An exonuclease-resistant chain-terminating nucleotide analogue targeting the SARS-CoV-2 replicase complex

Abstract

Nucleotide analogues (NA) are currently employed for treatment of several viral diseases, including COVID-19. NA prodrugs are intracellularly activated to the 5′-triphosphate form. They are incorporated into the viral RNA by the viral polymerase (SARS-CoV-2 nsp12), terminating or corrupting RNA synthesis. For Coronaviruses, natural resistance to NAs is provided by a viral 3′-to-5′ exonuclease heterodimer nsp14/nsp10, which can remove terminal analogues. Here, we show that the replacement of the α-phosphate of Bemnifosbuvir 5′-triphosphate form (AT-9010) by an α-thiophosphate renders it resistant to excision. The resulting α-thiotriphosphate, AT-9052, exists as two epimers (R_P/S_P). Through co-crystallization and activity assays, we show that the S_p isomer is preferentially used as a substrate by nucleotide diphosphate kinase (NDPK), and by SARS-CoV-2 nsp12, where its incorporation causes immediate chain-termination. The same -S_p isomer, once incorporated by nsp12, is also totally resistant to the excision by nsp10/nsp14 complex. However, unlike AT-9010, AT-9052-R_P/S_P no longer inhibits the N-terminal nucleotidylation domain of nsp12. We conclude that AT-9052-S_p exhibits a unique mechanism of action against SARS-CoV-2. Moreover, the thio modification provides a general approach to rescue existing NAs whose activity is hampered by coronavirus proofreading capacity.

Graphical Abstract

Introduction

In late 2019, the emergence of severe acute respiratory syndrome coronavirus type 2 (SARS-CoV-2, https://covid19.who.int/) started a worldwide crisis. This large genome, (+)RNA virus, first detected in Wuhan, China, belongs to the Sarbecovirus genus of the Coronaviridae family, order Nidovirales (1). Prevention of infection is generally achieved through wearing masks, social distancing, and vaccination; with several vaccines developed in record time. However, potent antiviral drugs are still needed for the treatment of infections and/or contact cases. Early intervention could curb viral load and reduce both disease course and transmission.

Among the existing drugs evaluated for treatment of SARS-CoV-2 are several nucleoside/tide analogues (NAs) (2–4). NAs are administered to patients as prodrugs, which are activated by various host enzymes into a 5′-triphosphate (5′-TP) form. The 5′-TP NA lures the viral RNA-dependent RNA polymerase (RdRp, CoV nsp12) and associated replication/transcription complex (RTC) for incorporation into the nascent viral RNA genome. Depending on the specific NA structure, incorporation of the activated 5′-TP form can cause either immediate or delayed chain termination, resulting in aborted/truncated RNA genomes (RNA ‘killers’). Alternatively, some NAs may be incorporated without significantly stalling synthesis, disrupting other biological functions of the RNA genome (RNA ‘corruptors’) (2,3). Incorporated NAs may disrupt RNA secondary structures, RNA-protein interactions, and other transcriptional/translational regulatory elements (chemical corruption), or increase viral mutation rates during further rounds of RNA synthesis, pushing the virus to error-catastrophe and extinction (genetic corruption) (3). NAs currently used in the treatment of RNA viral infections act through a variety of mechanisms, which are sometimes not easily predicted through examination of their chemical structure alone. Several of these terminate synthesis immediately, despite the presence of a 3′-OH (e.g. Sofosbuvir) and are therefore classed as ‘non-obligate chain terminators’. Other NAs have been reported to cause delayed-chain termination and/or transcriptional stalling (e.g. Remdesivir and 4′-FIU) (5,6), or viral lethal mutagenesis (e.g. Molnupiravir) (7–9). It must be noted that several of these mechanisms can occur simultaneously and contribute to the overall antiviral effect.

Unlike most RNA viruses, coronaviruses have evolved the capability to edit mistakes during copying of their genetic information through the acquisition of a 3′-to-5′ exonuclease (ExoN), located on nsp14 and activated by nsp10 (10–13). Upon mismatch insertion, the nsp14/nsp10 heterodimer complex excises the incorrect base, and synthesis resumes (14). In the absence of selective pressure, mutational inactivation of ExoN activity results in a ∼15-fold increase in mutation frequency throughout the CoV genome (12,15).

This natural proofreading and repair capacity thus poses an additional challenge for the treatment of CoVs with NAs, which must be considered during drug design and mechanism of action studies. One can logically propose that the ability of the nsp14-ExoN to access and remove an NA is related to the stall-time of the RTC following incorporation. For example, the 5′-TP of Molnupiravir is efficiently inserted into viral RNA and rapidly extended by the SARS-CoV-2 RTC (8). The potent, mutagenic effect of Molnupiravir suggests it is able to escape ExoN-mediated proofreading, as has been suggested for the related B-CoV, murine hepatitis virus (MHV) (7). In contrast, most chain-terminating NAs have so far shown limited success for the treatment of SARS-CoV-2 infections. Recently however, the guanosine analogue Bemnifosbuvir (AT-527) has shown therapeutic potential against COVID-19 (16,17). Like Sofosbuvir, Bemnifosbuvir possesses a 2′-fluoro, 2′-methyl modified ribose. The only difference is the nucleobase; uracil for Sofosbuvir and N2-amino-N6-methylamino-purine (metabolized to guanine) for Bemnifosbuvir. Its 5′-triphosphate form, AT-9010, is incorporated into RNA as a substitute for GTP, causing immediate termination of synthesis. The molecular mechanism, as determined by Cryo-EM and enzymatic studies, shows that the 2′-methyl group disrupts the alignment of the next nucleotide, blocking further nucleotide incorporation (17). Similarly, the 5′-TP form of Sofosbuvir triphosphate also causes immediate chain termination, but lacks significant activity against coronaviruses (16). The potency of Bemnifosbuvir can likely be attributed to the dual inhibition of the N-terminal pseudokinase-like domain of nsp12 (known as the NiRAN) (17), inhibiting an essential nucleotide-transferase activity, now shown to be involved in RNA capping (18).

Interestingly, AT-9010 shows a modest reduction in excision-rate by SARS-CoV-2 nsp14/nsp10, although complete excision is expected to occur (17). Suppressing the ExoN activity of the SARS-CoV-2 RTC may therefore potentiate the activity of this drug—along with many other nucleoside analogues. RdRp–ExoN combination therapies thus represent a promising research avenue for the treatment of COVID-19.

Rather than directly targeting nsp14 ExoN activity with a distinct inhibitor, we explored an alternative approach. Phosphorothioate linkages, in which the non-bridging oxygen of the phosphodiester bond is substituted by a sulfur atom, have been widely used in oligonucleotide research, in part due to their improved stability and resistance to ribonuclease degradation (19). Targeting the viral polymerase with activated 5′-α-thiotriphosphate NAs, may increase their potency by reducing their ability to be excised once incorporated in the RNA chain (20).

To this endeavor, we synthesized AT-9052, the 5′-α-thiotriphosphate version of AT-9010. Of note, the oxygen-to-sulfur substitution introduces chirality on the phosphorus atom, and thus AT-9052 exists as two isomers (R_P and S_P). Many enzymes show a strong stereoselectivity for a certain isomer, which must be taken into consideration for the development of these compounds. Stereospecificity does not only impact potential viral protein targets, but also the host enzymes involved in the activation pathway.

The R_P and S_P isomers were separated and purified using reverse phase chromatography, and the absolute configuration at the α-phosphorus atom was determined through enzymatic assays and co-crystallisation. This shows that the AT-9052-S_P isomer is stereoselectively accepted as a substrate by NDPKb, the protein presumably responsible for the final, di- to tri-phosphorylation step in the activation pathway. This same isomer is also preferred as a substrate for the SARS-CoV-2 nsp12 RdRp, where it is incorporated into RNA in the place of GTP, arresting synthesis. The resulting, terminated RNA is totally resistant to excision by the highly active SARS-CoV-2 nsp14/nsp10 ExoN complex. The α-thio modification appears thus as a possible route to rescue NAs whose antiviral potency is naturally jeopardized by the coronavirus RNA replicative mismatch repair pathway. However, unlike AT-9010, neither isomer of AT-9052 was found to target the NiRAN domain. This modification therefore alters the mechanism of action and antiviral target of this drug, and provides a new strategy for the treatment of COVID-19, and a promising option for combination therapies.

Materials and methods

Nucleoside analogue synthesis

Ribonucleotides (NTPs : ATP, GTP, UTP, CTP) were purchased as 100 mM solutions from Cytiva. AT-9010 was from ATEA Pharmaceuticals, synthesized and used as described in (17). AT-9052 was synthesised by NuBlocks (Oceanside, CA).

Analytical HPLC analysis of AT-9052-S_P and -R_P isomers

The racemic mixture of AT-9052 was analyzed by HPLC-DAD (Acquity Arc coupled to 2998 PDA detector, Waters) with a detection wavelength of 260 nm using an Acclaim Polar Advantage II column (modified C18, 3 μm, 3 × 150 mm, ThermoFisher) equipped with a guard column. Isomers were separated at a flow rate of 0.5 ml/min at 30°C via gradient elution, varying the concentration of acetonitrile from 3% to 80% over 30 min, in buffer A (50 mM triethylammonium bicarbonate [TEAB] pH 7.0). Precise gradient details for percent Buffer B (100% acetonitrile) are as follows: 0 min = 3%, 2 min = 3%, 12 min = 13%, 14 min = 80%, 16 min = 80%, 17 min = 3% and 30 min = 3%. The first eluted isomer, temporarily named AT-9052-Fast (for fast eluting isomer), was eluted at 7.4 min and the second, temporarily named AT-9052-Slow, was eluted at 9.1 min (Figure S1), both with a characteristic UV absorption spectrum, similar to GTP.

Purification of the isomers AT-9052-S_P and -R_P by preparative HPLC

Preparative purifications of the isomeric mixture of AT-9052-S_P and -R_P were achieved on HPLC-DAD system (Alliance 2695 coupled to 2998 PDA detector, Waters) on a semi-preparative X-Bridge C18 column (5μm, 10 × 250 mm) and a X-Bridge C18 pre-column (5μm, 10 × 10 mm), with a detection wavelength of 260 nm. Isomers were separated (4.5 ml/min at room temperature) using the following percentage of acetonitrile, in Buffer A (pH 7.1): 0 min = 2%, 20 min = 8%, 25 min = 80%, 27 min = 80%, 30 min = 2% and 48 min = 2%. The appropriate fractions were collected and lyophilized. The residue was dissolved in water and passed through a Dowex 50WX2 (Na⁺ form) column to give pure compound in sodium salt. As before, the purified isomers were temporarily named AT-9052-Fast and AT-9052-Slow relating to their time of elution from the column.

RNA synthesis and preparation

Hairpin RNA (HP-A, 5′6FAM-CGAGAGACUCGCGUAGUUUCUACGCG-3′), primer (P10, 5′cy5-GUCAUUCUCC-3′) and template (T20, 5′-UAGCUUCUUCGGAGAAUGAC-3′) RNA was purchased from biomers.net (HPLC grade). The primer–template pair (P10/T20), corresponds to the 3′ end of the SARS-CoV genome, with a single nucleotide change in the template, so as GTP is the first templated base for incorporation. P10/T20 were annealed at a molar ratio of 1:1.5 in 110 mM KCl at 70°C for 10 min, then cooled slowly to room temperature over several hours.

Expression and purification of NDPKb

The gene coding for the human recombinant enzyme nucleoside diphosphate kinase (NDPKb) was cloned into a pNC-ET28 expression vector (Twist Bioscience) with an N-terminal His-tag and the TEV cleavage site, ENLYFQ/G. NDPKb was expressed in Escherichia coli NEB C2566 cells (New England Biolabs) and the cells were grown at 37°C in TB medium containing 50 μg/ml Kanamycin until the absorbance at 600 nm reached 0.6–0.8. Induction was started with 0.5 mM isopropyl β-d-1-thiogalactopyranoside (IPTG), and expression was performed overnight at 17°C. Cells were harvested and the pellets were suspended in lysis buffer (50 mM Tris pH 8.0, 300 mM NaCl, 10 mM imidazole, 1 mM PMSF, 0.25 mg/ml lysozyme and 10 μg/ml DNase) before sonication. The soluble lysate was collected by centrifugation at 12 000g for 30 min at 10°C. The supernatant was loaded onto a Ni-NTA-beads (ThermoFisher), washed (50 mM Tris pH 8.0, 300 mM NaCl, 20 mM imidazole) and recombinant protein was eluted in the same buffer containing 250 mM imidazole. The N-terminal affinity tag was removed through cleavage with TEV protease (1:10 w/w ratio of TEV:protein) during overnight dialysis at room temperature (50 mM Tris pH 8, 150 mM NaCl, 1 mM DTT). A second purification step on Ni-NTA beads was performed to remove the His-tagged TEV protease and uncleaved protein, prior to final purification through size exclusion chromatography on a Superdex 75 16/600 GE column (GE Healthcare) in a final buffer containing 50 mM Tris pH 8, 150 mM NaCl, 1 mM DTT. Fractions containing NDPKb (as determined through SDS-PAGE) were pooled and concentrated to 10 mg/ml, aliquoted, flash-frozen and stored at –80°C.

X-ray crystallography of NDPKb in complex with AT-9052-S_P

Crystallization conditions were adapted from Morera et al. (21) using the sitting-drop vapor diffusion method. All crystals were grown at 277.15 K, using a 1:1 ratio of protein (5 mg/ml) mixed with AT-9052-Fast (final concentration 1 mM) to precipitant solution (10% PEG 3350, 50 mM Tris pH 8.4, 16% glycerol, 1 mM DTT). Crystals grew within 4 days and were directly frozen and stored in liquid nitrogen. The dataset of NDPKb in complex with AT-9052-Fast (S_P) was collected on the Proxima-2 beamline at Synchrotron SOLEIL. Images were processed with autoPROC toolbox (22) and further analyzed using the CCP4 suite (23). The phase was obtained using Molecular Replacement—PHASER (24) with the PDB entry 1NUE as a model. Ligands corresponding to the diphosphate form of AT-9052-S_P/R_P were generated using the AceDRG (25). Structure handling and refinement were done using Coot and Refmac5 respectively (26,27). Data—collection and refinement statistics are given Table 1. Structural analysis and figures were done using PyMOL (retrieved from http://www.pymol.org/pymol) and CCP4MG (28,29).

Table 1.

Data collection and refinement statistics

	NDPKb + AT-9052-Sp diphosphate (PDB: 8PYW)
Wavelength	0.9801
Resolution range	48.62–1.55 (1.61–1.55)
Space group	P 1 21 1
Unit cell a, b, c α, β, γ	54.22, 121.06, 70.76 90, 107.135, 90
Total reflections	478 528 (38 743)
Unique reflections	121 925 (11 757)
Multiplicity	3.9 (3.3)
Completeness (%)	97.16 (94.09)
Mean I/sigma (I)	11.56 (2.12)
Wilson B-factor	18.10
R-pim	0.03683 (0.2868)
CC1/2	0.997 (0.754)
Reflections used in refinement	121 745 (11 757)
Reflections used for R-free	6017 (576)
R-work	0.1706 (0.2380)
R-free	0.1995 (0.2647)
CC(work)	0.961 (0.864)
CC(free)	0.957 (0.829)
Number of non-hydrogen atoms	8348
Macromolecules	7309
Ligands	192
Solvent	847
Protein residues	912
RMS(bonds)	0.012
RMS(angles)	1.75
Ramachandran favored (%)	98.11
Ramachandran allowed (%)	1.22
Ramachandran outliers (%)	0.67
Rotamer outliers (%)	0.25
Clashscore	3.16
Average B-factor	23.09
Macromolecules	21.87
Ligands	45.76
Solvent	28.12

Statistics for the highest-resolution shell are shown in parentheses.

NDPKb phosphorylation assays

NDPKb (40 nM) activity was assayed using AT-9052-Fast, AT-9052-Slow, S_P-GTP-αS or R_P-GTP-αS (200 μM) as the phosphate donor, with GDP (1 mM) as phosphate acceptor in a reaction buffer containing 50 mM Tris pH 8, 50 mM KCl, 2 mM MgCl₂ and 1 mM DTT. Reactions were run at 37°C for 2 h, with aliquots collected at various time points and stopped in 16 mM EDTA, and heated at 95°C for 5 min. Samples were filtered on AcroPrep Advance 96-Well Filter Plates with 3K Omega membrane (Pall). Filtrates were diluted with 1 M TEAB (1:1) and injected onto a C18 reverse phase column (2.5 μm, 4.6 by 100 mm, X-Bridge C18 BEH Premier, Waters) equipped with a guard column, equilibrated in Buffer A (50 mM TEAB pH 7). The substrates and reaction products were eluted using several non-linear gradients of acetonitrile, varying the concentration from 0 to 80% over 25 min. The formation of GTP was quantified over time using an external calibration curve. As a control, NDPKb (40 nM) activity was assessed using S_P-GDP-αS or R_P-GDP-αS (200 μM) as the phosphate acceptor, with either GTP or ATP (1 mM) as the phosphate donor, using the same protocol. The formation of ADP or GDP was quantified over time using an external calibration curve.

Expression and purification of SARS-CoV-2 proteins

SARS-CoV-2 proteins were expressed and purified as previously described (17). Briefly, nsp7 and nsp8 were expressed from pQE30 vector, under the control of a T5-promoter, and nsp10 and nsp14 were purified from the pET28 + vector, under the control of the T7 promoter. All four proteins were N-terminally tagged with a 6-His sequence (nsp7 and 8) or 8-His sequence (nsp10 and 14), followed by the TEV cleavage site ENLYFQ/(N), where the N represents the first native amino acid of the coding sequence of the protein. Nsp7 and nsp8 proteins were expressed in Escherichia coli C2523 cells (New England Biolabs, NEB Express) in the presence of the pRARE2LacI (Novagen) plasmid. Expression was induced with 100 μM IPTG at an OD₆₀₀ of 0.5–0.6, and performed overnight at 17°C in the presence of ampicilin (100 μg/ml) and chloramphenicol (17 μg/ml). Cells were lysed by sonication (in 50 mM Tris–HCl pH 8, 300 mM NaCl, 10 mM imidazole, 20 mM MgSO₄, 0.25 mg/ml lysozyme, 10 μg/ml DNase, 1 mM PMSF) and protein was affinity-purified from the soluble extract using cobalt resin (TALON Superflow Cytiva) in a buffer containing 50 mM Tris–HCl pH 8, 300 mM NaCl, 2 mM β-mercaptoethanol, supplemented with 200 mM imidazole. The TEV protease was used to remove the N-terminal affinity tag during overnight dialysis, as described for NDPKb, at 4°C in a buffer containing no imidazole, supplemented with 1 mM DTT. Cleaved protein was repurified with cobalt resin, prior to purification with size exclusion chromatography (Cytiva Superdex S200), with a final buffer containing 25 mM HEPES pH 8, 150 mM NaCl, 5 mM MgCl₂ and 5 mM TCEP. Nsp10 and nsp14 were expressed in C2566 E. coli (New England Biolabs) in the presence of kanamycin (50 μg/ml). Expression was performed as described above with the following changes. Expression was induced with 500 μM and 50 μM IPTG for nsp10 and nsp14 respectively, in the presence of with kanamycin (50 μg/ml). Cells were lysed in a buffer of 50 mM Tris–HCl pH 8, 300 mM NaCl, 15 mM imidazole, 5 mM MgSO₄, 10% glycerol, 0.1% Triton X-100, supplemented with 5 mM β-mercaptoethanol, 1 mM PMSF, 0.25 mg/ml lysozyme, 10 μg/ml DNase. Following protein binding to cobalt resin (TALON Superflow Cytiva), the resin was washed with 10 CV of wash buffer (50 mM Tris–HCl pH 8, 300 mM NaCl, 15 mM imidazole, 5 mM MgSO₄, 10% glycerol, 1 mM β-mercaptoethanol) followed by a second wash with buffer supplemented with 1 M NaCl, prior to elution with 250 mM imidazole. For nsp10 the buffer was exchanged through dialysis to a final buffer of 50 mM Tris pH 8, 300 mM NaCl, 5 mM MgSO₄, 1 mM β-mercaptoethanol, 10% glycerol. Nsp14 was further purified on a HiLoad 16/60 Superdex 200 gel filtration column (GE Healthcare), in a final buffer of 50 mM Tris-HCl pH 6.8, 300 mM NaCl, 0.5 mM TCEP, 5% glycerol. SARS-CoV-2 nsp9 was expressed with an N-terminal ubiquitin cleavage sequence and a C-terminal 6-His tag from a pASK vector under the control of a tetracycline promoter. Protein was expressed in E. coli NEB Express C2523 cells (New England Biolabs) carrying the pCG1 plasmid for coexpression of the ubiquitin protease, allowing cleavage of the N-terminal tag during expression. Expression was performed overnight at 20°C (with 100 μg/ml ampicilin and 17 μg/ml chloramphenicol), following induction with 200 μg/L anhydrotetracycline at an OD₆₀₀ = 0.6–0.7. Cells were lysed by sonication (50 mM HEPES pH 7.5, 300 mM NaCl, 10 mM midazole, 5 mM MgSO₄, 1 mM β-mercaptoethanol, 0.25 mg/ml lysozyme, 10 μg/ml DNase, 0.1% triton and 1 mM PMSF) and lysed by sonication and purified first through affinity chromatography with cobalt resin, eluted in 100 mM imidazole, followed by size exclusion chromatography (GE Superdex S200) in a final buffer of 50 mM HEPES pH 7.5, 300 mM NaCl, 5 mM MgCl₂ and 1 mM β-mercaptoethanol. SARS-CoV-2 nsp12 was expressed with an N-terminal 8-His tag, followed by the TEV cleavage site (ENLYFQ/N) from a pJ404 vector in E. coli C2523 cells (New England Biolabs) carrying the pGT-f2 chaperone plasmid (Takara Bio). Protein was expressed at 23°C overnight, following induction at an OD₆₀₀ = 0.4–0.5, with 100 μM IPTG, in the presence of ampicilin (100 μg/ml) and chloramphenicol (17 μg/ml), plus 5 μg/l tetracycline and 2% ethanol to promote production of chaperone proteins (groES-groEL-tig). Cells were lysed over 60 min at 4°C in a buffer of 50 mM Tris pH 8, 300 mM NaCl, 5 mM MgSO₄, 10% glycerol, 1% CHAPS, supplemented with 5 mM 2-mercaptoethanol, 0.5 mg/ml lysozyme, 10 μg/ml DNase, 1 mM PMSF, 0.2 mM benzamidine. After 30 minutes, NaCl was slowly added to reach a final concentration of 1.2 M. The resulting lysate was centrifuged (30 000 × g, 30 min) and protein was purified from the soluble fraction using cobalt resin, firstly reducing the NaCl concentration to 300 mM through dilution. Resin was washed in a buffer containing 1 M NaCl, prior to elution in a buffer of 50 mM Tris pH 8, 300 mM NaCl, 0.5 mM TCEP, 500 mM imidazole and 10% glycerol. The affinity tag was cleaved overnight with TEV protease, as described above for nsp7 and nsp8, and re-purified using cobalt resin to remove the TEV. The final purification step was performed with size exclusion chromatography (GE Superdex S200 16/60) in a final buffer of 25 mM HEPES pH 8, 300 mM NaCl, 5 mM MgCl₂, 0.5 mM TCEP and 10% glycerol.

Thermal shift assays (TSA)

The influence of compound binding on protein stability was measured by Themofluor assay, using a CFX Connect BioRad real-time PCR machine. Reactions were run in thin-walled, 96-well plates, in a buffer composed of 10 mM HEPES pH 7.4, 150 mM NaCl, 5 mM MgCl₂ and 0.5 mM MnCl₂, in a final reaction volume of 20 μl. Compounds and NTP concentrations were fixed at 100 μM, with a final nsp12 protein concentration of 2 μM. SYPRO Orange dye was used at a five-fold final concentration. Melting-temperature (Tm) values given are the average and standard deviation of three experiments.

RNAylation inhibition assays

RNAylation of SARS-CoV-2 nsp9 by nsp12 was performed in a final buffer containing 50 mM Tris, pH 7.5, 1 mM DTT, 1 mM MnCl₂, 1 μg yeast inorganic pyrophosphatase (ThermoFisher EF0221), 20 μM nsp9 and 40 μM triphosphorylated RNA corresponding to the 5′ end of the SARS-CoV-2 genome (5′-pppAUUAAAGGUU-3′). The inhibitor was added at final concentrations of 1, 10 and 100 μM. Reactions were started through the addition of nsp12, final concentration 2 μM, and stopped after 30 min incubation at 25°C in 4X SDS loading die solution. Nsp9 (theoretical MW 12.4 kDa) and pRNA-nsp9 (theoretical MW 15.8 kDa) were separated on 4–20% gradient SDS-PAGE gels (NuSep), and stained with QuickBlue Protein Stain (Lubio Science). Gels were scanned on the Amersham ImageQuant800 machine to obtain density. Percent RNAylation of nsp9 was calculated as the ratio of the nsp9-pRNA¹⁰ band, relative to total nsp9 (nsp9 plus nsp9-pRNA¹⁰ bands). Inhibition of RNAylation was calculated as a percentage, relative to no inhibitor controls (n = 2).

Formation of an active nsp12:nsp7:nsp8 RTC

The active RTC was formed by first incubating nsp7 and 8 together at equimolar concentrations (100 μM), and nsp12 (15 μM) and nsp8 (45 μM) together (1:3 ratio), in the final buffer used for nsp12 purification. Following 30 min incubation at room temperature, all protein was mixed together for a further 15 min at room temperature. The final complex consisted of nsp12:7:8 at a 1:3:6 ratio, with 10 μM concentration of nsp12. The RTC was further incubated for 15 min at room temperature with the appropriate RNA (P10/T20 or hairpin RNA), in a final reaction buffer of 20 mM HEPES pH 7.5, 50 mM NaCl and 5 mM MgCl₂.

Steady-state primer extension polymerase assays

Steady-state polymerase assays were run with the annealed P10/T20 RNA. Single nucleotide incorporation reactions were initiated with 50 μM (final concentration) of GTP, AT-9010, AT-9052-S_P or AT-9052-R_P. The ability of the incorporated analogue to cause chain-termination was verified in parallel assays run in the presence of 50 μM ATP, the following nucleotide templated for incorporation. Final reaction concentrations were 0.5 μM nsp12, 0.4 μM RNA, in a final buffer of 20 mM HEPES pH 7.5, 50 mM NaCl and 5 mM MgCl₂. For competition experiments, protein-RNA complexes were prepared as described above, and reactions were initiated with all four native NTPs (50 μM each) in the absence (-ve control) or presence of 250 μM AT-9010, AT-9052-S_P or AT-9052-R_P. Reactions were quenched after indicated time-points (20 s–30 min) with 2× volume of FBD stop solution (formamide, 10 mM EDTA, 0.02% Bromophenol blue), and analysed using denaturing polyacrylamide gel electrophoresis, described below. All reactions were repeated in at least three independent experiments.

Pre-steady state quenched-flow kinetics

Experiments were carried out in a Bio-Logic QFM-400 rapid chemical quench-flow apparatus that controls reaction time by flow rate through a 2.5 μl chamber between the reaction mixer and the quencher mixer. The nsp12:7:8 RTC–RNA complex was prepared as described above, with the annealed P10/T20 RNA primer-template pair. Reactions were initiated in the rapid-quench machine by mixing equal volumes (18 μl each) of the protein-RNA complex with various concentrations of a single NTP substrate (GTP, AT-9010 or AT-9052-S_P) in the same reaction buffer. Final concentrations were 800 nM protein complex (based on nsp12 concentration), 150 nM RNA and 1–500 μM NTP. The excess enzyme over RNA was meant to eliminate steady-state conditions and recycling of the enzyme after incorporation. Reactions were quenched at various times (10–2000 ms) with 18 μl of 3M HCl (final concentration 1 M), and the pH immediately neutralized through the addition of an equal volume of 1 M NaOH, 300 mM MES. For each NTP substrate, 6 concentrations were measured over two independent experiments (i.e. protein complex and NTP dilution prepared separately), with 12 timepoints collected per concentration. Reactions products were diluted 1:1 in FBD stop solution prior to analysis.

Extension and excision assays

The active RTC (nsp12:7:8) was incubated with HP-A, as described above, and polymerisation was initiated through the addition 50 μM final concentration of ATP (first templated nucleotide), plus GTP, AT-9010, AT-9052-S_P or AT-9052-R_P (second templated nucleotide). Additional control reactions were run in the presence of UTP, as the third templated nucleotide. Final reaction concentrations were 0.5 μM nsp12, 0.4 μM RNA, in a buffer of 20 mM HEPES pH 7.5, 50 mM NaCl and 5 mM MgCl₂. Small aliquots of the reaction were stopped after indicated timepoints with 2× volume of FBD, to track incorporation. For the remainder of the reaction, polymerisation was stopped after 30 min by heating at 70°C for 10 min. The reaction was left at 30° for 30 min to allow reannealing of the hairpin RNA, prior to excision assays. The exonuclease complex was prepared by incubating nsp14 and nsp10 together at a 1:5 molar ratio for 5 min at 30°C in a buffer of 20 mM Tris pH 8, 1 mM MgCl₂, 5 mM DTT. Excision was initiated by diluting 5 μl of the ExoN complex in 20 μl of the stopped polymerase reaction, for a final concentration of 50 nM nsp14, 250 nM nsp10. Excision was stopped at indicated timepoints (2–120 min) via quenching in FBD.

Analysis of RNA products

Reaction products from RNA polymerase extension assays, rapid-quench experiments and excision assays were separated using denaturing polyacrylamide gel electrophoresis (20% acrylamide, 7 M urea, TBE buffer). All reactions were heated to 70°C for 10 min, then cooled on ice for 2 min prior to loading. Products were visualized using a Typhoon FluorImager, and analyzed with ImageQuant software. Graphs were plotted with GraphPad Prism. For all experiments, percent incorporation and/or percent excision was calculated by dividing the intensity of the band of interest by the sum of all bands in the lane, to account for loading error. For competition experiments, the preference for GTP insertion, relative to each analogue is referred to as the ‘discrimination factor’ whereby a higher value represents a higher preference for GTP. Discrimination values were calculated as the ratio between the intensity of each individual +1, +4 and +7 chain-terminated product (corresponding to AT-9010/AT-9052 incorporation products) with the sum of the intensity of all bands higher than each of these positions (e.g. the sum of +2 to +10 products, divided by +1 product). As analogue insertion causes chain termination, these higher bands represent elongation products derived from GTP incorporation at the same relative position. Discrimination was corrected to account for concentration difference between AT- analogue and GTP (5-fold difference). Values were calculated from three independent experiments, with 5 timepoints for each of the three products (+1, +4, +7), for a total of 15 data points per replicate.

Pre-steady state kinetic analysis

All kinetics experiments were plotted in GraphPad Prism, fit using nonlinear regression. GTP experiments were fit with a double exponential function (equation 1), while AT-9010 and AT-9052-S_P time course data were fit to single exponential function (equation 2). In both cases, the plateau was constrained to be as a shared value for all data sets, and Y0 was left unconstrained (>0). The obtained k_fast,k_slow and k_obs values were plotted against nucleotide concentration and fitted to equation 3. Of note, nucleotide concentration plots were also calculated using standard Michaelis–Menten kinetics and gave similar results. However, the curve fit was found to be better (higher R², lower predicted error) when fit to equation (3), which includes the Hill slope parameter. A Hill slope value of 1 is equivalent to mass action binding to one site, while values greater than one represent binding to multiple sites. Here, Hill slope values were found to be between 1.3 and 1.7. This may be caused by NiRAN binding, although this causes only a minor lag phase, due to the excess of NTP > enzyme (i.e. sufficient NTP for saturation of both sites).

$$\begin{eqnarray*} && Product = Y0 + {A}_{fast}\left[ {1 - exp\left( { - {k}_{fast}t} \right)} \right]\nonumber\\ && \quad +\, {A}_{slow}\left[ {1 - exp\left( { - {k}_{slow}t} \right)} \right]\end{eqnarray*}$$

(1)

where k_fast and k_slow are the observed rate constants of the fast and slow phases, and A_fast and A_slow are the amplitudes of product formation (plateau – Y0) for each phase.

$$\begin{equation*}Product = Y0 + A\left[ {1 - exp\left( { - {k}_{obs}t} \right)} \right]\end{equation*}$$

(2)

where k_obs is the observed single rate constant, and A is the amplitude of product formation (plateau – Y0).

$$\begin{equation*}\left. {{k}_{\left( {fast/slow/obs} \right)} = {k}_{max}{{\left[ {NTP} \right]}}^h/ \left( {{{\left[ {{K}_d} \right]}}^h + {{\left[ {NTP} \right]}}^h} \right)} \right)\end{equation*}$$

(3)

where k_max is the extrapolated maximal rate for NTP incorporation and K_d is the equilibrium dissociation constant for NTP binding and h is the Hill slope.

Results

Absolute Configuration of AT-9052, the 5′-α-thiotriphosphates of AT-9010

We have previously shown that AT-9010, the 5′-TP of the potent SARS-CoV-2 prodrug Bemnifosbuvir (AT-527), is efficiently incorporated into RNA by the RTC, however can be excised by nsp14/nsp10. Based on this, we synthesized AT-9052, the 5′-α-thiotriphosphate form of AT-9010, as a racemic mixture (R_P/S_P, Figure 1). The two isomers were separated and isolated using reverse phase HPLC (Supplementary Figure 1), and temporarily named Fast and Slow, relating to their retention time on the column.

Figure 1.

Structures of prodrug, 5′-triphosphate form and 5′-α-thiotriphosphate forms. (A) Phosphoramidate prodrug AT-527 (Bemnifosbuvir), with 2′-fluoro, 2′-methyl ribose and N₆-methyl guanine modifications. (B) AT-9010, the active 5′-triphosphate form of Bemnifosbuvir. (C) AT-9052-R_P and AT-9052-S_P, the 5′-a-thiotriphosphate diastereoisomeric forms of Bemnifosbuvir.

To determine the absolute configuration at the α-phosphorus atom, human NDPKb was co-crystallized with either AT-9052 Fast or Slow. The choice of NDPKb was based on its presumed role in the activation pathway of Bemnifosbuvir, where it is believed to perform the final phosphorylation step for formation of AT-9010 (3,16). As such, we anticipated that AT-9052 was also likely to be a substrate for NDPKb, potentially in a stereoselective manner. Crystallization of NDPKb alone, or in the presence of AT-9052-Slow proved to be difficult, and crystals were of poor quality and/or did not diffract. In contrast, cocrystallization of NDPKb and the AT-9052-Fast isomer yielded diffracting crystals, resolved to 1.5 Å, with one hexamer in the asymmetric unit (Table 1, Figure 2A). Its structure presents a ferredoxin fold (αβ sandwich), made up of a central β-sheet comprised of 4 antiparallel β-strands, surrounded by 3 parallel α-helices on one side, and 3 orthogonal α-helices on the other side, forming the nucleotide binding site. Density for AT-9052-Fast α-thiodiphosphate was found in all six subunits, bound in the same conformation allowing superimposition (Figure 2A). The additional density of the sulfur atom, relative to an oxygen, unambiguously shows that the α-thiophosphate is in the S_P configuration, according to the Cahn-Ingold-Prelog rules (Figure 2B, C). Comparison with an existing structure of NDPKb bound to GDP (PDB: 1NUE) shows a conserved base-stacking interaction between the guanine and Phe60, and a base check by Glu152 through hydrogen bonding with the N² of the purine base. In the GDP structure, the 2′ and 3′-OH of the ribose are stabilized through hydrogen bonding with Lys12 and Asn115. However, this is disrupted by the 2′-fluoro, 2′-methyl modification in AT-9052, with a noticeable shift in the overall position of the ribose (Supplementary Figure 2). In both structures (GDP or AT-9052), the hydroxyls of the β-phosphate are stabilized through hydrogen bonding with Thr94; however in the case of AT-9052-S_P, the general positioning is shifted by ∼3.5 Å. Nevertheless, the diphosphate group faces towards the catalytic histidine (His118) and Arg88, as is the case for GDP. In both structures, the α-phosphate or α-thiophosphate group is largely solvent exposed. In the case of GDP, the pro-R_P oxygen of the α-phosphate (referring to the theoretical position of the sulfur in AT-9052-R_P) contacts a metal ion in two out of the six subunits. We tentatively hypothesize that the closer proximity of the pro-R_P atom to metal ions and/or protein residues may explain the stereoselectivity for the S_P isomer of AT-9052.

Figure 2.

NDPK stereoselectivity for AT-9052-S_P. (A–D) NDPK cocrystallised with AT-9052-Fast isomer at 1.5 Å. (A) Superimposition of the subunits of the NDPK hexamer with bound AT-9052-Fast present in all subunits. (B) fo – fc omit map of AT-9052-Fast bound in the active-site of NDPK, contoured at ±3.0 σ. (C) Attempted fitting of AT-9052-S_P, AT-9052-R_P and AT-9010 into the additional density, showing the bound isomer corresponds to AT-9052-S_P. (D) Active site residues involved in interactions with the bound ligand. (E) Kinetics of conversion of GDP (phosphate acceptor) into GTP by NDPKb. Phosphodonor (NTPαS) used are AT-9052-Fast (black), AT-9052-Slow (green), S_P-GTP-αS (blue) or R_P-GTP-αS (red).

NDPKb plays a role in regulating intracellular nucleotide pool concentrations, by catalyzing the exchange of the γ-phosphate of nucleoside triphosphates (NTP, phosphodonor) to nucleoside diphosphates (NDP, phosphoacceptor). While the phosphodonor is principally ATP, and the phosphoacceptor is commonly GDP (30), NDPKb accepts all common nucleotide substrates (21), in accordance with the non-specific nucleobase binding provided by Phe60 stacking. Incubation of AT-9052-S_P (Fast) with NDPKb and GDP (as the phosphoacceptor) shows it is able to transfer the γ-phosphate, resulting in formation of GTP (65% conversion after 1 h). In contrast, GTP is not formed in the presence of AT-9052-R_P (Slow) under the same conditions (Figure 2E, Supplementary Figure 3). Similarly, only the S_P isomer of α-thiotriphosphate guanosine (S_P-GTP-αS) is efficiently used by NDPKb (Figure 2E), again supporting the finding that AT-9052-Fast corresponds to the S_P isomer. Notably, here we have shown that both AT-9052-S_P and S_P-GTP-αS isomers are selectively preferred by NDPKb as phosphate donors. Control reactions run with S_P- or R_P- GDP-αS, using either ATP or GTP as the phosphodonor, shows the reaction can also proceed in the other direction. Again, the S_P isomer is preferred, resulting in higher production of ADP or GDP (Supplementary Figure 4). We therefore presume this is also the case for AT-9052-S_P α-thiodiphosphate. Furthermore, given that NDPKb generally uses GDP as a phosphoacceptor (with ATP as the phosphodonor) we predict that AT-9052-S_P α-thiodiphosphate may also be a more efficient acceptor. This may also explain the low levels of activity also observed for R_P-GDP-αS. Importantly, these results show that the host enzyme NDPKb is stereospecific, which may play a role in the preferential activation of the prodrug to this specific S_P isomer.

Neither AT-9052-S_P nor AT-9052-R_P Interact with the NiRAN Domain of nsp12

The potency of Bemnifosbuvir/AT-9010 has been attributed to its dual targeting of both the RdRp, and NiRAN domains of SARS-CoV-2 nsp12 (17). The NiRAN is a pseudokinase-like enzyme, responsible for the covalent transfer of nucleoside mono- and di- phosphates (NMP and NDP, respectively) to protein cofactors. AT-9010 has been found to stably, and competitively bind to a guanine-specific pocket of the NiRAN, interfering with transferase activity. We tested the stabilization of nsp12 in the presence of GTP, AT-9010 or AT-9052-S_P/R_P isomers using thermal shift assays (TSA). We have previously shown that this can be used as an indicator of NTP/NA binding to the NiRAN, rather than the RdRp domain (17). Compared with AT-9010, stabilization of nsp12 by AT-9052 is reduced, irrespective of the α-thiophosphorus configuration (Figure 3A). Under these conditions, AT-9052-S_P and R_P binding is comparable to GTP, however the increase in thermal stability is very minor for all three nucleotides. In agreement with the previous finding, AT-9010 provides more thermal stability than its native nucleotide counterpart, and therefore remains the best ligand for the NiRAN domain. The difference between AT-9010 and GTP may be due to its preferential binding in the stabilized ‘base-in’ conformation, where the guanine base occupies a narrow pocket, formed as a result of its binding (17). In contrast, structures with GDP have shown it can also bind in a ‘base-out’ orientation, where the majority of interactions are formed with NDP phosphate group (31,32). Although we cannot confirm whether this is also possible for AT-9052, we conclude that the α-thiophosphorus modification results in a loss of stable binding compared with the parent compound.

Figure 3.

NiRAN inhibition by AT-9010 and AT-9052 (S_P/R_P). (A) Thermal shift stability assay of SARS-CoV-2 nsp12 incubated with GTP or NAs, where delta Tm represents the shift of the unfolding temperature following incubation with ligands. (B) Inhibition of NiRAN-mediated nsp9 RNAylation. Nsp12, nsp9 and triphosphorylated RNA(10mer) were incubated together for 30 min alone, or in the presence of increasing concentrations of GDP, AT-9010 or AT-9052-S_P/R_P. RNAylation of nsp9 results in a shift in molecular weight (nsp9 to nsp9-pRNA). Gel is representative of two independent experiments, run with three concentrations of each inhibitor. (C) Percent inhibition of RNAylation by each NA, and GDP was quantified from density analysis. (D) AT-9010 (PDB 7ED5) bound in NiRAN active pocket. Guanosine moiety is buried in small cavity, with diphosphates positioned in a narrow channel coordinated by two metal ions. The pro-S_P and pro-R_P oxygens indicate the theoretical position of the sulfur in the S_P and R_P forms of AT-9052, respectively.

To further investigate the inhibitory capacity of AT-9052, its ability to block RNAylation was investigated. NiRAN transferase activity has recently been linked with a role in the viral capping pathway through a two-step reaction. Using triphosphate-RNA as a substrate, the NiRAN first transfers the RNA to the N-terminal amino group of the viral cofactor protein nsp9, releasing pyrophosphate and forming a nsp9-pRNA intermediate (known as RNAylation). GDP is subsequently transferred to the monophosphate RNA, releasing nsp9 and forming a canonical GpppRNA cap structure (18). We tested the ability of AT-9010, and the corresponding 5′-α-thiotriphosphate AT-9052(S_P/R_P) to inhibit nsp9 RNAylation. GDP was used as a control, as it has been previously reported to inhibit this reaction. In contrast to AT-9010 and GDP, neither AT-9052 isomer significantly inhibits nsp9-RNAylation (Figure 3B). Based on this, we conclude that AT-9052 is unlikely to outcompete GDP for NiRAN binding, and thus does not significantly inhibit this domain.

The structure of AT-9010 diphosphate bound in the NiRAN active site (PDB: 7ED5) shows the guanine base and modified ribose are accommodated in a small pocket, in a ‘base-in’ conformation (Figure 3C). The diphosphate sits in a narrow groove, coordinated by two metal ions and an extensive hydrogen bonding network. While neither non-bridging oxygen of the alpha phosphate is directly involved in metal ion coordination, both are stabilized through interactions with K73 and R116. Replacement of the P–O bond with a longer P–S bond would likely interfere with these interactions, and subsequently affect the overall diphosphate positioning. Overall, we conclude that neither oxygen substitution is tolerated due to the narrow channel and tight interaction network, required for precise diphosphate alignment.

The AT-9052-S_P Isomer is a Substrate for the SARS-CoV-2 RTC

We have previously shown that AT-9010 is competitively inserted into RNA by the viral RdRp, as a substitute for GTP (17). To determine the effect of the thio-modification on incorporation by the viral polymerase, AT-9010, AT-9052-S_P and AT-9052-R_P were used as nucleotide substrates in a primer-extension assay (Figure 4A). The SARS-CoV-2 minimal RTC (nsp12-nsp7-nsp8₂) shows a clear stereoselectivity for the S_P isomer, which is incorporated at similar levels to GTP and AT-9010 (at the +1 position) under steady-state, multiple-turnover conditions. In contrast, the R_p isomer is poorly inserted into RNA.

Figure 4.

Incorporation of GTP, AT-9010 and AT-9052-(S_P/R_P) by SARS-CoV-2 nsp12-nsp7-nsp8₂ minimal RTC. (A) Timecourse of elongation of an annealed primer-template RNA (P10-T20) in the presence of a single NTP/NA (50 μM). Position of the +1 incorporation product (GTP/AT-9010/AT-9052-S_P/R_P) opposite a templating cytosine shown in red. For GTP, a +2 GTP:U mismatch is observed. Gel is representitive of three independent experiments. (B) Structure of nsp12 RdRp bound to RNA terminated with a 3′deoxy nucleotide (PDB 7UOB). GTP is poised for incorporation in the nucleotide-binding pocket, in the presence of two magnesium ions. The pro-S_P and pro-R_P oxygens indicate the theoretical position of the sulfur in the S_P and R_P forms of AT-9052, respectively. (C–E) Pre-steady state, rapid-quench kinetic experiments for single-nucleotide incorporation of GTP (C), AT-9010 (D) and AT-9052-S_P (E). Top panels show timecourse of individual concentrations, fit with a double exponential function for GTP, and single exponential function for AT-9010 and AT-9052-S_P. Bottom panels show rate concentration dependence plot, with calculated k_max (maximum rate of incoporation, s⁻¹) and K_d (dissociation constant, μm). Calculated kinetic values shown in Table 2.

In contrast to the GTP control, for which a +2 mismatch product (GTP incorporated opposite templating U) routinely appears, no further elongation products are seen following analogue incorporation. This indicates that insertion of AT-9052 halts synthesis, as is the case for AT-9010. This was confirmed through incubation of each analogue in the presence of ATP, the next natural nucleotide to be incorporated. While the expected +7 product can be seen for the GTP + ATP reaction, elongation is terminated following incorporation of all AT- analogues at the +1 position (Supplementary figure 5). The cryo-EM structure of incorporated AT-9010 (PDB: 7ED5) shows that the 2′-fluoro, 2′-methyl substituents of the ribose causes misalignment of the next incoming nucleotide, preventing further extension. This mechanism of action is likely analogous for AT-9052.

To investigate the structural basis behind the stereoselectivity of the nsp12 polymerase, we examined recently published cryo-EM structures of the SARS-CoV-2 RTC, bound to RNA and each of the native NTPs, poised for incorporation. Each complex was trapped in a stalled, pre-incorporation state following insertion of a 3′-deoxy nucleotide, allowing examination of nucleotide selectivity and alignment for catalysis (32). The position of the triphosphate is well conserved, showing that the pro-R_p oxygen of the α-phosphate faces towards the metal ion binding site (Figure 4B). For all NTPs, it contacts at least one of the two magnesium ions (Mg²⁺_B), and both in the presence of GTP. While this can only be used as a model, as structural rearrangement could occur in the active site to accommodate α-thiophosphate binding, we theorize that its replacement with sulfur would impede catalysis by disrupting metal ion coordination. This is consistent with biochemical data, casting light on why AT-9052-R_p is a poor substrate for the RdRp. The other non-bridging oxygen of the α-phosphate (pro-S_P) is located further from the metal ions and may interact with Arg555, a residue which is known to be highly flexible, potentially allowing accommodation of the sulfur substitution at this site (33).

We performed pre-steady state, rapid-quench kinetic experiments in order to further understand the molecular impact of the 2′ ribose and thiophosphate modifications on incorporation efficiency (Figure 4C–E, Table 2). Incorporation of GTP, AT-9010 and AT-9052-S_P was measured at the millisecond timescale (10–2000 ms) at a range of NTP concentrations (1–500 μM), quenching with 1M HCl (Figure 4C–E). Reactions were run with an excess of enzyme over RNA, and thus should represent a single nucleotide incorporation cycle, whereby RNA release and rebinding should not influence the overall reaction rates. The resulting time course of GTP incorporation best follows biphasic kinetics, with a dominant ‘fast’ phase (k_fast) representing ∼70% of the observed amplitude, followed by a second, ‘slow’ phase (Figure 4C, top panel). Interestingly, both show NTP concentration dependence, with maximal rates (k_max(f/s)) of 142.3 and 3.6 s⁻¹ for the fast and slow phases respectively (Figure 4C, Supplementary Figure 6). We predict that the faster rate represents the maximal speed of GTP incorporation for RNA–enzyme complexes in a ‘productive’, catalytically competent conformation. The slower phase suggests the presence of a second ‘less-productive’ population of RNA-bound complexes which require an additional rate-limiting step prior to incorporation, as reported for other polymerases (34,35). In contrast, a single phase of product formation is best modeled for both AT-9010 and AT-9052-S_P (Figure 4D, E, top panels, Table 2). This single phase is governed by the effect of the ribose modification on incorporation, which is thus rate-limiting. Fitting individual rates against substrate concentration (μM) gives maximal incorporation rates of 7.5 and 2.6 s⁻¹ for AT-9010 and AT-9052-S_P, respectively (Figure 4C–E, bottom panels, Table 2). These values are ∼20-fold and ∼55-fold lower than the calculated maximal rate for GTP incorporation (142.3 s⁻¹). In comparison, the calculated dissociation constants (K_d) are only ∼2–3-fold higher for AT-9010 (82.4 μM) and AT-9052-S_P (119.2 μM) compared with GTP (39.1 μM). This modest difference suggests that the chemistry step of phosphodiester bond formation, rather than nucleotide binding, plays a more significant role in the overall reduction in incorporation efficiency. This effect appears to be predominantly attributable to the 2′-fluoro, 2′-methyl ribose modification. The presence and the supposed orientation of the α-P–S bond of the AT-9052-S_P isomer relative to the α-P–O bond of AT-9010 translates into a further 2.8-fold decrease in the catalytic rate (k_max), which may well represent the thio elemental effect during catalysis. This results in a ∼4-fold difference in overall incorporation efficiency (k_max/K_d).

Table 2.

Kinetic parameters and Incorporation efficiencies of GTP and AT analogues

NTP	kmax (s−1)	Kd (μM)	Efficiency (E) kmax/Kd (s−1 μM−1)	Fold-reduction in efficiency	Discrimination (competition)
GTPfast	142.3 ± 6.9	39.1 ± 3.7	3.64 ± 0.39	1	1
GTPslow	3.6 ± 0.3	12.9 ± 2.7	0.28 ± 0.06	13.0	-
AT-9010	7.5 ± 0.5	82.4 ± 12	0.09 ± 0.01	39.9	7.3 ± 2.1
AT-9052-SP	2.6 ± 0.5	119.2 ± 45.6	0.02 ± 0.01	166.8	57.4 ± 16.88

Kinetic values (k_max and K_d) calculated from pre-steady state kinetic data (Figure 4), where efficiency (E) is k_max/K_d and fold-reduction in efficiency is calculated relative to GTPfast. Discrimination measured from multiple incorporation experiments, where each analogue is competing with GTP for insertion (Figure 5). The discrimination value represents the number of templating cytosines that would theoretically be required for the analogue to be inserted in the place of GTP, when at equimolar concentrations.

Overall, pre-steady state kinetics show that AT-9010 and AT-9052-S_P are incorporated into RNA ∼40-fold, and ∼166-fold less efficiently than GTP (Table 2). However, these values do not reflect whether these analogues can effectively compete with GTP for binding and insertion into RNA. In a cellular setting, the RTC is saturated with high intracellular concentrations (millimolar range) of native NTPs, which compete with NAs for incorporation (36). We performed multiple incorporation experiments in the presence of all four native NTPs (50 μM each), with a 5-fold excess of each NA. This shows that both AT-9010 and AT-9052-S_P are able to compete with GTP for insertion (Figure 5). Chain-terminated synthesis products are seen at positions + 1, +4 and + 7, opposite templating cytosines. The ratio between each AT-9052-S_P terminated band and larger elongation products reveals a discrimination factor (relating to the competition efficiency) of ∼60 ± 17, compared to GTP (Table 2). At equimolar concentrations, AT-9052-S_P is therefore expected to be incorporated in the place of GTP approximately 1 in every 60 times. This is ∼12-fold lower than AT-9010, with a calculated discrimination factor of 7.3 ± 2.1, comparable to what was previously found (17).

Figure 5.

Competition of AT-9010 and AT-9052-(S_P/R_P) with native NTPs for incorporation. Elongation of the P10 primer annealed to the T20 template, in the presence of all four native NTPs alone (-ve), or with a 5-fold excess of AT-9010, AT-9052-S_P or AT-9052-R_P. corresponding with templating cytosines. On the left, the identity of the incorporated (native) NTP is marked. Positions where GTP and/or analogues can be incorporated are shown in red, at the +1, +4 and +7 positions. In the presence of analogues, chain-terminated products, caused by competitive analogue incorporation in the place of GTP, are observed at these positions.

The calculated discrimination values for both AT-9010 and AT-9052-S_P suggest that the probability of incorporation is higher than what is estimated from pre-steady state kinetics alone. This may be due to the difference between single, versus multiple incorporation events, as well as factors relating to pre-steady state vs steady-state kinetics, discussed below.

We conclude that the SARS-CoV-2 RTC accepts the AT-9052-S_P isomer as a GTP substrate. While it is incorporated less efficiently than AT-9010, it is still able to compete with GTP for incorporation, where it causes immediate RNA chain termination.

RNA Terminated with Incorporated AT-9052 is Resistant to ExoN-Mediated Repair

To determine whether the resulting RNA product is subject to hydrolysis by the SARS-CoV-2 proofreading system, we measured excision of terminally incorporated GTP, AT-9010, and AT-9052-S_P/R_P by the SARS-CoV-2 3′-to-5′ exoribonuclease (nsp14). Incubation of the polymerization products with the nsp10/nsp14 heterodimer complex shows that RNAs terminated with the AT-9052-S_P isomer are completely resistant to cleavage (Figure 6A, B). In contrast, the R_P isomer is able to be excised, albeit at slower rates than the GTP control. Consistent with previous findings, incorporated AT-9010 is also excised slower than native nucleotides (17), although is a slightly better substrate for nsp14 than AT-9052-R_P, showing the α-thiophosphate substitution has an additive effect.

Figure 6.

Incorporation and excision of GTP, AT-9010 and AT-9052-(S_P/R_P). (A) Timecourse of incorporation of GTP, AT-9010, AT-9052-(S_P/R_P) by the SARS-CoV-2 minimal RTC on a fluorescently labeled hairpin RNA. Timepoints for polymerization (pol) shown in black. At 30 min, the reaction was stopped, and nsp14/nsp10 complex was added. Excision reactions (exo) stopped at indicated timepoints (gray). For the GTP control reaction, the +2 product represents a mismatch produced following GTP incorporated opposite U. For all nucleotides sensitive to excision, additional (unexplained) products appear following addition of nsp14/nsp10, indicated by (*). (B) quantification of excision products observed in (A) (n = 2). (C) Structure of SARS-CoV-2 nsp14 E191A active site mutant (PDB 7N0C), bound with RNA. Two metal ions are observed in the structure.

To confirm these findings, we performed incorporation-excision experiments using a more stable, hairpin RNA. Polymerization reactions were performed in the presence of ATP (first templated nucleotide) with each analogue (or GTP) 3′-terminally incorporated at the +2 position (Supplementary Figure 7a). Products containing incorporated AT-analogues migrate lower than expected on this RNA scaffold, making separation of +1 (ATP) and +2 (ATP + AT) products difficult. This effect is due to the 2′-fluoro 2′-methyl ribose modification, and does not occur with αS-GTP-S_P/R_P (Supplementary Figure 7b). Nevertheless, these results confirm that RNAs terminated with AT-9052-S_P are completely resistant to excision. Similarly, incorporation of αS-GTP-S_P also produces RNAs that are unable to be cleaved by nsp10/nsp14. This shows the S_P-α-thiophosphate modification alone is sufficient for blocking excision. In contrast, AT-9052-R_P shows only partial nsp14/nsp10 resistance, although its specific excision is difficult to determine on this RNA scaffold due to its poor insertion by the nsp12 RdRp, and co-migration with the +1 ATP band. Following two hours of incubation with nsp14/nsp10, ∼20% of the product remained unexcised. It seems likely that the remaining product is derived from the AT-9052-R_P terminated RNA, while the rapidly excised fraction represents the +1 ATP product. Incorporation of αS-GTP-R_P allows separation of the +1 and +2 products, resolving the ambiguity and confirming this to be the case (Supplementary Figure 7b). Rapid disappearance of the +1 ATP band is observed, while the +2, αS-GTP-R_P terminated product is excised at slower rates. We conclude that the RNA product produced following AT-9052-R_P incorporation are slower excised than native NTPs, however its poor incorporation means that this product is unlikely to be biologically relevant. In contrast, following rapid incorporation of the preferred AT-9052-S_P isomer, the RNA product is completely resistant to nsp14-mediated excision.

Based on research performed in the 1970s–1980s (37–40), it is now commonly accepted that nucleophilic attack on the α-phosphate for nucleotide incorporation leads to inversion of configuration, typical of S_N2 condensation reactions. As a result, incorporation of the S_P-NTP-αS is believed to yield the R_P isomer at the resulting phosphorothioate linkage, and vice versa (41–45). However much of this assumption is based on circumstantial evidence relying on the ‘known’ stereoselectivity of other enzymes and/or crude chromatography techniques (37–40). To visualize the stereoisomeric effect of the oxygen-to-sulfur substitution on exonuclease activity, we analyzed the cryo-EM structure of SARS-CoV-2 nsp10/nsp14 (E191A active-site mutant) in complex with RNA and Mg²⁺ (PDB: 7N0C) (46). The coronavirus 3′-to-5′ nsp14-ExoN belongs to the DEDD exonuclease superfamily, with a DEED catalytic motif which follows a two-metal ion catalysis mechanism (47). One ion is loaded into the catalytic site, coordinated by D90, E92 and D273 and the RNA phosphate group. The second ion is brought in by the remaining glutamate (E191) and the RNA phosphate, allowing activation of a water molecule for nucleophilic attack. Of note, in the 7N0C structure, the catalytic water is missing and the second metal-ion is displaced due to lack of proper coordination with E191. Nevertheless, the pro-S_P oxygen of the α-phosphate can be seen to contact both Mg²⁺ ions (Figure 6C), analogous to a previous report for the 3′-5′ exonuclease of the large fragment (or Klenow Fragment, KF) of DNA polymerase I (48). While this exonuclease is structurally distinct from that of the CoV ExoN, it uses a similar catalytic mechanism, with two divalent metal ions and a DEDD catalytic motif. Based on these existing structures, we predict that the replacement of the pro-S_P oxygen with sulfur disrupts metal ion coordination, abolishing enzyme activity, similar to the DNA pol I exonuclease. However, an inversion of configuration would mean that the sulfur atom of the incorporated AT-9052-S_P isomer is directed away from the metal ions (R_P isomer at the phosphorothioate linkage). From a structural perspective, it is difficult to determine why this would result in complete resistance to excision, although it is likely the bulkier sulfur group would have some impact on overall excision efficiency. In contrast, if retention of configuration was to occur, it is evident that exonuclease activity would be severely or totally abolished, compatible with our biochemical assay results.

Regardless of whether inversion or retention of configuration has occurred, here we show that following stereoselective incorporation of AT-9052-S_P isomer by the viral polymerase, the resulting RNA product is completely resistant to excision by nsp14-ExoN.

Discussion

Inhibition of the SARS-CoV-2 replication machinery by nucleos(t)ide analogues has been complicated by the presence of the virally encoded exonuclease (ExoN, nsp14), able to excise and repair non-canonical base-pairs in viral RNA (10–14). Treatment of CoV infections with NAs therefore depends on both the efficiency of their incorporation by the RdRp, and their resistance to excision by nsp14. Mutagenic NAs, such as the activated 5′-TP form of Molnupiravir, are rapidly incorporated and extended by nsp12, distancing them from the RTC active-site and limiting their exposure to ExoN proofreading. In comparison, chain-terminating NAs are theoretically more susceptible to excision, as stalling and/or backtracking of the RTC may trigger association of nsp14, while additionally providing a larger timeframe for repair (49). A multi-pronged approach, for dual inhibition of RdRp/ExoN activities, is thus a strategy of choice to capitalize on existing NAs (particularly chain-terminators) whose activity is hampered by this proofreading capacity. Here we investigated the possibility of using α-thio-modified NAs for the treatment of SARS-CoV-2. Incorporation of the activated 5′-thio-TP form would yield viral RNA containing phosphorothioate linkages, which are theoretically more resistant to exonuclease cleavage. This approach therefore provides a strategy for inhibition of both RdRp and ExoN activities by a single NA, circumventing the need for a separate, direct-acting ExoN antiviral.

Based on the potent, chain-terminating 5′-TP, AT-9010, we synthesized the racemic mixture (R_P/S_P) of its corresponding 5′-α-thiotriphosphate, AT-9052 (R_P/S_P). Through enzymatic assays and co-crystallization with the human kinase NDPKb, we were able to determine the absolute configuration of the separated isomers. Interestingly, this result also provided information on the stereospecificity of NDPKb. In the case of both GTP-αS and AT-9052, only the -S_P isomer is efficiently accepted as a substrate. This finding has important biological implications. During activation of Bemnifosbuvir (AT-527), NDPKb is able to perform the final phosphorylation step for formation of the 5′-TP, AT-9010 (unpublished data). Following administration of the corresponding thio-phosphoramidate of AT-527, we expect a similar metabolic pathway to occur. The stereoselectivity of NDPKb would direct phosphorylation of only the -S_P isomer, meaning that only this 5′-αS-TP form would be available to target viral enzymes.

Fortunately, the S_P isomer is also the preferred substrate for the SARS-CoV-2 RdRp. Similarly, other studies have also shown that most polymerases display S_P stereopreference for 5′-αS-NTPs (38,40,41,50). In contrast, AT-9052-R_P is poorly incorporated, explained by the fact that its sulfur atom would disrupting metal ion coordination. Like the parent compound, AT-9052-S_P is incorporated opposite cytosines in the template, where it promotes immediate chain-termination. Pre-steady state rapid-quench experiments show that the 2′-fluoro 2′-methyl ribose modification principally impacts the chemistry step of nucleotide incorporation, rather than NTP binding. Maximal nucleotide incorporation rates (k_max) are considerably reduced for both analogues compared with GTP. Furthermore, the presence of the S_P α-thiophosphodiester in AT-9052 translates to an approximately 4-fold lower overall incorporation efficiency (k_max/K_d) compared with the α-phosphodiester in AT-9010. While the S_P sulfur is well-accommodated in the active-site pocket, it may influence the positioning of Arg555, a residue located in the fingers domain and known to be highly flexible in nsp12 (33,51). The flexibility of this residue (seen when comparing the >40 SARS-CoV-2 RdRp structures) likely aids in the acceptance of a sulfur atom at this location, and the ability to use the AT-9052-S_P isomer as a substrate, albeit at a minor cost to incorporation efficiency. Arg555, along with palm-domain residue Ser759, has been shown to regulate the speed vs. fidelity balance during RNA replication in CoVs (33). The 2′-ribose and thiophosphate modifications would therefore have a double-effect, influencing interactions with both of these residues, accounting for the reduced catalytic rate.

We find that GTP has two binding/incorporation modes, resulting in two rates (fast and slow) of incorporation into RNA. Comparing the faster rate of GTP incorporation with the single phase obtained for AT-9010 and AT-9052-S_P shows a ∼40-fold and ∼166-fold reduction in overall incorporation efficiency (k_max/K_d), respectively. This is much higher than the discrimination values obtained from competition experiments performed under steady-state conditions, suggesting other factors are at play during multiple-incorporation events. When the SARS-CoV-2 RTC is provided with both GTP and AT-9052-S_P in direct competition experiments, AT-9052-S_P is incorporated in the place of GTP as frequently as 1 in every 60 times. Replication of the large (>30 kb) SARS-CoV genome, which contains >5400 templating cytosines, should allow ample opportunity for AT-9052-S_P insertion into viral RNA, provided significant pools are generated to compete with high (∼500 μM) intracellular concentrations of GTP (36).

Following insertion into RNA, the resulting terminal α-thiophosphodiester internucleosidic linkage is completely resistant to cleavage by the viral exonuclease nsp14. In comparison, incorporated AT-9010 is able to be excised, albeit at modestly slower rates than native nucleotides. The resistance of incorporated AT-9052-S_P to nsp14 cleavage therefore likely compensates for its lower incorporation rates. Even a single incorporation per genome would be sufficient, as the truncated RNA products would be unable to be rescued by the viral proofreading machinery.

Several other studies have also shown that following the incorporation of an S_P-NTP-αS into RNA, the resulting product is resistant to exonuclease degradation (42,45,52,53). This includes a recent study on SARS-CoV-2 ExoN, where incorporation of either S_P-CMP-αS or S_P-AMP-αS by the mitochondrial DNA-dependent RNA polymerase (DdRp), POLRMT, rendered products resistant to nsp14 cleavage (45). These studies all assume that the polymerization reaction proceeds with an inversion of configuration at the phosphorus atom, resulting in formation of an R_P α-thiophosphodiester linkage. However, this is somewhat contradictory with the fact that S_P-phosphorothioate linkages are generally known to be more resistant to 3′-5′ exonuclease degradation (48,54–56). In support of this, a previous structure-function study on the 3′-5′ exonuclease of the large fragment of E.coli DNA polymerase I shows that the complex bound to a DNA containing a S_P-phosphorothioate linkage causes severe structural rearrangements and prevents metal ion binding, thus inhibiting any excision (48). In contrast, the R_P-phosphorothioate linkage caused only minor structural rearrangements of the active site and metal ion coordination. This translated to only a ∼15-fold decrease in excision rate (48), similar to reports from other studies (53,57).

The frequent assumption that inversion of configuration occurs following phosphodiester bond formation (or cleavage), may need to be reconsidered. It is highly possible that not all polymerases, nor all exonucleases, share the same stereoselectivity and stereospecificity. The results obtained in this study suggest that inversion of configuration has not occurred. From a structural perspective, retention of the S_P configuration more logically explains the complete resistance to excision following incorporation of AT-9052-S_P.

Modifying existing ProTides (‘pronucleotide’ NAs containing transient α-phosphoramidate moiety) would allow direct delivery of α-thiophosphoramidate prodrugs. In support of this, ALS-2200; a racemic mixture of phosphorous diastereomers of a uridine nucleoside phosphoramidate prodrug, and VX-135; the pure S_P isomer of ALS-2200, have shown promising results in vitro (high potency and low toxicity) and in clinical trials against Hepatitis C Virus (HCV) (58,59). Further research should be directed at studying the activation and metabolism in host cells, including the stereoselectivity of host enzymes. Of importance, many ProTides require intracellular activation by Histidine triad nucleotide-binding protein 1 (HINT1), an enzyme which hydrolyses P–N bonds, for removal of the prodrug moiety and formation of the 5′-monophosphate intermediate. HINT1 is also known to catalyze the cleavage of P-S bonds, resulting in substitution of the sulfur with oxygen and conversion back to the parent (non-thio) monophosphate (60–62). Although this side-activity has been shown to be relatively slow, even for native nucleosides (63), the stability of AT-9052 and precursors remains to be tested in biological media.

Interestingly, in this specific case, partial desulfuration would result in the simultaneous production of AT-9010. We have previously shown that AT-9010 targets both the viral RdRp and its N-terminal NiRAN domain, where it likely impedes capping of the viral genome (17,18). In contrast, the sulfur substitution on AT-9052 results in a loss of NiRAN-inhibition and a concomitant gain of ExoN resistance. Partial desulfuration of its phosphorothioate prodrug may therefore result in the delivery of two different active triphosphates with distinct mechanisms of action—inhibiting NiRAN, RdRp and ExoN activities.

Here, we have provided evidence for a novel potential treatment strategy for SARS-CoV infections with α-phosphorothioate NAs. This strategy might potentiate or even rescue the activity of other chain-terminating NAs by increasing production of truncated, non-functional viral RNAs. In contrast to AT-9010, neither AT-9052-S_P nor R_P isomers are able to inhibit the RNAylation capping activity provided by the NiRAN domain. We thus conclude that the single atom modification of AT-9010 changes the antiviral target and mechanism of action of this drug. α-phosphorothioate modified prodrugs represents a promising strategy, worthy of further investigation for the treatment of all pathogenic viruses carrying exonuclease proofreading activity.

Data availability

Atomic coordinates and structure factors for the reported crystal structure has been deposited with the Protein Data bank under accession number 8PYW.

Supplementary data

Supplementary Data are available at NAR Online.

Acknowledgements

The authors acknowledge SOLEIL for provision of synchrotron radiation facilities and we would like to thank beamline teams of Proxima-1 and Proxima-2 for their assistance during data collection.

Funding

Fondation pour la Recherche Médicale (Aide aux Équipes); SCORE project H2020 SC1-PHE-Coronavirus-2020 [101003627]; ATEA Pharmaceuticals, the Innovative Medicines Initiative 2 Joint Undertaking (IMI-CARE) [101005077]; French Infrastructure for Integrated Structural Biology (FRISBI) [ANR-10-INSB-05-01]. Funding for open access charge: Innovative Medicines Initiative 2 Joint Undertaking (IMI-CARE) [101005077].

Conflict of interest statement. S.G., A.M. and J-P.S. are employees of ATEA Pharmaceuticals, Inc. The other authors declare no competing interests.