ARGONAUTE4 Control of Locus-Specific siRNA Accumulation and DNA and Histone Methylation

See allHide authors and affiliations

Science  31 Jan 2003:
Vol. 299, Issue 5607, pp. 716-719
DOI: 10.1126/science.1079695


Proteins of the ARGONAUTE family are important in diverse posttranscriptional RNA-mediated gene-silencing systems as well as in transcriptional gene silencing in Drosophila and fission yeast and in programmed DNA elimination in Tetrahymena. We cloned ARGONAUTE4 (AGO4) from a screen for mutants that suppress silencing of the Arabidopsis SUPERMAN(SUP) gene. The ago4-1 mutant reactivated silentSUP alleles and decreased CpNpG and asymmetric DNA methylation as well as histone H3 lysine-9 methylation. In addition,ago4-1 blocked histone and DNA methylation and the accumulation of 25-nucleotide small interfering RNAs (siRNAs) that correspond to the retroelement AtSN1. These results suggest that AGO4 and long siRNAs direct chromatin modifications, including histone methylation and non-CpG DNA methylation.

Members of the ARGONAUTE (AGO) protein family are important in RNA-mediated silencing systems such as posttranscriptional gene silencing (PTGS) in plants, RNA interference in animals, and quelling in fungi (1, 2). These systems use a ribonuclease III enzyme, DICER, to generate 21- to 22-nucleotide (nt) small interfering RNAs (siRNAs), which target the destruction of homologous RNA. In plants, PTGS is often associated with RNA-directed methylation of the corresponding DNA (3) and, conversely, plant chromatin mutants such as ddm1 andmet1 can affect PTGS (4). In addition, AGO family members have recently been implicated in histone modifications and transcriptional gene silencing. In fission yeast, deletion of argonaute, dicer, and RNA-dependent RNA polymerase homologs causes transcriptional derepression and loss of histone H3 lysine-9 (H3K9) methylation (5,6). In Tetrahymena, TIWI, a member of the PIWI subfamily of AGOs, is required for programmed DNA elimination, which is associated with 28-nt siRNAs and histone H3K9 methylation (7, 8). Finally, Drosophila PIWI is required for both posttranscriptional and transcriptional repression of alcohol dehydrogenase transgenes (9).

The clark kent (clk) mutants are epigenetic alleles of the Arabidopsis SUP gene caused by SUPgene silencing and extensive DNA methylation of CpG, CpNpG (where N is either A, C, T, or G), and asymmetric (CpHpH, where H is either A, C, or T) cytosines (10). The clk mutants are recessive and meiotically heritable, suggesting a primarily chromatin-based gene-silencing mechanism. Indeed, the initiation ofSUP silencing requires the DRM2 de novo DNA methyltransferase (11). Furthermore, by screening for suppressors of the clk-st allele, we isolated two chromatin modification enzymes required for the maintenance of SUPgene silencing, CHROMOMETHYLASE3 (CMT3) andKRYPTONITE (KYP). CMT3 encodes a DNA methyltransferase, and KYP encodes a histone H3K9–specific protein methyltransferase (12,13). kyp and cmt3 mutants both cause a loss of CpNpG methylation at SUP and all other loci tested. Here we describe the cloning of a third clk-st suppressor mutation in the AGO4 gene.

We identified one recessive allele of a clk-st suppressor gene that mapped to chromosome II. Other than suppression ofSUP silencing, we did not observe morphological defects in the homozygous mutant. By sequencing candidate genes, we identified a mutation in the AGO4 gene, previously named on the basis of its sequence similarity to AGO1 (1). The mutation destroyed a splice acceptor site, causing a deletion and frameshift that terminated translation after 595 amino acids (fig. S1). TheAGOs comprise a conserved family of eukaryotic genes (fig. S1) containing two domains of unknown function: an NH2-terminal PAZ domain and a COOH-terminal PIWI domain (14). Because the frameshift deleted almost the entire PIWI domain, the mutation is likely to cause severe loss ofAGO4 function. To confirm that the suppressor mutation is within AGO4, we transformed mutant plants with theAGO4 gene and found that the original clk-stphenotype was restored. Thus we named this suppressor mutationago4-1.

We analyzed the effect of ago4-1 on SUP DNA methylation using bisulfite genomic sequencing (Fig. 1A and table S1). Whereas CpG methylation levels were unchanged, ago4-1 showed a 2.8-fold reduction in CpNpG and a 4.5-fold reduction in asymmetric methylation. This methylation phenotype was similar to that of the cmt3 andkyp mutants, except that cmt3 and kypshowed a stronger reduction of CpNpG methylation than didago4-1. We previously found that, at all loci tested,cmt3 and kyp showed a reduction of CpNpG methylation but not of CpG methylation (12, 13). Therefore, we used Southern blot analysis with methylation-sensitive restriction enzymes to assay the effect of ago4-1 on both CpG and CpNpG methylation at three additional loci: the 180–base-pair (bp) centromeric repeat (CEN) sequence (Fig. 1B), theTa3 retrotransposon (Fig. 1C), and the FWA gene (Fig. 1D). The ago4-1 mutation did not affect either CpNpG or CpG methylation levels at these loci. The FWA locus also contains a substantial amount of asymmetric methylation (15), and bisulfite sequencing of FWA showed that the ago4-1 mutation did not reduce this methylation. Thus, the methylation phenotype of ago4-1 is locus-specific and different than that of the cmt3 and kyp mutants.

Figure 1

Methylation analysis of theago4-1 mutant. (A) Bisulfite sequencing results show the percent methylation in different sequence contexts ofSUP, MEA-ISR, and AtSN1 in different mutant backgrounds. (B to E) Southern blot analysis of genomic DNA cut with the indicated restriction enzyme and probed with (B) CEN, (C) Ta3, (D) FWA, and (E) MEA-ISR probes. Probes and restriction maps are described in (16). Hpa II and Msp I recognize the sequence CCGG. Hpa II is inhibited by methylation of either cytosine and allows detection of CpG and CpNpG methylation, whereas Msp I is only inhibited by methylation of the outer cytosine, allowing detection of CpNpG methylation only. Cfo I detects CpG methylation and Bgl II detects CpNpG methylation. ago4 A and B are different isolates of ago4-1. Controls include the met1 andcmt3-7 mutants in (B), (C), and (E), and the hypomethylatedfwa-1 mutant (15) in (D).

We found three other loci at which ago4-1 did have an effect on DNA methylation: MEA-ISR,AtSN1, and AtMu1. MEA-ISR is an approximately 183-bp sequence present in seven direct repeats in an intergenic region adjacent to the imprinted MEDEA gene (16). In the wild type, MEA-ISR locus contains 95% CpG, 58% CpNpG, and 26% asymmetric methylation (Fig. 1A and table S1). ago4-1 essentially eliminated the CpNpG and asymmetric methylation but did not affect the CpG methylation (Fig. 1A). We used Southern blot analysis with methylation-sensitive restriction enzymes to confirm these results. We found that CpNpG methylation was eliminated in ago4-1 but that CpG methylation was unaffected (Fig. 1E). AtSN1 is a retrotransposon sequence previously shown to be methylated (17). We found that the wild-type AtSN1 locus contains 75% CpG, 70% CpNpG, and 24% asymmetric methylation (Fig. 1A and table S1). ago4-1greatly reduced the non-CpG methylation to 14% CpNpG and 0.8% asymmetric methylation. The AtMu1 sequence is the 3′-terminal inverted repeat of the Arabidopsis DNA transposon Mu1 (18). We found that wild-typeAtMu1 shows 58% CpG, 35% CpNpG, and 11% asymmetric methylation. The ago4-1 mutation did not affect the CpG methylation but reduced the CpNpG methylation to 19% and the asymmetric methylation to 4.8% (table S1).

The locus-specific effect of ago4-1 shows that bothAGO4-dependent and AGO4-independent mechanisms control non-CpG methylation. CEN, Ta3, andFWA rely on an AGO4-independent mechanism,MEA-ISR on an AGO4-dependent mechanism, andSUP, AtSN1, and AtMu1 on both mechanisms. One explanation for AGO4-independent non-CpG methylation is that another AGO gene (nine of which are present in the Arabidopsis genome) could act redundantly with AGO4. Alternatively, pathways that do not involveAGO genes could function at some loci. For instance, if AGO4's primary role is to establish methylation, effects will only be visible at loci that require frequent establishment.

A comparison of the methylation phenotype of ago4-1 with those of mutants of CMT3 and DRM, the two types of DNA methyltransferase genes known to control non-CpG methylation, did not show a simple relationship (Fig. 1A). In particular,ago4-1 mimicked the drm1 drm2 double mutant atMEA-ISR (16). However, at both SUPand AtSN1, ago4-1 showed a reduction in CpNpG methylation that was intermediate between the effects of thecmt3-7 and drm1 drm2 mutants and a reduction of asymmetric methylation that was stronger than the effect of either (Fig. 1A). These results suggest that both CMT3 andDRM are involved in AGO4-dependent methylation.

To determine the relationship between AGO4, KYP, and CMT3, we performed chromatin immunoprecipitation (ChIP) experiments to examine histone H3K9 methylation levels at SUP. We previously found thatkyp, but not cmt3, reduced H3K9 methylation atSUP (19), suggesting that CMT3 acts downstream of KYP, because of targeting of CMT3 to methylated histones (13). Figure 2 shows thatago4-1 reduced H3K9 methylation of SUP relative to the wild-type strain clk-st. The simplest interpretation of these results is that AGO4 acts upstream of KYP to target H3K9 methylation. We also found that ago4-1 reduced H3K9 methylation at AtSN1, a locus where ago4-1 also reduced DNA methylation (Fig. 2). However, ago4-1 did not reduce H3K9 methylation of Ta3 (Fig. 2) or of theCEN repeats, where ago4-1 showed no DNA methylation effect. Thus, the effects of ago4-1 on H3K9 methylation are locus-specific and correlate with effects on DNA methylation.

Figure 2

ChIP analysis of H3K9 methylation, showing multiplex polymerase chain reaction analyses of SUP,AtSN1, and Ta3 together with ACTIN, a locus with a low level of H3K9 methylation (19). The input isclk-st chromatin before immunoprecipitation. “No AB” lanes are control immunoprecipitations with no antibody. The fold enrichment of the SUP, AtSN1, and Ta3signal over the ACTIN signal is shown.

We tested whether AGO4 function is associated with siRNAs by probing Northern blots of RNA preparations that had been enriched for small RNAs. AtSN1 was recently shown to be associated with a newly discovered class of long (approximately 25-nt) siRNAs (17). We could easily detect AtSN1 siRNAs in the wild-type Ler or clk-st strains and in thecmt3 or kyp mutant strains (Fig. 3A). However, these siRNAs were reduced to below the level of detection in ago4-1. We did not detect siRNAs specific for the SUP or AtMu1 sequences. This may be due to the limited sensitivity of Northern blot analysis, because siRNAs to AtSN1 (present in approximately 70 copies per genome) are probably easier to detect than siRNAs to low-copy-number genes such as SUP and AtMu1(17).

Figure 3

Effect of ago4-1 on AtSN1siRNAs. (A) Northern blot of small RNAs hybridized with a sense AtSN1 RNA probe. ∼25-nt siRNAs are found in all genetic backgrounds shown except for the ago4-1 mutant strain. Positions of 20- and 30-nt RNA markers are indicated. (B) Model for the function of AGO4 and long siRNAs in the control of histone and DNA methylation.

It has been shown that long siRNAs of tobacco TS SINE retroelements did not mediate resistance to a virus carrying TS SINE sequences, suggesting that, unlike the 21- to 22-nt siRNAs, long siRNAs do not participate in PTGS (17). In addition, mutants that affect RNA silencing were used to show a correlation of long siRNAs with DNA methylation. In particular, mutants inSDE1/SGS2 (an RNA-dependent RNA polymerase),SDE3 (an RNA helicase), and SGS3 (a novel gene) did not suppress the accumulation of long siRNAs or affect DNA methylation of AtSN1, but the sde4 mutant (not yet cloned) suppressed both long siRNAs and DNA methylation (17). ago4 and sde4 map to different chromosomes and are therefore not allelic (20).

Thus, AGO4 and SDE4 likely encode components of a silencing system that generates long siRNAs specialized for chromatin level gene silencing (Fig. 3B). Presumably, a Dicer-like enzyme (21) and possibly an RNA-dependent RNA polymerase are also involved in siRNA production. Once generated, the long siRNAs guide KYP-dependent histone methylation and CMT3- and DRM-dependent DNA methylation to specific regions of chromatin. The targeting of this system to transposable elements likely contributes to suppression of transposon proliferation and to genome stability.

Note added in proof: The long and short classes of small RNAs have been shown to be bona fide siRNAs and are likely made by distinct Dicer-like enzymes (22).

Supporting Online Material

Materials and Methods

Fig. S1

Table S1

References and Notes

  • * To whom correspondence should be addressed. E-mail: jacobsen{at}


Stay Connected to Science

Navigate This Article