Structural insights into dsRNA processing by Drosophila Dicer-2–Loqs-PD

Small interfering RNAs (siRNAs) are the key components for RNA interference (RNAi), a conserved RNA-silencing mechanism in many eukaryotes1,2. In Drosophila, an RNase III enzyme Dicer-2 (Dcr-2), aided by its cofactor Loquacious-PD (Loqs-PD), has an important role in generating 21 bp siRNA duplexes from long double-stranded RNAs (dsRNAs)3,4. ATP hydrolysis by the helicase domain of Dcr-2 is critical to the successful processing of a long dsRNA into consecutive siRNA duplexes5,6. Here we report the cryo-electron microscopy structures of Dcr-2–Loqs-PD in the apo state and in multiple states in which it is processing a 50 bp dsRNA substrate. The structures elucidated interactions between Dcr-2 and Loqs-PD, and substantial conformational changes of Dcr-2 during a dsRNA-processing cycle. The N-terminal helicase and domain of unknown function 283 (DUF283) domains undergo conformational changes after initial dsRNA binding, forming an ATP-binding pocket and a 5′-phosphate-binding pocket. The overall conformation of Dcr-2–Loqs-PD is relatively rigid during translocating along the dsRNA in the presence of ATP, whereas the interactions between the DUF283 and RIIIDb domains prevent non-specific cleavage during translocation by blocking the access of dsRNA to the RNase active centre. Additional ATP-dependent conformational changes are required to form an active dicing state and precisely cleave the dsRNA into a 21 bp siRNA duplex as confirmed by the structure in the post-dicing state. Collectively, this study revealed the molecular mechanism for the full cycle of ATP-dependent dsRNA processing by Dcr-2–Loqs-PD.


Overall structures
We used single-particle cryo-EM to study the structure and mechanism of Drosophila Dcr-2 in the dsRNA-processing cycle. We purified recombinant wild-type (WT) Dcr-2 and its catalytically inactive mutant Dcr-2(D1217N/D1476N) (hereafter Dcr-2(DDNN)), corresponding to Aquifex aeolicus RNaseIII (AaRNaseIII) D44N 15 , from insect cells, and mixed the Dcr-2 constructs with purified full-length Loqs-PD to form stable WT Dcr-2-Loqs-PD and Dcr-2 DDNN -Loqs-PD complexes, respectively (Extended Data Fig. 1). We also designed and purified a 50 bp palindromic sequence dsRNA with a 3′ two-nucleotide overhang and a 5′ monophosphate terminus as the substrate on the basis of previous studies 11,16 (Fig. 1a). Using a different combination of assembly components and buffers with or without ATP, we solved a series of cryo-EM structures of Dcr-2-Loqs-PD during the dsRNA-processing cycle at resolutions ranging from 3.1 Å to 4.6 Å. These structures include the Dcr-2 DDNN -Loqs-PD complex in the apo state and the dsRNA initial binding state without ATP; the dsRNA early-translocation and mid-translocation states and the active The cryo-EM structure of Dcr-2-Loqs-PD in the apo state at a resolution of 3.3 Å shows that Dcr-2 has the same L-shaped domain organization as human Dicer 9 , but with more structural elements solved (Fig. 1b). Similar to human Dicer, the N-terminal helicase region forms the base of the L-shaped structure and the DUF283 domain is located at the corner of the L and just below the catalytic active core. The previously unmodelled linker between the DUF283 and Platform domains (DP-linker) in human Dicer is now well resolved in Dcr-2 as a structural element that interacts with the RNase IIIa domain through two α-helices and one β-strand. The Platform and PAZ domains surrounding the Connector helix sit at the top of the core processing region. The intrinsic RNase III dimer comprising the RIIIDa and RIIIDb domains is located in the centre of the molecule with the insertion domains (RIIIDai and RIIIDbi) sitting in symmetric positions (Extended Data Figs. 4 and 5a-e). The C-terminal dsRBD interacts with the DP-linker and shields the RNase-activity centre (Extended Data Fig. 5e).
Compared with the apo state, the overall structures of Dcr-2-Loqs-PD complex undergo major conformational changes when interacting with the dsRNA substrate, especially on the helicase and DUF283 domains (Extended Data Fig. 6a-e). By contrast, the structures in the active dicing and post-dicing states demonstrate further conformational changes in the PAZ-Platform module. These structures and their mechanistic insights are discussed in more detail in the sections below.

Interaction between Loqs-PD and Dcr-2
In the absence of ATP, we found that Dcr-2-Loqs-PD tends to dimerize and forms an initial binding complex with dsRNA (Fig. 1c). The dimerization interface involves the RIIIDai domain of one Dcr-2 and the Hel2 domain of the other Dcr-2, and mutation of residues in the dimerization interface results in dissociation of dimer ( Fig. 1c and Extended Data Fig. 5f-k). The dimerization interface also disrupts the ATP-binding site of the helicase domain (Fig. 2a) that otherwise is well maintained to bind to ADP/ATP in the translocation and dicing states (Fig. 2b). The high-resolution (3.1 Å) cryo-EM structure of the initial binding complex is good enough for us to unambiguously identify the density of the C-terminal tail (amino acids 344-359) of Loqs-PD interacting with thehelicase domain of Dcr-2 ( Fig. 2c and Extended Data Fig. 6f,g). The other portions of the Loqs-PD protein cannot be distinguished in the cryo-EM maps probably due to its flexible nature. The short C-terminal tail extends and wraps over the surface of the Hel2i, Hel2 and Pincer domains (Fig. 2c,d), mainly through hydrophobic interactions and hydrogen bonds ( Fig. 2e-g). The C-terminal end residue of Loqs-PD, Ile359, is accommodated by the hydrophobic pocket comprised by Met222 from Hel2, Ile518 and Tyr519 from Pincer. Furthermore, the C-terminal carboxyl group of Ile359 of Loqs-PD forms a salt bridge interaction with the side chain of Arg522 from Pincer (Fig. 2e). Two hydrophobic residues identified previously 17 , Phe354 and Phe356 of Loqs-PD, bind to the hydrophobic patch on the Hel2 domain, whereas the aromatic ring of Phe356 stacks with the side chain of Arg369 (Fig. 2f). Tyr347 of Loqs-PD inserts into a hydrophobic cage comprising Leu232, Ile293 and Thr364 right at the interface of the Hel2 and Hel2i domains (Fig. 2g). Mutations to any of the three hydrophobic residues (Y347A, F356D, I359D) of Loqs-PD notably reduced its interaction with Dcr-2 as shown in the pull-down assay (Fig. 2h,i), indicating their essential roles in Dcr-2-Loqs-PD interactions.

Conformational changes after dsRNA binding
dsRNA binding to Dcr-2 induces a large conformational change of the helicase and DUF283 domains (Supplementary Videos 1 and 2). The N-terminal helicase domain of Dcr-2 belongs to the superfamily 2 group of helicase proteins, like RIG-I-like receptors (RLRs), and contains four subdomains: Hel1, Hel2i, Hel2 and Pincer 18 . However, in contrast to RLRs, of which the Pincer subdomain links the helicase domain to the protein's C-terminal domain, the Pincer subdomain of Dcr-2 links the helicase domain to the DUF283 domain. After binding to a dsRNA substrate, the helicase domain together with the DUF283 domain transform from a stretched conformation (Fig. 3a) to a closed one (Fig. 3b), wrapping around the RNA helix (Supplementary Video 2). Notably, the DUF283 domain rotates about 150° around Hel1 and directly contacts the dsRNA (Fig. 3c). Meanwhile, the second α-helix of Pincer rotates together with DUF283, forming the traditional V-shaped structure and clamping Hel1 (Fig. 3b,c). These conformational changes reorient the DUF283 domain between Hel1 and Hel2i and tightly interact with the two subdomains (Extended Data Fig. 6h,i).

Recognition of dsRNA by helicase-DUF283
The near-atomic high resolution (3.1 Å) structure of the initial binding state enabled us to examine the interactions between the Article helicase-DUF283 module and the dsRNA with precision and unambiguity at the residue level (Extended Data Fig. 7a). The dsRNA-recognition mode of the Hel1 and Hel2 subdomains is conserved with the RLR (Extended Data Fig. 7). Distinctly, a loop specific for the Dicer Hel2i subdomain forms intimate contacts with the dsRNA substrate, which corresponds to the second helix in Hel2i of RLRs, extending into the major groove of dsRNA duplex ( Fig. 3d and Extended Data Fig. 7d,f). Residues in this loop extensively interact with the backbone phosphates of the major groove that is widened subsequently (Fig. 3d), resulting in the dsRNA duplex bending around 10° towards the DUF283 domain (Fig. 3e). Moreover, the phosphate group of the 5′ strand terminus of the dsRNA is specifically recognized by Hel2i through a salt-bridge interaction between 5′-phosphate and Lys310 of Hel2i (Fig. 3d,g). The electrostatic potential surface suggests that there is a basic and conserved pocket around the 5′-phosphate of dsRNA in the Hel2i domain ( Fig. 3h,i). Furthermore, the DUF283 domain recognizes the minor and major groove of the dsRNA terminal duplex through Gln580 from region 1 and Asn635/Lys642 from region 3 of the dsRBD fold, respectively (Fig. 3f). In summary, the conformational changes of the helicase and DUF283 domains induced by dsRNA binding are coupled to the simultaneous recognitions of the major and minor groove of the dsRNA terminal duplex and the 5′-phosphate by Hel2i and DUF283, determining the initial binding of dsRNAs to the specific terminal duplex feature.

Structural states during translocation
Recent single-molecule analysis demonstrated that Dcr-2 translocates along the dsRNA after its helicase initially binds to the dsRNA terminus in the presence of ATP 19 . To capture the different steps during the translocation of Dcr-2-Loqs-PD along the dsRNA, we used cryo-EM to analyse the sample of Dcr-2 DDNN -Loqs-PD in complex with the dsRNA substrate under conditions with ATP and Mg 2+ . Through single-particle classification, we captured two representative cryo-EM structures of Dcr-2-Loqs-PD in complex with the dsRNA substrate in translocation states (Fig. 1d,e and Extended Data Fig. 3). In the first captured structure, termed the early-translocation state, about 8 bp of duplex threads through the helicase domain towards the catalytic centre of Dcr-2 (Fig. 1d). In comparison to the initial binding state (Fig. 1c)   Dcr-2-Loqs-PD complex in the early-translocation state has almost no conformational changes and no additional interactions with dsRNA, except that the cap and core modules shift about 3 Å away from the dsRNA (Extended Data Fig. 8a,b and Supplementary Video 1). The trajectory of dsRNA in the early-translocation state is same as in the initial binding state (Fig. 4a and Extended Data Fig. 8c). We captured another structure in the mid-translocation state, in which about 17 bp of a dsRNA duplex threads through the helicase domain further towards the catalytic centre of Dcr-2 (Fig. 4b). When superimposed onto structures of the initial binding and early translocation states, the cap-core modules sway back about 2 Å towards the dsRNA (Extended Data Fig. 8a,b and Supplementary Video 1). Importantly, in the mid-translocation state, the C-terminal dsRBD of Dcr-2 appears in the map and interacts with the dsRNA substrate (Extended Data Fig. 8d), resulting in further bending of the axis of the dsRNA about 25°, and the trajectory of the dsRNA is directly towards the PAZ domain ( Fig. 4c and Supplementary Video 1).
Removal of the C-terminal dsRBD of Dcr-2 completely abolished the cleavage ability of the enzyme on the dsRNA substrate and cryo-EM analysis revealed no further bending of the dsRNA during the translocation by the helicase domain (Extended Data Fig. 9), underlining the critical role of C-terminal dsRBD in ATP-dependent dsRNA processing by Dcr-2. During the translocation, the interactions between the DUF283 domain and RIIIDa/b domains (Fig. 4d), including relatively flexible interactions between the DUF283 and RIIIDb/RIIIDbi domains (Fig. 4e), and relatively rigid interactions between the DUF283 domain and RIIIDa through the DP-linker domain ( Fig. 4f and Extended Data Fig. 8f), remain almost the same as in the initial binding state (Extended Data Fig. 8a). This suggests that the overall domain configuration of Dcr-2 is relatively rigid during the translocation process until the dsRNA terminus reaches the Platform-PAZ domains. A comparison of the cryo-EM structures of the early-and midtranslocation states also demonstrates a very similar structure in their helicase domains, of which the ATP-hydrolytic pockets have an EM density for either ATP or ADP-Mg 2+ (Fig. 2b and Extended Data Fig. 8e). The current resolution is yet not high enough to distinguish the nucleotide state or separate the ATP-hydrolytic steps in the complex.

Article
However, the translocation of the dsRNA's terminus threading through the helicase domain towards the RNase active centre of Dcr-2 is clearly ATP driven, as none of the translocation intermediate states were observed in the sample without ATP (only the apo state or initial binding state were observed). It is expected that a deeper translocated state has no further conformational changes until the dsRNA terminus is bound by the Platform-PAZ domains. Thus, on the basis of the structure of mid-translocation state, we modelled the structure of a fully translocated state by threading a further 12 bp dsRNA through the helicase domain compared with the mid-translocation state ( Fig. 4g and Supplementary Video 1), which is similar to the recently reported structures of Arabidopsis thaliana DCL-1 (AtDCL-1) in complex with a primary miRNA substrate 13 (Extended Data Fig. 8g). However, when superimposing the modelled structure of the fully translocated state onto the active dicing structure of AtDCL3 in complex with a dsRNA substrate (Protein Data Bank (PDB): 7VG2) by the RIIIDa/b domains 14 , the interaction between the DUF283 and RIIIDb domains (Fig. 4d,e) has steric clashes with the dsRNA bound to the RNase active centre of RIIIDa/b of AtDCL3 in the active dicing state (Extended Data Fig. 8h), suggesting that this interaction would prevent the dsRNA substrate from accessing the RNase active centre of the RIIIDa/b domains during the translocation of Dcr-2 along dsRNA. This modelled structure is designated the pre-dicing state (Fig. 4g).

The structure of the active dicing state
We wondered how Dcr-2 is activated from the pre-dicing state to allow the dsRNA to reach the active centre of RIIIDa/b. We captured a structure of Dcr-2-Loqs-PD in complex with 50 bp dsRNA in an active dicing state under the same conditions as for the translocation states (Figs. 1f and 5a and Extended Data Fig. 3), in which the terminus of dsRNA was bound by the Platform-PAZ domains through recognition of the 3′ two-nucleotide overhang and 5′-phosphate, respectively (Extended Data Fig. 8i). The dsRNA is held in the RNase active centre formed by RIIIDa and RIIIDb, and two metal ions are bound in the two catalytic sites (Extended Data Fig. 8j), suggesting that the structure is in the active dicing state. As the Dcr-2 used here is a double mutant that has no dicing activity, the complex halts in a stable state, of which the structure was captured at a resolution (4 Å)-higher than the resolution of the structure in the mid-translocation state (4.2 Å). The C-terminal dsRBD in the active dicing state is also better defined for its interactions with the minor and major grooves of dsRNA mainly by residues of Arg1658, Lys1706 and Lys1702 (Extended Data Fig. 8k,l). The axis of dsRNA bound by the Dcr-2-Loqs-PD complex in the active dicing state is obviously distorted due to the interactions of dsRBD and RIIIDa/b (Fig. 5b). Moreover, compared with the structure in the pre-dicing state, the dsRNA in the active dicing state further threads about 4 bp through the helicase domain (Fig. 5a,c), resulting in the cap and core modules transforming about 10 Å away from the bottom module and rotating 40° (Fig. 5a,c and Supplementary Video 1). The keys for the conformational changes are the disruption of interactions between the DUF283 and RIIIDb domains (Fig. 5d), and the stretch of DP-linker in which a helix unwound about one turn ( Fig. 5e and Supplementary Video 1), eliminating the misalignment between the bottom and capcore modules of Dcr-2 and allowing the dsRNA duplex to shift about 10 Å (Fig. 5f) and position exactly in the catalytic active centre of the RIIID domains (Extended Data Fig. 8j). These conformational changes suggest that the ATP hydrolysis by the helicase domain is converted into a tension accumulated during the translocation of Dcr-2 on dsRNA, driving the transformation of the pre-dicing state to the active dicing state.

siRNA releasing in post-dicing state
To investigate the dynamic dicing process of Dcr-2, we mixed WT Dcr-2-Loqs-PD with a 50 bp dsRNA substrate, then added ATP and Mg 2+ just before generating the cryo-EM samples. From such an enzymatically active sample, we were able to capture a complex structure in the post-dicing state at a relatively lower resolution (4.6 Å) (Figs. 1g and 5g). In this structure, a clear breakage of the dsRNA after the dicing near to the catalytic centre was observed (Fig. 5g) and exactly 21 bp away from the PAZ-domain-binding terminus (Extended Data Fig. 8m). The cleavage of dsRNA disrupts the binding site for the C-terminal dsRBD in the active dicing state (Fig. 5b and Extended Data Fig. 8k-m), and probably results in the dsRBD becoming more flexible and losing its density in the averaged EM map of the post-dicing state (Fig. 5g,h), which may be also favourable to the release of cleaved siRNA products (Extended Data Fig. 8m and Supplementary Video 1). Compared with the structure in the active dicing state, the cap-core module with siRNA in the post-dicing state rotated about 20° (Fig. 5h), which is derived from the partial restoration of the spring helix in the DP-linker, probably accompanied by the release of the tension (Fig. 5i and Supplementary Video 1). The remaining dsRNA duplex bound by the helicase domain also returns to a conformation similar to the early-translocation state (Extended Data Fig. 8n), which enables the translocation of Dcr-2-Loqs-PD to enter the next processing cycle (Fig. 6 and Supplementary Video 1).

Discussion
Dcr-2 is a Dicer protein that requires ATP hydrolysis to process long dsRNA and produce siRNA duplex in vivo 20,21 , although later studies showed that human Dicer does not require ATP to cleave the dsRNA in vitro 22,23 . Loqs-PD is a cofactor protein that comprises two dsRBDs that are responsible for recruiting siRNA precursor substrates for Dcr-2. The interactions between the C-terminal tail of Loqs-PD and Dcr-2 provide a structural basis for the specific requirement of Loqs-PD for endo-siRNA production 3 . Its strong interaction with the Dcr-2 helicase domain lends a proximate convenience for loading siRNA precursors onto the helicase domain for initial binding. The two dsRBDs increase the dsRNA-binding affinity of the helicase, making it easier for the complex to reach the initial binding state (Extended Data Fig. 10a,b). However, the dsRBDs were not observed in the density of the two-dimensional average of the initial binding state, suggesting that the dsRBDs of Loqs-PD have only an assisting role in the initial dsRNA binding of the helicase domain of Dcr-2, which is consistent with the single-molecule result from a previous study 19 .
It is well known that Dcr-2 is an ATP-dependent enzyme. However, we found that Dcr-2 forms a homodimer when binding to dsRNA a substrate in the absence of ATP. Although we cannot conclude whether the dimerization is an in vitro artifact or whether it is physiologically relevant, this finding helped us to solve near-atomic high-resolution structures of Dcr-2-Loqs-PD in the apo and initial binding states under conditions without ATP. Interestingly, the conformational changes of the helicase and DUF283 domains induced by dsRNA binding, which are similar to RLRs, do not require ATP hydrolysis, instead, forming an ATP-binding pocket (Fig. 2a). Although the translocation requires ATP hydrolysis, Dcr-2-Loqs-PD does not undergo major conformational changes. However, in comparison to the modelled pre-dicing state, substantial conformational changes in the dicing state accompanied by an additional 4 bp dsRNA threading through helicase domain (Fig. 5) suggested that the transformation of Dcr-2-Loqs-PD into the active dicing state requires ATP hydrolysis. Thus, both the translocation and activation of Dcr-2-Loqs-PD are ATP-dependent (Fig. 6). Note that the helicase domain is responsible for reducing non-specific cleavage, indicating that the ATP-dependent translocation of dsRNA is the key property of Dcr-2 for controlling the precision of RNA processing (Extended Data Fig. 10c).
The DUF283 domain is a distinct domain that is found only in higher eukaryotic dicer proteins. The bioinformatic 24 and structural 25 studies suggested that DUF283 has a non-canonical dsRBD fold, and functions in protein-protein interactions 25 and dsRNA annealing 26 . Our high-resolution structures of Dcr-2-Loqs-PD in the apo and initial binding states, as well as the translocation and dicing states, revealed that DUF283 has a central role in conformational changes: from the apo to initial binding state and from the pre-dicing to active dicing state. In the transition from the apo to the initial binding state, the DUF283 domain undergoes conformational changes in a manner similar to the C-terminal domain in RLRs (Fig. 3). Moreover, DUF283 interacts with RIIIDb and RIIIDa, blocking the access of dsRNA to the catalytic centre of RIIIDa/b (Extended Data Fig. 8h) and preventing non-specific cleavage of dsRNA during the translocation. In the active dicing state, the interaction between DUF283 and RIIIDb is disrupted (Fig. 5d) and the DP-linker is stretched (Fig. 5e), allowing the dsRNA substrate to enter the active centres of RIIIDa/b for precise cleavage (Fig. 5f).
Recent research suggested that ATP is required for recognition and discrimination of dsRNA termini 12,27 . We have identified a 5′-phosphate-binding pocket in Hel2i in the initial binding state (Fig. 3g-i) under conditions without ATP. There are extensive studies on the 5′-triphosphate recognition by the C-terminal domain of RIG-I that has similar conformational change of helicase after dsRNA binding 28,29 . Dcr-2 may sense the 5′-triphosphate of RNAs using a similar binding pocket to that demonstrated previously 12,27 . Recent single-molecule analysis 19 supports the conclusion from our structures in this study that terminal loading of dsRNA as the initial step and translocation along the dsRNA are required to activate the processive cleavage activity of Dcr-2.

Summary
In conclusion, we used cryo-EM to solve the structures of Dicer-2 in complex with its cofactor Loqs-PD in its apo and initial dsRNA-binding states in the absence of ATP, and in multiple translocation and dicing states in the presence of ATP, deciphering the multiple ATP-dependent conformational changes during the full cycle of dsRNA processing by Dcr-2-Loqs-PD from initial dsRNA binding to dicing of the dsRNA substrate into the siRNA duplex ( Fig. 6 and Supplementary Video 1).

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-04911-x.

Data reporting
No statistical methods were used to predetermine sample size. The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment.

Protein expression and purification
The gene encoding full-length Drosophila melanogaster Dcr-2 (UniProt: A1ZAW0) was cloned from the recombinant pFastBac-Dcr-2 plasmid (gifted by the Q. Liu laboratory). Full-length DmLoqs-PD (UniProt: M9MRT5) was PCR amplified from Drosophila cDNA and cloned into a modified pET28a (with a 6×His-SUMO tag). The constructs of WT Dcr-2, Loqs-PD and other mutations were generated using a standard PCR-based cloning strategy and cloned into the corresponding vectors, and their identities were confirmed by sequencing analysis.
Dcr-2 or its mutants was expressed using the Bac-to-Bac baculovirus expression system (Invitrogen) in sf9 cells at 27 °C. One litre of cells (2 × 10 6 cells per ml, medium from Expression Systems) was infected with 20 ml baculovirus at 27 °C. After growth at 27 °C for 48 h, the cells were collected, resuspended in buffer A (150 mM NaCl, 20 mM Tris-HCl pH 8.0, 10% glycerol, 20 mM imidazole) with 0.5 mM PMSF and protease inhibitors, and lysed by adding 0.5% Triton X-100 and shaken gently for 30 min at 4 °C. Dcr-2 was purified to homogeneity using Ni-NTA affinity, Hitrap Q column (Cytiva), 2nd Ni-NTA affinity and size-exclusion chromatography using the Superdex 200 10/300 Increase column (Cytiva) (in that order).
Loqs-PD and its mutants were expressed in Escherichia coli BL21 (DE3). Loqs-PD was first purified by Ni-NTA affinity chromatography. Using protease ULP1 to remove the 6×His-SUMO tag, and dialysis was applied to remove imidazole. The sample was then applied to a second Ni-NTA chromatography and the flow-through was collected for size-exclusion chromatography using the Superdex 200 16/600 column (Cytiva). Fractions corresponding to the apo Loqs-PD were collected and concentrated to about 10 mg ml −1 .

Preparation of dsRNAs
The dsRNAs were in vitro transcribed using T7 RNA polymerase. The pUC19 plasmids containing target sequences with 3′-HDV ribozyme sequences were linearized by EcoRI, extracted with phenol-chloroform and precipitated with isopropanol. The in vitro transcription reaction was performed at 37 °C for 5 h in the buffer containing 100 mM HEPES-K (pH 7.9), 10 mM MgCl 2 , 10 mM dithiothreitol (DTT), 6 mM NTP each, 2 mM spermidine, 200 μg ml −1 linearized plasmid and 100 μg ml −1 T7 RNA polymerase. For the 5′-monophosphate RNA, 40 mM GMP was added in the transcription reactions. EDTA at a final concentration of 20 mM was added to the samples containing palindromic transcripts. The samples were heated to 95 °C for 5 min and then slowly cooled to room temperature. The annealed transcripts were purified by 8% denaturing urea PAGE, eluted from gel slices and precipitated with isopropanol. After centrifugation, the RNA precipitant was collected, washed twice with 70% ethanol and air-dried, and the RNA was dissolved in ultrapure water. We next used T4 PNK (NEB, M0201) to remove the 2′,3′ cyclic phosphate at the 3′ end of the RNA. The FAM-labelled dsRNA was produced by Silencer siRNA Labeling kit-FAM according to the manufacturer's instructions.

Pull-down assays
Pull-down assays were performed to detect Dcr-2-Loqs-PD interactions using His-tagged proteins purified from bacterial or insect cells. First, 1.25 μM His-tagged Loqs-PD and 0.6 μM untagged Dcr-2 were mixed and incubated on ice for 30 min. The protein mixture was then incubated with 15 μl Ni-NTA Agarose (Qiagen, 30210) in a total volume of 500 μl in the binding buffer (200 mM NaCl, 20 mM Tris pH 8.0, 5% glycerol, 20 mM imidazole) at 4 °C for 1 h with gentle rotation. After centrifugation at 500i for 1 min, the supernatant was removed and the beads were washed five times using wash buffer (200 mM NaCl, 20 mM Tris pH 8.0, 5% glycerol, 20 mM imidazole, 0.1% NP-40) by centrifugation, followed by SDS-PAGE analysis.

BS3/EDC-mediated cross-linking mass spectrometry
The purified complexes were incubated with 0.25 mM bis (sulfosuccinimidyl)suberate (BS3; Thermo Fisher Scientific, 21580) in the reaction buffer containing 50 mM HEPES pH 7.5, 80 mM NaCl and 5% glycerol at 25 °C for 2 h or 5 mM 1-ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (EDC; Thermo Fisher Scientific, PG82073) in the reaction buffer containing 50 mM HEPES pH 7.2, 80 mM NaCl and 5% glycerol at 25 °C for 2 h. Cross-linked complexes were further purified to remove oligomer and glycerol by size-exclusion chromatography. The proteins (10 μg) were precipitated and digested for 16 h at 37 °C by trypsin at an enzyme-to-substrate ratio of 1:50 (w/w). The tryptic digested peptides were desalted and loaded on an in-house packed capillary reverse-phase C18 column (40 cm length, 100 μM ID × 360 μM OD, 1.9 μM particle size, 120 Å pore diameter) connected to an Easy LC 1200 system. The samples were analysed with a 120 min high-performance liquid chromatography gradient from 6% to 35% buffer B (buffer A: 0.1% formic acid in water; buffer B: 0.1% formic acid in 80% acetonitrile) at 300 nl min −1 . The eluted peptides were ionized and directly introduced into a Q-Exactive mass spectrometer using a nano-spray source. Survey full-scan MS spectra (m/z = 300-1,800) were acquired in the Orbitrap analyzer with resolution r = 70,000 at m/z = 400. Cross-linked peptides were identified and evaluated using pLink2 software 30 .

Cryo-EM sample preparation and data collection
We used the same specimen preparation and data collection method for all of the cryo-EM datasets. An aliquot of 4 μl of purified or reaction sample was applied to a custom-made graphene grid 31 (Quantifoil Au 1.2/1.3, 300 mesh), which were glow-discharged (in a Harrick Plasma system) for 10 s at middle level after 2 min evacuation. The grids were then blotted by a couple of 55 mm filter papers (Ted Pella) for 0.5 s at 22 °C and 100% humidity, then flash-frozen in liquid ethane using the FEI Vitrobot Mark IV. Cryo-EM data were collected on different Titan Krios electron microscopes, all of which were operated at 300 kV, equipped with a Gatan K3 direct electron detector and a Gatan Quantum energy filter. All data were automatically recorded using AutoEMation 32 or EPU (post-dicing state dataset) in counting mode and defocus values ranged from −1.5 μm to −2.0 μm. The other parameters of each dataset are provided in Extended Data Table 1.

Image processing and 3D reconstruction
For all of the datasets, the image processing was adopted in similar steps. All of the raw dose-fractionated image stacks were 2× Fourier binned, aligned, dose-weighted and summed using MotionCorr2 (ref. 33 ). The following steps were then processed in RELION (v.3.1) 34 . The contrast transfer function parameters were estimated using CTFFIND4 (ref. 35 ). Approximately 2,000 particles were manually picked and 2D-classified to generate initial templates for automatic picking. A large number of particles were then automatically picked from raw micrographs on the basis of our templates. After one round of reference-free 2D classification and several rounds of 3D classification, using the initial 3D reference models obtained by ab initio calculation in RELION v.3.1, particles from good 3D classes, with better overall structure features, were selected for 3D refinement. The final high-resolution homogeneous refinement was performed in CryoSPARC 36 . The resolutions were determined by gold-standard Fourier shell correlation. Local resolution distribution was evaluated using blocres command in the Bsoft software package 37 . The detailed image processing of each dataset is provided in Extended Data Figs. 2 and 3.

Model building and refinement
The highest resolution EM density map of dimer status was used for initial model building, in which the quality of density was sufficient for de novo model building in COOT 38 . The initial model was separated into three parts (helicase-LoqsPD, DUF283 and other domains) and docked into EM 3D density maps of other states in Chimera 39 and then adjusted manually in ISOLDE 40 in Chimerax 41 and COOT. Finally, all of the models were refined against the EM map by PHENIX 42 in real space with secondary structure and geometry restraints. The final models were validated in PHENIX software package. The model statistics are summarized in Extended Data Table 1.

Statistics and reproducibility
For Extended Data Fig. 1a-c, experiments were repeated at least three times. For Extended Data Figs. 1d, 2a,f, 3a,j, 9a and 10a-c, experiments were repeated at least twice.

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
The atomic coordinates and structure factors in apo and initial dsRNA-binding states, early-and mid-translocation states, active dicing and post-dicing states in this study have been deposited at the RCSB PDB and Electron Microscopy Data Bank (EMDB) under EMD accession codes EMD-32236, EMD-32237, EMD-32238, EMD-32239, EMD-32240 and EMD-32241, and PDB accession codes 7W0A, 7W0B, 7W0C, 7W0D, 7W0E and 7W0F, respectively. The PDB and EMDB codes are also listed in Extended Data Table 1. Uncropped gel images are provided in Supplementary Fig. 1. Other structures used in this study were retrieved from PDB with accession code 7VG2 (AtDCL3), 2EZ6 (AaRNase III), 2FFL (GiDicer), 5ZAK (HsDicer), 6LXD (HsDrosha), 5F9H (HsRIG-I) and 7ELE (AtDCL1). The information of Dcr-2 and Loqs-PD is available in the UniProt database under accession codes A1ZAW0 and M9MRT5. Any other data or materials can be obtained from the corresponding authors on reasonable request.