Coronaviruses are positive-strand RNA viruses that pose a major health risk1: SARS-CoV-2 has caused a pandemic of the disease known as COVID-195,6. Coronaviruses use an RdRp complex for the replication of their genome and for the transcription of their genes2,3. This RdRp complex is the target of nucleoside analogue inhibitors—in particular, remdesivir7,8. Remdesivir inhibits the RdRp of multiple coronaviruses9,10, and shows antiviral activity in cell culture and animal models11. Remdesivir is currently being tested in the clinic in many countries12 and has recently been approved for emergency treatment of patients with COVID-19 in the United States4.

The RdRp of SARS-CoV-2 is composed of a catalytic subunit known as nsp1213 as well as two accessory subunits, nsp8 and nsp73,14. The structure of this RdRp has recently been reported15; it is highly similar to the RdRp of SARS-CoV16, a zoonotic coronavirus that spread into the human population in 20021. The nsp12 subunit contains an N-terminal nidovirus RdRp-associated nucleotidyltransferase (NiRAN) domain, an interface domain and a C-terminal RdRp domain15,16. The RdRp domain resembles a right hand, comprising the fingers, palm and thumb subdomains15,16 that are found in all single-subunit polymerases. Subunits nsp7 and nsp8 bind to the thumb, and an additional copy of nsp8 binds to the fingers domain15,16. Structural information is also available for nsp8–nsp7 complexes17,18.

To obtain the structure of the SARS-CoV-2 RdRp in its active form, we prepared recombinant nsp12, nsp8 and nsp7 (Fig. 1a, Methods). When added to a minimal RNA hairpin substrate (Fig. 1b), the purified proteins gave rise to RNA-dependent RNA extension activity, which depended on nsp8 and nsp7 (Fig. 1c). We assembled and purified a stable RdRp–RNA complex with the use of a self-annealing RNA, and collected single-particle cryo-electron microscopy (cryo-EM) data (Extended Data Fig. 1, Extended Data Table 1). Particle classification yielded a 3D reconstruction at a nominal resolution of 2.9 Å (Extended Data Fig. 1). This led to a refined structure of the RdRp–RNA complex that showed the RNA in the active centre in great detail (Extended Data Fig. 2).

Fig. 1: Preparation of active SARS-CoV-2 RdRp.
figure 1

a, SDS–PAGE analysis of the purified SARS-CoV-2 RdRp subunits nsp12, nsp8 and nsp7. The experiment was performed once. b, Minimal RNA substrate that folds into a hairpin with ‘template’ and ‘product’ regions. The RNA contains a 11-nucleotide, fluorescently labelled 5′ overhang. c, Incubation of the RdRp subunits (a) with RNA (b) leads to efficient RNA extension. RNAs were separated on a denaturing acrylamide gel and visualized with a Typhoon 95000 FLA Imager. Representative result of three independent technical replicates (Supplementary Fig. 1).

Our structure shows the RdRp enzyme engaged with over two turns of duplex RNA (Fig. 2, Supplementary Video 1). The structure resembles that of the free enzyme15, but additionally reveals a long protruding RNA and extended protein regions in nsp8 (Extended Data Fig. 3a). To our knowledge, these observations are unique: the RdRp complexes of hepatitis C virus19, poliovirus20 and norovirus21 contain only one turn of RNA, and show no features that resemble the newly observed nsp8 extensions (Extended Data Fig. 3b).

Fig. 2: Structure of complex between SARS-CoV-2 RdRp and RNA.
figure 2

a, Domain structure of nsp12, nsp8, and nsp7 subunits of RdRp. In nsp12, the conserved sequence motifs A–G16 are depicted. Regions included in the structure are indicated with black bars. b, Three views of the structure, related by 90° rotations (top, back view; middle, side view; bottom, top view). Colour code for nsp12 (NiRAN, interface, fingers, palm and thumb), nsp8, nsp7, RNA template (blue) and RNA product (red) used throughout. The magenta sphere depicts a modelled21 metal ion in the active site.

Our structure provides details of the interactions between the RdRp and RNA (Fig. 3). The nsp12 subunit binds to the first turn of RNA between its fingers and thumb subdomains (Fig. 3a, b). The active site is located on the palm subdomain, and is formed by five conserved nsp12 elements known as motifs A–E (Fig. 3b). Motif C binds to the RNA 3′ end and contains the residues D760 and D761, which are required for RNA synthesis10,14. The additional nsp12 motifs F and G reside in the fingers subdomain and position the RNA template (Fig. 3b). The observed contacts of nsp12 with the RNA product strand may retain short RNA during early steps of RNA synthesis.

Fig. 3: RdRp–RNA interactions.
figure 3

a, Schematic of protein–RNA interactions. Solid and hollow circles show nucleotides that were included in the structure or invisible, respectively. RNA is assembled from overlapping oligonucleotides (Extended Data Fig. 1c). RdRp residues in nsp12 within 4 Å of RNA are indicated and contacts are depicted with lines. Nsp8 residues and their putative RNA contact regions (horizontal lines, within 5 Å) are indicated even if side chain density is absent. b, Interactions of the RdRp active site with the first turn of RNA. Subunit nsp12 is in grey, and conserved motifs A–G are coloured. Active-site residues D760 and D761 shown as sticks. The magenta sphere depicts a modelled21 metal ion. c, Charged nsp8 residues that may interact with proximal RNA regions. Left, side view; right, inverted side view.

As the RNA duplex exits from the RdRp cleft, it forms a second helical turn that protrudes from the nsp12 surface (Fig. 3c). There are no structural elements in the RdRp that restrict the extension of the RNA duplex. These observations are consistent with the production of double-stranded RNA during replication. However, it is unclear whether replication in infected cells results in RNA duplexes or whether RNA strands are separated and—if so—how. It is also unknown when and how RNA strands are separated during the transcription of viral genes, which produces single-stranded product mRNAs that can be translated.

The protruding, exiting RNA duplex is flanked by long α-helical extensions that are formed by the highly conserved17 N-terminal regions in the two nsp8 subunits (Figs. 2, 3). These prominent nsp8 extensions reach up to 28 base pairs away from the active site and use positively charged residues that are positioned to interact with the RNA backbones (Fig. 3). The two nsp8 extensions differ with respect to their RNA interactions, which also argues for sequence-independent binding. The two nsp8 copies adopt different structures in the RdRp complex, and interact differently with nsp7 and nsp12 subdomains (Extended Data Fig. 3c). The nsp8 extensions also adopt different structures in crystals of nsp8–nsp7 complexes17,18, and are mobile in free RdRp15,16. This indicates that the nsp8 extensions are flexible in the RdRp complex and become ordered when an RNA duplex exits the enzyme.

The interactions of the nsp8 extensions with exiting RNA may explain the processivity of the RdRp, which is required for replicating the very long RNA genome of coronaviruses and other viruses of the Nidovirales order3. It is known that nsp8 and nsp7 confer processivity to nsp1214. It is also known that the substitution of the nsp8 residue K58 with alanine is lethal for the virus14. K58 is located in the nsp8 extension, and interacts with exiting RNA around the minor groove (Fig. 3c). The nsp8 extensions may be regarded as sliding poles, which slide along exiting RNA to prevent premature dissociation of the RdRp during replication. The sliding poles may serve a function similar to the ‘sliding clamps’ that confer processivity to DNA replication machines22.

To investigate how the RdRp binds to the incoming nucleoside triphosphate (NTP) substrate, we superimposed our structure onto the related structure of the norovirus RdRp–nucleic acid complex21. This suggested that the NTP-binding site is conserved, including putative contacts between nsp12 and the NTP (Extended Data Fig. 3d, Supplementary Video 2). Residues N691, S682 and D623 may recognize the 2′-OH group of the NTP, thereby rendering the RdRp specific for the synthesis of RNA rather than DNA. Our modelling is also consistent with binding of the triphosphorylated form of remdesivir to the NTP site, because there is space to accommodate the additional nitrile group that is present at the 1′ position of the ribose ring of remdesivir (Extended Data Fig. 3d).

While our manuscript was under review, the structure of another SARS-CoV-2 RdRp–RNA complex became available23 and was published soon thereafter24. Comparison of the two studies shows that the core structures are similar; however, we additionally observe a second turn of RNA and the nsp8 sliding poles. The other study suggests that remdesivir functions as an ‘immediate’ RNA-chain terminator23,24. However, published biochemistry has shown that several more nucleotides can be added to RNA after incorporation of remdesivir, leading to ‘delayed’ termination10,25. We note that this latter mechanism can explain how remdesivir escapes removal from the RNA 3′ end by the viral exonuclease nsp1426 that binds to the RdRp complex14. On the basis of the results presented here, mechanistic questions regarding coronavirus replication, transcription and antiviral targeting can now be investigated.


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.

Cloning and protein expression

The SARS-CoV-2 nsp12 gene was codon-optimized for expression in insect cells. The SARS-CoV-2 nsp8 and nsp7 genes were codon-optimized for expression in Escherichia coli. Synthesis of genes was performed by GeneArt (Thermo Fisher Scientific GENEART). The gene synthesis products of the respective genes were PCR amplified with ligation-independent cloning-compatible primer pairs (nsp12: forward primer: 5′-TACTTCCAATCCAATGCATCTGCTGACGCTCAGTCCTTCCTG-3′, reverse primer: 5′-TTATCCACTTCCAATGTTATTATTGCAGCACGGTGTGAGGGG-3′; nsp8: forward primer: 5′-TACTTCCAATCCAATGCAGCAATTGCAAGCGAATTTAGCAGCCTG-3′, reverse primer: 5′-TTATCCACTTCCAATGTTATTACTG CAGTTTAACTGCGCTATTTGCACG-3′; nsp7: forward primer: 5′-TACTTCCAATCCAATGCAAGCAAAATGTCCGATGTTAAATGCACCAGC-3′, reverse primer: 5′-TTATCCACTTCCAATGTTATTACTGCAGGGTTGCACGATTATCCAGC-3′). The PCR products for nsp8 and nsp7 were individually cloned into the pET-derived vector 14-B (a gift from S. Gradia; Addgene 48308). The two constructs for nsp8 and nsp7 contain an N-terminal 6×His tag and a tobacco etch virus (TEV) protease cleavage site. The PCR product containing codon-optimized nsp12 was cloned into the modified pFastBac vector 438-C (a gift from S. Gradia; Addgene 55220) via ligation-independent cloning. The nsp12 construct contained an N-terminal 6×His tag, followed by an MBP tag, a 10×Asp sequence and a TEV protease cleavage site. All constructs were verified by sequencing.

The SARS-CoV-2 nsp12 plasmid (500 ng) was transformed into DH10EMBacY cells using electroporation to generate a bacmid encoding full-length nsp12. Virus production and expression in insect cells was then performed as previously described27. Insect cell lines were obtained from Expression Systems (94-002F and 94-003F) or Thermo Fisher (12659017). Cell lines were not authenticated. No commonly misidentified cell lines were used. After 60 h of expression in Hi5 cells, cells were collected by centrifugation and resuspended in lysis buffer (300 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 30 mM imidazole pH 8.0, 3 mM MgCl2, 5 mM β-mercaptoethanol, 0.284 μg ml−1 leupeptin, 1.37 μg ml−1 pepstatin, 0.17 mg ml−1 PMSF, and 0.33 mg ml−1 benzamidine). The SARS-CoV-2 nsp8 and nsp7 plasmids were overexpressed in E. coli BL21 (DE3) RIL cells grown in LB medium. Cells were grown to an optical density at 600 nm of 0.6 at 37 °C and protein expression was subsequently induced with 0.5 mM isopropyl β-d-1-thiogalactopyranoside at 18 °C for 16 h. Cells were collected by centrifugation and resuspended in lysis buffer (300 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 30 mM imidazole pH 8.0, 5 mM β-mercaptoethanol, 0.284 μg ml−1 leupeptin, 1.37 μg ml−1 pepstatin, 0.17 mg ml−1 PMSF and 0.33 mg ml−1 benzamidine).

Protein purification

Protein purifications were performed at 4 °C. After collection and resuspension, cells of the SARS-CoV-2 nsp12 expression were immediately sonicated for cell lysis. Lysates were subsequently cleared by centrifugation (87,207g, 4 °C, 30 min) and ultracentrifugation (235,000g, 4 °C, 60 min). The supernatant containing nsp12 was filtered using a 5-μm syringe filter, followed by filtration with a 0.8-μm syringe filter (Millipore) and applied onto a HisTrap HP 5 ml (GE Healthcare), preequilibrated in lysis buffer (300 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 30 mM imidazole pH 8.0, 3 mM MgCl2, 5 mM β-mercaptoethanol, 0.284 μg ml−1 leupeptin, 1.37 μg ml−1 pepstatin, 0.17 mg ml−1 PMSF and 0.33 mg ml−1 benzamidine). After application of the sample, the column was washed with 6 column volumes (CV) high-salt buffer (1,000 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 30 mM imidazole pH 8.0, 3 mM MgCl2, 5 mM β-mercaptoethanol, 0.284 μg ml−1 leupeptin, 1.37 μg ml−1 pepstatin, 0.17 mg ml−1 PMSF and 0.33 mg ml-1 benzamidine), and 6 CV lysis buffer. The HisTrap was then attached to an XK column 16/20 (GE Healthcare), prepacked with amylose resin (New England Biolabs), which was pre-equilibrated in lysis buffer. The protein was eluted from the HisTrap column directly onto the amylose column using nickel elution buffer (300 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 500 mM imidazole pH 8.0, 3 mM MgCl2 and 5 mM β-mercaptoethanol). The HisTrap column was then removed and the amylose column was washed with 10 CV of lysis buffer. Protein was eluted from the amylose column using amylose elution buffer (300 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 116.9 mM maltose, 30 mM imidazole pH 8.0 and 5 mM β-mercaptoethanol). Peak fractions were assessed via SDS–PAGE and staining with Coomassie. Peak fractions containing nsp12 were pooled and mixed with 8 mg of His-tagged TEV protease (about 80% (w/w)). After 12 h of protease digestion at 4 °C, protein was applied to a HisTrap column equilibrated in lysis buffer to remove uncleaved nsp12, 6×His–MBP and TEV. Subsequently, the flow-through containing nsp12 was applied to a HiTrap Heparin 5 ml column (GE Healthcare). The flow-through containing nsp12 was collected and concentrated in a MWCO 50,000 Amicon Ultra Centrifugal Filter unit (Merck). The concentrated sample was applied to a HiLoad S200 16/600 pg equilibrated in size-exclusion buffer (300 mM NaCl, 20 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 1 mM MgCl2, 1 mM TCEP). Peak fractions were assessed by SDS–PAGE and Coomassie staining. Peak fractions were pooled and concentrated in a MWCO 50,000 Amicon Ultra Centrifugal Filter (Merck). The concentrated protein with a final concentration of 102 μM was aliquoted, flash-frozen and stored at −80 °C until use.

SARS-CoV-2 nsp8 and nsp7 were purified separately using the same purification procedure, as follows. After cell collection and resuspension in lysis buffer, the protein of interest was immediately sonicated. Lysates were subsequently cleared by centrifugation (87.200g, 4 °C, 30 min). The supernatant was applied to a HisTrap HP 5 ml column (GE Healthcare), preequilibrated in lysis buffer. The column was washed with 9.5 CV high-salt buffer (1,000 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 30 mM imidazole pH 8.0, 5 mM β-mercaptoethanol, 0.284 μg ml−1 leupeptin, 1.37 μg ml−1 pepstatin, 0.17 mg ml−1 PMSF and 0.33 mg ml−1 benzamidine), and 9.5 CV low-salt buffer (150 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 30 mM imidazole pH 8.0 and 5 mM β-mercaptoethanol). The sample was then eluted using nickel elution buffer (150 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol, 500 mM imidazole pH 8.0 and 5 mM β-mercaptoethanol). The eluted protein was dialysed in dialysis buffer (150 mM NaCl, 50 mM Na-HEPES pH 7.4, 10% (v/v) glycerol and 5 mM β-mercaptoethanol) in the presence of 2 mg His-tagged TEV protease (nsp7: about 10% (w/w), nsp8: about 6% (w/w)) at 4 °C. After 12 h, imidazole pH 8.0 was added to a final concentration of 30 mM. The dialysed sample was subsequently applied to a HisTrap HP 5 ml column (GE Healthcare), preequilibrated in dialysis buffer. The flow-through that contained the protein of interest was then applied to a HiTrap Q 5 ml column (GE Healthcare). The Q column flow-through containing nsp8 or nsp7 was concentrated using a MWCO 3,000 Amicon Ultra Centrifugal Filter (Merck) and applied to a HiLoad S200 16/600 pg equilibrated in size exclusion buffer (150 mM NaCl, 20 mM Na-HEPES pH 7.4, 5% (v/v) glycerol, 1 mM TCEP). Peak fractions were assessed by SDS–PAGE and Coomassie staining. Peak fractions were pooled. Nsp7 with a final concentration of 418 μM was aliquoted, flash-frozen and stored at −80 °C until use. Nsp8 with a final concentration of 250 μM was aliquoted, flash-frozen and stored at −80 °C until use. All protein identities were confirmed by mass spectrometry.

RNA extension assays

All RNA oligonucleotides were purchased from Integrated DNA Technologies. The RNA sequence used for the transcription assay is /56-FAM/rUrUrU rUrCrA rUrGrC rUrArC rGrCrG rUrArG rUrUr UrUrC rUrArC rGrCrG. We designed a minimal substrate by connecting the template RNA to the RNA primer by a tetraloop, to protect the blunt ends of the RNA duplex and to ensure efficient annealing. RNA was annealed in 50 mM NaCl and 10 mM Na-HEPES pH 7.5 by heating the solution to 75 °C and gradually cooling to 4 °C. RNA extension reactions contained RNA (5 μM), nsp12 (5 μM), nsp8 (15 μM) and nsp7 (15 μM) in 100 mM NaCl, 20 mM Na-HEPES pH 7.5, 5% (v/v) glycerol, 10 mM MgCl2 and 5 mM β-mercaptoethanol. Reactions were incubated at 37 °C for 5 min and the RNA extension was initiated by addition of NTPs (150 μM UTP, GTP and CTP, and 300 μM ATP). Reactions were stopped by the addition of 2× stop buffer (7 M urea, 50 mM EDTA pH 8.0, 1× TBE buffer). Samples were digested with proteinase K (New England Biolabs) and RNA products were separated on 8 cm × 8 cm 20% acrylamide gels in 1× TBE buffer supplemented with 8M urea. 6-FAM-labelled RNA products were visualized by Typhoon 95000 FLA Imager (GE Healthcare Life Sciences).

Cryo-EM sample preparation and data collection

An RNA scaffold for RdRP–RNA complex formation was annealed by mixing equimolar amounts of two RNA strands (5′-rUrUrU rUrCrA rUrGrC rUrArC rGrCrG rUrArG-3′; 56-FAM/rCrUrA rCrGrC rG-3′) (IDT Technologies) in annealing buffer (10 mM Na-HEPES pH 7.4, 50 mM NaCl) and heating to 75 °C, followed by step-wise cooling to 4 °C. For complex formation, 1.2 nmol of purified nsp12 was mixed with a 1.2-fold molar excess of RNA scaffold and sixfold molar excess of each nsp8 and nsp7. After incubation at room temperature for 10 min, the EC was subjected to size exclusion chromatography on a Superdex 200 Increase 3.2/300 equilibrated with complex buffer (20 mM Na-HEPES pH 7.4, 100 mM NaCl, 1 mM MgCl2, 1 mM TCEP). Peak fractions with a volume of 100 μl (absorbance at 280 nm of 0.155 AU, 2-mm path length) corresponding to a nucleic-acid-rich high-molecular weight population (as judged by absorbance at 260 nm) were pooled and concentrated in a MWCO 30,000 Vivaspin 500 concentrator (Sartorius) to approximately 20 μl. Three μl of RdRp–RNA complex were mixed with 0.5 μl of octyl β-d-glucopyranoside (0.003% final concentration) and applied to freshly glow-discharged R 2/1 holey carbon grids (Quantifoil). Prior to flash freezing in liquid ethane, the grid was blotted for 6 s with a blot force of 5 using a Vitrobot Mark IV (Thermo Fisher Scientific) at 4 °C and 100% humidity.

Cryo-EM data collection was performed with SerialEM28 using a Titan Krios transmission electron microscope (Thermo Fisher Scientific) operated at 300 keV. Images were acquired in EFTEM mode with a slit width of 20 eV using a GIF quantum energy filter and a K3 direct electron detector (Gatan) at a nominal magnification of 105,000× corresponding to a calibrated pixel size of 0.834 Å per pixel. Exposures were recorded in counting mode for 2.2 s with a dose rate of 19 e per pixel per s resulting in a total dose of 60 e per Å2 that was fractionated into 80 movie frames. Because initial processing showed that the particles adopted only a limited number of orientations in the vitreous ice layer, a total of 8,168 movies were collected at 30° stage tilt to yield a broader distribution of orientations. Untilted data were discarded. Motion correction, dose weighting, CTF estimation, particle picking and extraction were performed using Warp29.

Cryo-EM data processing and analysis

We exported the 1.3 million particles from Warp29 to cryoSPARC30, and the particles were subjected to 2D classification. Twenty-five per cent of the particles were selected from classes deemed to represent the polymerase, and refined against a synthetic reference prepared from the model with the Protein Data Bank (PDB) code 6M71. Ab initio refinement was performed using particles from bad 2D classes to obtain five 3D classes of ‘junk’. These five classes and the first polymerase reconstruction were used as starting references to sort the initial 1.3 million particles in supervised 3D classification rather than 2D, as the latter tended to exclude less abundant projection directions. Five hundred and fourteen thousand particles (39%) from the resulting polymerase class were subjected to another ab initio refinement to obtain five starting references containing four junk classes and the complex of interest. These classes were used as starting references in another supervised 3D classification. Four hundred and eighteen thousand particles (82%) from the class representing the complex were exported from cryoSPARC to RELION 3.031. There, all particles were refined in 3D against the reconstruction previously obtained in cryoSPARC, using a mask including only the core part of the polymerase and a short segment of upstream RNA to obtain a 3.1 Å reconstruction. CTF refinement and another round of 3D refinement improved the resolution further to 2.9 Å (map 1 in Extended Data Fig. 2a–c). Particles were re-extracted at 1.3 Å per pixel in a bigger box in Warp to accommodate distant parts of the RNA. Unsupervised 3D classification with local alignment was performed to obtain two classes (with nsp8b present and without). One hundred and seventy-two thousand particles with nsp8b present were finally subjected to global (map 2 in Extended Data Fig. 2a–c) and focused 3D refinement using a mask including the RNA, nsp8a and nsp8b (map 3 in Extended Data Fig. 2a–c).

Model building and refinement

To build the atomic model of the RdRp–RNA complex, we started from the structure of the free SARS-CoV-2 RdRp (PDB 6M71) that was recently adjusted by T. Croll (available through the Coronavirus Structural Task Force by A. Thorn at The structure was rigid-body fit into the cryo-EM reconstruction and adjusted manually in Coot32. Unmodelled density remained for helical segments in the N-terminal regions of both copies of nsp8. These nsp8 extensions were modelled by superimposing the nsp8 model (PDB 2AHM; chain H) from the crystal structure of the nsp7–nsp8 hexadecamer17, in which the far N-terminal region of nsp8 adopts the same fold. Nsp8a (chain B) showed weaker density than nsp8b (chain D), but the register was faithfully determined by superimposing well-resolved parts (residues 80–97). The most N-terminal helices in nsp8a and nsp8b (residues 6–31) were only visible after low-pass filtering of maps to the local resolution of 6 Å and were modelled by superposition of the crystal structure of nsp8 (PDB 2AHM; chain H) with residues 33–55, which positioned these helices within the density in the low-pass-filtered map. Side chains for residues 6–31 were subbed. Careful inspection of the remaining A-form RNA density revealed that in our complex, instead of the originally designed short template-primer duplex (see ‘Cryo-EM sample preparation and data collection’), four copies of one of the RNA oligonucleotides were annealed to form a pseudo-continuous long RNA duplex. Annealing was mediated by a 10-bp self-complementary region within this RNA oligonucleotide (Extended Data Fig. 1c). Nucleotides 5–18 of four RNA strands were modelled, whereas the flapped-out nucleotides 1–4 were invisible and excluded. The model was real-space-refined using phenix.refine33 against a composite map of the focused refinement (maps 1 and 3) and global reconstructions (map 2) generated in phenix.combine_focused_maps and shows excellent stereochemistry (Extended Data Table 1). Figures were prepared with PyMol and Chimera34.

Reporting summary

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