Structural insights into RNA processing by the human RISC-loading complex

Abstract

Targeted gene silencing by RNA interference (RNAi) requires loading of a short guide RNA (small interfering RNA (siRNA) or microRNA (miRNA)) onto an Argonaute protein to form the functional center of an RNA-induced silencing complex (RISC). In humans, Argonaute2 (AGO2) assembles with the guide RNA-generating enzyme Dicer and the RNA-binding protein TRBP to form a RISC-loading complex (RLC), which is necessary for efficient transfer of nascent siRNAs and miRNAs from Dicer to AGO2. Here, using single-particle EM analysis, we show that human Dicer has an L-shaped structure. The RLC Dicer's N-terminal DExH/D domain, located in a short 'base branch', interacts with TRBP, whereas its C-terminal catalytic domains in the main body are proximal to AGO2. A model generated by docking the available atomic structures of Dicer and Argonaute homologs into the RLC reconstruction suggests a mechanism for siRNA transfer from Dicer to AGO2.

Figure 1. Architecture of human Dicer

(a) Three-dimensional (3D) reconstruction of human Dicer shown in different orientations. The main structural elements are labeled. The channel in the middle of the volume is indicated by dashed lines. (b) The most biologically relevant of the top docking results of the atomic model of Giardia Dicer (color coded as in Supplementary Fig. 2a) and the human DEAD-box RNA helicase DDX3X (PDB 2i4i) (red ribbon). (c) The synthetic 3D model (gray solid surface) generated by removing the density corresponding to the base branch (blue wire) in the human Dicer reconstruction is used to match the reference-free two-dimensional (2D) class averages of the ΔDExH/D Dicer mutant sample. In each of the 10 panels, the first rows are reprojections (left) and the corresponding reference-free class averages (right) of the intact human Dicer; the second rows are reprojections from the synthetic Δbase-branch model (left) and the corresponding reference-free class averages (right) of the ΔDExH/D mutant (second row); the third rows are the difference maps of the 2D class averages between intact human Dicer and the mutant at 3σ threshold (right) and the superimposition of them (in red) to the corresponding projections of the synthetic model (left). Scale bars in all figures are 5 nm.

Figure 2. Architecture of the human RISC-loading complex (RLC)

(a) Comparison of the cross-linked RLC reference-free class averages (odd rows) with corresponding projection views of the 3D map of human Dicer (even rows). The difference maps of the 2D class averages between the cross-linked RLC and the apo-Dicer at 3σ level (in red) are superimposed onto the projection views. The class averages are classified into three distinct categories as discussed in the text. For categorization purposes, only the front L-shape views are shown here. (b) Comparison of the cross-linked Dicer-Ago2 complex reference-free class averages (top row) with corresponding projection views of the 3D map of human Dicer. The difference maps between the cross-linked Dicer-Ago2 complex and apo-Dicer class averages at 3σ threshold (red) are superimposed onto the projection views. Shown here are the three clearest class averages out of a total of 100. (c) Maximum-likelihood heterogeneity analysis of the uncross-linked RLC and Dicer-Ago2 samples. From about 4000 particles of uncross-linked RLC, we generated four subclasses and calculated their 3D reconstructions (upper panel). Among the four models, Model I has an obvious additional density between the platform and the base branch, while Model IV doesn't appear to have additional densities besides the apo-Dicer. About 2000 particle images of uncross-linked Dicer-Ago2 complex underwent similar analysis to generate two subclasses, each having its 3D model reconstructed (lower panel). Both models lack the additional density that exists in the RLC. The scale bars in all panels are 5 nm.

Figure 3. Reconstruction of GraFix-prepared human RLC

(a) 3D reconstruction of RLC shown in different orientations. (b) Docking of the atomic model of Thermus thermophilus Argonaute in the major part of the difference map (yellow transparent map) calculated between the 3D reconstructions of RLC shown in (a) and that of Dicer alone (shown as wire map). Argonaute domain code: cyan, N-terminal domain; orange, PAZ domain; pink, Mid domain; blue, PIWI domain. The docking is shown in four different views along the vertical axis. The front part of the RLC in the fourth view was removed to show the Ago docking more clearly. The scale bars represent 5 nm.

Figure 4. Proposed working model of the human RLC

The 3D density map of RLC is shown as a semi-transparent iso-surface. The atomic model of the DExH/D domain (red ribbon), the Giardia Dicer atomic model (gray-yellow-green-orange ribbon, color coding the same as in Figure 1b), and the Thermus thermophilus Argonaute (gray-cyan-orange-pink-blue, color coding the same as in Figure 3b) are docked in the density map. TRBP is illustrated as a string of three yellow spheres with a flexible linker connecting it to the DExH/D domain. Its motion range, based on our experimental results, is marked by the dashed yellow arrows. In (a), an atomic model of the siRNA (paired spirals with the guide strand in purple color and the passenger strand in yellow color) is aligned vertically between the Dicer's RNase III domain flat surface and Ago2's PAZ domain. This panel illustrates the state proposed in our model for the dicing of dsRNA by Dicer. In this state the PAZ domain of Ago2 could engage the newly diced end of the siRNA, as illustrated by the red arrow. In (b), the distance between the PAZ domains of Ago2 and Dicer allows a perfect accommodation of the 22 nt siRNA between them. This state of Ago2 could be stabilized by interactions with TRBP. Thus, this panel illustrates the hypothetical state after transfer of the newly diced siRNA's onto Ago2's PAZ domain while the other end remains bound to Dicer's PAZ domain. The flexible TRBP could help the transfer efficiency and correctness.