Revisiting RNA-directed DNA methylation

Abstract

RNA-directed DNA methylation (RdDM) involves sequence-specific guiding of the de novo methylation machinery to complementary genomic DNA by RNA molecules. It is still elusive whether guide RNAs bind directly to DNA or to nascent transcripts produced from it. Even the nature of the guide RNAs is not elucidated. RNA interference (RNAi) studies provided a link between RNAi and RdDM indicating that small interfering RNAs (siRNAs) trigger and guide cytosine methylation. The “siRNA hypothesis” is generally accepted. However, recent data demonstrated that RdDM is not always associated with the accumulation of corresponding siRNAs. RdDM triggers may differ from guide RNAs and further studies are needed to clarify if guide RNAs are small or long RNAs, if they are single or double stranded and if they target DNA or nascent transcript.

In plants, de novo DNA methylation was initially suggested to be triggered by pairing of homologous DNA.¹ This assumption was challenged when it was shown that replicating Potato spindle tuber viroid (PSTVd) efficiently triggered RdDM of homologous PSTVd cDNA transgenes in tobacco.² Since PSTVd is replicating via a RNA rolling circle mechanism without involvement of DNA intermediates, it was evident that de novo DNA methylation was actually RNA-directed. In the presence of RNA molecules guiding the RdDM machinery, RdDM entails the DOMAINS REARRANGED METYLTRANSFERHASE (DRM)-mediated methylation of cytosines (Cs) in asymmetric (CHH), quasi symmetric (CHG) and symmetric (CG) sequence contexts (H = A, C, T).³ In the absence of RdDM, asymmetric methylation is lost whereas CG and at some loci CHG methylation can be efficiently maintained by the MAINTENANCE METHYLTRANSFERASE1 (MET1) and CHROMOMETHYLTRANSFERASE3 (CMT3), respectively.^3,4

In addition to viroids, plant viruses and satellite RNA also induced RdDM of homologous genomic sequences.^5,6 Common to all of these RdDM inducers is their replication through dsRNA intermediates. In plants, the potency of dsRNA to trigger RdDM and transcriptional gene silencing (TGS) of transgene promoters was demonstrated by Matzke and coworkers.^7,8 Still, it was not clear whether dsRNA directly interacted with DNA or whether further intermediate processing steps were required. Jacobsen and coworkers provided a link between RNAi and RdDM, showing that processing of dsRNA by DICER-LIKE3 (DCL3) into 24-nt siRNAs and loading of these RNAs onto ARGONAUTE4 (AGO4) is required for RdDM.^9,10 Based on these findings, it was proposed that 24-nt siRNAs are recruiting DRM to homologous DNA to induce de novo cytosine methylation. How recruitment takes place is not fully understood but significant insights from Pikaard and coworkers indicated that DNA-DEPENDENT RNA POLYMERASE V (Pol V) is deeply involved in the process. AGO4 may directly interact with SUPPRESSOR OF TY INSERTION 5 (STP5) and with Pol V. Significantly, it was recently suggested that 24-nt siRNAs recognize nascent Pol V transcripts.^11-14

Although appearing convenient, the 24-nt siRNA-driven RdDM model fails to account for some observations. (1) When a Cauliflower Mosaic Virus (CaMV) 35S promoter-driven transgene was introduced into an Arabidopsis dcl3-knockout mutant, it was transcriptionally silenced, despite the fact that in such a mutant, 24-nt siRNAs are not produced.¹⁵ Importantly, TGS was associated with de novo methylation of the 35S promoter. (2) Recent reports suggested that 24-nt siRNAs move from Arabidopsis scions to rootstocks where they induce de novo methylation of target sequences.^16,17 In one grafting experiment, an Arabidopsis dcl3 mutant harboring the silencer (S) and the target (T) transgenes, was used.¹⁶ Although 24-nt siRNAs were not produced neither in the scion nor in the rootstock both containing S and T, the target sequence of the rootstock was significantly de novo methylated. (3) Arabidopsis methylome sequencing (methylC-seq) showed that 63% of methylated regions were not associated with siRNA clusters.¹⁸ It should be noted however that this study did not discriminate between symmetric and asymmetric methylation. Thus, methylation could not be assigned to the RdDM and the maintenance mechanisms. (4) Bisulfite sequencing data of an Arabidopsis dcl3 mutant showed that CHG and CHH methylation of AtSN1 (disperse repeat) was only moderately decreased. At the IR-71 (inverted repeat-71), another endogenous locus, CHH methylation was virtually not impaired.¹⁹ Obviously due to DCLs redundancy, other size classes of sRNAs could be produced in the absence of DCL3. However, 21–22-nt siRNAs are relatively ineffective for RdDM.^10,20

Notably, while on one hand, RdDM seems to occur in the absence of 24-nt siRNAs, on the other hand, at various loci, 24-nt siRNAs fail to trigger RdDM. Thus, (5) in Arabidopsis, methylC-seq combined with small RNA (sRNA) transcriptome sequencing (smRNA-seq) revealed 24-nt siRNAs mapping to more than one locus. However, some of these loci were methylated and some were not²¹ indicating that the presence of 24-nt siRNAs per se is not consistently inducing RdDM. Initiation of RdDM may require defined target locus features, such as Pol V accessibility. Pol V transcripts may act as a scaffold for 24-nt siRNA recognition.¹⁴ Yet, Pol V seems to be dispensable for RdDM.²² (6) MethylC-seq, smRNA-seq and Pol V chromatin immunoprecipitation sequencing (Pol V ChIP-seq) showed that the number of asymmetrically methylated Cs was the same in wild-type and pol v Arabidopsis mutants. However, compared with wild-type plants, asymmetric methylation was found in other chromosomal sites in pol v mutants. At 35% of the Pol V peaks, neither CHH methylation nor 24-nt siRNAs were detected. At 8% of the peaks, 24-nt siRNAs were generated but no overlap with CHH methylation was found.²²

In Arabidopsis, an RdDM pathway was identified that was independent of RNA-DIRECTED RNA POLYMERASE2 (RDR2) and 24-nt siRNAs but was dependent on RDR6.²³ Also, Potato virus X- and Plum pox virus-induced gene silencing and RdDM were compromised in RDR6-defective Nicotiana benthamiana.²⁴ These data showed that virus-derived dsRNA molecules can move from the cytoplasm into the nucleus and that production of RNA molecules that trigger RdDM may require RDR6 activity. Interestingly, the Arabidopsis IDN2, a SGS3-like double-stranded 5′-overhang RNA-binding protein was essential for RdDM.²⁵ One may thus speculate that longer than 24-nt RNA molecules guide the RdDM machinery. If so, what would be the minimal size requirement for the guide RNA? A Cucumber mosaic virus (CMV) vector carrying recombinant fragments with sizes smaller than 90 bp produced 24-nt siRNAs but failed to initiate RdDM and TGS of a homologous transgene in N. benthamiana.²⁶ Experimental evidence supporting the role of a longer than 24-nt RNA as the guide RdDM molecule emerged from work with intronic hairpin RNAs (int-hpRNAs).²⁷ Upon transcription, a classical hairpin (promoter-IR-polyadenylation signal sequence) becomes polyadenylated and transported to the cytoplasm. In order to retain an hpRNA in the nucleus, an hpRNA cDNA was inserted into an artificial intron located in a GFP transgene. After transcription and splicing, the GFP mRNA was exported into the cytoplasm, while the int-hpRNA appeared to be retained in the nucleus. In transgenic Nicotiana tabacum lines, the int-hpRNA was not processed into detectable amounts of siRNAs (due to the low sensitivity of the detection assay, their presence could not be fully excluded) but it extremely efficiently triggered cis and trans RdDM. By contrast, PTGS of a sensor construct was not initiated. The target region was de novo methylated to almost 100%. However, virtually no methylation was found in the regions directly flanking the region that shared homology with the int-hpRNA.²⁷ This implied the presence of a kind of “molecular ruler,” which would enable the measurement of regions to be methylated. Twenty-four-nt siRNAs can hardly act as such rulers, since they likely do not cover the full length of the target region and are not homogenously distributed along the target sequence.²⁸ Deep sequencing data revealed that thymine (T) is highly overrepresented at position -1 of 24-nt siRNA precursors, indicating that DCL3 processing occurs at preferential sites. More important, adenine (A) is the most common 5′ base of AGO4-bound 24-nt siRNAs.²¹ Molecules targeting a genomic DNA for RdDM need to precisely cover their targets at the nucleotide level. Thus, it is reasonable to assume that guide RNAs are not 24-nt siRNAs and are probably longer than 24-nt. Whether they are single or double stranded needs to be clarified. The fact that both DNA strands are uniformly methylated at asymmetrical sites suggests a double-stranded nature.²⁹

Likewise, RNAi, RdDM is triggered by dsRNA. However, how dsRNA mediates RdDM is still elusive. Long dsRNA molecules may be deeply involved in guiding the RdDM machinery to its target (DNA or nascent RNA). Twenty-four-nt siRNAs seem to have a case-dependent involvement in RdDM, but in contrast to the common view, they may be not involved in the guiding process. Twenty-four-nt siRNAs are predominantly located in the cytoplasm, where they could be involved in an intermediate cytoplasmic step of RdDM.³⁰ One may hypothesize that a high concentration of dsRNA in the nucleus, as is the case when using int-hpRNA constructs, mediates RdDM even in the absence of 24-nt siRNAs. In this view, it should be noted that a non-polyadenylated IR efficiently triggered trans RdDM of a transgenic promoter in an Arabidopsis ago4 mutant.³¹ Non-polyadenylated dsRNA may not be efficiently transported into the cytoplasm and would be retained in the nucleus. Thus, RdDM could be activated without the need of AGO4-loaded 24-nt siRNAs. If dsRNA insufficiently accumulates in the nucleus, e.g., in the case of conventional hpRNA constructs, inefficient RdDM may take place. During this step, Pol IV may transcribe partially methylated DNA at the region defined by dsRNA/DNA interaction or the dsRNA itself.³² Pol IV, but conceivably also scaffold transcripts produced by Pol II or Pol V,³³ may then be cleaved by AGO4-bound 24-nt siRNAs³⁴ and/or AGO1-bound 21-nt siRNAs. The cleavage products may finally serve as templates for RDR2³⁵ and/or RDR6²³ leading to production of additional dsRNA molecules (self-reinforcing loop), which would guide the RdDM machinery to amplify de novo methylation. It should be noted that numerous experimental data indicated the involvement of 24-nt siRNA in guiding the RdDM machinery. However, the selected exceptions from the “rule” discussed here clearly show that further experiments are needed to ascertain the actual nature of the guiding RNA molecule and the mechanistic details of its interaction with the target.