Piece of the Puzzle: Remdesivir disassemble the multimeric SARS-CoV-2 RNA-dependent RNA Polymerase Non-Structural Proteins (RdRp-NSPs) complex

The recently emerged SARS-like coronavirus (SARS-CoV-2) has continued to spread rapidly among humans with alarming upsurges in global mortality rates. A major key to tackling this virus is to disrupt its RNA replication process as previously reported for Remdesivir (Rem-P 3 ). For the rst time, we modeled the binding of Rem-P 3 to SARS-CoV-2 RdRp-NSPs complex, a multimeric assembly that drives viral RNA replication in human hosts. Findings revealed that while ATP-binding stabilized the replicative tripartite, Rem-P 3 disintegrated the RdRp-NSP complex, starting with the detachment of the NSP7-NSP8 heterodimer followed by minimal displacement of the second NSP8 subunit (NSP8 II ). More so, Rem-P 3 interacted with a relatively higher anity (ΔG bind ) while inducing high perturbations across the RdRp-NSP domains. D452, T556, V557, S682, and D760 were identied for their crucial roles in stacking the cyano-adenosine and 3,4-dihydroxyoxolan rings of Rem-P 3 while its exible P 3 tail extended towards the palm domain blocking D618 and K798; a residue-pair identied for essential roles in RNA replication. However, ATP folded away from D618 indicative of a more coordinated binding favorable for nucleotide polymerization. We believe ndings from this study will signicantly contribute to the structure-based design of novel disruptors of the SARS-CoV-2 RNA replicative machinery.


Introduction
Coronaviruses (CoVs) are a representative group of single-stranded, positive-sense RNA-viruses that are known to cause severe respiratory disorders, gastrointestinal and central nervous system (CNS) diseases in both human and animals. 1 Severe acute respiratory syndrome CoV-2 is a newly emerged viral strain from the Coronaviridae family currently causing a global pandemic with severe threats to millions, and peradventure, billions of the world's population. Middle East respiratory syndrome (MERS) CoV and Severe acute respiratory syndrome CoV are other members of the family earlier reported, with mortality rates far lesser than SARS-CoV-2. [2][3][4] Generically referred to as COVID-19, SARS-CoV-2 has recorded more mortality rates globally than MERS and SARS-CoV combined. 5 So far, deaths from SARS CoV-2 have been recently estimated at 31,412 with about 667,090 con rmed cases in about 183 countries of the world. (Coronavirus: Euronews) Presently, there is no FDA approved drug or vaccine for the treatment of SARS CoV-2 ( FDA, 2020), the highly urgent need to develop drugs and vaccines that will e ciently curtail its virulence.
SARS CoV-2 replication is very crucial to its pathogenesis even though quite complex. 8 Pivotal to the viral establishment in host cells is the assemblage of cytoplasmic and membrane protected replicationtranscription complexes (RTCs). 9,10 The RTCs most importantly coordinate the expression, replication, and ampli cation of the viral genome. They also create an environment for the virus to evade the hosts' immune system. 11 The production of viral sub-genomic mRNAs, synthesis of new genomic molecules and template strand required for replication are hampered without the RTCs. 12 At the center of the RTCs is the RNA-dependent RNA polymerase (RdRp) subunit 11 . In the CoV genome, the 5'-terminus ORF1a and ORF1b frames encode polyprotein 1a and 1b which are cleaved into 16 non-structural proteins (NSPs) that cumulatively drive the replication and transcription phases of the virus. [13][14][15][16][17] Several viral cycles mediated by the different RdRp-NSPs lead to the full replication and translation of the viral genome. 1 One of these steps is the nucleotide polymerization which involves at least three RdRp-NSPs namely NSP12, NSP7, and NSP8.
The NSP12 subunit is a highly essential RdRp of the viral replicative machinery and reportedly interacts with its cofactors; NSP7 and NSP8 which altogether drive genomic replication. 1,18 Previous reports have emphasized that NSP12 is unable to perform its function as a single entity 19 but must exist in a tripartite complex with NSP7 and nsp8 for activating the replication of long RNAs. 20,21 NSP8 has the capability of initiating the replication process, which accounts for its description as a primase. This subunit can de novo synthesize about 6 nucleotides in length, which could serve as a primer for RNA synthesis by NSP12 RdRp. On the other hand, NSP7 is integral to the CoV replicase polyprotein which also functions as a primase and binds to NSP12. 18,22,23 Together, the NSP7-NSP8 complex enhances the binding of NSP12 to RNA in addition to its enzymatic activity. 8 The functional multiunit complex of SARS-CoV has been structurally resolved, containing one molecule of NSP12, two molecules of NSP8 and one molecule of NSP7 1 . According to their ndings, NSP12 polymerase was bound to a heterodimer of NSP7-NSP-8 while a second NSP8 subunit was bound at a distinct NSP12 site. Therefore, since this assembly is crucial for RNA synthesis, we presume that targeting this complex could be an important strategy to interfere with the viral replication process. This could pave way for the design, synthesis, and repurposing of drugs that can disrupt this RdRp enzyme assembly, and consequently, viral replication.
GS-5734/Remdesivir (Rem), a nucleoside analogue, is a prodrug originally developed to combat Ebola virus, and functions by mimicking adenosine structure ( Figure 1). 24 This drug reportedly converts into a hydrolyzed and tri-phosphorylated (active) metabolite (Rem-P 3 ), a form that enhances its activity as a substrate for RdRp thereby replacing ATP. This event results in the termination of the polymerization process; regarded as 'chain termination'. 25 E cacies of Rem in COVID-19 treatment is currently been evaluated in clinical trials, in the US and China 26 , ever since the possibility was proposed in vitro. Rem exhibits broad-spectrum antiviral activities since most RNA viruses exhibit high structural similarities. 24,27 Remdesivir has been experimentally reported to inhibit the replication of SARS-COV-2 28 , a feat that could facilitate its adoption and approval for COVID-19 treatment. Although, recent studies have reported its NSP8 multimer revealed that the NSP7-NSP8 dimer was bound at loop interfaces of the nger-thumb domain (residues 408-414, 425-447, 552) of NSP12 while regions with residues 257-275, 322-348 and 367-407 served as a binding interface for the second NSP8 subunit. These details agreed with ndings from previous structural studies 1 and were used for exible protein-protein docking.
Molecular docking yielded 10 Rem-P 3 conformers among which the topmost ranked with a docking score of -7.4 kcal mol -1 was selected. This was aligned with the crystallized ATP molecule and yielded an RMSD of 0.2 Å. This further indicated that the inhibitor and substrate were properly aligned at the nucleoside pocket of SARS-CoV-2.
Remdesivir systematically disintegrates the SARS-CoV-2 RdRp assembly To understand possible effects of Rem-P 3 and ATP on the SARS-CoV-2 NSP complex, snapshots were taken from the resulting trajectories at various time-frames as the simulation proceeded. Comparative visual analysis was then carried out for the unbound, Rem-P 3 -and ATP-bound systems.
We observed that the RdRp-NSP multimer remained as a unit in the unbound and ATP-bound systems from the initial to the nal frames while dissimilar structural occurrences were observed in the Rem-P 3bound complex ( Figure 3).
Time-based conformational sampling revealed that the RdRp-NSP assembly was intact in the presence of Rem-P 3 until around 150ns when the NSP8 subunit was rstly detached from its NSP7 counterpart into the solvent environment. This was then followed by the displacement of the NSP7 subunit from the loop interface of the NSP12-thumb domain at the next 20ns time-frame (170ns).
Besides, we observed that the second NSP8 subunit was minimally displaced from its binding interface on NSP12. The disintegrated RdRp-NSP complex was maintained until the end of the simulation ( Figure   4).
Based on previous studies involving SARS-CoV, NSP7 and NSP8 play important roles in de novo initiation, primer extension, RNA synthesis, and replication, particularly when complexed with NSP12. 18,33,34 . Also, an assemblage of the NSP7-NSP8 heterodimer with NSP12 reportedly enhanced its RNA binding and polymerase activities. 8,18 Therefore, the ability of Rem-P 3 to disrupt the RdRp-NSP assembly in SARS-CoV-2 could correlate with its inhibitory activity against the virus. This is a major nding that possibly explains its reported e cacy in the treatment of COVID-19 infections.
We further investigated the effects of both compounds on the integrity of the NSP12 multimer over the simulation period by comparing parameters for the unbound, ATP-bound and Rem-P 3 -bound systems.
Firstly, we estimated the stability of the complexes using the RMSD metrics and result plotted in Figure  5a. As shown, the unbound and ATP-bound complexes were stable from the beginning to the end of the simulations. On the contrary, the Rem-P 3 -bound NSP complex was structurally stable (RMSD < 2Å) until 140ns where high instability was observed, characterized by an unusual spike in Cα motions after which the system was restabilized until the end of the run. This observation could correlate with the disintegration of the NSP7-NSP8 heterodimer from the NSP12 polymerase as reported above ( Figure 4) thereby explaining high conformational perturbations induced by Rem-P 3 when bound to the SARS-CoV-2 RdRp-NSP complex.
A possible explanation is that the disintegration of the NSP7-NSP8 heterodimer could further destabilize the NSP12 subunit, which could, in turn, affect binding interfaces for other NSPs necessary for RNA replication 1,18,34 . Another interesting and important nding was that the ATP-bound NSP-complex was more stable than the unbound system. This could imply that the binding of ATP to NSP12 further enhanced the integrity of SARS-CoV-2 NSP8-NSP12-NSP7-NSP8 complex. To minimize entropical effects, we de ned nally equilibrated RMSDs (FE-RMSDs) from the ultimate time frames (170-200ns) where the systems stabilized. As seen in Figure 5b, huge discrepancies exist between the stabilities of the simulated complexes, with a high FE-RMSD peak for the Rem-P 3 bound system. This could correlate with the systematic disintegration of the RdRp-NSP complex induced by Rem-P 3 relative to ATP which rather stabilized the complex.
Estimated mean FE-RMSDs of the systems were in the order ATP < Unbound < Rem-P 3 as presented in Supplementary Table 1. These stable time-frames were then utilized for subsequent conformational analyses.
Also, we projected the motions of the RdRp-NSP complexes along two principal components (PC 1 vs PC 2 ). From the PCA plot in Supplementary Figure S1, we observed similar directions of motions among the unbound and ATP-bound RdRp-NSPs while a more dispersed motion pattern was observed for the Rem-P 3 -bound complex.
We further monitored systemic uctuations among constituent residues of each NSP subunits using the FE-RMSF metrics, derived from the stable time-frames. Our ndings revealed that per-residual motions were relatively high in the Rem-P 3 -bound NSP complex compared to ATP-bound and unbound systems (Supplementary Figure S2). Collatively, the mean FE-RMSF was lowest in the ATP-system further indicative of the importance of ATP in stabilizing the RdRp-NSP multimeric complex. Estimated mean FE-RMSFs of the systems also followed in the ATP < Unbound < Rem-P 3 order.
We then mapped out the distinct subunits of the RdRp-NSP complex (including NSP12 sub-domains) and measured their uctuations with respect to the binding of the compounds.
Our ndings further emphasized the perturbative effects of Rem-P 3 across the NSP8-NSP12-NSP7-NSP8 assembly. The disruptive effects of Rem-P 3 binding were most pronounced in the NSP7-bound NSP8 heterodimer (FE-RMSF = 41.6Å) and could have led to their detachment from the nger loop interfaces, which was also highly perturbed (FE-RMSF = 9.0Å) according to our calculations ( Figure 6). Cumulatively, high Cα motions induced by Rem-P 3 at the NSP12 subdomains affected the binding and stability of associated NSPs at the interfaces (Supplementary Table 2). More so, high uctuations induced by Rem-P 3 among residues of the NSP12 N-terminal and nger subdomains could have disrupted interface interactions with the second NSP8 subunit (NSP8 II ), which also exhibited high structural perturbations (FE-RMSF = 11.9 Å, 9.0Å).
Interestingly, the NSP12 subdomains, in addition to other NSPs (7 and 8) exhibited minimal residual motions in the presence of ATP, indicative of the stability of the NSP complex, even compared to the unbound system. Hence, we can presume, that the binding of ATP further enhanced structural integrity across the RdRp-NSP assembly.
Differential stabilities of the NSP12-nucleoside sites were further evaluated, and as observed (Figure 5c), Rem-P 3 -bound nucleoside site was more unstable compared to the ATP-bound and unbound sites. This could suggest a possible correlation between the activities of Rem-P 3 at the nucleoside pocket and its disruptive effects on the entire RdRp-NSP complex.
Also, Rem-P 3 induced high Ca motions at the nucleoside site as further estimated using the FE-RoG metrics, which is also an indicator of structural compactness 35 . Relatively, the nucleoside side was more compact when bound by ATP, even compared to the unbound NSP12-pocket (Fig 5d).
Furthermore, we projected trajectorial motions of the compounds at the active pockets with respect to their binding activities. From the PCA plot, we could observe that Rem-P 3 exhibited highly unstable motions at the nucleoside pocket while ATP was coordinately bound with more compact motions indicative of its systemic stability.
Complementarily, we masked the compounds for FE-RMSD calculations to further determine their binding stability. As shown in Figure 7, Rem-P 3 was highly unstable (FE-RMSD = 3.5Å ± 0.5 ) while ATP demonstrated stable motions at the binding pocket (FE-RMSD = 1.5Å ± 0.8).
Also, we observed that ATP assumed a folded conformation at the hydrophobic pocket of the NSP12 subunit, which could have accounted for the compact site architecture (Figure 7d). On the contrary, Rem-P 3 exhibited an extended conformation with its cyano-adenosine ring stacked in between a visible hydrophobic patch (Figure 7b). The effect of this Rem-P 3 binding mode could be seen in the pocket which was less compact.
Using average structures for the Rem-P 3 -and ATP-bound complexes, we closely analyzed relative orientations at the NSP12-nucleoside site and complementary interactions ( Figure 8). As mentioned above, ATP ultimately assumed a compact conformation, with an inwardly folded P 3 tail while Rem-P 3 rather exhibited a linearly stretched conformation.
As observed, its cyano-nucleotide portion extended into the pocket created by D452, T556, V557, T680, S681, and S682 ( nger domain). This ring orientation could allow for high-a nity (NH---O) interactions with D452 and T556, which could, in turn, hold the cyano-adenosine ring in place. More so, its P 3 tail extends into the palm domain towards K798, D618, K621, R553, and D623. Also, the 3,4 dihydroxyoxolan ring was coordinated by N691, S759, and D760. These ndings, altogether, indicate an orientation that supports its inhibitory activities. Relatively, we could observe that the P 3 tail of ATP was folded away from D618 and K798, an important residue-pair responsible SARS-CoV RdRp polymerase activity as previously reported. 36 Binding free energy calculations and decomposition The extent to which the compounds bind to the RdRp-NSP complex was determined using the MM/PBSA method. 37,38 Stable time-frames (170-200ns) were also selected to minimize entropical effects and our results are presented in Table 1. Accordingly, Rem-P 3 was strongly bound than ATP with ∆G bind difference of -13.2 kcal mol -1 . This relatively higher a nity for the NSP12 nucleoside site further re ects its ability to competitively impede and replace the natural substrate. 25 Analysis of the binding components further revealed that the binding of Rem-P 3 was more unfavorable in the polar region as evidenced by higher a ∆G np,sol value. This could suggest that Rem-P 3 does not only block ATP binding but binds deeply in the hydrophobic nucleoside pocket of the NSP12 subunit.
This could favor its retention and involvement with more residues of the pocket. Also, vdW contributions to Rem-P 3 binding was relatively higher, which may compensate electrostatic (∆E ele ) effects that were negated by unfavorable polar solvation energies (∆G solv ). Cumulatively, ∆E vdW and ∆E ele highly favored gas-phase interactions (∆G gas ) of Rem-P 3 relative to ATP.
Also, we calculated the energies of individual residues and their respective contributions to the disparate binding of both compounds. Energy contributions > -1 kcal mol-1 were considered favorable and results are presented in Figure 9.
The roles of D618 and K798 to the binding of Rem-P 3 were further de ned with electrostatic contributions of -11.4 and -1.2 kcal mol -1 respectively. On the contrary, electrostatic effects by D618 were highly unfavorable for ATP (10.1 kcal mol -1 ).
This could suggest that ATP does not interfere with the roles mediated by D618 during nucleotide polymerization. 36 More so, while D452 electrostatically favored Rem-P 3 binding (-5.3 kcal mol -1 ), an unfavorable contribution of +12.5 kcal mol -1 was estimated for ATP. Similar Rem-P 3 → ATP (∆E ele ) transitions were observed for T556 (-5.4 → 1.9 kcal mol -1 ), D623 (-10.1 → 9.0 kcal mol -1 ), D760 (-11.2 → 9.2 kcal mol -1 ) and D761 (-8.3 → 7.4 kcal mol -1 ). Presumably, residues that favored Rem-P 3 binding at the nucleotide site of RdRp-NSP12 rather contributed unfavorably to ATP binding, which could limit ATP a nity in the region. Also, in Rem-P 3 , unfavorable ∆E ele contributions were highly compensated for by per-residue ∆E vdW energies which were favorable in R555 (-2.9 kcal mol -1 ), T556 (-1.4 kcal mol -1 ), V557 Taken together, we could presume that Rem-P 3 binds uncoordinatedly at the NSP12 nucleoside site, interacting with a high number of residues along its path. This binding pattern favorably enhanced its a nity and, perhaps, longer retention at the NSP12 site until it exerts maximum destabilizing effects su cient to disrupt the RdRp-NSP multimeric assembly.
Relatively, the ability of ATP to fold its exible P 3 tail away from D618 and K798 in a compact conformation indicates a selective and coordinated binding, favorable for the nucleotide polymerization process.
In this study, we investigated the differential binding of Rem-P 3 and ATP to the RdRp-NSP assembly. This involved subunit binding of an NSP7-NSP8 heterodimer to NSP12 while a second NSP8 (NSP8 II ) was bound at a distant site. The complex, although not currently available for SARS-CoV-2 at the time of ling this report, was modeled using a structurally similar SARS-CoV RdRp-NSP complex (PDB ID: 6NUR). The Rem-P 3 -, ATP-and Unbound complexes were subjected to GPU-accelerated molecular dynamics simulation of 600ns after which the trajectories were sampled conformationally. Findings revealed that the RdRp-NSP complex was more stable in the presence of ATP, even compared to the unbound system while on the other hand, Rem-P 3 appeared to disintegrate the complex. The NSP7-NSP8 heterodimer was rstly detached at ~150ns from the NSP12 while the second NSP8 (NSP8 II ) subunit was seen to be minimally displaced at the ultimate time-frame. Conformational analysis performed with RMSD and RoG parameters further revealed that Rem-P 3 induced a high degree of structural instability compared to the ATP and unbound system. RMSF analysis revealed relatively higher subdomain perturbations in the Rem-P 3 -bound RdRp-NSP complex. Also, analysis of ligand motions revealed that ATP exhibited a more stable and compact motion compared to Rem-P 3 , which could have also re ected on the active site, which was less deviated and more compact in the ATP-system. Analysis of binding dynamics revealed that Rem-P 3 exhibited an extended conformation that allowed it to interact with D542, T556, V557, T680, S681 and S682 of the NSP12-nger domain via its cyano-adenosine ring while its 3,4 dihydroxyoxolan ring oriented towards D760 and N961 forming high-a nity H-bonds. Also, its P 3 tail was extended towards the palm domain, interacting with D618 and K798. This binding pattern clearly interferes with the roles mediated by D618 and K798 in RNA replication process. On the contrary, ATP exhibited a more selective and coordinated binding wherein it assumed a folded (compact) conformation away from these residues.
Binding energy value of -33.4 kcal mol -1 was estimated for Rem-P 3 while ATP had a ΔG bind estimate of -20.5 kcal mol -1 . This further re ected the ability of Rem-P 3 to competitively replace ATP at the NSP12nucleoside pocket. Per-residue energy decomposition further emphasized the roles of D542, T556, V557, T680, S681, S682, D618 and K798 to the high-a nity binding of Rem-P 3 . We believe ndings from this study will signi cantly contribute to drug design endeavors, particularly structure-based approaches. This could yield novel compounds that can bind selectively and strongly to the NSP12-nucleoside site and disrupt interactions with other non-structural proteins.

Methods
Structural modeling of ligand and protein complexes Based on the rationale of this study, it was expedient to model the NSP8-NSP12-NSP7-NSP8 multisubunit complex for SARS-CoV-2 to which the inhibitor (Rem-P 3 ) and substrate (ATP) will be differentially bound. Homology models of each protein were built by MODELLER 9.18, a structural re nement tool, for which SARS-CoV NSP12, NSP7 and NSP8 crystal structures (PDB ID: 6NUR) 1 were used as templates. Amino acid sequences for the SARS-CoV-2 NSPs were retrieved from the NCBI with entries YP_009725303.1 (NSP7), YP_009725304.1 (NSP8) and YP_009725307.1 (NSP12). From the modeled structures, the ones with the best DOPE scores were selected. To obtain the tetramer assembly similar to the crystallized complex in 1 , modeled SARS-CoV-2 NSP12 was rst superimposed with a 6NUR_multimeric complex (SARS-CoV) after which active residues involved in NSP12, NSP7 and NSP8 interactions on SARS-CoV-2 were mapped out ( Figure 2). These coordinates were then used to generate the SARS-CoV-2 multimer in a three-stepped exible protein-protein docking procedure on the High Ambiguity Driven protein-protein DOCKing (HADDOCK 2.2) tool; step 1: NSP7-NSP8, step 2: the NSP7-NSP8-NSP12 and nally, step 3: NSP7-NSP8-NSP12-NSP8 multimeric complex after which 1000 energy minimization steps were performed to relax the model.
We then de ned the nucleoside region in the modeled SARS-CoV-2 multimer for binding simulation studies with ATP and Rem-P 3 . This was achieved by aligning the modeled complex with ATP-bound poliovirus RdRp (PDB ID: 2ILY). 29 Upon generation of the ATP-bound SARS-CoV-2 NSP7-NSP8-NSP12-NSP8 multimer, a grid box was centered on the ATP site to which optimized Rem-P 3 was docked using the Vina module. 39 Prior to this, Rem-P 3 was prepared using Gaussview after which the 2D structure was minimized at the B3LYP/6-311++G(d,p) theory level on Gaussian16 program package. 40 The best binding pose was aligned with the crystallized nucleoside in 2ILY, and was assumed as the starting structure for the ATP molecule and Rem-P 3 in the SARS-CoV-2 RdRp multimer.

GPU-accelerated molecular dynamics (MD) simulation
Having modeled the SARS-CoV-2 multimer and docked the compounds at the nucleoside binding site, the systems: NSP8-NSP12-NSP7-NSP8-ATP, NSP8-NSP12-NSP7-NSP8-Rem-P 3 and unbound NSP8-NSP12-NSP7-NSP8, were subjected to simulation runs of 600ns on Amber 18 Graphical Processor Unit 41 which enabled accelerated production runs. Parametrization of the individual receptor was performed on FF14SB force eld. Also, Antechamber and Parmchk modules were used to generate .frcmod les for ATP and Rem-P 3 . The LEAP module was then used to generate topology and parameter les for the complexes in addition to system neutralization and explicit solvation. 42 These complexes were then minimized for 2500 steps with a 500kcal/mol Å 2 restraint potential and also for 5000 steps with no restraints. Simultaneous heating (0-300k) and equilibration steps were performed followed by production runs that were restarted subsequently. Resulting trajectories were saved at every 1ns after which they were analyzed with the integrated CPPTRAJ module. 43 Relative stabilities of the SARS-CoV-2 multimers were determined by measuring the Cα-root mean square deviations (RMSDs) while other metrics such as the root mean square uctuation (RMSF), radius of gyration (RoG) and solvent accessibility surface area (SASA) were used to measure per-residual motions, structural compactness, and solvent-surface motions. [30][31][32]35,44,45 Trajectorial motions of the molecules were also projected along two principal components (PC1 and PC2) for insights into their dynamics and motion patterns. 46    Analysis of Rem-P3 and ATP motions, and orientations at the active site region of RdRp-NSP12. a. Ligand FE-RMSD b. Binding modes of Rem-P3 and d. ATP at the hydrophobic NSP12 pocket. Interacting residues are also shown in addition to the degree of hydrophobicity color coded from least → highly hydrophobic; -3.00 → 3.00. c. PCA clustering of ligand motions over the simulation period.

Figure 8
Relative orientations and interaction analyses of a. Rem-P3 and b. ATP at the NSP12 nucleoside site.
Complementary H-bond interactions are also pin-pointed.