Mechanism of SARS-CoV-2 polymerase stalling by remdesivir

Remdesivir is the only FDA-approved drug for the treatment of COVID-19 patients. The active form of remdesivir acts as a nucleoside analog and inhibits the RNA-dependent RNA polymerase (RdRp) of coronaviruses including SARS-CoV-2. Remdesivir is incorporated by the RdRp into the growing RNA product and allows for addition of three more nucleotides before RNA synthesis stalls. Here we use synthetic RNA chemistry, biochemistry and cryo-electron microscopy to establish the molecular mechanism of remdesivir-induced RdRp stalling. We show that addition of the fourth nucleotide following remdesivir incorporation into the RNA product is impaired by a barrier to further RNA translocation. This translocation barrier causes retention of the RNA 3ʹ-nucleotide in the substrate-binding site of the RdRp and interferes with entry of the next nucleoside triphosphate, thereby stalling RdRp. In the structure of the remdesivir-stalled state, the 3ʹ-nucleotide of the RNA product is matched and located with the template base in the active center, and this may impair proofreading by the viral 3ʹ-exonuclease. These mechanistic insights should facilitate the quest for improved antivirals that target coronavirus replication. Remdesivir is a nucleoside analog that inhibits the SARS-CoV-2 RNA dependent RNA polymerase (RdRp) and is used as a drug to treat COVID19 patients. Here, the authors provide insights into the mechanism of remdesivir-induced RdRp stalling by determining the cryo-EM structures of SARS-CoV-2 RdRp with bound RNA molecules that contain remdesivir at defined positions and observe that addition of the fourth nucleotide following remdesivir incorporation into the RNA product is impaired by a barrier to further RNA translocation.

C oronaviruses use an RdRp enzyme to carry out replication and transcription of their RNA genome [1][2][3][4][5] . The RdRp consists of three non-structural protein (nsp) subunits, the catalytic subunit nsp12 6 and the accessory subunits nsp8 and nsp7 3,7 . Structures of the RdRp of SARS-CoV-2 were obtained in free form 8 and with RNA template-product duplexes [9][10][11] . Together with a prior structure of SARS-CoV RdRp 12 , these results have elucidated the RdRp mechanism. For RNA-dependent RNA elongation, the 3ʹ-terminal nucleotide of the RNA product chain resides in the -1 site and the incoming nucleoside triphosphate (NTP) substrate binds to the adjacent +1 site. Catalytic nucleotide incorporation then triggers RNA translocation and liberates the +1 site for binding of the next incoming nucleoside triphosphate (NTP).
The nucleoside analog remdesivir is the only FDA-approved drug for the treatment of COVID-19 patients [13][14][15][16] . Remdesivir inhibits the RdRp of coronaviruses 10,[17][18][19][20][21][22] and shows antiviral activity in cell culture and animals 21,[23][24][25] . Remdesivir is a phosphoramidate prodrug that is metabolized in cells to yield an active NTP analog 21 that we refer to as remdesivir triphosphate (RTP). Biochemical studies showed that the RdRp can use RTP as a substrate, leading to the incorporation of remdesivir monophosphate (RMP) into the growing RNA product 10,20,22 . After RMP incorporation, the RdRp extends RNA by three more nucleotides before it stalls 10,20,22 . This stalling mechanism is specific to coronaviruses because the RdRp of Ebola virus can add five RNA nucleotides after RMP incorporation before it stalls 26 .
Recent structural studies showed RdRp-RNA complexes after remdesivir addition to the RNA product 3ʹ-end. One structure contained RMP in the +1 site 9 , whereas another structure contained RMP in the -1 site 10 . In both structures, RMP mimics adenosine monophosphate (AMP) and forms standard Watson-Crick base pairs with uridine monophosphate (UMP) in the RNA template strand. Thus, these studies explained how RMP is incorporated into RNA instead of AMP. However, they do not explain how remdesivir inhibits the RdRp because RdRp stalling occurs only after three more nucleotides have been added to the RNA 10,20 .

Results
Biochemical reconstitution of RdRp stalling by remdesivir. To investigate remdesivir-induced RdRp stalling, we first investigated how RTP (Fig. 1a) influences RdRp elongation activity on an RNA template-product scaffold (Fig. 1b) using a highly defined biochemical system (Methods). Consistent with recent studies 20,22 , we observed that RMP is readily incorporated into RNA and that the RNA is subsequently elongated by three more nucleotides before the RdRp stalls (Fig. 1c, d). At high NTP concentrations, RdRp stalling was largely overcome and the fulllength RNA product was formed despite the presence of RMP in the RNA product (Fig. 1c, d). Thus, the predominant mechanism of remdesivir action after its incorporation into the growing RNA is delayed RdRp stalling. Although we cannot exclude that a minor fraction of RdRp-RNA complexes may dissociate and terminate elongation, the stalling mechanism is also observed in a recent single-molecule study 27 .
To uncover the mechanistic basis of remdesivir-induced RdRp stalling, we aimed to determine structures of RdRp-RNA complexes containing RMP at defined positions in the RNA product strand. We prepared RMP-containing RNA oligonucleotides by solid-phase synthesis using 5ʹ-O-DMT-2ʹ-O-TBDMS-protected 3ʹ-cyanoethyl diisopropyl phosphoramidite (Rem-PA), which we synthesized from 1ʹ-cyano-4-aza-7,9-dideazaadenosine (Rem) in four steps (Fig. 2a, Methods, Supplementary Methods). The presence of RMP in the obtained RNA oligonucleotides was confirmed by denaturing HPLC and LC-MS after digestion into mononucleosides (Fig. 2b, c). We further confirmed that the presence of RMP inhibits RNA extension by RdRp on minimal RNA template-product scaffolds (Fig. 2d, e).
Structural analysis of RdRp stalling by remdesivir. The ability to prepare RNAs containing RMP at defined positions enabled us to structurally capture the two states of the RdRp complex that are relevant for understanding remdesivir-induced RdRp stalling. Specifically, we investigated RdRp-RNA complexes captured after addition of two or three nucleotides following RMP incorporation. We prepared RNA scaffolds containing RMP at positions -3 or -4 by annealing short RMP-containing oligonuclotides to long, loop-forming RNAs (scaffolds 1 and 2, respectively) (Methods). The annealed RNA scaffolds were then bound to purified RdRp and subjected to cryo-EM analysis as described 11 , resulting in two refined structures (Supplementary Table 1).
The first RdRp-RNA structure (structure 1) was resolved at 3.1 Å resolution ( Supplementary Fig. 1) and showed that the RMP was located at position -3 of the RNA product strand, as expected from the design of scaffold 1 (Fig. 3a, b). The RdRp-RNA complex adopted the post-translocated state. The RNA 3ʹend resided in the -1 site and the +1 site was free to bind the next NTP substrate. Comparison with our previous RdRp-RNA complex structure 11 did not reveal significant differences. The 1ʹ-cyano group of the RMP ribose moiety was located at position -3 and is accommodated there by an open space in the RNA product-binding site of the RdRp (Fig. 3a, b). Thus, structure 1 represents an active state of the elongation complex that is poised to add one more nucleotide to the RNA before stalling, consistent with biochemical results.
A translocation barrier underlies remdesivir-induced RdRp stalling. The second RdRp-RNA structure (structure 2) was resolved at 3.4 Å resolution ( Supplementary Fig. 1) and showed that the RMP moiety was not located at position -4, as was expected from our design of scaffold 2, but was instead located at position -3 ( Fig. 3a, b). The +1 site was no longer free, as observed in structure 1, but was occupied by the nucleotide at the RNA 3ʹ-end. The RdRp-RNA complex adopts the pre-translocated state and cannot bind the next NTP substrate. Thus, structure 2 indicates that the RMP moiety in the RNA product strand is not tolerated at position -4. These results suggested that remdesivirinduced stalling of the RdRp is due to impaired translocation of the RNA after the RMP reaches register -3.
To test the hypothesis that RdRp stalling results from a translocation barrier, we formed a third RdRp-RNA complex with an RNA scaffold that was identical to that in structure 2 except that RMP was replaced by AMP, and we determined the resulting structure 3 at 2.8 Å resolution (Fig. 3a, b and Supplementary Fig. 1). In structure 3, the RdRp-RNA complex adopted the post-translocation state and the +1 site was again free, as observed in structure 1. This shows that the unexpected pre-translocated state that we observed in structure 2 was indeed caused by the presence of RMP, which was not tolerated at position -4. In conclusion, the RMP moiety in the RNA product strand gives rise to a translocation barrier that impairs movement of the RMP from position -3 to position -4.

Discussion
Our results define the structural mechanism of remdesivirinduced RdRp stalling. They show directly that stalling is caused by a translocation barrier that the RdRp encounters after the addition of three more nucleotides following remdesivir incorporation into the growing RNA. Prior observations suggest that the translocation barrier that we define here is caused by the presence of the C1ʹ-cyano group in the remdesivir ribose moiety. First, this cyano group is critical for antiviral potency against Ebola virus 21 . Second, modeling the RMP at position -4 of the RNA product strand results in a steric clash with the side chain of serine-861 in nsp12 10,20 . Indeed, our structural data strongly support the modeling (Fig. 3c). Third, truncation of serine-861 to alanine 10,28 or glycine 28 renders the RdRp less sensitive or insensitive, respectively, to inhibition by remdesivir. We conclude that the translocation barrier results from the sterically impaired passage of the cyano group in RMP past the serine-861 side chain in the nsp12 subunit of RdRp.
We have summarized the mechanism of remdesivir-induced RdRp stalling in a molecular animation (Supplementary Movie 1). The remdesivir-stalled state is observed in our structure 2. In this structure, the RNA product 3ʹ-nucleotide is buried in the active center and is base-paired with the RNA template strand (Fig. 3a, b). This may explain why the RNA 3ʹ-end may at least partially escape proofreading by the viral exonuclease nsp14 29,30 . Nevertheless, some proofreading can occur and this renders remdesivir less efficient 19 , indicating that the viral exonuclease can remove several nucleotides from the base-paired RNA 3ʹ-end. Such removal of several RNA nucleotides may require RNA backtracking along the RdRp, and this may be induced by the viral helicase nsp13 31 .
Finally, although delayed RdRp stalling after remdesivir incorporation into growing RNA is likely to be an important mechanism of inhibition, another mechanism of remdesivir action based on RNA template-dependent inhibition of the RdRp

3' 5'
A R  Remdesivir triphosphate was recently proposed 28 . This alternative mechanism and the mechanisms of RdRp-dependent RNA proofreading will be studied further in the future. Meanwhile, the mechanistic insights presented here may facilitate the search for compounds with improved potential to interfere with coronavirus replication.
RTP synthesis is described in Supplementary Methods. All unmodified RNA oligonucleotides were purchased from Integrated DNA Technologies. The RNA sequence used for the RNA extension assay (Fig. 1c) is /56-FAM/rArArC rArGrG rArGrA rCrUrC rGrCrG rUrArG rUrUrUrU rCrUrA rCrGrC rG. The assay was performed as described 11 , except for the following changes. The final concentrations of nsp12, nsp8, nsp7, and RNA were 3 μM, 9 μM, 9 μM, and 2 μM, respectively. The highest concentration of NTPs was 0.5 mM for each nucleotide (ATP or RTP, GTP, UTP, and CTP), followed by a two-fold serial dilution. RNA products were resolved on a denaturing sequencing gel and visualized by Typhoon 95000 FLA Imager (GE Healthcare Life Sciences). The RNA sequences used for extending RMP-containing RNA oligonucleotides are: rUrGrA rGrCrC rUrArC rGrC-rA or rR-rGrUrG (product) and rArArA rCrArC rUrGrC rGrUrA/3ddC/ (template). The extension assay (Fig. 2e) was performed as described 11 , with minor changes. Reactions were started by addition of UTP (final concentrations: 6.25 μM, 12.5 μM, 25 μM, or 250 μM). RNA products were visualized by SYBR Gold (Thermo Fischer) staining and imaged with Typhoon 9500 FLA Imager (GE Healthcare Life Sciences).
Cryo-EM sample preparation and data collection. SARS-CoV-2 RdRp was prepared as described 11 . The RNA scaffolds used for structural studies were comprisesd of two RNA strands. The first RNA strand forms the RNA templateproduct hairpin and lacks the last 15 nt at the 3ʹ end. The second strand contains the missing 15 nt sequence. Upon scaffold annealing, a complete RNA templateproduct is formed with a single nick in the product RNA. This strategy had to be used because the length of the remdesivir-containing oligonucleotides was limited due to technical reasons. RNA scaffolds for RdRp-RNA complex formation were prepared by mixing equimolar amounts of two RNA strands in annealing buffer (10 mM Na-HEPES pH 7.4, 50 mM NaCl) and heating to 75°C, followed by stepwise cooling to 4°C. RNA sequences for RMP at position -3 (structure 1) are: rUrUrU rUrCrA rUrGrC rArUrC rGrCrG rUrArG rGrCrU rCrArU rArCrC rGrUrA rUrUrG rArGrA rCrCrU rUrUrU rGrGrU rCrUrC rArArU rArCrG rGrUrA and rUrGrA rGrCrC rUrArC rGrCrG rRrUrG. RNA sequences for AMP/ RMP at position -4 (structures 2 and 3) are: rUrUrU rUrCrA rUrGrC rArCrU rGrCrG rUrArG rGrCrU rCrArU rArCrC rGrUrA rUrUrG rArGrA rCrCrU rUrUrU rGrGrU rCrUrC rArArU rArCrG rGrUrA and rUrGrA rGrCrC rUrArC rGrC-rA/rR -rGrUrG. RdRp-RNA complexes were formed by mixing purified nsp12 (scaffold 1:2 nmol, scaffold 2:2 nmol, scaffold 3:1.6 nmol) with an equimolar amount of annealed RNA scaffold and threefold molar excess of each nsp8 and nsp7. After 10 min of incubation at room temperature, the mixture was applied to a Superdex 200 Increase 3.2/300 size exclusion chromatography column, which was equilibrated in complex buffer (20 mM Na-HEPES pH 7.4, 100 mM NaCl, 1 mM MgCl 2 , 1 mM TCEP) at 4°C. Peak fractions corresponding to the RdRp-RNA complex were pooled and diluted to~2 mg/ml. For structure 2, an additional 0.36 nmol of annealed RNA scaffold were spiked into the sample prior to grid preparation. Three microliters of the concentrated RdRp-RNA complex were mixed with 0.5 µl of octyl ß-D-glucopyranoside (0.003% final concentration) and applied to freshly glow discharged R 2/1 holey carbon grids (Quantifoil). The grids were blotted for 5 s using a Vitrobot MarkIV (Thermo Fischer Scientific) at 4°C and 100% humidity and plunge frozen in liquid ethane.
Cryo-EM data were collected using SerialEM 32 on a 300 keV Titan Krios transmission electron microscope (Thermo Fischer Scientific). Prior to detection, inelastically scattered electrons were filtered out with a GIF Quantum energy filter (Gatan) using a slit width of 20 eV. Images were acquired in counting mode (nonsuper resolution) on a K3 direct electron detector (Gatan) at a nominal magnification of 105,000x resulting in a calibrated pixel size of 0.834 Å/pixel. Images were exposed for a total of 2.2 s with a dose rate of 19 e − /px/s resulting in a total dose of 60 e − /Å 2 , which was fractionated into 80 frames. Previous cryo-EM analysis of the SARS-CoV2 RdRp showed strong preferred orientation of the RdRp particles in ice 11 . Therefore, all data were collected with 30°tilt to obtain more particle orientations. Motion correction, contrast transfer function (CTF)estimation, and particle picking and extraction were performed on the fly using Warp 33 . In total, 8004, 11,764, and 7043 movies were collected for structures 1, 2, and 3, respectively.
Cryo-EM data processing and analysis. For structure 1, 1.8 million particles were exported from Warp 33 1.0.9 to cryoSPARC 34 2.15. After ab initio refinement of five classes, an intermediate map from the previous processing of EMD-11007 11 was added as a 6th reference, and supervised three-dimensional (3D) classification was performed. 654k particles (37%) from the best class deemed to represent the polymerase were subjected to 3D refinement to obtain a 3.1 Å map. Half-maps and particle alignments were exported to M 35 1.0.9, where reference-based frame series alignment with a 2 × 2 image-warp grid, as well as CTF refinement were performed for two iterations. Although the resulting map had the same 3.1 Å nominal resolution, the features of interest were significantly cleaner.
For structure 2, 3.4 million particles were exported from Warp 1.0.9 to cryoSPARC 2.15. After ab initio refinement of 5 classes, the EMD-11007-related reference was added as a 6th reference, and supervised 3D classification was performed. To further clean up the resulting 881k particles (26%) from the best class deemed to represent the polymerase, another ab initio refinement of five classes was performed. Three of these classes and the EMD-11007-related reference were used for supervised 3D classification. 474k particles (54%) from the best class deemed to represent the polymerase were subjected to 3D refinement to obtain a 3.6 Å map. Half-maps and particle alignments were exported to M 1.0.9, where reference-based frame series alignment with a 2 × 2 image-warp grid, as well as CTF refinement were performed for two iterations to obtain a 3.4 Å map.
For structure 3, 2.2 million particles were exported from Warp 1.0.9 to cryoSPARC 2.15. Initial unsupervised 3D classification in three classes was NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-20542-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:279 | https://doi.org/10.1038/s41467-020-20542-0 | www.nature.com/naturecommunications performed using the EMD-11007-related reference. To further clean up the resulting 1.1 million particles (23%), ab initio refinement of five classes was performed. Four of these classes and the EMD-11007-related reference were used for supervised 3D classification. 819k particles (70%) from the best class deemed to represent the polymerase were subjected to 3D refinement to obtain a 3.1 Å map. Half-maps and particle alignments were exported to M 1.0.9, where referencebased frame series alignment with a 2 × 2 image-warp grid, as well as CTF refinement were performed for three iterations to obtain a 2.8 Å map.
Model building and refinement. Models were built using our previously published SARS-CoV-2 RdRp structure as starting model (PDB 6YYT [https://doi.org/ 10.2210/pdb6YYT/pdb]) 11 . For each of the structures 1-3, the model was first rigid-body fit into the density and subsequently adjusted in real-space in Coot 36 . Parts of the N-terminal NiRAN domain of nsp12, the N-terminal extension of nsp8a and the entire nsp8b molecule were removed due to absence or poor quality of density for these regions. Restraints for RMP were generated in phenix.elbow 37 and the structures were refined using phenix.real_space_refine 38 with appropriate secondary structure restraints. Model quality was assessed using MolProbity within Phenix 39 , which revealed excellent stereochemistry for all three structural models (Supplementary Table 1). Figures were prepared with PyMol and ChimeraX 40 .
Reporting summary. Further information on experimental design is available in the Nature Research Reporting Summary linked to this paper.

Data availibility
The electron density reconstructions and structure coordinates were deposited with the Electron Microscopy Database (EMDB) under accession codes EMD-11993, EMD-11994, and EMD-11995 and with the Protein Data Bank (PDB) under accession codes 7B3B, 7B3C, and 7B3D. Other data are available from corresponding authors upon reasonable request. Source data are provided with this paper.