Preferential remdesivir triphosphate incorporation by SARS-CoV-2 polymerase is altered to ATP by the S759A mutation

Abstract

Nucleoside analogs are successful in treating viral infections. dNTP analogs are primarily DNA chain terminators, while NTP analog remdesivir can inhibit RNA synthesis by delayed chain termination or when embedded in the template strand. Here, enzymatic assays, mass spectrometry, and cryo-EM structures demonstrate that SARS-CoV-2 RNA-dependent RNA polymerase (RdRp) preferentially incorporates remdesivir triphosphate (RTP), outcompeting 10-fold excess ATP; however, successive RTP incorporations are disfavored when ATP is present. The RdRp structures demonstrate that 1′-cyano-imposed conformational restriction of the remdesivir:UMP base-pair is resistant to translocation, reducing successive RTP incorporations. The S759A mutant confers RTP resistance. We show that the mutation switches the RdRp preference to ATP; RTP is incorporated only at 10-fold excess to ATP. The structures of S759A RdRp reveal that the primer 3′-end nucleotide repositioning and its altered ribose-ring conformation contribute to RTP resistance. These findings have implications for designing non-obligate nucleoside analogs with different inhibition mechanisms.

Using enzymatic assays, mass spectrometry, and cryo-EM structures, the authors demonstrate that SARS-CoV-2 RNA-dependent RNA polymerase (RdRp) preferentially incorporates remdesivir triphosphate (RTP), outcompeting ATP; whereas, the S759A mutant confers RTP resistance by switching the preference to ATP. These findings will guide the design of future non-obligate nucleoside analogs to overcome resistance mechanisms.

Introduction

Nucleoside/nucleotide analogs (NAs) constitute an important class of antiviral drugs used in the treatment of human immunodeficiency virus (HIV), hepatitis B virus (HBV), cytomegalovirus (CMV), herpes simplex virus (HSV), hepatitis C virus (HCV), and severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) infections. In their triphosphate (TP) form, these drugs act as substrates for viral polymerases, which catalytically incorporate NAs into the primer strand, to inhibit replication and/or transcription of the viral genome. However, the potency, specificity, and mechanisms of action can differ among NAs due to their chemical modifications relative to the natural NTP/dNTP substrates. dNTP analogs such as zidovudine (AZT), lamivudine (3TC), and tenofovir (TFV) target viral DNA polymerase or reverse transcriptase (RT), acting as obligate DNA chain terminators as these inhibitors lack the 3′-OH group and prevent the incorporation of the next nucleotide^1,2. However, the development of NAs against RNA viruses has been more challenging, and the approved NTP analog prodrugs, such as sofosbuvir (SOF), remdesivir (RDV), and molnupiravir (MOV), feature a 3′-OH group. While SOF has been developed for the treatment of HCV infection³, RDV and MOV have been approved in several jurisdictions for the treatment of COVID-19. Sofosbuvir prevents the incorporation of the next nucleotide by HCV RdRp; therefore, it functions as a non-obligate immediate RNA chain terminator⁴. Conversely, RDV and MOV do not immediately arrest RNA synthesis following incorporation; instead, they act through delayed inhibition^5–9 or lethal mutagenesis^10,11, respectively. Here, delayed inhibition is defined as RNA synthesis arrest after the polymerase has advanced at least two positions downstream during primer-strand extension and/or when the analog subsequently serves as a templating base. Coronaviruses, such as SARS-CoV-2, encode a 3′-to-5′ proof-reading exoribonuclease (ExoN) that can excise mismatched nucleotides or chain-terminating nucleotide analogs^12–15. The distinct mechanisms of inhibition of RDV and MOV help to evade SARS-CoV-2 proofreading^16,17.

The heterotrimeric SARS-CoV-2 RdRp complex, composed of the non-structural proteins Nsp7, Nsp8, and Nsp12, is responsible for viral RNA synthesis. Template binding and nucleotide incorporation are predominantly driven by regions in the viral polymerases referred to as motifs A through G, which are found within Nsp12. These motifs are sequence-derived and well-conserved among RdRps from diverse RNA viruses¹⁸. Nsp7 and Nsp8 appear to play structural roles. Kinetic and biochemical studies have demonstrated that RTP is a better substrate than ATP, its natural counterpart^5–7. Cryogenic-electron microscopy (Cryo-EM) structures of the SARS-CoV-2 RdRp active complex with RTP incorporated¹⁹ or bound as a substrate²⁰ have revealed the molecular interactions and provided the structural basis for RTP recognition. RDV functions as a delayed chain terminator, where inhibition occurs three nucleotides after the first RTP is incorporated^5,21–24. Several biochemical, structural, and mutagenesis studies support a mechanism that disfavors the translocation of the dsRNA containing remdesivir monophosphate (RMP), identifying a steric clash between the 1′-cyano group of RDV in the primer strand and Nsp12 residue S861 as the limiting factor^21,22,25,26. However, this translocation barrier can be easily overcome with increased NTP concentrations that better mimic cellular conditions^5,7,22. Thus, it is plausible that RMPs become embedded in the synthesized viral genomes and could serve as a templating base for subsequent rounds of replication. Indeed, the incorporation of UTP complementing RMP is compromised, resulting in template-dependent inhibition²⁵. Although RTP can effectively compete with ATP at a single site²⁷, it remains unclear how this competition is modulated for consecutive RTP incorporation events in the presence of ATP. Here, we examine the site-specificity of consecutive RTP incorporation by SARS-CoV-2 RdRp from a pool of NTPs, including ATP, and provide the structural features underlying consecutive RTP incorporations that may help rational design of improved nucleoside analogs.

We employed biochemical and mass spectrometry (MS) studies at varying RTP-to-ATP ratios and determined cryo-EM structures of the SARS-CoV-2 RdRp/dsRNA catalytic complexes with ATP or RTP to understand the incorporation preference from a heterogeneous mixture of RTP and ATP. We show that (I) RTP is preferentially incorporated at the first site even in the presence of 10-fold excess ATP, (II) successive incorporations are influenced by the relative ATP-to-RTP concentration, and (III) the remdesivir conferring resistant mutation, S759A, in Nsp12 repositions the primer 3-end with respect to the active site and alters the incorporation preference from RTP to ATP.

Results

Competition between ATP and RTP

Earlier studies have demonstrated that SARS-CoV-2 RdRp preferentially incorporates RTP 2 to 3 times more efficiently than ATP under noncompetitive conditions^5,6. One such study was conducted under pre-steady-state conditions using a dsRNA composed of a 40-mer template and a 6-carboxyfluorescein (FAM)-labelled 20-mer primer representing the 3′-end of the SARS-CoV-2 genome preceding the poly(A) tail⁶. Importantly, this sequence contains a homopolymeric stretch of uridines that can facilitate the sequential incorporation of four RTP or ATP molecules (Fig. 1A, top). Here, the first RTP molecule is incorporated more efficiently, whereas the second and third sequential RTP incorporations are less efficient than ATP⁶. However, treating a SARS-CoV-2-infected cell means that viral RNA synthesis occurs in the presence of both ATP and RTP, therefore, RTP competes with excess ATP for binding and incorporation. A biophysical technique that monitors nucleic acid synthesis by RNA template force-tension, found that RTP incorporation remains evident even at concentrations 25-fold lower than competing ATP⁷. Therefore, we hypothesized that the relative concentration of RTP to ATP is likely to influence which one is incorporated and if there is a site preference for one over the other.

Fig. 1. Competition between ATP and RTP for incorporation by wildtype (WT) and S759A mutant SARS-CoV-2 RdRp complex.

A RNA synthesis was evaluated on a dsRNA oligo that consists of a 40-nt template strand annealed to a 20-nt primer strand (top); a stretch of uridines spans the templated positions 23 through 26. The pattern of SARS-CoV-2-catalyzed RNA synthesis (bottom) in the presence of increasing concentrations of RTP (left) or ATP (right). When ATP concentration is reduced, or RTP concentration is increased, product formation at nucleotide position 26 (i-3) becomes more prominent, consistent with delayed chain termination for an RTP incorporated at nucleotide position 23 (i). The Heparin control lane is the resulting product when the SARS-CoV-2 RdRp complex is incubated with 4 mg/mL heparin prior to the addition of the RNA duplex and nucleotides. No observable RNA synthesis activity indicates that the concentration of heparin is sufficient to support single-turnover RNA catalysis. These reactions were allowed to proceed for 15 minutes following the addition of nucleotides and heparin. B Schematic depicting the method used to prepare SARS-CoV-2-catalyzed RNA primers for analysis by MALDI-TOF mass spectrometry (MS). C MALDI-TOF mass spectrum of the 20-nt RNA primer-strand and the 40-nt template-strand (left); their expected mass is shown in parentheses below the sequence. Polymerase extension reactions were performed by incubating UTP, 3’-dCTP, and ATP (right, top) or RTP (right, bottom) with the SARS-CoV-2 RdRp complex. The sequence of the extended portion of the RNA primer is shown above the graph; the size, indicated below in parenthesis, corresponds to the entire 28-nt RNA product. D RNA products synthesized by WT SARS-CoV-2 RdRp complex corresponding to different ratios of ATP and RTP that were supplemented to the reaction mixture. 30 µM UTP, 100 µM 3′dCTP, and 10 µM annealed RNA primer/template were also included in all reactions. On the right side of the gel are the likely products generated; “i” represents either an ATP or RTP incorporation event. RMP-containing RNA transcripts migrate more slowly than AMP-containing RNA transcripts because RTP is 24 daltons larger than ATP. The RNA synthesis reactions were performed n = 1. E The difference in observed masses of different RNA primers synthesized by WT (left) and S759A (right) SARS-CoV-2 RdRp complex in the presence of varying concentrations of ATP and RTP. The RNA size corresponding to 1:0 ATP-to-RTP concentration was used as the reference to calculate the size difference in response to increasing RTP concentration. The nucleotide concentrations that were supplemented to the SARS-CoV-2 RdRp complex for RNA synthesis, and the mass of the RNA, were determined by mass spectrometry and are shown in Tables S1 and S2 and Supplementary Figs. 1–4. The purified RNA product was analyzed using MALDI-TOF MS with a 2,4,6-trihydroxyacetophenone (THAP) matrix.

In this study, we used the same 40/20-nt genomic template/primer dsRNA as reported in previous structural and biochemistry studies^6,22. Here, positions 21 to 26 of the template sequence support the incorporation of two UTP molecules, followed by four successive RTP molecules, four ATP molecules, or a combination of both. We evaluated the pattern of RNA synthesis in response to different ATP to RTP concentrations (Fig. 1A, bottom). 10 µM of UTP and CTP were supplemented to the reaction to support the synthesis of a 32-nt RNA product. We observed that the first RTP incorporation at position 23 results in an intermediate product at position 26 (i-3), consistent with the delayed chain termination mechanism described earlier^5,21–24. However, additional termination products at site(s) 27 and 28, corresponding to i-3 sites for the successive RTP incorporations at positions 24 and 25, were not observed. This suggests—(I) multiple successive RTP incorporations may not occur when both RTP and ATP are present and/or (II) the i-3 delayed chain-termination does not hold true for successively incorporated RTPs. The 24 Da mass difference between RTP and ATP, Δ(RTP-ATP), is difficult to resolve using a traditional gel-based assay. Therefore, we evaluated the size of the RNA primer strand using a systematic MS study as outlined below. In this experiment, the SARS-CoV-2 RdRp carried out RNA synthesis in the presence of ATP and RTP at different molar ratios.

Mass spectrometry revealed site-specific RTP incorporations when ATP is present

Building on the above gel-based assay, we first optimized RNA synthesis conditions to achieve the desired 28-mer dsRNA at high yield for MS (Fig. 1B; see Methods). Here, the RNA strands were terminated with 3′-dCTP instead of CTP to avoid potential misincorporation, simplifying the readout to focus on the homopolymeric region of interest. Each RNA product was isolated via phenol-chloroform extraction, followed by precipitation in sodium acetate and isopropanol. The purified RNA products were then subjected to Matrix-Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF) MS to measure the precise mass of respective RNA. The labelled dsRNA substrate prior to RNA synthesis resulted in a spectrum with a single primary peak at a m/z of 6852 Da, 3 Da smaller than the predicted mass of the labelled 20-mer primer (Fig. 1C, left). Conversely, two less intense peaks are observed at m/z values of 6360 Da and 12,715 Da. The 6360 Da peak presumably represents a population of unlabeled RNA primer or a degraded product. The 12,715 Da peak corresponds to the RNA template with a calculated mass of 12,704 Da. Extension of the RNA primer with four ATP or four RTP alone produced 28-mer RNAs with observed masses of 9371 Da or 9458 Da, respectively, which are close to their expected masses (Fig. 1C, right). A mass difference of 87 Da between the two RNA products is within 7 Da of the theoretical value of 4Δ(RTP-ATP). Subsequently, the 20-nt RNA primers were extended to 28-nt RNAs under different competitive ATP to RTP ratios, and the masses of the primers were interpreted to evaluate the number of RTP incorporations (Fig. 1D, E, Supplementary Figs. 1, 2, Supplementary Table 1).

RTP was preferentially incorporated at the first site, even at an ATP to RTP ratio of 10:1; however, surprisingly, the three subsequent incorporations were ATP. Here, the observed RNA mass is higher by 1Δ(RTP-ATP) compared to the reference RNA with ATP. At a 3:1 ATP to RTP ratio, the analog was preferentially incorporated at two positions, and when the ratio was 1:1, there were three RTP incorporations. The mass differences in all three cases are near multiples of ~24 Da, Δ(RTP-ATP), suggesting that RTP incorporation is preferential rather than random. At ATP to RTP concentration of 1:3, only one or no ATP is incorporated, and at a 1:10 ratio of ATP to RTP, only RTP is incorporated at all four positions. It is important to note that our experiment does not show the preferred positions between sites 24 and 26 for RTP. Together, these data reveal that the incorporation of multiple RTP molecules is contingent upon the relative concentration of RTP to ATP.

Previous studies have demonstrated that the SARS-CoV-2 RdRp complex harboring the Nsp12-S759A substitution incorporates RTP less efficiently than ATP^20,28,29. It is therefore unsurprising that the mutant enzyme is less susceptible to RTP-induced delayed-chain termination under competitive conditions (Supplementary Figs. 2,3A). Consistent with this, our MS study of the RNA product generated by the S759A mutant RdRp showed a significantly reduced preference for RTP incorporation (Fig. 1E, right). Here, the RNA product from experiments with RTP-to-ATP ratios of 1:10, 1:3, 1:1, and 3:1 showed only ATP incorporations by the mutant enzyme (Supplementary Fig. 3, Supplementary Fig. 4, Supplementary Table 2). In fact, an increase in RNA molecular weight due to a single RTP incorporation event is only evident when RTP is at 10-fold molar excess to ATP. Together, the MS data clearly show that the S759A substitution alters the preferential incorporation from RTP to ATP.

Structures of wild-type and S759A SARS-CoV-2 RdRp complexes

To understand the molecular basis for the selective incorporation of RTP, we determined the structures of SARS-CoV-2 RdRp complexes incorporating RTP or ATP. In parallel, we also determined the analogous structure of the Nsp12-S759A mutant complex to understand the molecular mechanism of the resistance mutation (Fig. 2). The same 40/20-nt dsRNA was used for the structural studies. The wild-type (WT) and mutant complexes were incubated with UTP and ATP to catalytically extend the primer by seven nucleotides, complementing the template overhang sequence “AAUUUUA”. The primer is extended by seven nucleotides, two UTP, four ATP, followed by one UTP, and the 3′-end of this elongated 27-nt primer is translocated to the post-translocation site (P-site), leaving the empty nucleotide-binding site (N-site) ready to bind a CTP that was not supplied (Fig. 2B). In contrast, when incubated with UTP and RTP, both wild-type and the mutant complexes stalled after incorporation of five nucleotides, “UURRR”, and the 3′-end RMP was positioned at the pre-translocated N-site, preventing the binding of the next RTP (Fig. 2C); identical ATP and RTP incorporations were also observed for the S759A mutant complex. The structures of the wild-type RdRp with ATP (WT-AMP) and RTP (WT-RMP), and the S759A mutant with ATP (S759A-AMP) and with RTP (S759A-RMP) were obtained at 2.8, 2.9, 3.1, and 2.9 Å resolution, respectively (Table 1, Supplementary Fig. 5). Supplementary Fig. 6 compares the positioning of the 3’-end UMP or RMP in our structures, and Fig. 3A shows the 3’-end ribose ring conformations.

Fig. 2. Structures of WT and S759A mutant RdRp complexes.

A Structural model (left) and cryo-EM density map (right) of the WT RdRp/RNA complex (WT-AMP). RdRp complex subunits and Nsp12 domains are colored as follows: Nsp7 (yellow); Nsp8 (gold); Nsp12 domains thumb (green), palm (magenta), fingers (blue), NiRAN (pink), interface (grey); RNA template (cyan) and primer (red) strands. Active site region is boxed in grey. B Schematic representation of the dsRNA present in the WT-AMP and S759A-AMP structures. The bases of template and primer strands are colored in grey and incorporated nucleotides in yellow (UMP) and cyan (AMP). The WT (light blue) and S759A (wheat) are shown in cartoon representation. C Schematic representation of the dsRNA present in the WT-RMP and S759A-RMP structures, with protein color scheme and annotations as in (B); RMP is in magenta. The key amino acid residues in the active site regions are shown as sticks in (B, C).

Table. 1.

Single particle cryo-EM data and structure analysis statistics

	WT-AMP (EMD-54695) (PDB 9SAR)	WT-RMP (EMD-54693) (PDB 9SAP)	S759A-AMP (EMD-54694) (PDB 9SAQ)	S759A-RMP (EMD-54692) (PDB 9SAO)
Data collection and processing
Magnification	130,000x	130,000x	130,000x	130,000x
Voltage (kV)	200	200	200	200
Electron exposure (e-/Å2)	40	40	40	40
Defocus range (um)	−0.6 to −1.8	−0.6 to −1.8	−0.6 to −1.8	−0.6 to −1.8
Pixel size	0.9	0.9	0.9	0.9
Symmetry imposed	C1	C1	C1	C1
Initial particle images (no.)	654,460	1,237,567	1,002,711	738,355
Final particle images (no.)	109,207	145,169	93,716	167,051
Map resolution (Å)	2.8	2.9	3.1	2.9
FSC threshold	0.143	0.143	0.143	0.143
Local resolution estimation (Å)	6.2–2.8	6.3–2.9	6.2–3.1	6.5–2.9
Refinement
initial model used (PDB ID)	7DOK	9SAR	9SAR	9SAR
Model resolution (Å)	3.0	2.6	2.7	2.8
FSC threshold	0.143	0.143	0.143	0.143
Model resolution range (Å)	100 to 3.0	100 to 2.6	100 to 2.7	100 to 2.8
Map sharpening B factor (Å2)	-94	-98	-82	-105
Model composition
Non-hydrogen atoms	9330	8663	9305	8697
Protein residues	1084	1007	1082	1011
Nucleotide residues	31	25	31	25
Ligands	Zn:2	F86:3 Zn:2	Zn:2	F86:3 Zn:2
B factors (Å2)
Protein	74.86	45.04	25.78	54.58
Nucleotide	109.08	43.81	75.59	54.18
Ligands	82.91	19.27	26.64	17.24
R.m.s. deviations
Bond lengths (Å)	0.007	0.005	0.006	0.004
Bond angles (°)	0.937	0.793	0.873	0.618
Validation
MolProbity score	1.42	1.78	1.75	1.80
Clash score	4.30	5.46	3.21	5.97
Poor rotamers (%)	0.42	2.13	3.67	2.24
Ramachandran plot
Favored (%)	96.65	96.48	96.74	96.70
Allowed (%)	3.17	3.52	3.26	3.30
Disfavored (%)	0.19	0.00	0.00	0.00

Fig. 3. Effect of remdesivir incorporation on translocation.

A Conformations of the sugar rings of the last incorporated nucleotide occupying P-site is 3′-endo and 2′-endo in the WT-AMP (yellow) and S759A-AMP (green) structures, respectively. The conformation of the sugar ring of all RMPs (magenta) structures is in 3′-endo. B The available SARS-CoV-2 RdRp structures containing remdesivir at different positions in the primer strand are indicated in the scheme by their PDB IDs and the structures reported in this study are labeled. Remdesivir (RMP) position is indicated as a black dot and NMP as an x. C Minor groove distance is plotted for each bp position between +1 to −5 for all the structures listed in the scheme in B with curves following the same color code. D Distances between C1′-C1’ atoms of bps are plotted for each position between +1 and −5; curves follow the same color code of the scheme in (B). E Minor groove distances are plotted for each bp position between +1 to -5 for all the structures obtained in this study.

Structures of the SARS-CoV-2 RdRp complex with incorporated RMP or incoming RTP have been reported^19–22,24. For comparison of the structures, the upstream positions of the nucleotide analog are described in reference to the N-site, denoted as position +1, and the P-site is −1. The published structures have RTP positioned at +1, prior to incorporation (PDB: 7UO4) and after incorporation (PDB: 7BV2), incorporated RMP translocated by one nucleotide to −1 (PDB: 7C2K), and translocated by three nucleotides to −3 site (PDB: 7B3B, 7B3C and 7L1F); the upstream nth site is counted backward from -1 site (Fig. 2B, C, Supplementary Fig. 6). In comparison, our RMP complexes are stalled after three RTP incorporations, with RMPs occupying the −2, −1, and +1 positions of the primer; the experimental density for the RNA substrate unambiguously traced the sequence (Supplementary Figs. 7 and 8).

Remdesivir impairs translocation

Our WT-RMP structure is consistent with a prior pre-steady-state kinetic study, confirming the incorporation of three consecutive RMPs into the same 40/20 RNA duplex⁶. Unlike a follow-up structural study on this dsRNA with four incorporated RMPs²², we did not observe the incorporation of the fourth RTP. Therefore, our structure, which captures RMP at the −2 position, completes the sequence of structures with the first incorporated RMP occupying each position, starting from −3 to +1. The delayed chain termination mechanism specific to SARS-CoV-2 occurs when RMP occupies the −3 position, obstructing translocation to the −4 position^5,21–24. Unrelated to this mechanism of inhibition, the structural data, including ours, suggest that RMP resists translocation from the +1 position, and the incoming NTPs above a threshold concentration push the RNA up to three incorporations.

To assess the potential structural impacts of the RMP at different positions along the RNA-primer on the overall conformation of the RdRp complex, we aligned available structures based on respective Nsp12 Cα superpositions. The superposition of the four structures reported in this study aligns well with a root mean square deviation (RMSD) of ~0.5 Å, indicating no significant overall differences. These structures also align well with earlier reported structures (PDB IDs: 7BV2, 7C2K, 7B3B, 7B3C, 7L1F); Nsp12 Cα superimpose with an RMSD of ~0.5 Å for any pair of structures except for 7L1F, for which the RMSD is ~1.2 Å. The low RMSD among the structures suggests a well-defined RNA-binding tunnel of the RdRp that maintains its shape to guide the translocation of the RNA duplex rather than being perturbed in the translocation process.

A nucleotide ribose ring in dsRNA favors a 3′-endo conformation; however, it can transiently switch to a 2′-endo conformation³⁰, as observed in S759A-AMP structure. In contrast, as observed in all structures, the ribose rings of RMPs in the RNA are in a 3′-endo conformation, primarily restricted by the C1′-cyano group (Fig. 3A). This locked 3′-endo conformation may provide a lower entropic penalty for RTP binding at the N-site or +1 site³¹, contributing to RTP’s selectivity over ATP. However, this conformation could come at a cost, as this locked 3’-endo conformation may decrease the translocation efficiency of the R:U bp compared to an A:U bp. To understand the implications of R:U bp at different positions of the dsRNA substrate, we calculated the minor groove and backbone C1′-C1’ distances of each bp from positions −5 to +1 for the WT-AMP and WT-RMP structures (Fig. 3B). The RMP-incorporated structures that have actively translocated the dsRNA show an increase of the minor groove width around −1 site (Fig. 3C). For the backbone distances, the most significant difference is observed at position −4, where the A:U bp exhibits a change in distance of about 0.5 Å at C1′-C1′ atoms compared to the RMP structures (Fig. 3D). This suggests a difference in how the R:U would migrate compared to an A:U base pair. The minor groove widths in our RMP- and AMP-incorporated structures show two distinct trends, with little influence from the mutation (Fig. 3E). The structure with four incorporated RMPs to the same template/primer (PDB: 7L1F) appears to be an outlier from the remaining structures, including ours. Comparison of this −3 structure with our −2 RMP complex suggests that the RdRp must undergo larger conformational rearrangements to translocate the R:U base pair from −2 to −3, although such translocation remains achievable. Moreover, partially ordered density for the 4th RMP at +1 site in 7L1F structure (Supplementary Fig. 9) suggests less efficient incorporation of the 4^th RMP. Also, NTP incorporation at the −1 and −2 positions following RMP is hindered under pre-steady-state conditions⁶ or at a lower reaction temperature²⁶.

Successive incorporations of RMP are not favored

To ascertain that a pause occurs after three sequential RMP incorporation events, as observed in our structures, RNA synthesis by the WT and S759A SARS-CoV-2 RdRp complexes was monitored on the labelled dsRNA duplex (Fig. 4A). Here, the RdRp complex was equilibrated with RNA substrate and MgCl₂ before catalysis, which was initiated by the simultaneous addition of NTPs and heparin. The final concentration of heparin was 4 mg/mL to ensure the RNA products were synthesized under single-turnover conditions³². For the WT SARS-CoV-2 RdRp complex, the addition of 1 µM ATP and 1 µM UTP could readily generate the expected 27-mer RNA product, as CTP was not supplied in this experiment (Fig. 4B). Alternatively, substituting 1 µM RTP for ATP resulted in product formation at position 25 (−2) within the first 30 s, after which a conversion to a 26-nt product occurred. This transition signifies the rapid incorporation of the first three RTP molecules, followed by a slowdown in RNA catalysis due to the inefficient incorporation of the fourth RTP molecule, as reflected in 7L1F structure (Supplementary Fig. 9). Indeed, similar previous observations indicate that the fourth RTP was incorporated much more slowly than the first three^6,22. Increasing RTP concentration to 10 µM accelerated the incorporation of the fourth RTP at position 26 within the first 30 s. Subsequently, the subtle formation of a 27-mer product was observed when UTP concentration was increased to 10 µM. The combination of our structural and biochemical data, along with previous kinetic data, suggests that the incorporation of the first three RMP is highly favoured. However, the incorporation of the next two nucleotides is possible when the RTP and/or NTP concentration is increased.

Fig. 4. RNA synthesis assay of WT and S759A SARS-CoV-2 RdRp complexes.

A The RNA template/primer supporting four successive incorporations of ATP or RTP from nucleotide positions 23 (i) to 26 (i-3) in our study. B Migration pattern of RNA products synthesized by WT SARS-CoV-2 RdRp complex. The 5′-6-FAM labelled primer serves as the marker. The Heparin control lane is the resulting product when the SARS-CoV-2 RdRp complex is incubated with 4 mg/mL heparin prior to the addition of the RNA duplex and ATP/UTP (10 µM). This reaction was equivalent to the longest time point tested (600 s). No product formation indicates that the heparin concentration used is sufficient to monitor RNA synthesis under single-turnover conditions. The 27-nt RNA product generated in the presence of ATP alone migrates the same distance as a 26-nt RNA product generated in the presence of RTP alone. This is likely due to differences in the chemical structures of ATP and RTP, resulting in different mass-to-charge ratios and subsequent migration patterns. C Migration pattern of RNA products synthesized by S759A SARS-CoV-2 RdRp complex. Under reaction conditions identical to B, the S759A complex does not incorporate the 4th RTP molecule. The RNA synthesis reactions were performed n = 1, and the original gel is in Supplementary Fig. 10.

In contrast, the S759A SARS-CoV-2 RdRp complex could not readily generate the expected 27-mer RNA product in the presence of 1 µM ATP and 1 µM UTP, as indicated by the formation of intermediate products at earlier time points (Fig. 4C, ATP-containing reactions). Synthesis of 27-mer RNA products could be rescued with increased reaction time and/or nucleotide concentration. This observation agrees with S759A imposing a global cost on RNA catalysis by SARS-CoV-2 RdRp^28,29. Notably, this mutation appears to exacerbate the translocation deficiency during sequential RTP incorporation events (Fig. 4C, RTP-containing reactions). Here, 10 µM of RTP is required to support consecutive incorporation events at positions 24 and 25 (i-2). However, the 26-nt product (i-3) is not formed with increased RTP concentration. Therefore, the stalling and likely dissociation at the i-2 position are due to reduced RTP incorporation efficiency^28,29, which is compounded by an overall reduction in processive RNA catalysis.

Structural impact of the S759A mutation on NTP incorporation and RNA translocation

S759 is located in the conserved catalytic hairpin S759-D760-D761 (SDD) within motif C and is unique to the Nidovirales order; other positive-sense ssRNA virus RdRps have a glycine (GDD) instead of this serine³³. Another notable difference, a conserved alanine (A547; SARS-CoV-2 numbering) in motif F of coronavirus RdRp, is a glutamate in positive-sense RNA viruses with smaller genomes, such as poliovirus. This glutamate with a conserved arginine (R555; SARS-CoV-2 numbering) helps recruit the correct NTP by poliovirus RdRp^34,35. A mutagenesis study³⁶ concluded that (I) increased R555 flexibility due to missing glutamate in SARS-CoV-2 reduces the NTP selection fidelity and in turn increases the RNA synthesis efficiency required to replicate the large ~30 kb coronavirus genome and (II) the interaction of S759 at the active-site hairpin with the 2′-OH of the primer 3′-end nucleotide provides an alternative fidelity checkpoint.

We compare the structures of S759A-RMP and S759A-AMP complexes with their respective wild-type complexes (Fig. 5, Supplementary Fig. 6). The primer 3′-end in the S759A-RMP and the WT-RMP complex is locked at the N-site following three RTP incorporations. The structure comparison showed reduced interaction of RMP at N-site in the mutant structure compared to that in the WT-RMP structure (Fig. 5A). Moreover, our S759A-AMP complex with its primer 3′-end nucleotide ribose ring positioned over the mutation site reveals important differences from the WT-AMP structure. In the WT-AMP structure, the ribose ring of the primer 3′-end nucleotide has a 3′-endo conformation, and, in agreement with previously published SARS-CoV-2 RdRp complex structures, the 2′-OH group is recognized by the H-bond interaction with the Oγ atom of S759^20,24. In contrast, the H-bond is lost in the S759A-AMP structure. Interestingly, in the mutant complex, the ribose ring at the primer 3′-end adopts a 2′-endo conformation (Fig. 5B) and is also shifted toward the catalytic aspartate D760 when compared to the wild-type complex (Fig. 5C); the primer terminal 3′-OH is shifted by ~1.3 Å, and 2′-OH is shifted by ~2.3 Å. These structural findings explain the impact of the S759A mutation on SARS-CoV-2 RdRp fidelity as discussed later.

Fig. 5. Impact of the S759A mutation on the substrate interaction, and processive RNA synthesis under single turnover conditions by WT and S759A SARS-CoV-2 RdRp complex.

A Interactions of RMP (magenta) in the N site in the WT-RMP (left, grey) and S759A-RMP (right, cyan) structures. For clarity, only RMP and the interacting residues are shown. Distances are in Å and indicated by dotted lines; the red dotted lines indicate the interactions that are lost in the S759A mutant and in black the ones that differ between the two structures. B Hydrogen bond network of the primer 3′-end nucleotide at P-site in the WT-AMP (left, grey) and in S759A-AMP (right, green) structures. The active site hairpin residues (SDD/ADD) and the U27 are depicted as sticks, with all the distances below 4.0 Å indicated as dashed lines. C Overlap of the P strand 3′-OH between WT-AMP (grey) and S759A-AMP (green) structures. The U:A bp and the SDD/ADD residues are shown as sticks. The shift of U27 between the two structures is indicated as dashed lines with the distances annotated in Å. D The sequence of the RNA template/primer supporting RNA synthesis in our experiment is at the top. The addition of ATP, CTP, and UTP enables the synthesis of a 32-mer product. Below is the migration pattern of RNA products synthesized by the WT (left) or S759A (right) SARS-CoV-2 RdRp complex. The Heparin control block is the resulting product when the SARS-CoV-2 RdRp complex is incubated with 4 mg/mL heparin prior to the addition of the RNA duplex and nucleotides. No product formation at any time point indicates that the heparin concentration used is sufficient to monitor RNA synthesis under single-turnover conditions. Red asterisks indicate natural intermediate products formed early during RNA synthesis that are extended over time; in the presence of heparin, these products suggest that natural stalling occurs without dissociation. The uncropped gels are in Supplementary Fig. 11. E Quantification of the RNA synthesis reaction products in (D) shows the lower catalytic efficiency of the mutant RdRp. The plotted data points for each replicate (n = 3) are shown in the graph, and the mean value across three replicates is represented by their respective line.

To evaluate the catalytic impact of the mispositioned primer, we considered processive elongation by the WT and S759A complex (Fig. 5D, E). These reactions were initiated by the simultaneous addition of ATP, CTP, UTP, and heparin. For the WT RdRp complex, intermediate products of ~24-nt form at earlier reaction timepoints (5 and 15 s) before extension to the anticipated 32-nt RNA product (Fig. 5D, left, red stars). A prior study, using the same genomic template sequence, also observed this pattern, where intermediate products form at 10 s with the WT SARS-CoV-2 RdRp complex³⁵. Due to the presence of heparin, we can be certain that this observation is not due to the dissociation of the enzyme but instead suggests that the WT RdRp complex changes speed during RNA synthesis. For the S759A RdRp complex (Fig. 5D, right), RNA catalysis is slower compared to the WT (Fig. 5E). The repositioning of the primer 3′-end with respect to the catalytic hairpin in the S759A-ATP structure (Fig. 5C) likely contributes to the decreased catalytic rate, and the reduced viral fitness observed in cell culture²⁸.

Discussion

Site-specific preference for RTP incorporation

To elicit an antiviral effect, NAs must efficiently compete with natural NTP pools for incorporation by a viral polymerase. However, several factors can diminish an NA’s efficacy, which are often not captured in simplified biochemical assays. ATP concentration in the cell has been reported to range from 1 to 10 mM³⁷. Conversely, the intracellular concentration of RTP is much lower, ~10 µM^38,39. Even though biochemically, RTP incorporation is preferred over ATP^5,6, the comparatively higher concentration of ATP may limit the number of RTP incorporations during minus-strand RNA synthesis. In agreement with a previous study⁷ our MALDI-TOF MS analysis demonstrates that RTP is preferentially incorporated at a single site when ATP is present at a higher concentration; however, successive incorporations appear unfavorable. Together, these assays demonstrate that RTP is competitively superior to ATP, which is reflected in the therapeutic efficacy of RDV against SARS-CoV-2. Structurally, the ribose ring of RMP has a relatively fixed 3′-endo conformation at all positions compared to AMP, where the ribose ring can switch between 3′-endo and 2′-endo (Fig. 3A). The locked 3′-endo conformation of RMP appears to fit better to the N-site, suggesting an entropically favorable binding of RTP over ATP to an empty N-site. However, RTP reduces subsequent NTP incorporation efficiency because the required primer 3′-end translocation from the N-site is hindered by RTP’s less adaptable sugar ring and the 1′-cyano group. Therefore, multiple sequential RMP residues are likely to compound inefficient translocation. This explains why both WT- and S759A-RMP structures are captured at the −2 position and why partially ordered density at the −3 position was previously observed with the same RNA substrate²².

Potential RMP sites in a synthesized negative-strand RNA

To evaluate the greater impact of the translocation deficiency introduced by the R:U bp, we considered the SARS-CoV-2 genome (accession: NC_045512.2) and the number of opportunities for RTP incorporation during transcription. Out of the ~30 kb genome, UMP comprises 9594 bases of which 2439 sites are followed by a subsequent UMP, which will favor ATP incorporation and, likely reducing the chances for RTP incorporation by approximately 25%. Of those, four successive templating uridines appear 237 times across the SARS-CoV-2 genome. Furthermore, our MALDI-TOF data show that RTP incorporation is also hindered when RMP occupies the -2 and -3 positions. There are 5597 instances where one or two bases separate two UMPs in the viral genome. Overall, the sequence-dependent translocation effect could decrease the opportunities for RTP incorporation by about 84% when the ATP concentration is 10x or higher than RTP. It should be noted that this extrapolation of RMP-induced translocation deficiency is based on four successive templating uridines. It is likely that different local templating sequences may alter successive RTP incorporation events. Additionally, the presence of the 3′-to-5′ proofreading ExoN will further reduce the number of RTP incorporations during minus-strand synthesis^16,40. Together, these observations provide a holistic perspective on RTP, encompassing how its chemical modifications, although necessary for its antiviral effect, can negatively and selectively influence incorporation events.

The structural implications of S759A on RTP resistance and RNA synthesis

Two independent in vitro studies have shown that the S759A substitution in Nsp12 confers ~10-fold resistance to RDV^28,29. Moreover, the analogous mutation in murine hepatitis virus, an orthologous coronavirus, also conferred resistance²⁸. Previous biochemical and structural studies have demonstrated that S759 plays an integral role in the efficient incorporation of RTP^20,28,29. The structural reorganization, primarily of the Nsp12 617–624 loop away from the RMP (Fig. 6), might weaken the stabilization of RMP, which may partially explain the lower affinity of this mutant for remdesivir. The primer 3′-end nucleotide at P-site in WT-ATP structure is significantly impacted by the S759A mutation (Fig. 5B, C), and the repositioned 3′-end appears to be less favorable for next nucleotide addition.

Fig. 6. Shift of the palm domain loops in the S759A structures.

A Overlap of the palm domain segments between the WT-AMP (gray) and S759A-RMP (cyan). B Overlap of the palm domain segments between the WT-AMP (gray) and S759A-AMP (green). Residues of the active site hairpin (758-761) are labelled and shown as sticks. The loops showing the largest movements are depicted, with Cα represented as sphere. C Measured change in the local residue (left column) Cα distance (Å) between WT-AMP complex and the S759A-RMP (middle column) or the S759-AMP complex (right column). Changes greater than 0.5 Å distance are bolded.

Resistance to RDV is often reported in immunosuppressed patients. Resistant mutations in Nsp12 include, but are not limited to, V792I, M794I, E796D/K/G, C799F/Y, E802D, and T803I^41–46. Notably, these mutations surround K798, which interacts with the γ−phosphate of the incoming NTP²⁰. On their own, these substitutions demonstrate little to no resistance in cell culture^46,47. However, an additive effect has been observed for V792I when combined with S759A^28,47. Structurally, the conserved active-site hairpin (SDD) conformation remains unaffected by the S759A mutation (ADD), regardless of the observed shift in the primer 3′-end. D618, Y619, D623, and K798 are key residues in this region that coordinate the incoming NTP²⁰. We observed the largest shift in Cα distance at D618 ( ~ 1.5 Å) and K798 ( ~ 1.8 Å) for the mutant in both S759A-AMP and S759A-RMP structures (Fig. 6); thus, the structural change in this mutant is observed irrespective of whether the N-site is occupied or empty and may have implications for associated mutations. It is likely that the repositioning of K798 and D618 are compensatory readjustments of the active-site region in response to the S759A mutation induced positioning of the primer 3’-end.

The S759A substitution exacerbates delayed-chain termination, thereby potentially increasing the likelihood of excision²⁹. In this instance, the resistant phenotype is amplified by augmented stalling associated with the RMP residue buried 3 nucleotides upstream, which could promote RdRp backtracking and potentially increase exposure of the incorporated analog to ExoN-mediated excision^7,29,40,48. Substituting S759 with glycine incurs a fidelity cost, implicating this residue in an alternative fidelity checkpoint³⁶. Given that the primer terminus is perturbed in the S759A-ATP structure, it is intriguing to speculate that the conserved S759 plays an important role in coordinating coronavirus RNA synthesis and the repositioning of the 3′-end of the RNA primer at the RdRp active site may facilitate backtracking for ExoN cleavage. Notably, ExoN activity has yet to be investigated in the context of SARS-CoV-2 RdRp complexes harboring the S759A or S759G substitution.

SARS-CoV-2 Nsp12 S759A vs HIV RT M184V/I mutations

In HIV RT, a viral DNA polymerase, M184 is the positional equivalent of S759 in SARS-CoV-2 Nsp12. The RT mutation M184V/I confers resistance to nucleoside drugs containing a L-ribose ring, such as 3TC and FTC⁴⁹. Structures and biochemical studies have explained that the M184V/I mutation introduces a wider bifurcated sidechain that results in a steric conflict with the altered L-ribose ring conformation of 3TC/FTC⁵⁰, and hepatitis B (HBV) DNA polymerase also develops 3TC/FTC resistance by a similar mechanism⁵¹. Conversely, S759A resistance is caused by a loss of interaction, which is likely in response to the very efficient rates of incorporation for RTP. These contrasting resistance mechanisms illustrate how active-site mutations can confer resistance through different molecular mechanisms. This underscores how structural and biochemical properties of a given nucleotide analog can elicit various antiviral resistance mechanisms.

Implications of this study

Overall, this study integrates biochemistry, MS, and cryo-EM findings to provide a comprehensive view of nucleotide analog incorporation by the SARS-CoV-2 RdRp complex. The finding that SARS-CoV-2 RdRp preferentially incorporates RTP even in the presence of higher ATP concentration is a favorable characteristic of a nucleoside drug. However, an incorporated RTP molecule can immediately impede translocation, offering a new perspective on the molecular basis of non-obligate chain termination and highlighting the kinetic challenge a nucleotide analog must overcome. Understanding this translocation barrier is especially pertinent for evading coronavirus ExoN-mediated excision. RDV is also active against non-segmented negative-sense ssRNA viruses, including Nipah virus, Ebola virus, and respiratory syncytial virus. Biochemical studies of these viral polymerases have shown that RTP incorporation can elicit different patterns of delayed-chain termination, and multiple RTP incorporations markedly enhance primer-strand inhibition^8,9,52,53. In all instances, RTP inhibition can be overcome by increasing the NTP substrate concentration, allowing RMP to be positioned in subsequent viral genome strands^{7,8,21,22,24,29}. The sequence- and concentration-dependent selective incorporation of RTP offers additional insight into the design of NAs against ssRNA viruses.

The loss of key contacts caused by the S759A substitution clarifies its role in remdesivir resistance and underscores the essential contribution of this conserved serine residue to the catalytic efficiency required for coronavirus replication. Together, these insights advance our mechanistic understanding of RdRp inhibition and inform the rational design of next-generation nucleotide analogs. The unique experimental platform combining an optimized enzymatic synthesis of RNAs from a heterogeneous mixture of RTP and ATP, and reliable MS analyses of these RNAs can be expanded to other systems and NAs.

Materials and methods

Protein cloning, expression and purification

Plasmid pRSFDuet-1 (nsp8-nsp7)(nsp12) expressing the wild-type form of SARS-CoV-2 RdRp was obtained from Addgene (ref. 165451). The S759A point mutation was introduced by PCR using 2 overlapping primers (P1_S759A 5′- GGCATCATCTGCCAGAATCATCATGCTGAAG-3’ and P2_S759A 5′- ATGATTCTGGCAGATGATGCCGTTGTTTGC-3’). One ng of wild-type plasmid was used as template, and the PCR reaction was performed by adding 0.5 μL of Q5 High-Fidelity DNA Polymerase (New England Biolabs) following the manufacturer’s protocol in a final volume of 50 μL. PCR products (5 μL) were loaded on 1% TAE agarose gel for quality control, then 20 μL were digested with Dpn1 (Thermo Scientific) for 1 h at 37 °C. Ten μL of the digested product was used to transform DH5α E. Coli (Invitrogen) for plasmid selection and amplification. The presence of the S759A point mutation was confirmed by sequencing (Macrogen). Both wild-type and S759A mutant of SARS-CoV-2 RdRp were expressed and purified as described in Madru et al., 2021⁵⁴.

RNA synthesis assay

The template and primer used in the gel-based RNA synthesis studies were 5′-phosphorylated and purchased from Dharmacon. The template (5′-CUAUCCCCAUGUGAUUUUAAUAGCUUCUUAGGAGAAUGAC-3’) and primer (5′-6-carboxyfluorescein-GUCAUUCUCCUAAGAAGCUA-3’) were annealed by mixing at a 1:1.5 molar ratio (primer to template) in buffer (50 mM Tris-HCl pH 8 and 50 mM NaCl) and heating to 95°C before allowing to cool to room temperature.

All reactions were performed at room temperature using the purified SARS-CoV-2 complex described above. Briefly, 2 µM RdRp complex and 100 nM RNA substrate were incubated in reaction buffer (25 mM Tris-HCl pH 8, 5 mM MgCl₂) for 15 min. RNA synthesis was initiated by the addition of nucleotides (various concentrations) and heparin (4 mg/mL) and stopped at indicated time points using equal parts formamide/EDTA (50 mM). The reaction products were incubated at 95 °C for 5 min and then resolved by 20% Urea-PAGE, and the fluorescent signal was scanned using Typhoon PhosphorImager (Cytiva). The data were analyzed using GraphPad Prism 7.0 (GraphPad Software, Inc., https://www.graphpad.com).

MALDI-TOF mass spectrometry

Generation of the dsRNA for MS analysis was performed similarly to the RNA synthesis assay described above, with slight variations. Here, 2 µM RdRp complex and 10 µM RNA substrate were incubated in the reaction buffer for 15 min. RNA synthesis was initiated by the addition of nucleotides (various concentrations, shown in Tables S1 and S2) and allowed to proceed for ~16 h. The reaction was stopped with an equal volume of phenol-chloroform. The aqueous layer was then passed through a Micro Bio-Spin® P-30 column (Bio-Rad, #7326223) that was equilibrated with H₂O. The collected flow-through underwent precipitation with sodium acetate (0.3 M final) and isopropanol before being washed twice with 70% ethanol. The pellet was resuspended in 10 µL Milli-Q H₂O to achieve a final concentration of ~15 µM. The RNA synthesis products were then subjected to MALDI-TOF MS with a 2,4,6-trihydroxyacetophenone (THAP) matrix (Dr. Randy Whittal and Dr. Joseph Utomo, Mass Spectrometry Facility, Edmonton, AB, Canada).

Assembly of RdRp/RNA complexes and substrate incorporation for cryo-EM

Template (5′- CUAUCCCCAUGUGAUUUUAAUAGCUUCUUAGGAGAAUGAC-3’) and primer (5′- GUCAUUCUCCUAAGAAGCUA-3’) RNAs were purchased from Integrate DNA technologies, resuspended separately in H₂O and mixed in a 1:1 molar ratio in TE buffer. The mixture was incubated at 60 °C for 5 minutes then cooled down at room temperature to favor prime-template annealing. RdRp (wild-type or S759A) purified by anion exchange (HiTrap Q HP, Cytvia) was incubated with annealed RNA (in a complex:RNA molar ratio of 1:1.5) and UTP (50x molar excess to RNA) to incorporate the first two nucleotides on the primer strand, in a buffer containing 20 mM HEPES pH 7.5, 75 mM NaCl, 5 mM MgCl₂ and 5 mM DTT. Complex was incubated at room temperature for 10–15 min and then purified by SEC on a Superdex 200 Increase 10/300 GL column connected to an AKTA PURE 25 FPLC system (GE Healthcare Life Sciences) maintained at 6 °C. Fractions of interest were pooled and concentrated using a 10-kDa-cutoff Amicon ultra-centrifugal filter (Merck Millipore) and stored at −80 °C prior to further use. 3 µM of SEC-purified RdRp/RNA complex (wild-type or S759A) was used for substrate incorporation by adding 300 µM of either ATP or RTP and 150 µM UTP. Samples were incubated at room temperature for 10–15 min, then kept on ice prior to grid preparation.

Cryo-EM grid preparation and data collection

Vitrification RdRp/RNA complexes were done on Quantifoil R 1.2/1.3 holey carbon grids (Au300) using a Leica EM GP (Leica Microsystems). The grids were glow-discharged for 45 s at 25 mA, with the chamber pressure set to 0.3 mbar (PELCO easi-Glow; Ted Pella). Glow-discharged grids were mounted in the sample chamber of a Leica EM GP at 8 °C and 95% relative humidity. Optimized grids for all samples were obtained by pipetting 3 μL of sample at ~3 μM. The sample (in 20 mM HEPES pH 7.5, 75 mM NaCl, 5 mM MgCl₂, 5 mM DTT) was incubated on the grids for 10 s before back-blotting for 12 s using two pieces of Whatman Grade 1 filter paper, and the plunge-freezing was carried out by dipping the grids in liquid ethane at a temperature of −172 °C. The grids were clipped and mounted on a 200-keV Glacios cryo-transmission electron microscope (Thermo Fisher) with an autoloader and Falcon 4i direct electron detector equipped with a Selectris energy filter as installed in our laboratory. High-resolution datasets were collected on the Glacios using EPU software version 3.5.1 (Thermo Fisher). Electron microscopy data were recorded as movies in counting mode at a nominal magnification of ×130.000 yielding a pixel size of 0.9 Å. The total exposure time was 8.13 s for WT-AMP, 8.18 s for WT-RMP and S759A-AMP and 15.56 s for S759A-RMP for a total dose of 40 e⁻¹/A² for all datasets, fractionated to 2043 frames for WT-AMP, 2052 frames for WT-RMP, S759A-AMP and S759A-RMP. The data collection parameters for all structures are listed in Supplementary Table 1.

Cryo-EM data processing

An overview of the workflow used to process each structure is provided in Supplementary Fig. 5. For each dataset, individual video frames were motion-corrected and aligned using MotionCor2⁵⁵ as implemented in the Relion 4 package⁵⁶ and the contrast transfer function parameters were estimated by CTFFIND-4⁵⁷. The particles were automatically picked using the reference-free Laplacian-of-Gaussian routine in Relion 4. The picked particles were cleaned by cycles of two-dimensional (2D) and 3D classifications. A density map for RdRp (EMD-30210) was blurred to 40 Å resolution and used as the reference for initial 3D classifications to eliminate partially disordered and incomplete particles. The homogeneous particle sets from the best 3D class were used to calculate gold-standard 3D auto-refined maps and corresponding masks. For each structure, the particles were repicked at the 3D auto-refinement stage and classified further within its mask. Particles were extracted with a box size of 256 pixels. Final 3D classification generated a distinct class (Supplementary Fig. 5). The final set of particles for each structure was used to calculate gold-standard auto-refined maps, further improved by Bayesian polishing and contrast transfer function refinement. All data processing steps were carried out using Relion 4. The local-resolution maps were calculated using ResMap and the orientation plots were generated by Relion 4.

Model building

Data processing yielded 2.8 Å, 2.9 Å, 3.1 Å, and 2.9 Å density maps for WT-AMP, WT-RMP, S759A-AMP and S759A-RMP, respectively, and these were used to fit the atomic models for the respective structures. Previously published SARS-CoV-2 RdRp in complex with RNA (PDB ID: 7DOK) was used as reference for building the model of the WT-AMP structure and then this model was used as initial reference for the other three structures. All model building was carried out manually using COOT⁵⁸ coupled with iterative rounds of real-space structure refinement using Phenix 1.19.2⁵⁹. In all structures, a single copy of Nsp8 could be modelled with residues 191-198 of this subunit missing from all four structures. In addition, residues 1-75 are missing from WT-AMP, 1-77 from WT-RMP, 1-76 from S759A-AMP and 1-77 from S759A-RMP structures. Nsp7 residues 63-83 are missing in WT-AMP and S759A-AMP, 1 and 62-83 in WT-RMP, 1 and 66-83 in S759A-RMP structures. Nsp12 missing residues are the following: 1-5, 107-110, 898-912 and 931-932 in the WT-AMP structure; 1-28, 53-66, 99-114, 898-912 and 931-932 in the WT-RMP structure; 1-6, 107-110, 898-912 and 931-932 in the S759A-AMP structure; 1-28, 55-68, 100-114, 898-912 and 931-932 in the S759A-RDV structure. All structure figures were generated using PyMol (https://pymol.org/2/), Chimera⁶⁰ and ChimeraX⁶¹.

Statistics and reproducibility

No statistical analysis was performed; replicates are described in the figure legend. All reaction samples that underwent MS analysis were performed once (n = 1). Quantification of RNA synthesis patterns in Fig. 5E was based on three independent experiments (n = 3).

Reporting summary

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

Author contributions

Conceptualization: C.J.G., M.G., K.D.; Investigation and Methodology: C.J.G. and H.W.L.; Biochemistry: C.J.G.; MS: C.J.G.; Protein purification: M.F.O.; Cryo-EM and structural studies: M.F.O., Q.G., and B.D-W.; Data analysis: C.J.G., H.W.L., M.G.; Structural biology: M.F.O., Q.G., and K.D.; Funding acquisition: K.D. and M.G.; Project administration and supervision: K.D. Original draft written by: C.J.G., M.F.O., K.D.; Reviewed and edited by all authors.

Peer review

Peer review information

Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: Joanna Timmins and Laura Rodríguez Pérez. A peer review file is available.

Data availability

The data that support this study are available from the corresponding author upon reasonable request. The coordinates and cryo-EM density maps for the structures WT-AMP, WT-RMP, S759A-AMP, and S759A-RMP are deposited with PDB accession codes/EMDB codes 9SAR /EMD-54695, 9SAP/EMD-54693, 9SAQ/EMD-54694, and 9SAO/EMD-54692, respectively. All MS data can be obtained from the Open Science Framework (OSF) online repository at 10.17605/OSF.IO/YDRT3.

Competing interests

The authors declare no conflicts of interest.

Supplementary information

The online version contains supplementary material available at 10.1038/s42003-026-09844-z.