Cytosine DNA methylation is an epigenetic modification of DNA that is usually associated with the stable and heritable repression of transcription. Most methylated sequences found in genomes are transposable elements, indicating the likely ancestral role of DNA methylation in genome defense. DNA methylation is also important in imprinting, X chromosome inactivation, and the epigenetic regulation of genes. While DNA methylation is widespread in plants, fungi and animals, it has been curiously lost in some of the most well studied genetic organisms including S. cerevisiae, S. pombe, C. elegans and Drosophila.

We study DNA methylation in the model plant Arabidopsis thaliana because of its facile genetics, small size, and trim genome. Furthermore, unlike other organisms like mouse, where DNA methylation mutants are inviable, Arabidopsis can tolerate mutations that virtually eliminate methylation, allowing for further study. Arabidopsis methylation mutants display developmental abnormalities because of defects in the methylation of several key genes that regulate development. We have taken advantage of the stable methylation present at two such developmental genes, in order to perform genetic screens for mutants affecting DNA methylation. Our studies have revealed that DNA methylation is controlled by 1) the specificity of several different DNA methyltransferases, 2) targeting by other chromatin modifications such as the methylation of histone tails, and 3) targeting of specific DNA sequences by small interfering RNAs (siRNAs)

At the heart of our work is the use of so-called epigenetic mutations. These are alleles of developmentally important genes that are stably silent as a result of DNA methylation. Unlike real genetic alleles, epigenetic alleles have a DNA sequence that is identical to wild type. Nonetheless, these alleles are inherited in a Mendelian fashion and can be used for classical genetic studies. SUPERMAN mutants show an altered floral structure, and thus the phenotype can be easily monitored by the naked eye. The SUPERMAN epigenetic alleles are caused by dense hypermethylation and silencing of the SUPERMAN gene, which is otherwise unmethylated in wild type plants. This hypermethylation effect is meiotically heritable and causes a recessive loss-of-function phenotype.

The second set of epigenetic mutants we study are at the imprinted FWA locus. In wild type, FWA is methylated and silent in all adult tissues of the plant. fwa hypomethylated mutant strains show a dominant late flowering phenotype due to a permanent loss of methylation present within two direct repeats in the FWA promoter, which causes ectopic expression of the gene.

Thus, SUPERMAN and FWA can adopt two heritable epigenetic states, either methylated and silent, or unmethylated and active.

DNA methylation is found at cytosine residues in three different sequence contexts, CG, CNG, and asymmetric (all sites not in CG or CNG). Furthermore DNA methylation can be classified as the initial establishment of methylation (de novo methylation) or maintenance of preexisting methylation. The main enzyme that maintains preexisting CG methylation is MET1, a homolog of mammalian DNMT1. Our work helped to define the enzymes involved in non-CG methylation and the enzymes controlling de novo methylation.

First, we performed a mutant screen to identify genes required for the maintenance of SUPERMAN DNA methylation and silencing. This screen uncovered nine loss-of-function alleles of the CHROMOMETHYLASE3 gene, which encodes a novel type of CNG specific DNA methyltransferase. chromomethylase3 mutants cause a genome wide reduction of CNG DNA methylation and result in the reactivation of SUPERMAN as well as previously silent retrotransposons. These findings show that CNG methylation is important for gene silencing, and clearly define CHROMOMETHYLASE3 as the main enzyme controlling this modification.

Second, we used the Arabidopsis genome sequence, coupled with reverse genetics, to show that the DOMAINS REARRANGED METHYLASE2 (DRM2) gene (the plant ortholog of mammalian DNMT3) encodes the major de novo methyltransferase in Arabidopsis. To do this, we first needed to develop in vivo assays for de novo methylation. First, we discovered that FWA is an efficient substrate for de novo methylation and transgene silencing when it is transformed into wild type plants. However, when transformed into drm2 mutants, FWA de novo methylation is blocked. Second, we discovered that a transgenic inverted repeat of the SUPERMAN locus causes de novo methylation of the endogenous SUPERMAN gene. This de novo methylation and gene silencing is also blocked in the drm2 mutant. Importantly, drm2 does not block gene silencing of preexisting silent FWA or SUPERMAN alleles, demonstrating that the DRM2 is important for de novo methylation, but is dispensable for the maintenance of preexisting methylation.

To more fully characterize non-CG methylation, we studied the drm2 and cmt3 mutants, as well as a drm2 cmt3 double mutant. This work showed that DRM2 and CMT3 act in a partially redundant and locus specific manner to control both asymmetric and CNG methylation. While DRM2 controls the majority of asymmetric methylation, and CMT3 controls the majority of CNG methylation, only in the drm2 cmt3 double mutant was all non-CG methylation eliminated.

2) The role of chromatin modifications. Our screen for suppressors of SUPERMAN gene silencing uncovered a second gene with a silencing phenotype remarkably similar to that of the CNG specific DNA methyltransferase CHROMOMETHYLASE3. This gene, named KRYPTONITE, encodes a member of the Su(var)3-9 class of histone methyltransferases, and like other members of this group, methylates lysine 9 of histone H3. Because CNG DNA methylation is lost in kyp mutants, this suggests that CNG methylation is controlled by histone methylation. Indeed, using chromatin immunoprecipitation we found that the KRYPTONITE mutants, but not CHROMOMETHYLASE3 mutants, reduce lysine 9 methylation at silent loci, clearly placing the action of histone lysine 9 methylation upstream of the targeting of DNA methylation. We have also found that histone H3 lysine 27 methylation appears to be important in the control of DNA methylation and we are currently trying to identify the methyltransferase enzyme(s) involved.

3) The role of small RNAs. Our screen for suppressors of SUPERMAN gene silencing uncovered a third gene called, ARGONAUTE4 (AGO4). ago4 mutants reduce both non-CG DNA methylation and histone H3 lysine 9 methylation at SUPERMAN and other affected loci. Since AGO proteins were only known to be involved in RNA interference and microRNA pathways that target mRNAs post-transcriptionally, it was initially surprising to find an AGO required for SUPERMAN transcriptional gene silencing. However, recent evidence from S. pombe, Tetrahymena, and Drosophila, as well as work from several Arabidopsis laboratories, now support the involvement of AGO proteins and small RNAs in the targeting of chromatin modifications. This work is important because, until now, we have not known of a mechanism that can explain how one gene gets methylated while another gene does not. Having siRNAs as a guide provides a highly specific targeting mechanism that likely involves the pairing of siRNAs with either DNA or nascent RNA transcripts.

A second approach again brought us to the conclusion that small RNAs are at the heart of DNA methylation control. We performed a reverse genetic screen for mutants that mimic drm2, and block de novo methylation of FWA. We found that four mutants involved in RNA silencing – rna dependent rna polymerase2 (rdr2), dicer-like3 (dcl3), silencing defective4 (sde4) and ago4 –failed to establish DNA methylation at FWA. Thus, a canonical RNA silencing pathway mediates de novo DNA methylation.

Our current work is focused on the mechanisms involved in the targeting of DNA methylation by chromatin modifications and small RNAs, using both genetic and biochemical approaches. We are also developing new techniques for studying genome wide patterns of DNA methylation, which should allow us to study wild type and mutants in unprecedented detail.