999). In our conditions cellularization begins at about 5 days after pollination (5 DAP). Several days before cellularization three regions of endosperm can be identified: central peripheral (composed of regularly spaced nuclei with associated cytoplasm lining the central region of the embryo sac); micropylar peripheral (nuclei embedded in a common cytoplasm surrounding the suspensor); and chalazal (a dense multinucleate tissue at the chalazal pole). The free-nuclear central peripheral endosperm often forms enlarged and sometimes multinucleate ‘nodules’ near the chalazal endosperm. An Arabidopsis seed at heart stage is shown schematically in Fig. 3.

Following interploidy crosses resulting in maternal excess, few endosperm nuclei are produced, peripheral endosperm cellularizes early (beginning at 4 DAP), chalazal endosperm is underdeveloped, no nodules are seen, and embryo differentiation is delayed (Fig. 4, [4x × 2x] and [6x × 2x]; Scott et al., 1998). In contrast, seeds with paternal excess produce peripheral endosperms with large numbers of nuclei which cellularize late (from 6 DAP) or never, and massively overgrown chalazal endosperm and nodules (Fig. 4, [2x × 4x] and [2x × 6x]; Scott et al., 1998).

Confocal microscopy shows that [METI a/s × METI a/s] seeds are not identical to wild type [2x × 2x]. [METI a/s ×METI a/s] seeds show some features of seeds with maternal excess – for example small chalazal endosperm – but cellularization (at 5-6 DAP) is not early and proliferation of peripheral endosperm is not inhibited as in [4x × 2x] seeds (Fig. 4). Crosses in which a METI a/s plant is only one of the parents are more directly comparable with interploidy crosses. In [2x × METI a/s] seeds the endosperm underproliferates and cellularizes early (3-4 DAP), and these phenotypes are more extreme in [4x × METI a/s] crosses, with cellularization occurring at 2-3 DAP. In contrast, [METI a/s × 2x] seeds produce large peripheral endosperms with delayed cytokinesis (6-7 DAP), and overgrown chalazal endosperms and nodules, while [METI a/s × 4x] seeds have an even more pronounced paternal excess phenotype, with no cellularization at 10 DAP.

Endosperm development is quantified in Fig. 5, which shows numbers of peripheral endosperm nuclei counted after initiation of cellularization but before the embryo has begun to consume the endosperm. This stage occurs at a different number of days after pollination for each cross, ranging from 5 to 7 DAP, but is intended to reflect the maximum extent of endosperm proliferation. [METI a/s × METI a/s] seeds produce 598±126 (mean±s.e.m.) (n=3) peripheral endosperm nuclei, compared with 429±31 (n=3) for [2x × 2x] seeds. [2x × METI a/s] seeds produce less than half the number of peripheral endosperm nuclei observed in [METI a/s × METI a/s] seeds (227±17; n=6), while [METI a/s X 2x] seeds have more than twice the number (1,365±90; n=3). [4x × METI a/s] seeds have even fewer peripheral endosperm nuclei than [2x × METI a/s] – less than half the number again (97±10; n=4). On average, [METI a/s × 4x] seeds have slightly fewer peripheral endosperm nuclei than [METI a/s × 2x] (1,291±386; n=3), but in contrast to the latter, [METI a/s × 4x] seeds are mainly inviable. A similar trend is seen in [2x × 6x] compared with [2x × 4x] endosperms (Scott et al., 1998).

For imprinting to affect gene expression in the developing seed, the different imprints on maternal and paternal chromosomes inherited by the primary endosperm nucleus must be propagated during endosperm proliferation. This raises the possibility that a uniparentally transmitted METI a/s transgene could interfere with propagation of methylation imprints on all chromosomes in the endosperm, masking parent-specific effects. Therefore we tested whether presence of the transgene per se affected seed development, through reciprocal crosses between hemizygous METI a/s (hemiMETI a/s) and wild-type 2x plants. (METI a/s plants that are hemizygous for the transgene remain hypomethylated, though to a lesser extent than homozygotes; Finnegan et al., 1996; and Fig. 1.) The results of these crosses are shown in Fig. 6. In 11 progeny of a [hemiMETI a/s × 2x] cross, 6 inherited a copy of the transgene (Fig. 6A). Although approximately half of the seeds contained the transgene, all the seeds produced by the [hemiMETI a/s × 2x] cross were similar in size, morphology, and development (Fig. 6B,C). Likewise, we saw no differences among seeds produced by the [2x × hemiMETI a/s] cross. However, there were differences between the crosses. [2x × hemiMETI a/s] crosses produced small seeds (mean weight 10.9 μg; n=26) with phenotypes similar to those of [2x × (homozygous) METI a/s] seeds, while [hemiMETI a/s × 2x] seeds were large (28.6 μg; n=39) and resembled [METI a/s × 2x] seeds (Fig. 6B-D; cf. Figs 2, 4, 5). Endosperm proliferation in crosses using one hemiMETI a/s parent followed the same trends as crosses in which one parent was homozygous for the endosperm has an excess of imprinted alleles that behave as if they were inherited from the father, thus phenocopying an excess of paternal genomes. [hemiMETI a/s × 2x] seeds produced 810±23 (mean±s.e.m.) peripheral endosperm nuclei (n=6), higher than the mean for [METI a/s × METI a/s] crosses, while [2x × hemiMETI a/s] seeds generated 242±28 peripheral endosperm nuclei (n=6), lower than the [METI a/s × METI a/s] mean (Figs 5 and 6C).

In crosses using a METI a/s plant as only one parent, seed weights, germination frequencies, and developmental patterns including endosperm proliferation and timing of cellularization all show that hypomethylation closely phenocopies the effects of interploidy crosses (Figs 2, 4, 5). [METI a/s × 2x] seeds have a strong paternal excess phenotype with high seed weight, many endosperm nuclei, delayed endosperm cellularization, and overgrown chalazal endosperm, although both parents are diploid, and the seed is nourished by a hypomethylated mother which suffers a variety of defects in vegetative and floral development (Finnegan et al., 1996). This behaviour is consistent with a model in which hypomethylation of the maternal genome in METI a/s plants has prevented silencing of endosperm-promoting genes which would normally only be expressed from the paternal genome (Haig and Westoby, 1989, 1991; Scott et al., 1998). Meanwhile, the wild-type paternal genome contributes its normal complement of silenced endosperm-inhibiting genes and active endosperm-promoting genes. The net effect according to the model is that the endosperm has an excess of imprinted alleles that behave as if they were inherited from the father, thus phenocopying an excess of paternal genomes.

The reciprocal phenotypes of [4x × 2x] and [2x × 4x] crosses can only be explained if female and male gametes contribute different sets of active alleles or else different complements of gene products (i.e. cytoplasmic factors) to the seed. (In the first case the alleles are not necessarily expressed in the gametophyte, but need to be transmitted to the seed with a potential for expression – although some may fall into both categories, like MEA, which is transcribed in the female gametophyte as well as being transmitted to the seed with maternal alleles competent for expression; Grossniklaus et al., 1998.) Similarly, the reciprocal phenotypes of [2x × METI a/s] and [METI a/s × 2x] crosses can only be explained if uniparental hypomethylation affects sex-specific gene expression in a way that closely phenocopies interploidy crosses. Formally it is possible that the interploidy cross phenomena are due to a dosage imbalance of genes expressed exclusively in the central cell and sperm whose products are carried over to the endosperm. It is then possible that hypomethylation allows ectopic expression of these gametophytic genes in the wrong sex, so that sperm-specific genes are activated in the central cell and vice versa, and that deregulated gametophytic expression alone is responsible for the phenocopy of interploidy crosses. These scenarios would be consistent with the reciprocal phenotypes without involving imprinting. However, we consider this an unlikely explanation for all of our findings, particularly those resulting in paternal excess and its phenocopy. Plant sperm and generative cells (sperm precursors) are characterized by condensed chromatin, little cytoplasm, and few organelles, and very few generative cell- or sperm-specific proteins have been identified (McCormick, 1993; Blomstedt et al., 1996). It is difficult to conceive how the paternal excess phenotype seen following interploidy crosses could be explained solely by an overdose of gene products specific to sperm, or how ectopic production of sperm-specific gene products in the embryo sac could have the observed effects on seed development. Furthermore, gametes produced by METI a/s plants are fertile, which one might not expect if there was general deregulation of sex-specific genes. Consequently, we consider disruption to the balance of expression of imprinted genes to be the most likely explanation for the results of interploidy crosses and uniparental hypomethylation.

Based on the above hypothesis that a hypomethylated genome has a similar effect to adding a genome of the opposite sex, one can predict that crossing a hypomethylated plant with a polyploid plant should have even more severe consequences for seed development. To test this we performed reciprocal crosses between hypomethylated diploid plants and normally methylated tetraploid plants. Seeds produced by [METI a/s × 4x] crosses usually abort and all have a strong paternal excess phenotype, resembling offspring of normally methylated [2x × 6x] crosses. In the [METI a/s × 4x] cross, a hypomethylated maternal genome, which phenocopies excess paternal genomes, is added to a real excess of paternal genomes, apparently pushing this cross towards more extreme paternal excess. Similarly, seeds from [4x × METI a/s] crosses have a stronger maternal excess phenotype than [2x × METI a/s] seeds.

So far the few studies of the role of methylation in parent-of-origin effects on seed development have focussed on single imprinted loci, e.g. zein genes (Lund et al., 1995), the R locus (Kermicle, 1970; Kermicle and Alleman, 1990; Finnegan et al., 1998), and MEA (Vielle-Calzada et al., 1999). The first two genes are not involved in seed morphogenesis, and the latter study is complicated by likely effects of ddm1 on chromatin configuration as well as methylation (Jeddeloh et al., 1999). To our knowledge the work here presents the first evidence that methylation has a general role in parent-of-origin effects in plants, most likely reflecting its role in regulating expression of many imprinted genes on both maternal and paternal chromosomes.

For some mammalian imprinted genes, parent-specific methylation has been traced from the sperm and eggs of parents to the somatic tissues of offspring, indicating that the methylation patterns inherited from each parent are maintained after fertilization (Jaenisch, 1997; Tilghman, 1999). Therefore we envisaged the METI a/s transgene as having several possible consequences for parent-of-origin effects: it could prevent establishment or maintenance of methylation during gametogenesis, or it could prevent maintenance of parent-specific methylation in the a/s] phenocopying [4x × 2x] crosses and [hemiMETI a/s × 2x] resembling [2x × 4x] (Fig. 6B-D). Strikingly, within each cross with a hemiMETI a/s parent, there was a single class of seed as measured by size and weight, morphology, and number of peripheral endosperm nuclei, although only half the progeny inherited the transgene (Fig. 6A). Therefore, the seed phenotypes were consistent with imprints being maintained, whether the seed contained a METI a/s construct or had a wild-type genotype. The Arabidopsis genome has been found to regain methylation slowly after a METI a/s transgene or ddm1 mutation is segregated away (Vongs et al., 1993; Finnegan et al., 1996), which has been interpreted as reflecting a slow rate of de novo methylation. It could be argued that our results reflect lack of remethylation in seeds only one generation after losing the transgene. However, de novo methylation would not be required to propagate imprints on methylated DNA inherited from a wild-type parent, and the reciprocal phenotypes of [2x ´ hemiMETI a/s] and [hemiMETI a/s ´ 2x] crosses again imply that imprints are maintained on chromosomes transmitted by these plants. Therefore we conclude the hemiMETI a/s results show that the transgene per se does not affect maintenance of\ methylation imprints in the seed. Taken together our results show that the METI a/s transgene prevents establishment or propagation during gametogenesis of methylation associated with parent-of-origin effects. This is in contrast to the results described by Vielle-Calzada et al. (1999), who concluded that the ddm1 mutation abolished maintenance of MEA imprinting in the seed rather than establishment of the imprint in pollen.

Our results are consistent with a role for METI in establishing imprinting-associated methylation, but it is not known when and where during reproductive development this enzyme is active. In mouse, investigation of the major methyltransferase Dnmt1, among other evidence, indicated that imprints are most likel imposed during meiosis (Mertineit et al., 1998; Brannan and Bartolomei, 1999). Nothing is known of when imprinting might be set in plants, but as it must occur when male and female gametes or their precursors are separated, it could be any time between floral organ differentiation and fertilization. It is notable that following crosses between hemiMETI a/s and wild-type plants, each seed develops according to the methylation status of its parents regardless of whether it inherits a transgene; one explanation is that at least some element of parent-specific methylation may be set before the nuclear divisions of meiosis. Another possibility is that gametes do not normally express METI but inherit METI protein from the diploid spore mother cells: this could also explain why the genotype of the DNA inherited from a wild-type parent, and the reciprocal phenotypes of [2x × hemiMETI a/s] parent plant rather than the meiotic product is reflected in the seed phenotype. Analysis is further complicated by lack of information about endogenous METI expression as well as the timing and location of METI a/s activity: e.g. the 35S promoter driving the antisense construct is probably not active during pollen development (Wilkinson et al., 1997). METI mRNA and protein localization in wild-type and transgenic plants would help distinguish between the alternatives.

Following [METI a/s × METI a/s] crosses 90% of seeds are viable, although the parent plants are hypomethylated by 85% (Finnegan et al., 1996). Therefore it is possible in plants to test the effects of relaxing methylation-dependent imprinting – an experiment impossible in animals, as mice with methylation and [hemiMETI a/s × 2x] crosses again imply that imprints are maintained on chromosomes transmitted by these plants. Therefore we conclude the hemiMETI a/s results show that the transgene per se does not affect maintenance of methylation imprints in the seed. Taken together our results show that the METI a/s transgene prevents establishment or propagation during gametogenesis of methylation associated with parent-of-origin effects. This is in contrast to the results described by Vielle-Calzada et al. (1999), who concluded that the ddm1 mutation abolished maintenance of MEA imprinting in the seed rather than establishment of the imprint in pollen. reduced by 70% through targeted mutation of Dnmt1 die early in embryogenesis (Li et al., 1992). We found that seeds from [METI a/s × METI a/s] crosses were abnormal, but in seed weight, extent of endosperm proliferation, and timing of endosperm cellularization more closely resembled [2x × 2x] seeds than those produced by crosses between one METI a/s and one wild-type 2x plant (Figs 2, 4, 5). This observation, along with the complementary phenotypes of crosses in which only one parent is hypomethylated, suggests that the METI a/s transgene deregulates the sets of antagonistic growth control genes predicted to be subject to parent-specific expression by the parental conflict theory (Haig and Westoby, 1989, 1991; Moore and Haig, 1991). Jaenisch (1997) proposed that according to the theory, “removal of all imprints should have no ill effect”, and “Imprinting has no intrinsic role in mammalian development…Imprinted genes are viewed as ‘paired sets’ of genes involved in the same pathway where removal of the set has little or no developmental consequences.” Our data is consistent with the hypothesis that removal of imprinting in parents indeed has little effect on development (compared with removal of imprinting in just one parent). We have no evidence concerning removal of the sets of genes per se; instead, we infer that biparental hypomethylation in effect adds sets of antagonistic genes.

Although [METI a/s × METI a/s] crosses produce seeds that appear more normal than [2x × METI a/s ] or [METI a/s × 2x] crosses, they contain small chalazal endosperms (Fig. 4) and weigh less than wild-type seeds (Fig. 2), both features of seeds from interploidy crosses which have inherited extra maternal genomes (Scott et al., 1998). This is an unexpected result, as one might predict a paternal rather than maternal excess phenotype for the following reason. If imprinting-specific methylation is lost equally on the maternal and paternal genomes in METI a/s plants, then the effective genome ratio in a [METI a/s × METI a/s] endosperm should be 3m (2m from the central cell and 1m equivalent from the hypomethylated sperm):3p (1p from the sperm and 2p equivalent from the hypomethylated central cell), making it equivalent to the 2m:2p ratio found in a [2x × 4x] cross. One aspect of the phenotype, however, is consistent with this prediction: the number of peripheral endosperm nuclei in [METI a/s X METI a/s] crosses is higher than in [2x × 2x] and about six-fold greater than in [4x × 2x] crosses (Fig. 5 and Scott et al., 1998).

We do not know why [METI a/s × METI a/s] seeds appear to show a combination of maternal and paternal excess phenotypes, but there are several factors which may contribute to this. Demethylation is not complete in METI a/s plants (Finnegan et al., 1996), perhaps in part because of other methyltransferases in the Arabidopsis genome which are not affected by the METI a/s transgene (Genger et al., 1999). Partial demethylation could affect gamete genomes or individual sequences unequally. In addition, some genes may even become hypermethylated in a METI a/s background, like the (non-imprinted) SUPERMAN locus (Jacobsen and Meyerowitz, 1997). Finally, due to the complicated regulation of imprinted genes, global DNA hypomethylation in mouse can repress as well as activate imprinted alleles (Li et al., 1993; Tilghman, 1999). It is conceivable the same occurs in plants, although evidence presented above suggests that the overwhelming effect of hypomethylation is to activate normally silent imprinted alleles.

The results above show that DNA methylation is an important part of the parent-of-origin effect in Arabidopsis following interploidy crosses, consistent with an essential role for methylation in the parental imprinting mechanism in flowering plants. Global DNA hypomethylation appears to derepress genes contributed to the seed by the polar nuclei that would normally be active only in the male genome, and derepress genes contributed by the sperm that would normally be female-specific. This has an effect of ‘paternalizing’ the female genome and ‘maternalizing’ the male genome (Fig. 7A). The phenotypic consequences are shown in Fig. 7B. We conclude that it is possible through uniparental hypomethylation to modify seed development and ultimately, size, most likely through lifting the silencing on parentally imprinted genes.

The following are gratefully acknowledged: Jean Finnegan for METI a/s seed; Eric Richards for the 180 bp centromeric repeat clone; Andrew Ward for comments on the manuscript. S. A. was supported by a BBSRC Case Studentship and Biogemma UK Ltd; R. V. by BBSRC grant PO8575; M. S. by BBSRC grant P6696. The confocal microscope was purchased by The Wellcome Trust (grant 049452).