Demethylation of transposons can activate the expression of nearby genes and cause imprinted gene expression in the endosperm; this demethylation is hypothesized to lead to expression of transposon small interfering RNAs (siRNAs) that reinforce silencing in the next generation through transfer either into egg or embryo. Here we describe maize (Zea mays) maternal derepression of r1 (mdr1), which encodes a DNA glycosylase with homology to Arabidopsis thaliana DEMETER and which is partially responsible for demethylation of thousands of regions in endosperm. Instead of promoting siRNA expression in endosperm, MDR1 activity inhibits it. Methylation of most repetitive DNA elements in endosperm is not significantly affected by MDR1, with an exception of Helitrons. While maternally-expressed imprinted genes preferentially overlap with MDR1 demethylated regions, the majority of genes that overlap demethylated regions are not imprinted. Double mutant megagametophytes lacking both MDR1 and its close homolog DNG102 result in early seed failure, and double mutant microgametophytes fail pre-fertilization. These data establish DNA demethylation by glycosylases as essential in maize endosperm and pollen and suggest that neither transposon repression nor genomic imprinting are its main function in endosperm.

Fertilization and seed development is a critical time in the plant life cycle, and coordinated development of the embryo and endosperm are required to produce a viable seed. In the endosperm, some genes show imprinted expression where transcripts are derived primarily from one parental genome. Imprinted gene expression has been observed across many flowering plant species, though only a small proportion of genes are imprinted. Understanding how imprinted expression arises has been complicated by the reliance on single nucleotide polymorphisms between alleles to enable testing for imprinting. Here, we develop a method to use whole genome assemblies of multiple genotypes to assess for imprinting of both shared and variable portions of the genome using data from reciprocal crosses. This reveals widespread maternal expression of genes and transposable elements with presence-absence variation within maize and across species. Most maternally expressed features are expressed primarily in the endosperm, suggesting that maternal de-repression in the central cell facilitates expression. Furthermore, maternally expressed TEs are enriched for maternal expression of the nearest gene, and read alignments over maternal TE-gene pairs indicate that these are fused rather than independent transcripts.

Demethylation of transposons can activate expression of nearby genes and cause imprinted gene expression in endosperm, and it is hypothesized to lead to expression of transposon siRNAs that reinforce silencing in the next generation through transfer either into egg or embryo. Here we describe maternal derepression of R1 (mdr1), a DNA glycosylase with homology to Arabidopsis DEMETER that is partially responsible for demethylation of thousands of regions in endosperm. Maternally-expressed imprinted genes were enriched strongly enriched for overlap with demethylated regions, but the majority of genes that overlapped demethylated regions were not imprinted. Demethylated regions were depleted from the majority of repetitive DNA in the genome but enriched in a set of transposon families accounting for about a tenth of the total demethylated regions. Demethylated regions produced few siRNAs and were not associated with excess CHH methylation in endosperm or other tissues. mdr1 and its close homolog dng102 are essential factors in maternal and paternal fertility in maize, as neither double mutant microgametophytes nor megagametophytes gave rise to seeds. These data establish DNA demethylation by glycosylases as essential in maize endosperm and pollen and suggest that neither transposon regulation nor genomic imprinting are its main function.

Fertilization and seed development is a critical time in the plant life cycle, and coordinated development of the embryo and endosperm are required to produce a viable seed. In the endosperm, some genes show imprinted expression where transcripts are derived primarily from one parental genome. Imprinted gene expression has been observed across many flowering plant species, though only a small proportion of genes are imprinted. Understanding the rate of turnover for gain or loss of imprinted expression has been complicated by the reliance on single nucleotide polymorphisms between alleles to enable testing for imprinting. Here, we develop a method to use whole genome assemblies of multiple genotypes to assess for imprinting of both shared and variable portions of the genome using data from reciprocal crosses. This reveals widespread maternal expression of genes and transposable elements with presence-absence variation within maize and across species. Most maternally expressed features are expressed primarily in the endosperm, suggesting that maternal de-repression in the central cell facilitates expression. Furthermore, maternally expressed TEs are enriched for maternal expression of the nearest gene. Read alignments over maternal TE-gene pairs indicate fused transcripts, suggesting that variable TEs contribute imprinted expression of nearby genes.