In silico structural elucidation of Nipah virus L protein and targeting RNA-dependent RNA polymerase domain by nucleoside analogs

Abstract The large (L) protein of Mononegavirales is a multi-domain protein that performs transcription and genome replication. One of the important domains in L is the RNA-dependent RNA polymerase (RdRp), a promising target for antiviral drugs. In this work, we employed rigorous computational comparative modeling to predict the structure of L protein of Nipah virus (NiV). The RdRp domain was targeted by a panel of nucleotide analogs, previously reported to inhibit different viral RNA polymerases, using molecular docking. Best binder compounds were subjected to molecular dynamics simulation to validate their binding. Molecular mechanics/generalized-born surface area (MM/GBSA) calculations estimated the binding free energy. The predicted model of NiV L has an excellent quality as judged by physics- and knowledge-based validation tests. Galidesivir, AT-9010 and Norov-29 scored the top nucleotide analogs to bind to the RdRp. Their binding free energies obtained by MM/GBSA (−31.01 ± 3.9 to −38.37 ± 4.8 kcal/mol) ranked Norov-29 as the best potential inhibitor. Purine nucleotide analogs are expected to harbor the scaffold for an effective drug against NiV. Finally, this study is expected to provide a start point for medicinal chemistry and drug discovery campaigns toward identification of effective chemotherapeutic agent(s) against NiV. Communicated by Ramaswamy H. Sarma


Introduction
Nipah virus (NiV) is a negative-sense, nonsegmented enveloped RNA virus discovered in 1998 as a deadly zoonotic agent within the Henipahvirus genus, after Hendra virus (Family: Paramyxoviridae). The first documented outbreak of NiV occurred in Malaysia in the form of encephalitis cases among pigs (Chua et al., 1999). The infection was subsequently transmitted to humans in close contact with infected pigs leading to 265 cases and 105 deaths by the end of May 1999 (Chua et al., 2000). Later outbreaks occurred in Singapore, the Philippines, Bangladesh and India with a case fatality rate approaching 80% (Pillai et al., 2020). In Bangladesh, frequent outbreaks are reported, especially during winter (Sharma et al., 2019). In addition, epidemiological and viral studies identified fruit bats (Pteropus sp.) (a.k.a. flying foxes) as the natural host of NiV and Hendra virus, the causative agent of outbreaks among horses in Australia (Ang et al., 2018).
In 2018, the World Health Organization (WHO) placed NiV among the top 10 infectious diseases that pose public health risks due to their pandemic potential (WHO, 2018). The pathophysiology of NiV infection is still largely unknown. However, inflammation of vasculature is prominent in the lungs, brain, heart and kidneys . Following an average incubation period of 10-14 days (range: 2-120 days), clinical features of NiV infection manifest as fever, headache, vomiting, cough, respiratory difficulties and neurological abnormalities (Ang et al., 2018;Chua et al., 2000). Symptoms seem to be strain specific, as respiratory manifestations were rare in the Malaysian outbreak, but prevalent in subsequent outbreaks in India and Bangladesh, where the infections were caused by a second strain .
NiV genome is 18.2 kb, encoding six structural proteins (N, P, M, F, G and L) and additional three nonstructural proteins (V, C, W). L gene encodes the largest protein (2244 amino acid) responsible for transcription and genome replication. The structure of L is still unsolved; however, experimentally solved L proteins from different viruses within the order of Mononegavirales share a similar multidomain arrangement [reviewed in (Liang, 2020)]. The N-terminal domain of L is the RNA-dependent RNA polymerase (RdRp) that shares a similar cupped right-hand conformation; palm, finger and thumb with other polymerases. The second domain of L is a poly-ribonucleotidyltransferase (PRNTase) believed to be responsible for the 5 0 capping of mRNA. The methyltransferase (MTase) domain methylates the 5 0 cap of mRNA transcripts. The remaining parts are expected to provide structural support for the enzymatically active domains (Velthuis et al., 2021). Viral genomic RNA is wrapped in helical assemblies of nucleoprotein (N) monomers that protect the viral genome (and antigenomes) from cellular nucleases (Ker et al., 2021;Ogino & Green, 2019). In Mononegavirales, the ribonucleoprotein complex, rather than the RNA alone, is the actual template for the transcription/replication complex, during which the RNA becomes naked only for transient periods during which it passes through the L protein. Upon leaving the L, viral genome (and antigenome) is re-capsidated by N monomers that are immediately assembled in close proximity to L protein.
The wrapping and unwrapping of genomic RNA is a dynamic process mediated by the phosphoprotein (P) that associates with L and N proteins during RNA synthesis (Ogino & Green, 2019). Indeed, P protein is an important cofactor for replication and transcription of Mononegavirales genomes.
Due to its pivotal role in virus replication, the RdRp domain is commonly targeted in drug discovery studies (Elfiky, 2022;Elfiky et al., 2022;Ganeshpurkar et al., 2019). The conserved nature of the RdRp and lack of homologous protein in humans add an extra advantage for developing safe, specific and wide-spectrum antiviral drugs. Nucleoside analogs have been successfully used to inhibit RNA polymerization in different viruses (Geraghty et al., 2021). Ribavirin is a nucleoside analog that has been used in treatments of infections caused by Respiratory Syncytial Virus (RSV), Hepatitis C Virus (HCV) and other hemorrhagic viruses (Graci & Cameron, 2006). It has also been used for NiV infections in humans in Malaysia but its therapeutic value is unclear (Banerjee et al., 2019). In hamster model, ribavirin did not reduce the mortalities but only delayed the death of the experimentally challenged animals (Freiberg et al., 2010).
Recent developments in structural biology and computational chemistry have improved the process of drug discovery and drug repurposing for the treatment of different diseases, including viral infections (Li & Peng, 2021). Promising antiviral compounds have been identified computationally for the treatment of epidemic viruses such as influenza virus (Du et al., 2012), Ebola (Shurtleff et al., 2012), Zika (Pattnaik et al., 2018), Dengue (Zhou et al., 2008) and SARS-CoV-2 (Adem et al., 2022;Elgohary et al., 2022;Elshemey et al., 2022). Since no approved antiviral drugs or vaccines are available to control NiV, this in silico study aimed at modeling L protein and targeting the RdRp domain via repurposing different nucleoside analogs as potential inhibitors of NiV.

Domains detection and prediction of L protein structure
The amino acid sequence of L protein was retrieved from NCBI database (ID: NP_112028.1) and subjected to a position-specific iterative BLAST (PSI-BLAST) search against Protein Data Bank (PDB) entries. The L protein of Human Parainfluenza virus 5 (HPIV-5) (PDB ID: 6V85) was the most homolog sequence available with a pairwise identity of 32.39% with NiV L protein sequence. Primary full-length models of L protein were built by SWISS-MODEL (Waterhouse et al., 2018) and Phyre2 (Kelley et al., 2015) based on the crystal structure of HPIV-5. To delineate the domains of NiV L protein, multiple sequences of L proteins from different viruses within Mononegavirales were aligned by ClustalX tool (Sievers et al., 2011;Supplementary Figure  S1). Structural superimposition of generated primary models and results of the multiple sequence alignment were used to delineate the boundaries of domains in NiV L protein.
The RdRp domain was found to be 971 residues. The residues P604-G709 in the RdRp domain have no homologous structure in the Protein Data Bank; hence, this part of sequence was not predicted in a convenient fold in all the primary models. Therefore, this stretch of residues was deleted before constructing the model of the RdRp domain. The residue numbering corresponds to the full amino acid sequence is in parentheses throughout the manuscript. All domains of L protein (RdRp, PRNTase and CD-MTase-CTD) were modeled separately by AlphaFold2 (Jumper et al., 2021;Mirdita et al., 2022), based on solved L proteins of HPIV-5 (PDB ID:6V85), HMPV (PDB ID: 6U5O), Rabies virus (PDB ID: 6UEB), Vesicular stomatitis virus (PDB ID: 5A22 and 6U1X) and RSV (PDB ID: 6PZK and 6UEN). The final full model was constructed by integrating analogous template alignments with deep-learning predicted inter-domain spatial restraints approach in the DEMO server (Zhou et al., 2022). The models were subjected to quality assessment via MolProbity (Davis et al., 2007), Ramachandran plots (Supplementary Figure S2) and Z-score (Supplementary Figure S3) (Wiederstein & Sippl, 2007). The distribution of secondary structures across the L protein is depicted graphically in Supplementary Figure S4.

Molecular dynamic simulation of RdRp
Prior to molecular dynamic simulation, the crystal structure of HCV (PDB ID: 4WTD) was used as a guide for adding divalent metals (Zn 2þ ) into the active site of NiV RdRp. To mimic the coordination geometry of metals seen in the guide structure, side chains of Asp722 and Asn723 (832 and 833) were modified via Dunbrack library of rotamers in UCSF Chimera (Pettersen et al., 2004;Shapovalov & Dunbrack, 2011). The model was finally subjected to all-atoms molecular dynamic (MD) simulation by GROMACS 2021.3 package (Abraham et al., 2015) using the CHARMM36 forcefield. The solvated system with an explicit TIP3P water model and ions (Na þ Cl À at 0.15 M) in a water box was minimized by the steepest descent energy algorithm for a maximum of 5000 steps. The LINC algorithm was used to constrain all bonds containing hydrogen atoms, whereas long-range electrostatic interactions were calculated by the Particle Mesh Ewald algorithm.
System equilibrium was performed in two steps. First, the constant particle number, volume and temperature (NVT) ensemble was applied for 2 ns where the temperature was coupled by the Berendsen thermostat (Berendsen et al., 1984). In the second step, the constant particle number, pressure and temperature (NPT) dynamics were applied for two nanoseconds where the velocity-rescaling algorithm and Parrinello-Rahman barostat-controlled temperature and pressure, respectively (Bussi et al., 2007;Parrinello & Rahman, 1981). For the production run, all restraints were removed, and temperature and pressure were coupled as in the NPT ensemble for 100 ns at 310 K and 1 bar. MD simulation trajectory was analyzed by the built-in modules of GROMACS. Simulation trajectory was clustered and a representative frame per cluster were obtained (Tubiana et al., 2018).

Receptor and drugs preparation
Representative models resulting from clustering were cleaned and protonated in the Hþþ server (Gordon et al., 2005) at the physiological pH (7.4), internal and external dielectrics of 10 and 80 and salinity of 0.15 M. The protonated structures were prepared in AutoDock Tools 1.5.6 as described previously (Forli et al., 2016). Residues of the active site (aspartate, and asparagine) were treated as flexible during the molecular docking process. On the other hand, a panel of nucleosides analogs known to inhibit RdRp of different viruses were collected from literature and downloaded in 2D structure files (SDF) from PubChem (Supplementary Table  S1). The triphosphate moiety was added to the drugs via Marvin Sketch and the overall structure was corrected via Structure Checker (version 21.20.0, ChemAxon https://www. chemaxon.com). The 3D conformers were generated and minimized by OpenBabel software (O'Boyle et al., 2011) using MMFF94 forcefield (Halgren, 1996). Drug protonation at a pH of 7.4 was done during PDBQT file generation.

Molecular docking of nucleosides analogs
The drugs were docked into the RdRp active site by AutoDock Vina (Eberhardt et al., 2021;Trott & Olson, 2010). The grid box was centered at the active site and set to 30 � 30 � 30 Å with exhaustiveness of 256. Other parameters were accepted in their default values. Docking results were inspected visually to choose the most accurate poses in which the triphosphate moiety interacts with the active site (including the divalent metals) while the nitrogenous base is protruding towards the cavity of the receptor or any side of the cavity. The four types of ribonucleotides, as well as amifloxacin and indoprofen were also docked to serves as positive and negative controls, respectively (Elfiky, 2020). Best docking complexes were analyzed by the PLIP server (Adasme et al., 2021).

MD simulation of RdRp-analogs complexes
Systems of the top three analogs and the GTP complexes with the RdRp after docking were prepared by CHARMM-GUI . During systems preparation, drug topology files were obtained from CGenFF server (Vanommeslaeghe et al., 2010) during systems preparation. The complexes were solvated via the TIP3P water model and Na þ Cl À ions were added at the physiological concentration (0.15 M). Other MD parameters were similar to the previously described for the RdRp simulation. Post-dynamic analyses of trajectories were performed by built-in modules in the GROMACS package.

Binding energy calculations
Binding free energy evaluation of drugs was performed by calculations of molecular mechanics/Generalized-Born Surface Area (Wang et al., 2019) in the gmx_MMPBSA program (Vald� es-Tresanco et al., 2021). The last 70 ns of the simulation was used to estimate binding free energy using the Mongan-Simmerling model of Generalized Born (gb ¼ 8) (Mongan et al., 2007).

NiV L overall structure
Five domains in the L protein of NiV were delineated by multiple sequence alignment and structural superimposition of NiV primary models with the L from HPIV-5 (Figure 1(A)). NiV L model showed domains arrangement similar to other viruses within the Mononegavirales order with a similar structural folding (Figure 1(B)). All L proteins of Mononegavirales are large (�250 kDa), multifunctional complex structures that carry out transcription and genome replication processes (Liang, 2020). Structurally, L proteins have five conserved domains: (i) RdRp that catalyzes phosphodiester bond formation between ribonucleotides, (ii) polynucleotidetransferase (PRNTase) adding G-cap to the 5 0 -end of mRNA transcripts, (iii) connector domain (CD) required for structural stability, (iv) methyltransferase (MTase) for methylation of the 5 0 -caps on mRNA products, and (v) the C-terminal domain (CTD) whose function is still unknown (Velthuis et al., 2021). The sequence alignment of multiple L proteins also revealed five conserved regions (CRs), the first three regions (CR-I to CRIII) are parts of the RdRp domain, while CR-IV and CR-V are in the PRNTase domain. In contrast, the MTase domain harbors the least conserved region (Supplementary Figure S1). The overall structure of NiV L showed domain architecture very similar to previously solved structures of VSV (PDB ID: 5A22) and rabies virus (PDB ID: 6UEB) (Figure 2).
The predicted model for L has a good quality judged by the Ramachandran plot (Supplementary Figure S2), Z-score of À 20 (Supplementary Figure S3) and MolProbity score of 1.10, equivalent to the X-ray resolution value of experimentally solved structures (Davis et al., 2007). Z-score is an overall indication of protein model structure quality generated by contrasting the total energy deviation of the model against a distribution range calculated from experimentally solved structures deposited in the PDB (Wiederstein & Sippl, 2007). Validation criteria are summarized in Table 1.
The structural superimposition of NiV L with its counterparts in RSV, Rabies and HPIV-5 shows differences in the spatial arrangements of the last three domains (Figure 2). These differences are likely triggered by conformational changes during the adoption of transcription or replication competent conformation (Abdella et al., 2020;Ruedas & Perrault, 2014). Indeed, these domains experience large-scale rearrangements and dissociation from the RdRp-PRNTase core or even be lost during protein preparation for cryo-EM, as seen in RSV maps (Abdella et al., 2020;Cao et al., 2020;Gilman et al., 2019). The observed motion of the last three domains in HPIV-5 and Rabies virus L proteins suggested that the movement of CD, MTase and CTD domains away from the RdRp-PRNTase module is required for capping and methylation (Abdella et al., 2020;Horwitz et al., 2020). Additionally, another replication model of viral genome suggests that MTase and CTD exchange their positions in relation to the RdRp-PRNTase module to provide a chance for N protein monomers to be recruited and assembled near the exit tunnel of genome/antigenome in the replicative conformer. However, another arrangement is expected to fulfill methylation needs for mRNA transcripts (between MTase and PRNTase) (Abdella et al., 2020). Nonetheless, the RdRp and PRNTase domains are highly conserved in their fold and subdomains (RMSDs: 3.323, 5.138 and 7.392 Å for HPIV-5, RSV and rabies virus, respectively) despite the low sequence identity (20-32%) between these proteins. These findings support previous expectations for the conserved architecture of RdRp despite low sequence similarities (Venkataraman et al., 2018).
The capping of mRNA in Mononegavirales is mediated by the covalent binding of the first nucleotide to a conserved histidine residue (HR motif within the CR-V) in the PRNTase domain (Liang, 2020). In the NiV L model, this histidine (His1237) is located under the CD domain. The priming loop (1145-1171) containing the GxxT motif is modeled away from the central cavity as in the solved structures of RSV and VSV, suggesting a noninitiation state conformation. In Mononegavirales, the GxxT motif is conserved as GSxT. In addition to its essential role in binding the cap guanosine, this motif is also involved in the initiation of transcription. Indeed, alanine mutagenesis of the glycine residue resulted in defects in the initiation of replication, elongation and capping reactions (Braun et al., 2017). Side chains of the A/ GxGxG motif residues (1843-1847) located within the CR VI are grouped together to form the conserved binding site of S-adenosylmethionine, the methyl group donor. The KDKE motif is a catalytic tetrad for the methylation reaction (Paesen et al., 2015). In the NiV L model, these residues are K1711, D1830, K1866 and E1903 (corresponding to K1821, D1940, K1976 and E2013 in the nontrimmed amino acid sequence). Finally, the CD and CTD parts have no enzymatic activities and, consequently, have the least conserved sequences. The CD domain consists of 11 A-helices, two parallel b-sheets and a long linker loops flanking the domain  (�30 residues at both ends). Similarly, the CTD domain has seven A-helices and two antiparallel b-sheets.

The RdRp domain
As in other polymerases, the NiV RdRp domain has the canonical ''fingers, palm and thumb' configuration with the conserved motifs (Figure 1(C) and (D)). Helices are the predominant secondary structure within fingers and thumb subdomains, and these parts have the largest interface area with the PRNTase domain. The conserved glycine residue (G789) that allows a small degree of flexibility to the palm and thumb subdomains is located at a b-turn as seen in RSV and enterovirus (Gilman et al., 2019). The conserved active site (GDN) is located at a short loop separating the two b-sheets of motif B (Figure 1(D)). In motif A, Asp612 is responsible for coordinating metal ions with the active site residues (Asp722 and Asn723) in a similar fashion predicted in RSV structure (Braun et  Despite being an essential cofactor for RNA stability and phosphodiester bond formation, divalent metal ions were absent in all solved structures of L protein from viruses within the Mononegavirales class. The RdRp of NiV shares a similar local arrangement in the active site with the solved polymerase of HCV; hence, it was used to guide the addition of two zinc ions before MD simulation of the structure. To assess model stability during MD simulation, the root-meansquare deviation (RMSD) of backbone atoms was analyzed.
After the first 20 ns of simulation, the model of NiV RdRp showed a stable conformation with only minor deviations of 0.28 Å around the average RMSD (2.01 Å). The first 13 residues at the N-terminal end were excluded from RMSD calculations due to their extreme fluctuations and complete absence of secondary structures. Over the simulation period, up to 97% of frames have a RMSD between 1.60 and 2.36 Å. RdRp radius of gyration and solvent accessible surface area remained stable during the simulation; both confirm the structural stability of the model. The clustering of the RdRp simulation trajectory resulted in six clusters and a representative frame for each (Figure 3(A)). The average RMSD between frames was 2.27 Å (N terminal 13 residues included). The majority of backbone atoms deviations occurred in a narrow period during the simulation course, specifically between 65 and 75 ns corresponding to frames ca. 650-750 ( Figure 3B). The third and the fifth clusters had the most variations in RMSDs, as shown in the clustering dendrogram in Figure 3(C).

Purine nucleoside analogs are the most promising drugs
Binding affinities of repurposed drugs to NiV RdRp ranged from À 5.4 to À 9.1 kcal/mol. The affinities of the best drugs are depicted graphically in Figure 4 and their 2D chemical structures are drawn in Figure 5. Galidesivir, AT-9010 and Norov-29 have the highest affinity score to the binding site. It should be mentioned that any of the other drugs predicted to have lower binding affinities might be the actual inhibitor of NiV RdRp. This is because, in the first place, these compounds compete with the natural substrates (nucleotide triphosphates) for complementary binding to the template RNA and being incorporated into the growing RNA chain rather than occupying the active site permanently (Shannon et al., 2022). Yet, drugs with higher affinity to the active site are more likely to be incorporated into the chain being synthesized.
Unfortunately, AT-9010 and Norov-29 have not been evaluated against NiV or a closely related virus. However, Galidesivir was tested against a plethora of viruses including NiV, RSV and measles virus (MeV) by cell-based assays (HeLa cell line). The EC 50 for NiV was 41.9 mM while RSV and MeV had EC 50 values of 11.0 and 6.19 mM, respectively (Warren et al., 2014). Figure 6). This observation is expected since the cavity of the RdRp is highly charged to accommodate the viral genome and the continually shuttled triphosphate nucleotides (Venkataraman et al., 2018). Positively charged and polar amino acids stabilize the sugar-phosphate backbones of RNA strands passing through the cavity of the RdRp (Gong & Peersen, 2010). It is worth mentioning that cellular membranes are not permeable to the triphosphate forms of nucleotides analogs as they are highly charged. Hence, these drugs are administered as neutral or slightly charged prodrugs that are intracellularly metabolized to the active (triphosphate) forms recognized by the polymerases (Picarazzi et al., 2020).

Most interactions between the docked drugs and the protein involve hydrogen bonds and salt bridges (Table 2 and
Galidesivir triphosphate is an analog of adenosine nucleotide originally developed against HCV but later showed wide-spectrum activities against Marburg virus, Zika virus, West Nile virus and Measles virus in different animal models Warren et al., 2014). The binding affinities of galidesivir and uprifosbuvir obtained in this work are consistent with findings obtained in a computational repurposing study on HCV and SARS-CoV-2 (Elfiky, 2020). Very recently, a pharmacokinetic assessment of galidesivir in health human subjects showed the drug to be safe and welltolerated when administered intravenously or intramuscularly (Mathis et al., 2022).
AT-527 is a prodrug of AT-9010 that showed a promising efficacy against the HCV when administered orally (Berliba et al., 2019;Mungur et al., 2020). AT-752 is another prodrug of AT-9010 that showed an EC 50 of 0.48-0.77 mM against dengue virus and other flaviviruses including Yellow fever virus. It was found to significantly reduce viremia and increase the survival among challenged animal models (Good et al., 2021;Lin et al., 2022). Recently, cryo-EM of AT-9010 bound to SARS-CoV-2 polymerase revealed that the methyl group at the 2 0 site of the ribose is the critical part that causes RNA chain termination by creating a repulsive hydrophobic interaction with the ribose oxygen atom of the incoming nucleotide (Shannon et al., 2022).
On the other hand, NoroV-29 is a synthetic adenosine analog recently identified as a potent inhibitor of the RdRp of human Norovirus with an EC 50 of 0.015 mM, albeit cytotoxic properties were observed in a murine macrophage cell line (Li et al., 2020). No further optimization or pharmacokinetic assessments on animal models were performed. Mericitabine is a cytidine analog with an EC 50 ¼ 0.99 mM against HCV determined by a cell-based assay (Le Pogam et al., 2012). It has a safe clinical profile when administered intramuscularly in a phase-II clinical trial to treat HCV (Wedemeyer et al., 2016).
Remdesivir had been evaluated in African green monkeys and was found to provide full survival from lethal infections by NiV (Lo et al., 2019). This is in good agreement with our findings of molecular docking in which remdesivir had an affinity score higher than its parent scaffold (Adenosine) (Figure 4). In contrast, favipiravir was reported to protect the Syrian hamster model challenged with a lethal dose of NiV (Dawes et al., 2018). Nonetheless, the drug showed a docking score (À 7.23 kcal/mol) lesser than the physiological substrates of the RdRp (À 7.8 to À 8.0 kcal/mol) (Supplementary Table S1).
On the contrary, remdesivir is the first-in-class nucleoside analog to be approved by the FDA for treating of COVID-19 patients in hospitals (Lamb, 2020). In SARS-CoV-2, kinetic assays had found remdesivir to be incorporated in a 2-fold preference over its adenine parent (Dangerfield et al., 2020). The RdRp of SARS-CoV-2 recognizes remdesivir triphosphate and incorporate it into the growing RNA chain. After adding three nucleotides following remdesivir, the RdRp is stalled, and RdRp translocation is inhibited (Kokic et al., 2021).
Cell-based replicon assays of INX-08189, the prodrug of BMS-986049, against HCV revealed high potency in which EC 50 values ranged from 0.9 nM to 12 nM (Vernachio et al., 2011). Its pharmacokinetic properties were also accepted as an oral drug in monkey models. Nonetheless, it is still an investigational drug, and its phase-II clinical trial was discontinued in 2012. Regarding the ribavirin drug, our results are consistent with the previously reported low efficiency of ribavirin observed among NiV patients during the Malaysian outbreak in 1998, where the mortality was slightly decreased among treated cases (Chong et al., 2001). Lastly, 2 0 -C-methylcytidine was developed as an inhibitor for NS5B in HCV (Coelmont et al., 2009). It has also exhibited antiviral activities against Norovirus with EC 50 of 0.03-0.1 mM (Q. Li et al., 2020).

The drugs are stable during MD simulations
For accurate estimation of binding mode and energy, the top three analogs as well as the positive control (GTP), were subjected to all-atoms MD simulation for 100 ns. RdRp backbone atoms showed a stable conformation for the last 70 ns. The RMSD of RdRp-drug complexes (except RdRp-AT-9010 complex) changed in a range of 0.7-1.2 Å ( Figure 7A). The stability is also confirmed by conserved radii of gyration in all systems, which changed only less than 1 Å (Figure 7(E)).
Stable gyration and solvent-accessible surface area of the simulated protein provide evidence of structural stability and absence of unfolding events . It is still obvious that the presence of nucleoside analogs (and GTP) induced changes in the conformations of the RdRp, especially during the first 30 ns of the simulation period. These changes might have occurred as responses to changes in ligands RMSDs.
Regarding the flexibility of RdRp parts, residues fluctuations were monitored during the simulation via calculating the RMSF (Figure 7(B)). The major flexibility was observed in the first 15 residues at the N-terminal end of the RdRp. These residues have neither a secondary structure nor stable interactions with other parts within the RdRp. Other peaks in RMSF are residues at b-turns between different helices and sheets, especially in the fingers subdomain (dark cyan in Figure 1(C)). When compared to stable parts, loops and free  terminal residues normally experience higher degrees of motion due to the continuous movement of solvent molecules (Salsbury, 2010). Ligands binding stability was evaluated by their RMSD over the simulation period. Except remdesivir, all nucleoside analogs and the GTP reached a stable position during the first 20 ns of the simulation. Norov-29 took an additional time (� 15 ns) to reach a stable state, while the terminal phosphate group of remdesivir experienced intermittent small-scale motion (Figure 7(C)). Ligand fluctuation is normal during the search for a lower energy pose having stable interactions with the receptor (Ganesan et al., 2017;Nair & Miners, 2014). The number of hydrogen bonds between compounds and the RdRp in each system is depicted graphically in Figure 7(D). Lastly, electrostatic surface potential rendering ( Figure 8) revealed that most established interactions between analogs and the RdRp involved positively charged residues (Lysine and arginine). This observation is expected since the triphosphate moiety has multiple hydroxyl groups.

The binding energy of top analogs
Binding energy estimation by MM/GBSA approach is well documented in evaluating antiviral RNA inhibitors with a good correlation with experimental measurements (Peddi et al., 2018). The MM/GBSA results of binding free energy of each analog are detailed in Table 3 and summarized in Figure 9. The highly charged cavity of the RdRp domain, especially in proximity to the GDN active site, accounts for the multiple noncovalent interactions with the triphosphate form of the tested analogs. The observed binding free energies may reflect such interactions. The DG obtained from MM/GBSA re-ranked Norov-29 on top of galidesivir and AT-9010. This rescoring is usually used to overcome the limitations of quick scoring of docking programs (Wang et al., 2019).

Conclusion
Nipah virus is an alarming viral pathogen that requires attention of scientific and medical communities before causing larger outbreaks or even a global pandemic. The computational modeling of L protein in NiV has shown a conserved architecture seen in other Mononegavirales. The RdRp in NiV has structural folding and conserved motifs similar to other paramyxoviruses, especially HPIV and rabies virus. Based on molecular docking and MD simulations, guanine nucleoside analogs are the most promising inhibitors for NiV RdRp. Experimental evaluation of these analogs, after solving the structure, is expected to shed light on the successful development of wide-spectrum antiviral agents possibly affecting other notorious viruses within the Mononegavirales class such as rabies virus, Measles virus, RSV, Hendra virus and mumps virus.