The mechanisms of siRNA selection by plant Argonaute proteins triggering DNA methylation

Abstract

The model plant Arabidopsis thaliana encodes as many as ten Argonaute proteins (AGO1-10) with different functions. Each AGO selectively loads a set of small RNAs by recognizing their length and 5' nucleotide identity to properly regulate target genes. Previous studies showed that AGO4 and AGO6, key factors in DNA methylation, incorporate 24-nt small-interfering RNAs with 5' adenine (24A siRNAs). However, it has been unclear how these AGOs specifically load 24A siRNAs. Here, we biochemically investigated the siRNA preference of AGO4, AGO6 and their chimeric mutants. We found that AGO4 and AGO6 use distinct mechanisms to preferentially load 24A siRNAs. Moreover, we showed that the 5' A specificity of AGO4 and AGO6 is not determined by the previously known nucleotide specificity loop in the MID domain but rather by the coordination of the MID and PIWI domains. These findings advance our mechanistic understanding of how small RNAs are accurately sorted into different AGO proteins in plants.

Figure 1.

P4RNAs with 5′ A and/or unstable ends are successfully converted into mature siRNAs. (A) Schematic of 24-nt heterochromatic-siRNA biogenesis in Arabidopsis. P4RNA, a 30–40 nt single stranded RNA is first transcribed by Pol IV. Subsequently, a complementary strand is synthesized by RDR2, forming a long double-stranded RNA (dsRNA) with an extra nucleotide at the 3′ end of RDR2 strand. The dsRNA is subsequently diced by DCL3 into a 24-nt siRNA duplex, which is loaded into AGO4 or AGO6 to form RISC. The passenger strand is cleaved and ejected by AGO, while the guide strand remains in mature RISC. (B) P4RNAs are classified into four bins according to conversion efficiency (CE), from lowest bin 1 to highest bin 4. P4RNAs with higher CE produce more siRNAs. (C) An example scatter plot of AGO4-bound siRNAs and P4RNAs. Each spot represents one P4RNA. All P4RNAs were grouped into four bins with CEs ranging from low (bin 1) to high (bin 4). (D, E) First nucleotide frequency of AGO4- (D) and AGO6- (E) bound small RNAs. Number of P4RNAs in each bin is shown on the right side. P4RNAs with higher CE tend to have a stronger 5′ A bias in both AGO4 and AGO6. See also Supplementary Table 3. (F, G) Second nucleotide frequency of AGO4- (F) and AGO6- (G) bound 5′ A small RNAs. Number of P4RNAs in each bin is shown on the right side. P4RNAs with higher CE tend to have an A/U nucleotide compared to a C/G nucleotide at the g2 position in both AGO4 and AGO6. See also Supplementary Table 3.

Figure 2.

AGO4 and AGO6 actively select 24A siRNAs via distinct mechanisms. (A) Schematic of RISC assembly in vitro. To form RISC, 3 × FLAG-tagged AGO mRNAs were translated in tobacco cell lysate, then incubated with radiolabeled small RNA duplexes containing specific 5′ nucleotides. Co-immunoprecipitation was performed using anti-FLAG coated beads to purify AGO4/6-bound siRNAs. The radioactivity was quantified after separation by a denaturing UREA PAGE. (B) AGO4/6–RISC assembly using siRNA duplexes bearing different 5′ nucleotide on the guide strand. AGO4 and AGO6 actively select 24A siRNAs. (Top) Schematic for siRNA duplexes used in this study. The guide strand was 5′ radiolabeled. The passenger strand has an amino C6 linker at the 5′ end that prevents the passenger strand from being loaded onto AGO4/6. (Middle) Small RNAs loaded into AGO4 (upper panel) and western blotting of AGO4 proteins expressed in the tobacco BY-2 lysate after immunoprecipitation with anti-FLAG antibody (lower panel). (Bottom) Small RNAs loaded into AGO6 and western blotting of AGO6 proteins expressed in the tobacco BY-2 lysate after immunoprecipitation with anti-FLAG antibody. (C, E) Quantification of loaded siRNAs in (B) and (D), respectively. The band intensity of siRNAs was normalized to the value of 5′ A. The graphs show the mean ± SD from three technically independent experiments. Loading efficiency of each experiment of AGO4 and AGO6 was shown as gray and black dots. See also Supplementary Table 3. (D) 5′ nucleotide preference of AGO4 and AGO6 when thermodynamic stability and nucleotide identity are separated. AGO4 accepted all four siRNAs with a slight preference for the siRNAs with purine nucleotides, whereas siRNAs with 5′ A were still predominantly incorporated into AGO6. (Top) 5-nitroindole (i) was introduced at the nucleotide position of the passenger strand facing the 5′ terminal nucleotide of the guide strand. See also (B).

Figure 3.

AGO6 forms RISC more efficiently with shorter siRNAs than AGO4. (A) The 19–25-nt siRNA duplexes used in this study. The 5′ end of the guide strand was radiolabeled with ³²P. (B) In vitro AGO4/6-RISC assembly with 21- and 24-nt siRNAs. AGO4 preferentially bound 24-nt siRNAs compared to 21-nt siRNAs, whereas AGO6 was able to load both. (C) In vitro AGO4/6-RISC assembly using a mixture of 19–25-nt siRNA duplexes. Although both AGO4 and AGO6 preferentially bound to long (23–25 nt) siRNAs, AGO6 also bound short (21–22 nt) siRNAs compared to AGO4. (D) Quantification of loaded siRNAs in (C). The relative band intensity of each length of siRNA was calculated with the total band intensity as 1. The graphs show the mean ± SD from three technically independent experiments (AGO4, black dots; AGO6, gray dots). Benjamini–Hochberg procedure (false discovery rate approach)-corrected P values from multiple paired t tests are as follows: *P = 0.029415, **P = 0.029415, ***P = 0.006793, ****P = 0.048948. ns, not significant. See also Supplementary Table 3. (E) AGO6-RISC assembly using 21-nt and 24-nt siRNA duplexes bearing different 5′ nucleotide on the guide strand. The guide strand was 5′ radiolabeled. The passenger strand has an amino C6 linker at the 5′ end that prevents the passenger strand from being loaded onto AGO6. AGO6 strictly selected siRNAs with 5′ A, regardless of the siRNA length. (F) Quantification of loaded siRNAs in (E). The band intensity of siRNAs was normalized to the value of 5′ A. The graphs show the mean ± SD from three technically independent experiments. See also Supplementary Table 3.

Figure 4.

The nucleotide specificity loop in the MID domain does not determine the 5′-nucleotide preference of AGO4 and AGO6. (A) Schematic domain architecture of Argonaute protein. The 5′ end of small RNA interacts with the 5′ end binding pocket formed by the MID and PIWI domains. The nucleotide specificity loop is in the MID domain. The sequences of nucleotide specificity loops of AGO1, AGO4 and AGO6 are shown inside the orange frame. (B) Schematic of chimeric AGO proteins whereby the nucleotide-specific loops of AGO4 and AGO6 were swapped. (C) Schematic for siRNA duplexes used in this study. The guide strand was 5′ radiolabeled. The passenger strand has an amino C6 linker at the 5′ end that prevents the passenger strand from being loaded onto AGO. 5-Nitroindole (i) was introduced at the nucleotide position of the passenger strand facing the 5′ terminal nucleotide of the guide strand. (D) In vitro RISC assembly with AGO4/6 chimeric proteins. AGO4L6, like AGO4, accepted all four siRNAs with a slight 5′ purine preference. AGO6L4 predominantly showed a 5′ A preference as in AGO6. (E) Quantification of loaded siRNAs in (D). The band intensity of siRNAs was normalized to the value of 5′ A. The graphs show the mean ± SD from five technically independent experiments. Note that the experiments with chimeric AGO4/6s in Figures 4E, 5C and E were performed simultaneously. For the experimental controls, wild-type AGO4 and AGO6, the samples were electrophoresed in three different gels corresponding to Figures 4E, 5C and E, resulting in similar quantitative results in those Figures. See also Supplementary Table 3.

Figure 5.

Both MID and PIWI domains are involved in 5′ nucleotide selection. (A) Schematic diagram of chimeric proteins with MID or PIWI domains swapped between AGO4 and AGO6. (B) In vitro RISC assembly with AGO4/6 chimeric proteins. AGO4M6 showed a predominant preference for 5′ A and 5′ G. AGO6M4 showed a predominant 5′ A preference as in AGO6. (C) Quantification of loaded siRNAs in (B). The band intensity of siRNAs was normalized to the value of 5′ A. The graphs show the mean ± SD from five technically independent experiments. See also Figure 4E and Supplementary Table 3. (D) In vitro RISC assembly with a AGO4/6 chimeric protein. AGO4P6 showed a predominant 5′ A preference as in AGO6. (E) Quantification of loaded siRNAs in (D). The band intensity of siRNAs was normalized to the value of 5′ A. The graphs show the mean ± SD from five technically independent experiments. See also Figure 4E and Supplementary Table 3.

Figure 6.

AGO4 and AGO6 have distinct small RNA selection mechanisms. (A) Schematic of distinct 5′ A selection mechanisms of AGO4 and AGO6. AGO6 mainly selects the 5′ A siRNAs by recognizing the nucleotide identity, while AGO4 loads 5′ A sRNAs through a combination of 5′ purine preference and preference for weak 5′ terminal base pair. (B) A model for 21U miRNA rejection by AGO4 and AGO6. Because of the property of AGO4 to bind only 23–25 nt long siRNAs and AGO6 to form RISC with only 5′ A siRNAs, these AGOs are unable to bind 21U miRNAs, avoiding unwanted transcriptional repression of the miRNA locus.