HDA6, a putative histone deacetylase needed to enhance DNA methylation induced by double-stranded RNA

Abstract

Introduction

RNA silencing refers to epigenetic gene silencing that is triggered by double-stranded RNA (dsRNA) (Matzke et al., 2001; Wang and Waterhouse, 2001). RNA silencing is usually considered with respect to post-transcriptional processes of RNA degradation that occur primarily in the cytoplasm, such as post-transcriptional gene silencing (PTGS) in plants (Vance and Vaucheret, 2001; Voinnet, 2001), quelling in Neurospora crassa (Cogoni, 2001) and RNA interference (RNAi) in animals (Hannon, 2002; Zamore, 2002). A central step in the PTGS/RNAi mechanism is processing of dsRNA into 21–24 nucleotide (nt) RNAs, which are believed to guide ribonuclease (RNase) cleavage of the complementary mRNA. Despite the current emphasis on post-transcriptional varieties of RNA silencing, there is increasing evidence that dsRNA can also induce transcriptional silencing and epigenetic changes at the genome level.

RNA-directed DNA methylation (RdDM) in plants provided the first example of an RNA-mediated epigenetic alteration of homologous nuclear DNA (Wassenegger et al., 1994; Wassenegger, 2000). RdDM involves the de novo methylation of almost all cytosines (Cs), not just those in symmetrical CG or CNG nucleotide groups, within a region of sequence identity between the triggering RNA and the target DNA (Pélissier et al., 1999). Experiments using Cre/lox-mediated recombination to convert a transcribed DNA direct repeat into an inverted repeat (IR), a step that initiates transcription of a hairpin RNA, confirmed that RdDM requires dsRNA (Mette et al., 2000). Moreover, if a dsRNA contains promoter sequences, de novo methylation and transcriptional gene silencing (TGS) of homologous promoters in trans can occur (Jones et al., 1999, 2001; Mette et al., 2000; Sijen et al., 2001). Similarly to the dsRNA involved in PTGS/RNAi, promoter dsRNAs are processed into short RNAs a little over 20 nt in length. It is not yet certain whether the short RNAs or the dsRNA precursors direct methylation of homologous DNA sequences, although the length of short RNAs is close to the minimal DNA target size of RdDM (∼30 bp) (Pélissier and Wassenegger, 2000). In addition, experiments using a virus-induced gene silencing system suggested that RNAs as short as 33 nt are sufficient to direct de novo methylation (Thomas et al., 2001). Evidence for functionally distinct sub-populations of short RNAs in plants was reported recently: ‘longer’ short RNAs 24–26 nt in length were implicated in RdDM, whereas ‘shorter’ short RNAs (21–22 nt) were involved in initiating PTGS (Hamilton et al., 2002).

Although RdDM is well established in plants, the extent to which it occurs in other organisms is not yet clear. The unusual pattern of methylation induced by RNA, in which Cs in any sequence context become modified, contrasts to the distribution in animal genomes, where methylation is present predominantly in CG dinucleotides. In a recent study, however, significant non-CG methylation was detected in mammalian embryonic stem cells (Ramsahoye et al., 2000). This finding hints that RdDM might occur in mammals but be restricted to an early period of development. Even in organisms that do not substantially methylate their DNA, RNAs might direct chromatin modifications to specific regions of the genome. Genetic evidence from Drosophila (Pal-Bhadra et al., 2002), fission yeast (Hall et al., 2002; Volpe et al., 2002) and Caenorhabditis elegans (Dudley et al., 2002) indi cates that TGS and/or chromatin modifications require components of the RNAi machinery. If RNA-guided genome modifications occur regularly in diverse organisms, then RdDM would simply represent a special type that is prevalent in plants. Because RdDM occurs throughout plant development and produces a signature pattern of DNA methylation, it is amenable to analysis in plants.

We have established a system in Arabidopsis for studying RNA-mediated promoter methylation and TGS of the constitutive nopaline synthase promoter (NOSpro). Silencing and methylation of the target NOSpro are initiated or substantially reversed within one generation after introducing or removing the silencing locus, which encodes a NOSpro dsRNA that is processed to short RNAs (Aufsatz et al., 2002). We are currently carrying out a genetic analysis of this system to identify components of the RdDM machinery and to assess the role of chromatin alterations in establishing and/or maintaining the distinctive pattern of methylation induced by RNA. As reported here, this mutant screen has identified a histone deacetylase, HDA6, that might act preferentially in the RNA-directed pathway of genome modifications.

Results

The NOSpro trans-silencing system has been described for tobacco (Mette et al., 2000) and Arabidopsis (Aufsatz et al., 2002). It consists of a target locus K (Kanamycin resistance), which contains NOSpro–NPTII and NOS genes (Figure 1, construct K), and an unlinked silencing complex H (Hygromycin resistance), which comprises a NOSpro IR downstream of a 35S promoter (35Spro) and a 19S-HPT gene (Figure 1, construct H). The target NOSpro–NPTII gene becomes transcriptionally silenced and methylated at symmetrical (CG and CNG) and asymmetrical (CNN) cytosines in the NOSpro region when NOSpro dsRNA is transcribed from the NOSpro IR. The NOSpro dsRNA is processed into short sense and antisense RNAs 21–24 nt in length (Aufsatz et al., 2002).

Open in a separate window

Fig. 1. Transgene constructs. The target construct K (Kanamycin resistance) contains a nopaline synthase (NOS) promoter (white N)–neomycin phosphotransferase (NPTII) gene and a NOS gene. The silencing construct H (Hygromycin resistance) contains a hygromycin phosphotransferase (HPT) gene driven by the 19S promoter and a NOSpro IR, separated by 260 bp spacer containing the α′ promoter of the soybean storage protein β-conglycinin gene (Chen et al., 1986), under the control of the 35S promoter. The M (Mutator) construct contains a phosphinothricin acetyltransferase (PAT) gene, encoding resistance to Basta, under the control of the mannopine synthase promoter (white M). Four copies of the B1 enhancer of tobacco endogenous pararetrovirus (Mette et al., 2002a) on the right allow for possible activation tagging (Weigel et al., 2000) in addition to insertional mutagenesis. A bacterial origin of replication (pUC18) can be used to clone ‘M’ inserts by plasmid rescue. Abbreviations: LB, RB: left and right T-DNA borders, respectively; Kbac, a Kan-resistance marker used in bacterial cloning steps; T, NOS terminator; OT, octopine synthase terminator.

To obtain mutants impaired in NOSpro RNA-mediated TGS, T-DNA insertional mutagenesis was performed. Plants doubly homozygous for the target K locus and silencing H locus were transformed with the M (Mutator) T-DNA construct containing a PAT gene, which encodes resistance to the herbicide Basta, under the control of a mannopine synthase promoter (Figure 1, construct M). The ‘M’ T-DNA contains no sequence homology in regulatory or protein-coding regions to either the K or H construct and therefore has no potential to interfere with expression of genes at the K or H loci by homology-dependent silencing. Plants that had taken up the ‘M’ T-DNA were selected by spraying with Basta. Herbicide-resistant plants (M1 generation) were allowed to selffertilize and pools of M2 seeds, which are still doubly homozygous for the target and silencing loci, were plated on medium containing kanamycin (Kan) to select for recovery of NOSpro–NPTII gene expression.

One Kan-resistant mutant line, rts1 (RNA-mediated transcriptional silencing), exhibited genetic behavior consistent with a recessive mutation. Homozygous rts1 progeny displayed moderate and relatively uniform resistance to Kan at a concentration of 40 mg/l (Figure 2A), appearing only slightly less resistant than wild-type seedlings containing the target locus in the unsilenced state (Figure 2C). The moderate Kan resistance of rts1 seedlings was accompanied by partial recovery of NOSpro–NPTII gene expression. RNase protection assays revealed that the steady-state level of NPTII RNA in rts1 mutants is ∼20–30% of that detected in the non-silenced target line in a wild-type background, and this value remained relatively stable into advanced generations (Figure 3, NPTII blot, lanes 6–8). Silencing is reinitiated in rts1/RTS1 heterozygotes as indicated by a return to undetectable levels of NPTII RNA in first generation backcross progeny (Figure 3, NPTII blot, lanes 4 and 5). Accordingly, rts1/RTS1 heterozygous seedlings are Kan sensitive (data not shown).

Open in a separate window

Fig. 2. Growth of rts1 mutant seedlings on Kan. Seedlings that are triply homozygous for the target K locus, silencing H locus and the rts1 mutation (A) grow almost as well on medium containing 40 mg/l Kan as wild-type seedlings that are homozygous for the unsilenced target K locus (C). Wild-type seedlings that are doubly homozygous for the target K locus and the silencing H locus are sensitive to Kan (B).

Open in a separate window

Fig. 3. RNase protection to analyze NPTII expression in rts1 mutants. Top: the map shows the location of the protected 190 bp NPTII fragment with respect to the NOSpro–NPTII gene. As input RNA, in vitro transcribed antisense RNA (black line) from a subcloned NPTII fragment upstream of the internal PstI (P) site was used. Bottom: NPTII expression was observed in wild-type T4 plants containing the unsilenced K locus (lane 2), but not detectable in T4 plants containing the silencing H locus (lane 3). Expression recovers 20–30% in rts1 plants from the M3, M4 and M5 generations (lanes 6–8), which is sufficient for resistance at the relatively low concentration of Kan used (40 mg/l). Silencing is rapidly reinitiated in first generation backcross (BC1) progeny, which are heterozygous for the rts1 mutation (lane 4, K and H loci are hemizygous; lane 5, K and H loci are homozygous). The percentage expression in rts1 mutants was determined by scanning blots with a Pharmacia Image Master using the residual input RNA after RNase digest (top band) as a reference standard. Lower band: positive control showing constitutive expression of the eIF-4A gene indicated by a 300 bp protected RNA fragment. The top band represents residual input RNA after RNase digest. R1 and r1: short versions of RTS1 and rts1, respectively.

NOSpro dsRNA is synthesized at roughly wild-type levels in rts1 mutants (Figure 4A), confirming that recovery of NOSpro–NPTII target gene expression is not due to a block in transcription of the NOSpro IR. Examination of NOSpro short RNAs in the rts1 mutant revealed a slight enhancement of all size classes of sense and antisense short RNAs derived from the NOSpro dsRNA (Figure 4B), suggesting that the rts1 mutation has a subtle influence on short RNA metabolism.

Open in a separate window

Fig. 4. NOSpro dsRNA and short RNAs in rts1 mutants. (A) The 0.3 kb NOSpro dsRNA, which is not made in plants lacking the silencing H locus (lane 1), is synthesized from the H locus in rts1 mutants (three shown, lanes 3–5) at levels comparable to those of wild-type plants (lane 2). (B) Sense and antisense NOSpro short RNAs 21–24 nt in length, which are not produced in plants lacking the silencing H locus (lanes 1 and 4) are somewhat enhanced in rts1 mutants (lanes 3 and 6) compared with wild-type plants (lanes 2 and 5). T4, M3 and M4 indicate the plant generation tested. Genotypes of plants are indicated above each lane. R1 and r1 are shortened versions of RTS1 and rts1, respectively. Arrows refer to the position of RNA size markers.

Methylation-sensitive restriction enzymes and DNA blot analysis were used to examine methylation in the target NOSpro in rts1 mutants. In wild-type plants, the target NOSpro is active and unmethylated in the absence of the silencing locus (Figure 5A, left, K/K;–/–; R1/R1). Following introduction of the silencing complex, the target NOSpro becomes heavily methylated at the sites recognized by BstUI and SacII, which are sensitive to CG and C(N)G methylation, respectively, and ∼50% methylated at the site recognized by NheI, which is sensitive to methylation in an asymmetrical CNN on the top strand and a CNG on the bottom strand (Figure 5A, right, K/K;H/H;R1/R1) (Aufsatz et al., 2002). In rts1 mutants, a unique pattern was reproducibly observed in all M3 plants tested: methylation was clearly reduced at the SacII and BstUI sites, but stayed essentially unchanged at the NheI site, resulting in similar proportions of methylated and unmethylated fragments for each of the three enzymes (Figure 5B, K/K;H/H;r1/r1; four plants). This pattern persisted into the M4 generation (Figure 5C, K/K;H/H;r1/r1; two plants), demonstrating that a stable methylation state had been attained. The reduction in methylation at the symmetrical Cs in the SacII and BstUI sites (∼30–50%) in rts1 mutants is consistent with the extent of recovery of NPTII gene expression (∼20–30%). The rts1 methylation pattern differed from those observed previously in met1 and ddm1 mutants, which are defective in a DNA methyltransferase (DMTase) and a putative chromatin remodeling protein, respectively. The effects of these mutations were non-uniform in our system, but some met1 and ddm1 mutant plants sustained substantial losses of methylation in both symmetric and asymmetric Cs (Aufsatz et al., 2002).

Open in a separate window

Fig. 5. DNA methylation analysis. As shown in the map below panel C, methylation in the target NOSpro was examined by performing an EcoRI/PstI (E/P) double digest and adding as indicated a methylation-sensitive restriction enzyme: SacII (S) (^mC^mCGCGG), BstUI (B) (^mCG^mCG) or NheI (N) (top strand: GCTAG^mCaa; bottom: GCT AG^mCtg). (A superscript ‘m’ indicates a methylated C that can inhibit cleavage.) The probe consisted of NPTII-coding sequences. (A) Left: the target NOSpro at the K locus is normally unmethylated in the absence of the silencing H locus, as indicated by shifts to the smaller fragments when S, B or N are used. Right: when the silencing H locus is introduced, heavy methylation at the S and B sites and ∼50% methylation at the N site is observed. In individual rts1 plants of the M3 (B) and M4 (C) generations, methylation at the S and B sites is reduced on average to ∼50%. This is similar to the level of methylation at N sites, which remains unchanged in rts1 mutants compared with the wild-type. (D) CG methylation at centromeric and rDNA repeats was examined in rts1 and ddm1 mutants and wild-type (wt) plants using the isoschizomers HpaII (H) (^mC^mCGG) and MspI (M) (^mCCGG). The patterns with H and M appear identical only in the ddm1 mutants, indicating loss of CG methylation. Genotypes of individual plants are indicated above each panel. R1 and r1 are shortened versions of RTS1 and rts1, respectively.

A bisulfite sequencing analysis of the target NOSpro in wild-type and rts1 mutant plants generally confirmed the results obtained with methylation-sensitive restriction enzymes. In wild-type plants, methylation was heavier in symmetric Cs than in asymmetric Cs, and in the rts1 mutant, methylation was partially reduced primarily at CGs and CNGs (Figure 6). The fractional decreases in C(N)G methylation must be considered in the context of continuous de novo methylation induced by NOSpro dsRNA/short RNAs (Aufsatz et al., 2002). Indeed, it is noteworthy that the reductions occur primarily in symmetric Cs where methylation can be maintained by MET1 (CG) (Jones et al., 2001; Kishimoto et al., 2001) and CMT3 (CNG) (Bartee et al., 2001; Lindroth et al., 2001; Papa et al., 2001). In contrast, methylation at asymmetric Cs is not thought to be maintained by any known DMTase (Pélissier et al., 1999) and this view is consistent with the observation that asymmetric C methylation in the NOSpro TGS system requires the continuous presence of NOSpro dsRNA (Aufsatz et al., 2002). Therefore, the level of asymmetric C methylation, which ranges from ∼25–50% on the two strands of the target NOSpro (Figure 6), can be taken as an indication of ongoing de novo methylation catalyzed by a de novo DMTase(s) in the presence of NOSpro RNAs. Both the restriction enzyme analysis and the bisulfite sequencing experiments showed that the major change in rts1 mutants is a shift downward in the amount of C(N)G methylation to a level approaching that in CNN nucleotide groups. This suggests that the rts1 mutation does not severely perturb de novo methylation but impedes primarily the reinforcement of C(N)G methylation above the ‘de novo’ level of ∼25–50% resulting from continuous RdDM.

Open in a separate window

Fig. 6. Summary of bisulfite sequencing of the target NOSpro in wild-type plants and rts1 mutants. Bottom: region of the NOSpro homologous to NOSpro dsRNA showing sites for methylation-sensitive restriction enzymes (Figure 5), the transcription start site (arrow) and start of the NPTII-coding region. Middle: cytosine methylation in the bottom DNA strand of the NOSpro (black: CG; blue: CNG; red: CNN). The height of the vertical lines shows the percentage of methylated Cs per total PCR clones sequenced. The dotted line indicates 50%. Top: bar graph showing reductions in methylation in the bottom strand for specific nucleotide groups in the rts1 mutant. In the top DNA strand, methylation was reduced from 64 to 55% (14%) in CGs; from 47 to 46% (2%) in CNGs; and was slightly increased (from 49 to 51%) in CNNs (not shown). Approximately 20 cloned PCR fragments were sequenced from each strand for both wild-type and rts1 mutant plants (see Supplementary data available at The EMBO Journal Online). DNA used for bisulfite treatment was isolated from pooled rts1 plants of the M4 generation. The data represent an average of this population.

To ascertain the effects of the rts1 mutation on global methylation patterns, CG methylation in centromeric and rDNA repeats was examined using the isoschizomers HpaII and MspI. Again in contrast to the effects of the ddm1 (Figure 5D, ddm1) and met1 (not shown) mutations, the rts1 mutation did not reduce CG methylation of either repeat family, as indicated by a pattern of digestion identical to that observed in wild-type plants (Figure 5D, rts1 and wt). This suggests that RTS1 activity is targeted to the NOSpro.

Attempts to isolate the RTS1 gene by screening λ genomic clones for inserts containing ‘M’ T-DNA–plant DNA junctions were unsuccessful. PCR analysis revealed that several unlinked ‘M’ T-DNAs were integrated into the genome of the rts1 mutant, but none of these cosegregated with the Kan-resistant phenotype of rts1 plants. Therefore, the RTS1 gene was isolated by positional cloning using codominant PCR-based markers; these detect polymorphisms between ecotype Columbia, which was used to set up the NOSpro silencing system, and the mapping ecotype Landsberg (Konieczny and Ausubel, 1993).

The rts1-1 mutation was mapped within a 98 kb physical interval on chromosome 5 represented in BAC clones MQB2, MJH22 and MDC12 (Figure 7). This genetically defined interval contains 28 annotated open reading frames (ORFs) (www.arabidopsis.org). PCR amplification of sets of overlapping fragments from rts1-1 mutant and wild-type plants identified a PCR fragment of MDC12.7 that showed a length polymorphism between mutant and wild-type plants. MDC12.7 encodes HDA6 (protein accession No. {"type":"entrez-protein","attrs":{"text":"BAB10553.1","term_id":"10177292"}}BAB10553.1), which belongs to the RPD3 family of histone deacetylases (Murfett et al., 2001; Plant Chromatin database; www.chromdb.biosci.arizona. edu).

Open in a separate window

Fig. 7. Mapping of RTS1. The rts1 phenotype (Kan resistance in the presence of the H silencing locus) cosegregated with markers LFY3 and CER454886 at the bottom of chromosome 5. In a mapping population of 624 plants there were nine recombination events between LFY3 and RTS1 and six between RTS1 and CER454886. Five additional markers were created within the LFY3/CER454886 genetic interval for fine mapping of RTS1. There were two recombination events between CER456729 and RTS1 and one between RTS1 and CER455379, locating RTS1 on a 98 kb physical interval. This sequence is represented in overlapping BAC clones MBQ2, MJH22 and MDC12. The bottom part shows organization of the RTS1/HDA6 gene. The black boxes represent exons, the dashed lines denote introns. Coordinates of translational start and stop, exon–intron boundaries and the positions of the mutation in the rts1-1 and rts1-2 alleles are shown with respect to the numbering of BAC clone MDC12 (accession No. {"type":"entrez-nucleotide","attrs":{"text":"AB008265","term_id":"2618600"}}AB008265). The start and end of the mRNA are demarcated according to accession Nos {"type":"entrez-nucleotide","attrs":{"text":"AF195548","term_id":"11066140"}}AF195548 and {"type":"entrez-nucleotide","attrs":{"text":"AY072201","term_id":"18176318"}}AY072201. The direction of the black arrow indicates that RTS1/HDA6 is in complementary orientation to MDC12.

Two rts1 alleles were recovered in separate M2 pools: rts1-1 had a 37 bp deletion 21 bp downstream of the start codon; rts1-2 sustained a 14 bp deletion in the last exon (Figure 7). The rts1-1 mutation is presumably a null allele since it disrupts the HDA6-coding region after the first seven amino acids. Although the rts1-2 allele only lacks the last 13 amino acids of the HDA6 protein, the strength of Kan resistance in rts1-2 and rts1-1 plants is comparable, suggesting that rts1-2 might also be a null allele.

To confirm the correspondence between HDA6 and the rts1 mutation, the wild-type HDA6 gene containing ∼1 kb 5′ and ∼160 bp 3′ flanking sequences was placed on a T-DNA encoding resistance to Basta and reintroduced into rts1 plants. RNase protection experiments performed using RNA prepared from Basta-resistant seedlings indicated that NPTII RNA was no longer detectable (Figure 8, lanes 2 and 3), demonstrating reinitiation of silencing. The wild-type HDA6 gene was thus able to complement the rts1 mutation.

Open in a separate window

Fig. 8. Complementation of the rts1 mutation with the wild-type HDA6 gene analyzed by RNase protection. Top: transformation of rts1-1 plants, which express the NOSpro–NPTII target gene, as evidenced by a 190 bp protected fragment (lane 1), with a 3.5 kb genomic HDA6 fragment results in re-silencing of the NPTII gene (lanes 2 and 3) comparable to wild-type plants (lane 4). Two representatives of ten complemented (+) T1 plants are shown. Wild-type expression of the NPTII gene in plants containing the unsilenced target K locus is shown in lane 5. Middle: positive control showing constitutive expression of the eIF-4A gene (300 bp protected fragment). Bottom: HDA6 genotypes. The rts1-1 allele contains a 37 bp deletion resulting in higher mobility of a PCR-amplified fragment on agarose gels (lane 1) compared with the PCR fragment from wild-type plants (lanes 4 and 5). Both bands can be detected in rts1 plants complemented with wild-type HDA6 (lanes 2 and 3). The K, H and RTS1 genotypes for individual plants are shown above each lane. R1 and r1 are shortened versions of RTS1 and rts1, respectively.

The hda6^rts1 mutant plants display a very mild phenotype. Even in plants of the M5 generation, when the rts1-1 mutation has been homozygous for four generations, the only visible defect is somewhat reduced fertility of the first flowers (data not shown).

Discussion

In a genetic screen designed to recover mutants impaired in TGS and RdDM of a NOSpro–NPTII target gene, we have identified a mutant, rts1, that is deficient in a putative Rpd3-like histone deacetylase, HDA6. Analysis with methylation-sensitive restriction enzymes and bisulfite sequencing revealed that the primary methylation defect in rts1 mutants is a reduction in methylation at symmetric CGs and CNGs, while levels of CNN methylation remain relatively unchanged. The partial loss of C(N)G methyl ation in rts1 mutants is associated with a moderate recovery of NOSpro–NPTII target gene expression, indicating that enhanced C(N)G methylation occurring as a consequence of HDA6 activity contributes to full silencing. The fact that the hda6^rts1 mutation does not affect CG methylation in centromeric or ribosomal DNA repeats suggests that HDA6 might be specialized for the RNA-directed pathway of chromatin modification. If so, HDA6 would be the first unique component of the RdDM pathway to be identified.

In mammals, histone deacetylases (HDACs) and DMTases act together to repress transcription through the action of methyl-binding proteins, which are com ponents of, or recruit, HDAC complexes to methylated DNA (Burgers et al., 2002; for a dissenting view, see Goll and Bestor, 2002). There is evidence for similar links between DNA methylation and histone deacetylation in plants (Chen and Pikaard, 1997; Li et al., 2002). Nevertheless, knowledge about how DMTases and HDACs are targeted to specific regions of the genome is limited. In our system, it is clear that de novo methylation at symmetrical and asymmetrical Cs in the target NOSpro is directed by NOSpro RNAs (Aufsatz et al., 2002). This information allows us to propose a hypothetical sequence of events in which HDA6 acts in between the de novo and C(N)G maintenance (or reinforcement) steps (Figure 9).

Open in a separate window

Fig. 9. Hypothetical model for RTS1 activity. One or more de novo DMTases in conjunction with NOSpro RNAs (wavy lines) catalyze an intermediate level of uniform methylation (m) of Cs in all sequence contexts in the target NOSpro, resulting in partial silencing. Although a known de novo DMTase in Arabidopsis, DRM (Domain-Rearranged Methyltransferase; Cao and Jacobsen, 2002), remains a good candidate for this step, other candidate DMTases (MET1, CMT3, Dnmt2; Finnegan and Kovac, 2000) cannot yet be ruled out. HDA6 is envisioned to recognize the partially methylated DNA and to recruit maintenance (or reinforcement) DMTases such as MET1 and CMT3, which enhance CG (bold capital M) and to a lesser extent CNG (capital M) methylation. This step helps to lock in the silent state, probably in conjunction with additional histone modifications.

We assume that the degree of methylation in asymmetrical Cs (∼25–50%), which is not maintained following cessation of dsRNA synthesis (Aufsatz et al., 2002), reflects continuous RdDM catalyzed by one or more de novo DMTase(s), which remain to be identified, in the presence of NOSpro RNAs. The rts1 mutation does not appear to interfere substantially with this step, which leads to roughly comparable, intermediate levels of methylation in symmetric and asymmetric Cs and to partial silencing. Instead, the rts1 mutation appears to inhibit the enhancement of methylation, predominantly in C(N)G nucleotide groups, to a degree in excess of the ‘de novo’ level. The augmentation of C(N)G methylation is apparently necessary to lock in the silent state, probably in conjunction with additional histone modifications. In this hypothetical model, HDA6 is envisaged to recognize the partial methylation at most Cs resulting from the de novo step of RdDM; it then recruits MET1 and CMT3, the traditional CG and CNG maintenance activities, respectively (Bartee et al., 2001; Kishimoto et al., 2001; Lindroth et al., 2001; Papa et al., 2001), to reinforce preferentially C(N)G methylation. This scenario would be consistent with reported associations between Dnmt1, the mammalian homolog of MET1, and HDACs in mammals (Burgers et al., 2002). Furthermore, a role for HDA6 in maintenance rather than establishment of transcriptional repression is consistent with a similar maintenance role proposed for an Rpd3-like HDAC in Drosophila (Wheeler et al., 2002).

HDA6 was recovered in a separate mutant screen that was initially based on deregulation of an auxin-responsive transgene (Murfett et al., 2001). It turned out, though, that the hda6 mutation was not interfering with an auxin response, but was alleviating silencing of a multicopy transgene locus. It is interesting to consider why hda6 mutations were repeatedly recovered in our screen and in the one carried out by Murfett and coworkers, particularly since this HDAC has not been identified in other genetic screens to find mutants defective in DNA methylation (Vongs et al., 1993; Bartee et al., 2001; Lindroth et al., 2001) or cis-TGS (Amedeo et al., 2000). The NOSpro trans-TGS system relies on promoter dsRNA produced by transcribing a NOSpro IR. Intriguingly, the transgene construct used by Murfett and colleagues also contained a promoter IR of ∼260 bp. While the authors attributed the spontaneous silencing of their transgene locus to its repetitive nature, it is conceivable that a promoter dsRNA was unknowingly transcribed from the promoter IR in their system.

The Arabidopsis thaliana genome encodes 15 HDACs; four of these, including HDA6, are similar to the Saccharomyces cerevisiae Rpd3 family. The amino acid sequences of all four Rpd3-like HDACs in Arabidopsis are highly similar except in the C-termini, which diverge significantly. This suggests that the four enzymes have distinct functions. Two of the genes, HDA1 and HDA6, are expressed in seedlings, which is the stage at which both we and Murfett and coworkers selected for mutants. HDA1 is expressed more highly than HDA6 in seedlings (Plant Chromatin Database; http://chromdb.biosci.arizona.edu), indicating that the frequent recovery of hda6 mutants in each genetic screen was not simply due to selection for the HDAC expressed most strongly at this stage of development. Although the enzymatic activity of HDA6 has not yet been confirmed, the HDA6 gene encodes all of the amino acids that are important for HDAC activity and/or transcriptional repressor function (Hassig et al., 1998; Kadosh and Struhl, 1998). Antisense suppression of the HDA1 gene leads to pleiotropic effects on Arabidopsis development, owing in some cases to ectopic expression of tissue-specific genes (Tian and Chen, 2001). In contrast, hda6^rts1 mutants appear essentially normal, even into advanced generations.

NOSpro short RNAs were slightly but reproducibly enhanced in the hda6^rts1 mutant. Until the RNA species involved in RdDM is clarified, it is difficult to judge the significance of this observation. Nevertheless, an influence on NOSpro short RNA metabolism is consistent with the involvement of guide RNAs in NOSpro silencing and methylation, which in turn is compatible with the recruitment of HDA6 to the target NOSpro. The idea that HDA6 might function preferentially in an RNA-directed pathway of genome modifications to reinforce primarily C(N)G methylation implies a degree of HDAC specialization that is in agreement with recent results from yeast. The five HDACs in S.cerevisiae are targeted to different areas of the genome and overlap only slightly in function: Hos1/Hos3 and Hos2 affect rDNA and ribosomal protein genes, respectively; Sir2 represses subtelomeric heterochromatin; and Rpd3 and Hda1 are involved mainly in deacetylating specific promoters by recruitment mechanisms that remain to be discovered (Robyr et al., 2002). Our observation that Rpd3-like HDA6 does not act on centromeric or rDNA repeats but appears to be recruited to the NOSpro as part of an RNA-directed pathway of silencing opens up the possibility that guide RNAs might similarly target specific HDACs to cognate promoters in yeasts and other eukaryotes. Promoter regions are not generally thought to be transcribed. However, further analysis of the numerous microRNAs (miRNAs) derived from dsRNAs encoded by imperfect IRs in intergenic regions in Arabidopsis (Llave et al., 2002a, b; Mette et al., 2002b; Park et al., 2002; Reinhart et al., 2002; Rhoades et al., 2002) and animals (McManus and Sharp, 2002) might reveal that some of these contain promoter sequences. Indeed, seven miRNAs that possibly target promoters of endogenous genes were identified recently in Arabidopsis (Park et al., 2002). Given the extraordinary specificity that can be achieved through RNA–DNA base pairing, even with very short RNAs (Eddy, 2002; Storz, 2002), the potential for guide RNAs to play a major role in directing epigenetic modifications to identical sites in eukaryotic genomes is considerable.

Materials and methods

Acknowledgements

References