DRD1 is a SNF2‐like protein previously identified in a screen for mutants defective in RNA‐directed DNA methylation of a seed promoter in Arabidopsis. Although the initial study established a role for DRD1 in RNA‐directed DNA methylation, it did not address whether DRD1 is needed for de novo or maintenance methylation, or whether it is required for methylation of other target sequences. We show here that DRD1 is essential for RNA‐directed de novo methylation and acts on different target promoters. In addition, an unanticipated role for DRD1 in erasure of CG methylation was shown when investigating maintenance methylation after segregating away the silencing trigger. DRD1 is unique among known SNF2‐like proteins in facilitating not only de novo methylation of target sequences in response to RNA signals, but also loss of methylation when the silencing inducer is withdrawn. The opposing roles of DRD1 could contribute to the dynamic regulation of DNA methylation.
RNA‐directed DNA methylation has been documented in plants (Mathieu & Bender, 2004) and in human cells (Kawasaki & Taira, 2004; Morris et al, 2004). Genetic analyses in Arabidopsis have identified the DNA methyltransferases that catalyse de novo cytosine (C) methylation in response to RNA signals (Cao et al, 2003; Aufsatz et al, 2004) and maintain symmetrical C(N)G methylation (N is A, T or C) in cooperation with histone‐modifying enzymes (Jones et al, 2001; Aufsatz et al, 2002a; Jackson et al, 2002; Malagnac et al, 2002). A plant‐specific SNF2‐like putative chromatin‐remodelling protein, DRD1, was identified in a screen for mutants impaired in RNA‐directed DNA methylation and silencing of the seed‐specific α′ promoter (Kanno et al, 2004). From the initial genetic screen, it was unclear whether drd1 mutants are deficient in de novo methylation or in maintenance methylation. In the case of RNA‐directed DNA methylation, de novo methylation leads to the modification of Cs in all sequence contexts within the region of RNA–DNA sequence identity, whereas maintenance methylation refers to perpetuation of symmetrical C(N)G methylation during DNA replication in the absence of the RNA trigger (Matzke et al, 2004). To further understand the role of DRD1 in RNA‐directed DNA methylation, we examined separately the effects of a drd1 mutation on RNA‐directed de novo methylation and on maintenance methylation. We report here that DRD1 influences both of these processes in a manner that facilitates the dynamic regulation of DNA methylation.
Results and Discussion
Transgenic systems to study RNA‐mediated silencing and methylation of the seed‐specific α′ promoter (Kanno et al, 2004) and the constitutive nopaline synthase (NOS) promoter (Aufsatz et al, 2002b) have been described previously. These systems rely on a double‐stranded RNA—encoded at a silencer H complex (hygromycin resistance)—that is homologous to the respective target promoter sequence. The double‐stranded RNAs are processed to short RNAs, 21–24 nucleotides in length, which are thought to trigger de novo methylation and silencing of the homologous target promoter at an unlinked target K complex (kanamycin resistance; supplementary Fig 1 online). In the absence of the silencer, the respective target promoter is active and unmethylated (Fig 1A: left, α′ promoter; right, NOS promoter), whereas it is repressed and methylated in the presence of the silencer (Fig 1B: left, α′ promoter; right, NOS promoter).
In drd1 mutants, non‐CG methylation of the target α′ promoter is greatly reduced, as evidenced by complete digestion with methylation‐sensitive restriction enzymes abbreviated F, Sc, Ps, Pa, E and Ba (Fig 1C, left: data are for third‐generation drd1‐6 plants). By contrast, substantial CG methylation is retained, as indicated by the persistent double band generated by the restriction enzyme abbreviated H (Fig 1C, left). Similar findings for drd1‐1 led to the suggestion that DRD1 is important primarily for non‐CG methylation (Kanno et al, 2004).
To analyse whether DRD1 is needed for RNA‐directed de novo methylation of target sequences, crosses were made to generate F1 plants in which a naive target K complex was combined with the silencer H complex in either wild‐type (D/D; Fig 2D) or homozygous drd1‐6 (d/d) plants (Fig 2E). Methylation of the target promoter was then examined in the resulting F1 progeny.
In wild‐type F1 plants, the target α′ promoter showed increased methylation in CGs and in non‐CGs after introducing the silencer H complex (Fig 1D, left). The level of methylation observed in wild‐type F1 plants approximated that seen in plants in which the target complex and silencer complex had been together in the same genome for several generations (Fig 1B, left). The similar methylation patterns indicate that the maximum attainable level of RNA‐directed methylation of the target α′ promoter is essentially reached in the first generation containing both transgene complexes. By contrast, the target α′ promoter did not acquire detectable methylation after being combined with the silencer H complex in homozygous drd1‐6 plants (Fig 1E, left), which showed a hybridization pattern identical to that of the unmethylated target α′ promoter (Fig 1A, left). The lack of methylation was not because of inadequate production of RNA signals, as indicated by the continued presence of α′ promoter short RNAs in drd1‐6 plants (Fig 3A).
The target NOS promoter also becomes methylated de novo at CGs and non‐CGs in wild‐type plants after introducing the respective silencer H construct (Fig 1D, right; Aufsatz et al, 2004), but it failed to acquire measurable methylation in the drd1‐6 mutant (Fig 1E, right). Bisulphite sequencing confirmed that no detectable methylation was induced at the target NOS promoter in drd1‐6 mutants, whereas methylation was observed at Cs in all sequence contexts within the region of RNA–DNA sequence identity in wild‐type plants (supplementary Fig 2 online). NOS promoter short RNAs were detected in the F1 plants (Fig 3B), indicating that RNA signals were available but unable to induce methylation in the drd1‐6 mutants. In contrast to the α′ promoter, the level of NOS promoter methylation was less in F1 progeny than in plants in which the target and silencer had been together for several generations (compare Fig 1D, right, with Fig 1B, right). Despite this difference, the results show that DRD1 is indispensable for RNA‐directed de novo methylation of two distinct promoters.
The efficiency of maintenance methylation was examined in wild‐type and drd1‐6 plants after crossing out the respective silencer H complexes to remove the source of the RNA signals. In wild‐type F2 progeny descended from DRD1 wild‐type parents (Fig 2F), the target α′ promoter lost both CG and non‐CG methylation after segregating away the silencer complex (Fig 4F, left), resulting in a hybridization pattern that is indistinguishable from the unmethylated α′ promoter (Fig 1A, left). An identical pattern of methylation (Fig 4G, left) was observed in wild‐type F2 progeny descended from the drd1‐6 mutant (Fig 2G). Bisulphite sequencing confirmed the loss of non‐CG methylation and showed only some residual methylation at two CG dinucleotides (supplementary Fig 3A online). Thus, in the α′ promoter silencing system, almost all methylation is lost in wild‐type progeny when the source of the RNA signal is withdrawn. Unexpectedly, however, in drd1‐6 progeny lacking the silencer H complex (Fig 2H,I), substantial CG methylation was detected, even though non‐CG methylation was lost. Retention of CG methylation is again exemplified by the persistent double band generated by the restriction enzyme abbreviated H. This double band was observed in drd1‐6 progeny that are homozygous (Fig 4H, left) and hemizygous (Fig 4I, left) for the target K complex, indicating no dependence on dosage of the target promoter. Bisulphite sequencing confirmed the enhanced maintenance of CG methylation in the drd1‐6 mutant (supplementary Fig 3B online) relative to wild‐type plants (supplementary Fig 3A online).
A comparable result was obtained when analysing maintenance methylation in the NOS promoter system. In contrast to the α′ promoter, the target NOS promoter retains significant methylation in the absence of the silencer H complex in wild‐type plants: although non‐CG methylation seems to be lost (exemplified by complete digestion with the enzyme abbreviated N (NheI)), considerable CG methylation remains, as shown by only partial digestion with restriction enzymes abbreviated P (Psp1406I) and B (BstUI) (Fig 4F, right). Similarly to the α′ promoter, however, the NOS promoter shows increased CG methylation in drd1‐6 progeny. This is evidenced by poorer digestion with enzymes P and B in drd1‐6 mutants (Fig 4H,I, right) than in wild‐type plants (Fig 4F,G, right). By contrast, cleavage by enzyme N, which is diagnostic of non‐CG methylation, is equivalent in both drd1‐6 mutants and wild‐type plants (Fig 4F–I, right). Collectively, these findings show a previously unsuspected role for DRD1 in the complete erasure of CG methylation after segregating away the silencer H complex that encodes the RNA trigger.
The results of the experiments reported here, which have examined separately de novo and maintenance methylation, necessitate a revision of the original proposal that DRD1 is primarily important for non‐CG methylation (Kanno et al, 2004). DRD1 is required for RNA‐directed de novo methylation of Cs in all sequence contexts, including methylation of CG dinucleotides (Fig 5). However, DRD1 is also necessary for efficient loss of methylation, particularly CG methylation, once the source of the RNA signal is removed. The latter requirement is more evident for the target α′ promoter, which loses almost all CG methylation in wild‐type plants after segregating away the silencer complex but shows elevated CG methylation in the absence of the silencer in the drd1‐6 mutant. Therefore, the methylation pattern observed in third‐generation drd1 plants can be explained as an additive defect caused by inadequate erasure of CG methylation coupled to failed de novo methylation, which affects predominantly non‐CG nucleotide groups (Fig 5). Whether the remaining CG methylation in drd1 mutants is due to more robust maintenance or to a deficiency in either passive loss or active removal of CG methylation is not known. Passive loss occurs when there is a failure to maintain methylation during successive rounds of DNA replication. By contrast, active demethylation requires a demethylase activity and can take place in nondividing cells (Santos et al, 2002). In animal systems, DNA glycosylases, which are normally involved in base excision repair, have been reported to be involved in active demethylation of CG dinucleotides (Kress et al, 2001). In Arabidopsis, two large proteins that contain DNA glycosylase domains, DEMETER (Xiao et al, 2003; Kinoshita et al, 2004) and REPRESSOR OF SILENCING 1 (Gong et al, 2002), have been implicated in the removal of CG methylation and the re‐activation of silenced genes. Additional work is required to clarify how DRD1 contributes to the loss of CG methylation.
Although DRD1 seems to be a common requirement for RNA‐directed DNA methylation, the two target promoters tested here differ in the rate of their response to the introduction or withdrawal of RNA signals in wild‐type plants. The target α′ promoter is very responsive, acquiring a full level of methylation in the first generation that combines the target and silencer complexes and losing most methylation in the first generation after crossing out the silencer complex. By contrast, the target NOS promoter does not attain the full level of RNA‐directed methylation in the F1 generation and retains substantial CG methylation after removing the source of the trigger RNA. The reasons for these differences in response time are not known. In both systems, the respective short RNAs seem to be present in comparable amounts. Variations in the sequence composition, structure or chromatin condensation state of the two target complexes might alter their sensitivity to RNA signals (Matzke et al, 2004). For example, there are several copies of the NOS promoter at the corresponding target K complex (Aufsatz et al, 2002b), whereas only a single copy of the α′ promoter is present at the corresponding target K complex (Kanno et al, 2004). The higher density of CG dinucleotides in the NOS promoter (supplementary Fig 2 online) may contribute to more efficient maintenance of CG methylation at this promoter. By contrast, the α′ promoter, which is comparatively deficient in CG dinucleotides (supplementary Fig 3 online), is not repressed by CG methylation but instead seems to be silenced by non‐CG methylation (Kanno et al, 2004). As shown here, this non‐CG methylation must be established de novo in each generation by a DRD1‐dependent, RNA‐directed pathway. The fact that the α′ promoter is silenced by non‐CG methylation probably allowed us to identify the drd1 mutant in the corresponding genetic screen. By contrast, proteins required for CG methylation, the DNA methyltransferase MET1 (Aufsatz et al, 2002a) and histone deacetylase 6 (Aufsatz et al, 2002b), were identified in a genetic screen carried out on the NOS promoter silencing system.
Several other SNF2‐like proteins, including DDM1 (Jeddeloh et al, 1999), Lsh (Dennis et al, 2001) and ATRX (Gibbons et al, 2000), can regulate DNA methylation. However, DRD1 is the only one of these factors that has been implicated both in RNA‐directed de novo methylation of cytosines in all sequence contexts and in erasure of CG methylation. The drd1 alleles identified in the original genetic screen contain mutations in functionally implicated regions of the SWI2/SNF2 ATPase domain (Kanno et al, 2004), suggesting that DRD1 functions as a chromatin‐remodelling protein to disrupt histone–DNA contacts and/or displace nucleosomes. One possibility is that DRD1 is specialized to allow RNA signals to access homologous target DNA in a chromatin context. Depending on the availability of RNA signals and various DNA‐modifying enzymes in different cell types, DRD1 activity could facilitate RNA‐guided de novo methylation catalysed by DNA methyltransferases or demethylation of CG dinucleotides catalysed by DNA glycosylases. In this model, DRD1 is a pivotal factor in the dynamic regulation of promoter activity, which might contribute to developmental plasticity, epigenetic reprogramming or adaptation of plants to the environment. Future research will have to focus on determining the full scope of DRD1 activity by using genomics approaches to identify endogenous targets in the genome.
In view of the proposed role of DRD1 in opening chromatin to RNA signals to create a substrate for de novo methylation, one can ask whether loss of CG methylation, which also requires DRD1, occurs at DNA sequences targeted by short RNAs (Fig 5, short RNAs in parentheses). Although the silencing complex that encodes the trigger RNA was segregated away in plants used to assess maintenance methylation, we cannot exclude the possibility that parental RNA signals are carried over into cells of the next generation, providing a short time period early in development when RNA‐guided demethylation could occur. For example, DEMETER acts only in the central cell of the female gametophyte to remove CG methylation (Xiao et al, 2003; Kinoshita et al, 2004). Thus, whether RNA‐directed de novo methylation or RNA‐guided demethylation occurs would be governed by the differential cell‐type‐specific availability of either DNA methyltransferases or DNA glycosylases, respectively.
Plant material and genotyping. All experiments were performed on drd1‐6 mutant (Kanno et al, 2004) or wild‐type Arabidopsis thaliana (ecotype Columbia‐0). The genotypes of plants used for molecular analyses were determined by PCR‐based screens, as described in the supplementary information online.
DNA methylation analysis. DNA was isolated from genotyped plants and analysed for cytosine methylation, using methods described previously (Aufsatz et al, 2002a, 2002b, 2004; Kanno et al, 2004). Southern data results were reproduced for at least two plants of each genotype. For the drd1‐6 mutant shown in Fig 1C (left), DNA was isolated from plants of the M3 generation (genotype K/K;H/H;d/d).
Supplementary information is available at EMBO reports online (http://www.nature.com/embor/journal/vaop/ncurrent/extref/7400446‐s1.pdf).
We acknowledge financial support from the Austrian Fonds zur Förderung der wissenschaftlichen Forschung (Grant P‐15611‐B07) and the European Union (contract no. HPRN‐CT‐2002‐00257).
- Copyright © 2005 European Molecular Biology Organization