RNA-directed DNA methylation requires stepwise binding of silencing factors to long non-coding RNA

Abstract

Ribonucleic acid-mediated transcriptional gene silencing (known as RNA-directed DNA methylation, or RdDM, in Arabidopsis thaliana) is important for influencing gene expression and the inhibition of transposons by the deposition of repressive chromatin marks such as histone modifications and DNA methylation. A key event in de novo methylation of DNA by RdDM is the production of long non-coding RNA (lncRNA) by RNA polymerase V (Pol V). Little is known about the events that connect Pol V transcription to the establishment of repressive chromatin modifications. Using RNA immunoprecipitation, we elucidated the order of events downstream of lncRNA production and discovered interdependency between lncRNA-associated proteins. We found that the effector protein ARGONAUTE4 (AGO4) binds lncRNA independent of the RNA-binding protein INVOLVED IN DE NOVO2 (IDN2). In contrast, IDN2 binds lncRNA in an AGO4-dependent manner. We further found that the de novo DNA methyltransferase DOMAINS REARRANGED METHYLTRANSFERASE2 (DRM2) also associates with lncRNA produced by Pol V and that this event depends on AGO4 and IDN2. We propose a model where the silencing proteins AGO4, IDN2 and DRM2 bind to lncRNA in a stepwise manner, resulting in DNA methylation of RdDM target loci.

Keywords: Arabidopsis thaliana; Argonaute; DNA methyltransferase; INVOLVED IN DE NOVO2; non-coding RNA.

Figure 1

A subset of AGO4-bound RNA-directed DNA methylation (RdDM) targets requires IDN2. (a) Overlaps of AGO4-bound regions of differential CHH methylation (DMRs) between Col-0 and specified mutants. Most nrpe1 DMRs (beige) also have reduced CHH methylation in drm1/drm2 (blue). These overlap with idn2 (purple) DMRs. Total DMRs are 558 nrpe1, 573 drm1/drm2 or 216 idn2 out of 820 AGO4-bound loci. (b) Comparison of methylation levels on AGO4-bound nrpe1 DMRs. Most nrpe1 DMRs have reduced CHH methylation in rdr2, ago4 and drm1/drm2. A subset of nrpe1 DMRs has reduction in idn2 or in the idn2/idnl1/idnl2 triple mutant, but most have no or intermediate changes. nrpe1 DMRs were defined as having at least a four-fold reduction in CHH methylation. All values are represented as a ratio to Col-0. (c) Classification of AGO4-bound nrpe1 DMRs. Any DMRs overlapping promoters of protein-coding genes, gene bodies, transposons, centromeric regions or intergenic regions were counted. Total AGO4-bound nrpe1 DMRs were plotted as a baseline (black). nrpe1 DMRs were then divided into regions that have losses in CHH methylation in idn2/idnl1/idnl2 (dark grey) and regions that have no change (grey). nrpe1 DMRs with CHH methylation losses in drm1/drm2 were also plotted (light grey). Bars represent the percentage of each category that overlapped a particular genomic feature. *P < 0.05, **P < 0.001 (chi-square test). (d) Production of small interfering RNA (siRNA) on IDN2/IDNL1/IDNL2-dependent regions requires Pol IV, Pol V and AGO4. Total nrpe1 DMRs were classified as either idn2/idnl1/idnl2 changed (blue, meCHH ≤25% of Col-0) or unchanged (green, meCHH ≥75% of Col-0). The 24 nt siRNAs in either nrpd1, nrpe1 or ago4 overlapping each category were counted and plotted as a ratio to Col-0 with reads per million normalization.

Figure 2

Binding of AGO4 to long non-coding RNA (lncRNA) does not require IDN2. (a)–(f) RNA immunoprecipitation (RIP) experiments using an α-AGO4 antibody and nuclei isolated from wild type and mutant seedlings. Mean values and standard deviations of at least two biological repetitions are shown. Experiments were performed on the following loci: IGN29 (b), IGN34 (c), IGN35 (d), and IGN36 (e). Pol II-transcribed ACTIN2 served as a control for equal total RNA levels (a) and for potential contamination with genomic DNA (f). (g–l) Quantitative reverse-transcriptase polymerase chain reaction (qRT-PCR) with total RNA isolated from seedlings. Mean values and standard deviations of three biological repetitions are shown. Four Pol V-transcribed loci with reduced DNA methylation in idn2 were analysed: IGN29 (h), IGN34 (i), IGN35 (j), and IGN36 (k). Pol II-transcribed ACTIN2 served as a control for equal total RNA levels (g) and potential contamination with genomic DNA (l).

Figure 3

Binding of long non-coding RNA (lncRNA) by IDN2 depends on small interfering RNA (siRNA) and AGO4. (a–f) RNA immunoprecipitation (RIP) experiments with an α-IDN2 antibody and nuclei isolated from wild type and RNA-directed DNA methylation (RdDM) mutant seedlings. Mean values and standard deviations of at least three biological replicates are shown. Experiments were performed on the following loci: IGN29 (b), IGN34 (c), IGN35 (d), and IGN36 (e). Pol II-transcribed ACTIN2 served as a control for equal total RNA levels (a) and for potential contamination with genomic DNA (f). (g) Western blot with an α-IDN2 antibody and proteins isolated from wild type (wt) and RdDM mutant seedlings. Ponceau staining of the membrane shows comparable loading of protein extracts. The experiment was performed twice and one representative biological replicate is depicted.

Figure 4

Association of DRM2 with Pol V transcripts requires small interfering RNA (siRNA), AGO4 and IDN2. (a–f) RNA immunoprecipitation (RIP) experiments with an α-DRM2 antibody and nuclei isolated from wild type and RNA-directed DNA methylation (RdDM) mutant seedlings. Mean values and standard deviations of at least two biological replicates are shown. Experiments were performed on the following loci: IGN29 (b), IGN34 (c), IGN35 (d), and IGN36 (e). Pol II-transcribed ACTIN2 served as a control for equal total RNA levels (a) and for potential contamination with genomic DNA (f). (g) Western blot with an α-DRM2 antibody and proteins isolated from wild type (wt) and RdDM mutant seedlings. Ponceau staining of the membrane shows comparable loading of protein extracts. The experiment was performed twice and one representative biological replicate is depicted.

Figure 5

Pol V transcripts, small interfering RNA (siRNA), AGO4, IDN2 and DRM2 are necessary for CHH methylation at IDN2-dependent loci. (a) Chop-PCR with genomic DNA isolated from wild type and RNA-directed DNA methylation (RdDM) mutant seedlings. The DNA was digested with methylation-sensitive restriction enzymes and amplified by PCR. Three Pol V-transcribed loci with reduced DNA methylation levels in idn2 were analysed in addition to a control locus (At5G58510) that has been previously shown to have Pol V- and AGO4-dependent CHH methylation (Zheng et al., 2013). Regions lacking a restriction site were used as a loading control. The experiment was performed three times and one representative biological replicate is shown. (b) Methylation status of AGO4-bound nrpe1 differentially methylated regions (DMRs) as classified by the idn2/idnl1/idnl2 triple mutant. Those DMRs with a dependency on idn2/idnl1/idnl2 were selected and methylation levels in rdr2, ago4, idn2 and drm1/drm2 relative to the wild type were plotted. Dependency was defined as at least a four-fold reduction in CHH methylation compared with Col-0.

Figure 6

Model of stepwise association of silencing proteins with long non-coding RNA (lncRNA) produced by Pol V to mediate CHH methylation. At IDN2/IDNL1/IDNL2-dependent loci, AGO4 interacts with lncRNA produced by Pol V in a manner at least partially dependent on small interfering RNA (siRNA). In turn, IDN2 binds to lncRNA possibly recognizing a 5′-overhang of a double-stranded RNA consisting of siRNA and lncRNA. Finally, DRM2 directly or indirectly associates with the Pol V transcript in a way dependent on IDN2 and AGO4, and the DNA at the locus is methylated.